Tunnel-and-Towns Model Anatomy

What the model is, and how to read this page

One question, two cities, three linked pieces of arithmetic.

Progress Ireland's tunnel-and-towns model asks one question. If the Dublin region has to house a fixed number of extra homes by 2050, is a rail-led city cheaper to serve with infrastructure, and worth more, than a roads-led one? The model builds both cities on the same housing total and compares them. Nothing in it claims that a railway creates homes.

The Road Towns option builds the railway that is already committed (DART+, FourNorth, MetroLink and, since v3.9, the Navan Railway), fills it, and serves every other home with roads, parking and utilities as a car suburb. The Rail Towns option adds a ladder of rail investment on top: the through-running tunnel from Heuston to the Docklands, four-tracking to Naas, more throughput on the Northern line, electrified extensions, new stations and twenty greenfield rail towns of 15,000 homes each. Whatever still does not fit on rail is a car-suburb home, exactly as in the Road Towns option.

Three pieces of arithmetic do the work.

The capital model turns train seats into homes, fills those homes in a fixed order, and prices what the remaining car-suburb homes cost to serve.
The demand engine lets behaviour respond to the quality of the railway: a town on a terminating, low-frequency line generates fewer rail trips, more car trips and a lower density than one on a through-running network.
The cost–benefit analysis values the Rail Towns option minus the Road Towns option under the Department of Public Expenditure's appraisal conventions.

This page's numbers are model v5.7, read from the model at build time. Central case for Dublin, Meath and Kildare: 264,000 more homes rail-served by 2050 under Rail Towns than under Road Towns, with room for 321,000 more once the works operate; the State's transport bill €31.4bn against €23.5bn, so Road Towns €7.9bn cheaper; all modelled infrastructure €50.7bn against €48.4bn; net present value €9.1bn on transport benefits and €9.1bn with land use. The Low scenario deepens the state-transport gap, and every quotation of the central case has to say so. Since v3.12 replacement and vacancy homes no longer load the network, which removes the state-transport saving the v3.9 page reported; the rail-led city's case now rests on parking, utilities and the CBA's benefits.

How to read this page. Each section below opens with a click. Every section, at every level, stays under 500 words; longer topics are cut into smaller sections. The lowest level is the register of inputs: each one shows its Low, Central and High values, its status letter (P published, D derived, C convention, A assumption, U unverified), the source it rests on, what it drives, how it was tested and what evidence would replace it. The model, the honesty ledger and the register text quoted at the lowest level call the two futures Option 1 (Road Towns) and Option 2 (Rail Towns).

490 words.

Version 4.2 turns the problems with Bundle 6, and with every plan, into levers that default to today's behaviour: population, household size, the opening year and a delay rule, cost maturity, land, flat junctions, and two measures of reach to jobs.

Bundle 6 was appraised in 2018 on 2013 population projections for 2035, on a fixed land-use map, with a service plan its own authors said needed timetable verification, and with works costed before design. The committed railway that came out of it is the reference case for everything the model says about the tunnel. So the questions Seán put on 22 September 2026 apply to every plan: what if there are more people, in smaller households; what if the works come late, cost more, and need land nobody has priced; what do the lines actually reach.

Version 4.2 answers with levers, each inert at its default so that Options 1 and 2 still replay the frozen v3.12 basis: a population multiplier and a household-size trend that move the need through the Housing Commission's own household identity; an opening year and build length for the Bundle 6 works, with the tunnel's nine-year spend curve now an input; a delay rule under which homes that complete before the railway can serve them stay road-served; a cost-maturity uplift on every pre-design step and a land allowance on every surface work in built fabric, in every option; a reliability penalty on legs through flat junctions; a switch that takes Bundle 6's Docklands reach away until MetroLink exists; and a labour-market module.

The results are read on a grid of three demand cases against three readings of how much of Bundle 6 gets built. On the 2018 study's own reading, Bundle 6 serves 184,000 homes by rail against Road Towns' 126,000 and Rail Towns' 390,000, and its case against Road Towns is worth €1.93bn. On Duggan's reading it serves 135,000, the same as Road Towns, and the works are built for nothing: −€0.59bn. As consented it is Road Towns, exactly: €0.00bn. The tunnel programme keeps a positive value against every reading, €7.19bn against the study's Bundle 6 and €9.07bn against Duggan's, and pays the same cost and junction stresses on its own pre-design steps.

Two things the grid concedes first. Bundle 6 opens in 2036 against the tunnel's 2038, and against Road Towns it pays for itself at almost no net cost. And the one figure the whole comparison turns on, how many more Northern-line trains the Bundle 6 city centre can turn without a tunnel, is a 2018 engineering review, not a timetable.

The demand levers move the need through the Commission's own arithmetic; they hit the seat-bound plans first and they work through the road tiers.

The model's need was pinned to the Housing Commission's highest population column, 7.25 million nationally at 2.2 persons per home. Growth since 2022 has run ahead of every plan the railway was sized on: Dublin had 76 per cent of the NTA's 2016 to 2042 growth by 2026, and the State is adding about 95,000 people a year against the 72,000 a year the Commission column implies. The population lever multiplies the region's 2050 population; the household-size lever sets persons per home in 2050 directly. Both move the need through one identity: the Commission's need is linear in one over persons per home, with a slope of population over one less the vacancy rate, so the population each column implies can be backed out of the columns themselves (3,190,000 for the Dublin region) and the two levers are exact and inert at their defaults.

With eight per cent more people and existing demand growing 45 per cent, the homes to serve rise from 548,000 to 663,000. Road Towns still serves 120,000 by rail: its city end has no seats to give. Bundle 6 falls from 184,000 to 154,000, because the growth of existing demand takes the Northern-line seats its city centre freed. Rail Towns serves 391,000. Everything else goes to roads, and both of the seat-bound plans are then in the dearest road tier, so the state-transport gaps widen to €0.2bn in Bundle 6's favour against Road Towns and €7.8bn in Rail Towns' favour. With 2.0 persons per home the homes to serve are 692,000; each home makes fewer trips, so the railway carries slightly more of them, and the gaps are €0.1bn and €7.9bn.

Read this honestly. The demand stresses are symmetric and they favour whichever plan houses more people by rail; they do not single out Bundle 6. They work through the road tiers, which Michael Dnes reads as far too generous to the road-led plans, so the absolute gaps under stress are only as good as those bounds. What they do show is which plans run out: Road Towns first, Bundle 6 next.

Link to this section

The Bundle 6 works get their own opening and build length, a delay rule that makes lateness cost homes, a cost-maturity uplift and a land allowance on every plan's pre-design surface works, and a reliability penalty at flat junctions.

Version 4.1 opened Bundle 6 in 2033 with the tunnel's nine-year spend curve, so two spend years fell before the window and the pair failed its own spend-weight check. Version 4.2 gives the Bundle 6 pair an opening year of 2036 (no Railway Order exists for Connolly through-platforms or the Newcomen drop lock; DART+ West's took 2022 to 2024 and became operative in 2025 after a judicial review; MetroLink's took 2022 to 2026) and a five-year build, and resamples the curve to the build length.

The cost-benefit analysis had rewarded delay: a later opening deferred cost while the homes still arrived by 2050. The delay rule holds completions at the central rate, so homes that complete before the railway can serve them stay road-served. With it, Bundle 6 opening in 2040 is worth €1.24bn instead of €1.93bn, 13,000 homes lost; the tunnel opening in 2040 is worth €6.84bn instead of €9.11bn, 33,000 lost; without the rule the same slip read €6.84bn. Front-loaded completions, 65 per cent by 2038, cost the tunnel programme more than Bundle 6 for the same reason: €6.49bn against €1.68bn.

The 2018 works were Jacobs' ball-park figures with a ten per cent risk allowance, against 25 per cent on the tunnel. Since v5 Connolly and Tara Street carry the lever's default 35 per cent over the indexed figure, Newcomen and the stations 30 per cent, the fleet rows none. The lever raises that 35 to the Oxford Global Projects mean rail overrun at strategic outline stage, 56 per cent, on the same class of step in every plan: Connolly and Tara Street, the tunnel's ancillary steps, extensions and access; not the fleet rows, Newcomen or the stations, which are already priced. Rail Towns then costs €8.8bn more than Bundle 6 on state transport instead of €8.0bn. A fifteen per cent land and consent allowance on surface works in built fabric moves the same gap to €8.1bn; nobody has priced that land, and the input is status U. Five generalised minutes of unreliability at the flat junctions at Islandbridge and Ossory Road take Road Towns to 126,000 rail homes and Bundle 6 to 177,000; the tunnel's legs are grade-separated and keep 390,000.

Duggan's reading puts all of this together with the city end at two trains: Bundle 6 serves 135,000, Road Towns 126,000, and the works are worth −€0.59bn. Rail Towns pays the same uplifts and the same penalty and is worth €8.48bn against Road Towns under them, €9.07bn against that Bundle 6. If FourNorth does not proceed, as Duggan expects, Bundle 6 falls to 134,000 and Rail Towns to 390,000.

Values, Low / Central / High0.35 / 0.35 / 0.35 ×

WhereGlobal inputs, key ob_predesign

StatusA — ASSUMPTION: no direct source; a sensitivity is run where it matters

Source and basis

Oxford Global Projects for the UK DfT, 'Updating the evidence behind the optimism bias uplifts for transport appraisals' (2020), Table 5: rail capital cost overrun, mean, at the strategic outline business case 56% (international, n = 355; OBC 33%, FBC 30%; 80% certainty at SOBC needs +121%), UK rail 65% at SOBC (n = 18); TII Rail/Road Cost Forecasting: rail +38% mean overrun; DART+ South West base €400m (2019) → budget €900m (×2.25, Duggan); SYSTRA 2018 §5.2.16 carried 10% risk on the Bundle 6 elements against 25% on the tunnel; Jacobs' 2017 figures are 'ball park' (App H). The bored tunnel is on Jacobs' out-turn basis (optimism bias inside) and the DART+ rows are budget class, so the uplift does not apply to them. Positions 0.37 (TII mean) and 0.56 (OGP rail at SOBC). Of the Bundle 6 works only Newcomen (44 v 34) and Cabra/Kylemore (32 v 25) carry +30% at central; Connolly (204) and Tara Street (23.5) are at Jacobs' 2017 figure ×1.0, 75% of the works by value (v4.4 correction, fisk). Inputs ob_predesign_scope = 1 exempts the rows already priced or uplifted. Inert at 0 · v5 DEFAULT (6 Oct 2026, Seán; ledger #93): lo/ce/hi 0.35/0.35/0.35, was 0/0/0 in v4.4 (the replay basis keeps v4.4: model/inputs/legacy-v44.ts)

Why this value

v4.2: the model priced the Bundle 6 works on Jacobs' 2017 ball-park figures and Rail Towns' ancillary steps on PI estimates, while the bored tunnel carries Jacobs' out-turn basis with optimism bias inside. This puts every pre-design step on the same footing.

What it drives

The active cost of every step in MATURITY.predesign (the four Bundle 6 works and fleet, the tunnel's ancillary steps, the extensions, stations and access), in every option.

How it was tested

Sensitivities 0.37 (TII rail mean) and 0.56 (Oxford Global Projects rail at SOBC); the Duggan delivery case.

What would replace it

Costed designs; the NTA's own estimate for a Connolly rebuild.

Link to this section

Bertaud's two measures on the engine's three job nodes: the jobs a new rail home can reach, and the workers an employer can reach, under each plan.

The demand engine already prices reach: its accessibility index drives trips, density and pace. Version 4.2 reports it in Alain Bertaud's terms. The employee side is the jobs a home can reach within a commute budget; the employer side is the workers within the budget of a job. The model holds three job nodes, Heuston, the centre and the Docklands, 300,000 jobs in all, and the door-to-door clock minutes from each corridor to each node under each plan's timetable, walk, wait and changes at weight one, and by car with eight minutes to park. A home counts a node as reachable through a soft threshold, 45 minutes with a spread of 12 minutes, because the homes on a corridor are not all at one distance; 60 minutes is reported beside it.

Per new rail-served home the three networks are close. Within 45 minutes: 67,000 jobs from a Road Towns rail home, 71,000 from a Bundle 6 home, 74,000 from a Rail Towns home. Within 60: 168,000, 176,000 and 180,000; Bundle 6 matches the tunnel network here because its homes sit on the committed corridors, closer in, while Rail Towns' towns sit at Naas, Enfield and Drogheda. On the Kildare corridor, where Bundle 6's Docklands service and MetroLink change apply, the figures are 54,000, 74,000 and 110,000.

The labour market is where the plans separate. Workers within 45 minutes of the average job, at 1.2 workers a home: 69,000 under Road Towns, 83,000 under Bundle 6, 128,000 under Rail Towns. Rail Towns leads because it houses more people by rail, and the measure sums them; within 60 minutes the three are 247,000, 270,000 and 341,000.

Bundle 6's reach depends on MetroLink: its Kildare-line trains reach the centre by a change at Glasnevin. On the committed reach, until that interchange exists, Bundle 6 serves 179,000 homes by rail instead of 184,000. At the NTA model's own fifteen-minute interchange penalty it serves 179,000. A Docklands-heavy job mix lifts it to 187,000; a centre-heavy one lowers it to 181,000.

These figures rank the networks for the new homes. They are shares of three nodes, not of Dublin's jobs; they say nothing about existing residents; and they are not comparable with the NTA's own count of 863,000 people within 45 minutes of the city centre, which is the whole region on the NTA's model.

Link to this section

On 6 October 2026 Seán made the September review's corrections the default, with the continuous road-cost curve, Jacobs' 20 trains an hour through the tunnel (26 since v5.1), time savings on the seat-sizing trip rate, the committed Wicklow DART and four sourced recheck fixes. Each decision is reversed below, so its value stays on the page.

Version 4.4 built the September review's corrections as switches, each off by default, so the published figures did not move. On 6 October 2026 Seán decided which to adopt, and version 5 makes them the default: all sixteen work packages, including the continuous road-cost curve; Jacobs' 20 trains an hour through the tunnel, which needs no intermediate ventilation shafts; time savings capped at the capital model's own rail trip rate; the committed battery-electric DART to Wicklow in every option; the four recheck fixes whose values come from a count or a dataset; and one car time for each Maynooth-line town. The v4.4 values are kept as the replay basis, so the model still reproduces v3.12.

On the v5.3 basis, Road Towns serves 126,158 rail homes, Rail Towns 389,823 and Bundle 6 183,731. NPV1, transport only and the headline since v5.3, is €9.11bn (2v1), €1.93bn (3v1) and €7.19bn (2v3); the table gives NPV2.

Decision reversed, alone	Rail Towns rail homes	2v1	3v1	2v3
The whole v4.4 basis	325,251	€6.58bn	€3.62bn	€2.44bn
Road-cost staircase, no curve	389,823	€8.58bn	€1.71bn	€6.87bn
20 trains an hour (v5.0)	389,823	€9.51bn	€1.93bn	€7.58bn
Navan forecast grown again by a fifth (v5.1)	390,596	€9.03bn	€1.84bn	€7.18bn
Navan existing load 900	388,497	€8.36bn	€1.32bn	€7.04bn
Rail Towns wires Greystones to Wicklow	389,823	€8.83bn	€1.93bn	€6.90bn
One leg M car time, Enfield's	389,823	€8.74bn	€2.23bn	€6.52bn
The four recheck fixes off	389,461	€8.73bn	€1.72bn	€7.02bn
Time savings on 360 trips a year (moves rung 1 only)	389,823	€11.38bn	€2.16bn	€9.23bn

Not adopted: the vacancy fix, which waits for the Housing Commission's Appendix 2C; the parking fees and running costs; and the Maynooth–Kilcock town. That town, 36,000 homes on 300 hectares in place of the three Kilcock–Enfield towns, gives Rail Towns 389,689 rail homes and €7.68bn against Bundle 6; if the shared delivery-pace pool counts it as one standard town rather than 2.4, 389,689 and €7.68bn (the loss spreads across the lines, because the model has no pace cap for one town).

In v5.0 the corrected road rates and curve put the state-transport gap at +€4.0bn; since v5.3, with no urban road tunnels, Road Towns €7.9bn cheaper on state transport. Two readings decide the money. Against Road Towns, the value turns on where Dublin's car homes start to sprawl beyond the urban edge, near 500,000 car-home equivalents, where Road Towns sits and the tier bounds have no source; against one common do-minimum with no tier 4 or 5 roads (Michael Dnes's review) it is €6.40bn. Against Bundle 6, it turns on the tunnel's ventilation design and on the trains Connolly can turn back.

Three changes since 6 October: the ventilation shafts back in v5.1, Irish Rail's Navan forecast counted once in v5.2, and in v5.3 no urban road tunnels, land use on matched evidence and the fixes of two Codex reviews. Version 5.4 has its own section.

Version 5.1 (14 Oct). The shafts are back: 26 trains an hour. No homes move; NPV2 falls to €8.16bn.

Version 5.2 (20 Oct). Irish Rail's Navan forecast is a 2050 figure, so the model deducts it as given and no longer grows it by a fifth (navan_grow; ledger #97). Rail Towns now fits 9,000 of the branch's 9,000 planned homes and serves 389,823 rail homes; the three Navan towns still fit none. NPV2 rose to €8.48bn against Road Towns and €3.26bn against Bundle 6.

Version 5.3 (22 Oct). Three of Seán's decisions and two reviews. Dublin builds no urban road tunnels: car homes beyond the urban edge take urban-edge roads further out, and the cost–benefit analysis counts their longer drives (road_sprawl; ledger #100). Land use is valued on a matched rail-access premium, less the transport benefits the residents already get; at central that adds nothing for Rail Towns against Road Towns, so each pair now leads with the transport-only NPV1 (land_method; #101). A Codex review then found seven faults, and a second review eight more; all are fixed (#102, #103). The main one: Rail Towns against Bundle 6 is now Rail Towns against Road Towns less Bundle 6 against Road Towns, line by line (pair_common_base); appraised directly, it had missed Bundle 6's own benefits. Sprawl now follows the road-cost bound year by year. NPV1 was €1.75bn against Road Towns, €1.63bn for Bundle 6 against Road Towns and €0.12bn against Bundle 6; on state transport Road Towns €3.1bn cheaper.

Values, Low / Central / High8 / 8 / 8 tph

WhereGlobal inputs, key vent_tph

StatusA — ASSUMPTION: no direct source; a sensitivity is run where it matters

Source and basis

v4.3 carries 26 tph from the Spencer Dock options vent table (PI estimate). Jacobs 2021 §5.4.3 (p117) is the only engineering source: one train per ventilation section, '20 trains per hour could feasibly operate' at 3-minute headways. The programme-sequence note lists the intermediate shafts 'only if >20 tph', so at 2 (20 tph) the shafts cost nothing. Seats (6,960 at 8), the 10-car row's headroom (15 tph at 8) and the Northern ceiling's tunnel trains scale with it. Band pinned at 8 so the lo/hi scenarios reproduce v4.3; sensitivity vent_20tph runs 2 · v5 DEFAULT (6 Oct 2026, Seán; ledger #93): lo/ce/hi 2/2/2, was 8/8/8 in v4.4 (the replay basis keeps v4.4: model/inputs/legacy-v44.ts) · v5.1 DEFAULT (14 Oct 2026, Seán: 'add the shaft'; ledger #95): lo/ce/hi 8/8/8 again, the two intermediate ventilation shafts and 26 tph back in; the 20 tph design is the named sensitivity v51_vent_20

Why this value

v4.4 (fisk F17g). v4.3 takes the tunnel from 18 to 26 tph with two ventilation shafts on a PI estimate (Spencer Dock options vent table). The only engineering source, Jacobs 2021 §5.4.3 (p117), finds '20 trains per hour could feasibly operate' with one train per ventilation section. The programme-sequence note lists the intermediate shafts 'only if >20 tph'. At 2 the tunnel runs 20 tph and the shafts cost nothing.

What it drives

The seats of dublin.tunnel_vent (6,960 × vent_tph ÷ 8), the 10-car row's headroom (3,270 × (7 + vent_tph) ÷ 15), the Northern ceiling's tunnel trains (northern.ts), frequency.ts, the shaft cost (0 at or below 20 tph) and, with fleet_one_basis = 1, dublin.tunnel_extra_fleet. At 2 and central on the v4.4 basis: no homes move; NPV2 2v1 +€0.88bn, 2v3 +€0.88bn (the shafts buy nothing at central). At the hi band: Option 2 −61,508 rail homes, NPV2 2v1 −€23.8bn. On the v5.1 basis (default 8, ledger #95) at 2 and central: no homes move; NPV2 2v1 and 2v3 +€1.51bn each; the room for homes falls from 227,235 to 209,089.

How it was tested

Sensitivities vent_20tph and vent_20tph_fleet (since v4.4) and v51_vent_20 (since v5.1, the reversed v5.1 decision). Band pinned at 8 so the lo/hi scenarios reproduce v4.3; v5.0 set 2/2/2 and v5.1 restored 8/8/8.

What would replace it

A ventilation and fire-safety study for the R09 tunnel at 24–26 tph, or a timetable study (Thameslink and Crossrail run 24 tph).

Link to this section

Values, Low / Central / High600 / 1370 / 3000 pphpd

WhereGlobal inputs, key navan_load

StatusD — derived by arithmetic from published figures

Source and basis

IÉ EPR forecast: up to 4.3m passenger movements a year by 2050 (Navan Central 1.9m, Dunshaughlin 1.4m, Navan North 0.8m, Kilmessan 0.2m) ÷ 300–330 days × 8–12% peak share = 1,040–1,720 pphpd (central ~1,370). No ÷ 2 for direction: peak_share is a share of all-direction daily volume, the convention t_town uses (Jacobs daily volumes are both directions combined); each branch trip counts once at its branch-end station. If IÉ 'movements' count boardings only, the load is about twice this. v4.3 applied ÷ 2 (520–860) and kept central 900; default unchanged, fisk profile 1,370 (recheck R-A-1, 27 Sep 2026). High 3,000 = the three 'large' park-and-rides drawing the M3 corridor (42,500 Meath residents leave by car and bus daily, RTÉ 24 May 2026; NX bus +28% since 2023). Today's Dunboyne-spur riders (Hansfield 504 + Dunboyne 328 + M3 Parkway 561 = 1,393 Dublin-bound boardings, Rail Census 2024) are already inside the Maynooth-line background growth · v5 DEFAULT (6 Oct 2026, Seán; ledger #93): lo/ce/hi 600/1370/3000, was 600/900/3000 in v4.4 (the replay basis keeps v4.4: model/inputs/legacy-v44.ts)

Why this value

The line's first job is existing Meath commuters, and they board the same trains the new homes would use. IÉ forecasts up to 4.3m passenger movements a year by 2050 across the four stations, i.e. 1,040–1,720 peak-direction passengers in the peak hour (central ~1,370) at the model's 8–12% peak share of all-direction daily volume, the convention t_town uses (recheck R-A-1, 27 Sep 2026: the earlier ÷ 2 for direction gave 520–860 and halved the load); the High band (3,000) is the three 'large' park-and-rides drawing the M3 corridor's 42,500 daily out-commuters. Deducted from the branch's usable seats (core slice and branch cap), on top of the Maynooth-line background growth that already holds today's Dunboyne-spur riders. Since v5.2 it is deducted as given: the forecast is a 2050 figure, so it is not grown again by bg_growth (navan_grow).

What it drives

Leg V usable seats. Corrected 27 Sep 2026 (fisk F12c): the load bites at the central band, because the Option 2 Navan branch is seat-bound (util_V 92%); the earlier note that it bites only above ~2,000 pphpd was wrong. The central 900 sits BELOW the consistent 1,040–1,720 derivation in 'why' (recheck R-A-1). On v4.3 defaults, 1,370 takes 8,464 Option 2 rail homes and €0.72bn off NPV2 2v1 and 2v3 (3v1 unchanged); 1,040 takes €0.21bn; 1,720 also takes 5,652 Option 1 homes and fails one outer-seat check (outer_all 1 in the profile resolves it). The fisk profile model/profiles/fisk-2026-09/navan.ts sets 1,370; F12c's 690 (+€0.32bn) moved the value the wrong way.

How it was tested

600 / 3,000 sensitivities; navan_load_690, navan_load_1040, navan_load_1370, navan_load_1720 (since v4.4).

What would replace it

IÉ's demand forecast by section (the EPR's CBA), and the park-and-ride space counts.

Link to this section

Values, Low / Central / High0 / 0 / 0 switch

WhereGlobal inputs, key navan_grow

StatusC — convention (a reasonable modelling standard, not a measurement)

Source and basis

v5.2 (20 Oct 2026, Seán: 'remove the second growth'; ledger #97).md, finding 4). navan_load comes from IÉ's EPR forecast of up to 4.3m passenger movements a year BY 2050, so it already stands at the 2050 horizon; the leg V core slice (seats_V) and the branch cap (outer_V) then multiplied it by 1 + bg_growth, the growth of today's riders from 2026 to 2050, so the same years of growth counted twice. The outer sections of legs K, N and M start from Rail Census 2024 counts and keep their growth. Today's Dunboyne-spur riders stay inside the Maynooth-line background growth (navan_load source note). Central effect (v5.1 → v5.2): Rail Towns' Navan branch keeps 589 usable seats in place of 315 and fits 7,204 of its 9,000 committed homes (was 3,854); Bundle 6 fits all 9,000 (was 7,799); the three Navan towns still fit none (they need 1,914 places by 2050); rail homes 126,158 / 309,068 / 182,530 → 126,158 / 312,418 / 183,731; room for homes 227,235 → 230,584 (tunnel 159,010 → 162,359); NPV2 2v1 / 3v1 / 2v3 8.16 / 4.74 / 3.08 → 8.48 / 4.88 / 3.26; NPV1 6.16 / 3.57 / 1.69 → 6.44 / 3.68 / 1.85. In the Low and High bands no rail home or net value moves (Low: the branch has seats to spare; High: leg V's core slice is empty either way) · v5.2 DEFAULT (20 Oct 2026, Seán; ledger #97): lo/ce/hi 0/0/0; the v4.4 replay basis keeps 1 (model/inputs/legacy-v44.ts); reversed as the named sensitivity v52_navan_grow

Why this value

Irish Rail's Navan forecast (4.3m passenger movements a year) is a 2050 figure, so the line's riders already stand at the model's horizon. Since v5.2 the load is deducted as given.

What it drives

Leg V usable seats (core slice) and the leg V branch cap. Central v5.1 → v5.2: Rail Towns fits 7,204 of the branch's 9,000 committed homes (was 3,854) and Bundle 6 all 9,000 (was 7,799); rail homes 312,418 / 183,731 (+3,350 / +1,201); NPV2 2v1 8.16 → 8.48, 3v1 4.74 → 4.88, 2v3 3.08 → 3.26. The Navan towns still fit none.

How it was tested

v52_navan_grow (since v5.2) puts the second growth back.

What would replace it

IÉ's demand forecast by section and year (the EPR's CBA), which would say whether its 2050 figure assumes the 2050 population.

Link to this section

Values, Low / Central / High1 / 1 / 1 switch

WhereGlobal inputs, key road_sprawl

StatusC — convention (a reasonable modelling standard, not a measurement)

Source and basis

Seán, 22 Oct 2026: 'I think we omit the urban tunnel -- nobody's gonna build that -- and just accept more sprawl'; 'Please correct the cost side (capex) and the benefit side (traffic, speed, agglomeration)'. Before this, Road Towns' 501,416 central equivalents sat 1,416 above the bound, and the smoothing ramp toward the €245m-a-km tunnel rate (tier_curve_on) added €5.8bn to its road bill. The v4.4 replay basis keeps 0

Why this value

Ledger #100 (22 Oct 2026). Seán: 'I think we omit the urban tunnel -- nobody's gonna build that -- and just accept more sprawl', then 'omit tunnel in all cases. Please correct the cost side (capex) and the benefit side (traffic, speed, agglomeration)'. v5.2 priced Dublin's car-home equivalents above the urban-edge bound (Dublin.b4, 500,000) as urban tunnel capacity at €245m a route-km; Road Towns sat 1,416 above the bound at central, and the smoothing ramp toward that rate added €5.8bn to its road bill. the difference arrives year by year (the base option's equivalents on the region's completions, the test option's fewer as its rail homes are delivered, delay factor included), from 2045 at central, not from 2026; tier 5's embodied carbon is the urban-edge factor; sprawl_pv is the whole NPV1 effect (carbon and the fuel-tax premium included), sprawl_pv3 adds the agglomeration share.

What it drives

Dublin.t5flag → 0 (derived.ts), so tier 5 and the smoothing ramp take the tier-4 urban-edge rate; the CBA's sprawl lines (c_sprawl1/2: equivalents beyond the bound; sprawl_trip_km and sprawl_kph: longer commutes) add to car-km avoided (decongestion, vehicle costs, safety, carbon, air and noise), to time savings and so to the wider economic benefits (NPV3).

How it was tested

model/v5_basis.test.ts (DECIDED_CAP); model/sprawl_timing.test.ts (ledger #102: the cost removed = (r5 − r4) × the beyond-edge count; sprawl_pv reconciles to a run with no sprawl travel); the v4.4 replay basis keeps 0 (legacy-v44.ts); CBA checks 'Ledger #102: sprawl is zero …'.

What would replace it

A network model run of an unconstrained car-led settlement pattern for the GDA (NTA ERM), and Census 2022 POWSCAR commute distances by ring.

Link to this section

Values, Low / Central / High1 / 1 / 1 switch

WhereCBA inputs, key land_method

StatusC — convention (a reasonable modelling standard, not a measurement)

Source and basis

22 Oct 2026 review of the land line: the v5.2 margin (€65,000) is a gap between CSO county median prices, which mix dwelling type, size, schools and distance to Dublin and do not isolate rail access; no published appraisal uses one. Method 1 keeps the location-margin logic (a fixed total of homes, so the gain is the value difference between sites) on matched evidence; Seán: 'apply all the fixes in that order'): in this fixed-home model Road Towns houses the same households on converted farmland too, so the farmland-to-housing gain arises in both options and only the difference between sites counts (what methods 0 and 1 measure); and TAG A2.2 D.2.2 takes the dependent households' transport benefits out of the transport case, which this model keeps. Its +€1.19bn / +€2.08bn (2 v 1, 50%) were double counts. The v4.4 replay basis keeps 0

Why this value

Ledger #101 (22 Oct 2026). Seán asked for the basis of the location margin, who else uses it and whether it is defensible, then 'Let's go with your recommendation'. The v5.2 margin was a gap between county median prices; no published appraisal uses one. Method 1 keeps the location-margin logic on matched evidence (a rail-access premium, other things equal); method 2 is the dependent-development reading of TAG A2.2 and TAF §7.2.3(e).

What it drives

b_hvlu, so PVB2, NPV2, BCR2 and the public-accounts and funding ratios; NPV1 and every transport line are untouched. Each pair now leads with NPV1 (CBA_HEADLINE_ROWS).

How it was tested

CBA sensitivities land_county_gap, land_premium_* (since v5.2); the v4.4 replay keeps 0. TAG A2.2 D.2.2 also takes the dependent households' transport benefits out of the transport case, which this model keeps.

What would replace it

A Dublin hedonic study of rail access for new homes in the outer counties. A dependent-development appraisal needs a different design: homes that exist only with the railway, with their transport benefits taken out of the transport case.

Link to this section

Road Towns builds 150 km of new motorway and the rest of its traffic congests at €0.25 a car-km; the tunnel's fleet runs on one through-running service; a third review priced the east-side turnback and took the step trains' cars out of the 10-car step. The headline against Road Towns is about zero.

Version 5.4 (22 Oct). Four changes. Seán set the new road: Road Towns builds about 150 km of new motorway, an outer ring from Naas to the M1 and two radials, not the 550 km that level of service D would take; the rest of its traffic runs on the existing network (road_built_share; ledger #105). Every option builds the same share, priced at the scarcity tiers. A car-km on the congested network is valued at €0.25, not €0.15 (mec_cong; #106). The tunnel's fleet runs on one through-running service, so its step trains run 28 km, not 48 (fleet_service; #104). A third review priced the east-side turnback that pattern needs and took the step trains' extra cars out of the 10-car step, because the fleet rows already buy them (#107). NPV1 is €9.11bn against Road Towns, €1.93bn for Bundle 6 against Road Towns and €7.19bn against Bundle 6; on state transport Road Towns €7.9bn cheaper.

v5.3 or v5.4 decision reversed, alone (NPV2)	Rail Towns rail homes	2v1	3v1	2v3
Urban road tunnels priced again (v5.2 roads)	389,823	€10.14bn	€2.32bn	€7.83bn
Land on the county price gap (v5.2 method)	389,823	€10.78bn	€2.88bn	€7.90bn
Rail Towns v Bundle 6 appraised directly (v5.2 arithmetic)	389,823	€9.11bn	€1.93bn	€7.19bn
All the road level of service D asks for is built (v5.3 roads)	389,823	€14.35bn	€2.70bn	€11.65bn
Congestion at €0.15 a car-km (v5.3 value)	389,823	€6.40bn	€1.43bn	€4.97bn
The tunnel's fleet on the end-to-end route rule (v5.3)	389,823	€9.09bn	€1.93bn	€7.16bn
The 10-car step at its full estimate (v5.3; the default again since v5.6)	389,823	€9.11bn	€1.93bn	€7.19bn

Values, Low / Central / High0.15 / 0.25 / 0.35 €/km

WhereCBA inputs, key mec_cong

StatusC — convention (a reasonable modelling standard, not a measurement)

Source and basis

v5.4 DEFAULT (22 Oct 2026, Seán; ledger #106): 'lets assume 25c per km congestion cost'. With 150 km of new road (road_built_share) Road Towns' traffic runs on a congested network, so a car-km taken off it is worth more than the €0.15 weighted for today's network (the v5.3 central, now the band's low end); the high end €0.35 sits between UK TAG A5.4's 'other' 27p and conurbation A-road 38p. The v4.4 replay basis keeps 0.08 / 0.15 / 0.25 (model/inputs/legacy-v44.ts). Earlier source: UK TAG A5.4 / IFS: conurbation A-roads 38p, other 27p, urban A 15p, rural 1–3p per km (2015, 2019–20 prices); weighted for Dublin commuter-belt trips (mix of rural motorway, urban A and M50) ≈ €0.15. DoT: GDA aggravated congestion €350m (2017) → €2bn (2033) supports the order of magnitude

Why this value

UK TAG A5.4 / IFS marginal external congestion costs: conurbation A-roads 38p/km, other 27p, urban A 15p, rural 1–3p; weighted for a Dublin commuter-belt trip ≈ €0.15. DoT's GDA congestion cost (€350m 2017 → €2bn 2033) supports the order of magnitude. No Irish marginal-cost schedule exists.

What it drives

Decongestion (v4.3 central 2v1: €2.95bn PV).

How it was tested

v4.3 central 2v1: €0.08 / €0.25 → NPV1 €4.00bn / €7.34bn; NPV2 €5.20bn / €8.54bn.

What would replace it

An NTA ERM marginal-cost run on the radials.

History

v4.4 (fisk F02): with dm_mode 0 this MEC applies in full to car-km on a network that Road Towns has already expanded, so the CBA credits the avoided roads and the congestion they would have relieved. dm_mode 1 scales it by (1 − cong_overlap × road_built_share) (ledger #105;4 review, finding 8: €0.25 × (1 − 0.27 × 0.2728) = €0.2316/km at central); dm_mode 2 removes tiers 4–5 and scales it by tol_mec_mult. Sensitivity overlap_full_027 holds the deduction at the unscaled 27%.

Link to this section

Values, Low / Central / High1 / 1 / 1 switch

WhereGlobal inputs, key fleet_service

StatusC — convention (a reasonable modelling standard, not a measurement)

Source and basis

Ledger #104 (22 Oct 2026; Seán: Option A, Rail Towns pays for the base trains). Every published service pattern for this tunnel runs it through from the Kildare line to the Northern line, with short-turns at both ends: DART Underground Business Case (IÉ 2012) §2.1.4 and Table A7 (16 tph: Drogheda – Inchicore 8, Balbriggan – Hazelhatch 4, Clongriffin – Hazelhatch 4; turnbacks at Inchicore, Clongriffin and Balbriggan); DART Expansion Options Assessment Addendum (Aug 2018) App. A (Bundle 2: every tunnel train Hazelhatch to Drogheda, Dundalk, Malahide or Clongriffin); Jacobs 2021 §5.2.2.4 (R09: 11 Northern DART services linked with the Kildare line). The model's own counts set the rest: kildare_inner = naas_4track_tph (13 of the 26 tunnel trains), the west-side turnbacks (dublin.tunnel_turnbacks), the seat credit's north_share. Used only when fleet_one_basis = 1. v5 1, the v4.4 replay basis 0 (inputs/legacy-v44.ts). Note docs/tunnel-fleet-vs-service-2026-10-22.md

Link to this section

Values, Low / Central / High0 / 0 / 0 switch

WhereGlobal inputs, key road_map_mix

StatusA — ASSUMPTION: no direct source; a sensitivity is run where it matters

Source and basis

4 review, finding 5). Ledger #105 priced the 150 km at the model's scarcity tiers (Road Towns central: 33 km rural dual carriageway, 50 km orbital, 67 km urban-edge, 0.5 km tunnel tier, 16.3 interchange-equivalents, €9.0bn); the road-towns-map network it stands for is 60 km of ring and 88 km of radials, costed at the same unit rates as 83 km rural, 60 km ring at the orbital rate, 4.7 km urban-edge and 10 interchanges, €5.8bn. The switch prices every option at the map's mix (sensitivity road_map_mix); the default keeps the tier mix until Seán chooses. Both bases default to 0

Why this value

4 review, finding 5). Ledger #105 scaled the route-km to 150 km but left the tier mix and the junction count to the scarcity tiers: Road Towns' 150 km fall as 33 km rural, 50 km orbital, 67 km urban-edge and 0.5 km tunnel tier, with 16.3 interchange-equivalents, €9.0bn. The road-towns-map network the 150 km stand for is 60 km of ring and 88 km of radials: 83 km rural, 60 km at the orbital rate, 4.7 km urban-edge and 10 interchanges, €5.8bn at the same unit rates. The switch prices every option's built km at the map's mix (56.2% rural DC, 40.7% orbital, 3.2% urban-edge, one interchange per 14.8 km). Off by default until Seán chooses the costing basis.

What it drives

tierRate in model.ts: tiers 2–5 take r_mix (derived.ts) and tier 1 costs nothing, in every city and option; so roads_tiers, road_capex, renewals, residuals and the CBA's embodied road carbon (road_emb_tot1/2: the map's km mix and junction count). Homes, rail, car-home equivalents, the sprawl lines and the decongestion credit do not move. At central Road Towns €9.0bn → €5.8bn, Bundle 6 €8.5bn → €5.5bn, Rail Towns €6.7bn → €4.4bn.

How it was tested

model/road_built.test.ts (the bill under the switch, the default unchanged, checks at every band); sensitivity road_map_mix; both bases default to 0, so legacy-v44.ts carries no row.

What would replace it

A scheme-level cost estimate of the ring and the two radials (TII route options), which would replace both the tier mix and the per-km junction allowance.

Link to this section

Since 23 October the Rail Towns option builds towns on every seat its trains leave empty, with no ceiling on the towns building at once, and runs 10-car trains that it buys and runs. Road Towns and Bundle 6 keep their programmes. The v5.4 Rail Towns is a sensitivity.

Version 5.5 (23 Oct). Seán set the Rail Towns case for the report (ledger #108). The Rail Towns option now adds towns on every seat its trains leave empty, with no ceiling on the towns building at once (fill_committed, fill_towns_max), and its counted trains carry 1,088 places as 10-car trains that it buys and runs (tencar_all, tencar_cost). In v5.5 it put 451,854 homes on rail by 2050, building 31.5 towns at once. Road Towns and Bundle 6 keep the programmes they had. NPV1 was €9.43bn against Road Towns, €1.27bn for Bundle 6 against Road Towns and €8.16bn against Bundle 6 (version 5.7 below has the current values).

The comparison is one-sided by decision: only Rail Towns fills its trains and runs longer trains. The table shows the v5.4 Rail Towns and the matched cases beside it.

v5.5 decision reversed (NPV2)	Rail Towns rail homes	2v1	3v1	2v3
Towns limited by pace, 870 places a train, no homes after 2050 (the v5.4 Rail Towns)	312,418	€1.46bn	€1.84bn	−€0.38bn
At most 20 towns building at once	341,823	€4.74bn	€1.93bn	€2.81bn
8-car trains, towns still filling them	370,766	€6.08bn	€1.93bn	€4.15bn
Every option fills its trains (the matched comparison)	389,823	€1.09bn	€2.68bn	−€1.59bn

Values, Low / Central / High2 / 2 / 2 switch

WhereGlobal inputs, key fill_committed

StatusC — convention (a reasonable modelling standard, not a measurement)

Source and basis

v5.5 DEFAULT (23 Oct 2026, Seán; ledger #108: 'don't refer to it as the maxed out case. just call it rail towns'): 2, Rail Towns alone; v5.2 to v5.4 kept 0 (the replay basis too); reversal v55_paced; the matched comparison is fill mode 3 (v55_fair_fill). History: 21 Oct 2026, Seán: 'imagine we expand towns as needed to absorb train capacity. How does this impact BCR'; then 'all relevant numbers ... assuming 10 car trains and larger towns to absorb the extra capacity'. A read-only probe the same day (town potential ×20, Option 2's town-scaled costs pinned) found the tunnel's lead rests on the alternatives leaving committed DART+ and MetroLink seats empty. With the switch on, an option that fills keeps the homes its default towns complete on each leg and adds towns on the room left (room_X − townsfit0_X), every fill leg filling the same share of it; the town potential grows to the room; the room left is held to the homes left for towns; idle default towns (Rail Towns' three Navan towns, which have no seats) are freed; added towns build for Rail Towns' years (pace_years2) in every option, at the option's own pace factor (Inputs fill_years_own); and every option pays Rail Towns' station and access cost per added town home (Options 1 and 3 through rail_fillworks; Option 2 through its own rows, which then scale per home and no longer dilute). Off: every v5.2 number bit for bit

Link to this section

Values, Low / Central / High20 / 25 / 25 towns

WhereGlobal inputs, key fill_towns_max

StatusA — ASSUMPTION: no direct source; a sensitivity is run where it matters

Source and basis

v5.7 DEFAULT (24 Oct 2026, Seán; ledger #110, on the building-speed question: '25 towns, 2055'): at most 25 towns build at once at central and High; Low keeps 20. At central, with fill_end 2055, the 25 towns fill Rail Towns' trains by 2055 (389,823 rail homes in 2050, the room of 462,654 in 2055). At High the ceiling does not bind (7.3 towns at once). Without it (v5.6) Rail Towns needed 31.5 towns at once to put every home in by 2050. Still not drawn in the Monte Carlo (MC_EXCLUDED), so every draw holds 25. History: v5.6 (23 Oct 2026, Seán; ledger #109, on the v5.5 open call 'a ceiling at Low': 'ceiling of 20 towns at low'): the Low band holds Rail Towns to 20 towns building at once (the programme's count); at Low pace (300 a town a year) no ceiling had meant 84 towns at once. Central and High keep 1,000 (no ceiling). Not drawn in the Monte Carlo (uncertainty.ts MC_EXCLUDED): a band rule, not a quantity with a distribution. History: v5.5 DEFAULT (23 Oct 2026, Seán; ledger #108): 1,000, no ceiling: Rail Towns builds towns on every seat its trains leave empty (31.5 towns building at once at central). 20 (the programme's count, v5.2 to v5.4) is the sensitivity v55_towns_20; the replay basis keeps 20. History: ASSUMPTION (21 Oct 2026). The pace note calls 600 homes a town a year 'an ambition above every Irish outturn'; 20 towns at that pace is 12,000 greenfield town homes a year in the Dublin region (6,000 at the Low pace, 20,000 at the High), against about 20,000–25,000 completions a year in all forms recently. No source sets a ceiling; 20 holds every option to the delivery Rail Towns already claims. The read-only probe of 21 Oct had no ceiling. At 20 no option fills its trains by 2050 at central, so the result moves with the ceiling: the probe reports 17, 20, 25, 30 and none (1,000)

Link to this section

Values, Low / Central / High2 / 2 / 2 switch

WhereGlobal inputs, key tencar_all

StatusC — convention (a reasonable modelling standard, not a measurement)

Source and basis

v5.5 DEFAULT (23 Oct 2026, Seán; ledger #108): 2, Rail Towns alone; v5.2 to v5.4 kept 0 (the replay basis too); reversal v55_eight_car. History: 21 Oct 2026, Seán: 'assuming 10 car trains'. The new fleet runs 10-car trains at most: a DART+ X'trapolis 5-car unit is 'equivalent in length to the 4-carriage train sets in the current DART fleet', and two make 'the longest train size operable by the new fleet on current infrastructure', carrying 'at least 1,100' (Iarnród Éireann, 2 Dec 2022; Alstom, 22 Dec 2025), about 164 m, the length of today's 8-car DART. The DART+ business case counts each EMU service as an '8-car Electric Multiple Unit (EMU) – 1200 passengers', a figure it gives 'in the absence of any known new fleet capacities' (AECOM, Sep 2021, Table 5.7 footnote 33); its fleet plan pairs half-length units to at most 168 m (footnote 38), so we read its 8-car EMU as a full-length train. The model converts it at 870 ÷ 1,200, then applies α 0.9: 783 usable places, 71% of the rated 1,100. Two readings. (1) If 870 is a full-length train, the default applies a comfort discount twice and this switch removes one: no carriages are needed. (2) The fleet ordered so far, 285 carriages (28.5 ten-car trains), may not run every committed DART+ path at full length; then 870 is an average of 5-car and 10-car trains, and this switch means longer trains on the committed paths. Their carriages are not costed here; they are common to every option and cancel in every pair. The model buys 10-car sets only for the paths it adds. Trains longer than 164 m would need three units (about 246 m) and longer platforms on every line; not modelled. Rows written at 870 a train (the freed Loop Line paths, the tunnel's fleet and ventilation trains) follow the rated load; dublin.tunnel_10car then adds no seats and keeps its cost (how much of it is tunnel platform work is open). Today's trains are not credited in any option. Off: every v5.2 number bit for bit

Link to this section

Values, Low / Central / High1 / 1 / 1 switch

WhereGlobal inputs, key tencar_cost

StatusC — convention (a reasonable modelling standard, not a measurement)

Source and basis

v5.5 DEFAULT (23 Oct 2026, Seán; ledger #108): 1, longer trains bought and run: Rail Towns pays for the two extra cars on every counted train, their depot, stabling and traction power, platform lengthening and running cost (v5.2 to v5.4 kept 0; the replay basis keeps 0; the sensitivity v55_load_only reads 870 → 1,088 as a load standard with nothing bought). History: 22 Oct 2026, Seán: 'Add a cost for the extra carriages and platforms' (for 'moving from 8 to 10 cars'). The reading is the second of the two in tencar_all's source: the committed fleet (285 carriages) cannot run every counted path at full length, so 870 is an average of 5-car and 10-car trains and every train at 10 cars needs a quarter more cars. If the first reading holds (870 is already a full-length train), set this to 0: there is nothing to buy, and the gain is a change of load standard open to every option

Link to this section

Since 23 October the appraisal values the riders who use rail in both options, adds 10 congested minutes to each car trip a home gives up, prices the 10-car step at its full estimate and charges Rail Towns for the Northern works its own trains need. At the Low band at most 20 towns build at once.

Version 5.6 (23 Oct). Seán decided the nine open calls of v5.5 (ledger #109). Five change the numbers:

The riders who use rail in both options gain the change in frequency, reach and through running between the two networks (er_on, CBA line b_riders).
A car trip that a home gives up saves 10 more minutes for the congested base network (cong_dt; Low 0, High 20).
The 10-car step is priced at its full estimate (tunnel_10car_basis). This counts about €162m of carriages twice; the net row shows the effect.
Rail Towns pays for the Northern works its own trains need, to a city turnback whose site is not specified (north_works).
At the Low band at most 20 towns build at once (fill_towns_max). This value is outside the Monte Carlo.

In v5.6 NPV1 was €11.88bn against Road Towns, €1.84bn for Bundle 6 against Road Towns and €10.05bn against Bundle 6. The comparison stays one-sided by decision. The table reverses each decision on the current basis.

v5.6 decision reversed (NPV2)	Rail Towns rail homes	2v1	3v1	2v3
The 10-car step net of the carriages the option already buys	389,823	€9.31bn	€1.93bn	€7.38bn
Riders on both networks not valued (v5.5)	389,823	€7.62bn	€1.66bn	€5.96bn
No congested car minutes (v5.5)	389,823	€7.17bn	€1.92bn	€5.25bn
No Northern works charged	389,823	€10.12bn	€1.93bn	€8.19bn

Values, Low / Central / High0 / 10 / 20 min

WhereCBA inputs, key cong_dt

StatusC — convention (a reasonable modelling standard, not a measurement)

Source and basis

v5.6 DEFAULT (23 Oct 2026, Seán; ledger #109: '+10 base case'): 10, inside the congestion probe's range (docs/congestion-150-2026-10-22.out.md: about 2 minutes at today's trip rate, 10–17 in the middle, 20 and more at the model's rate; queues under a hypothetical loading of all strategic demand on the new links, not network-wide bounds). Low 0 = v5.5 (car minutes at today's congested rings, the decongestion value the only treatment); High 20. Added after the engine travel factor: a home that drives in the base option and rides in the test option saves cong_dt more minutes on each switched trip (not scaled by the rail network factor), as the probe's Δt. The land-use step deducts the residents' larger user benefit (res_ben_home), so NPV2 rises by less than NPV1. The replay basis keeps 0

Why this value

Ledger #109 (23 Oct 2026). Seán, on the v5.5 open call 'congested minutes': '+10 base case'. Until v5.5 a car commute in Road Towns took today's congested ring times (d_min, from the detailed workbook's access rings), although Road Towns puts several hundred thousand more cars on a road network it builds only in part (road_built_share); the decongestion value (mec_cong) was the only treatment, and it prices the delay a car imposes on others, not the delay its own driver suffers. cong_dt is the extra minutes a one-way peak car commute takes in the congested base network: a home that drives in the base option and rides in the test option saves that much more on each switched trip.

What it drives

d_min after the engine travel factor (cba.ts B()), so b_time; through b_time the agglomeration base (b_webs) and the residents' deduction (res_ben_home) that the land-use step nets off. Every pair: 3 v 1 as well as 2 v 1, because Bundle 6's switched homes also leave the congested roads.

How it was tested

model/riders.test.ts: at 0 every v5.5 number returns; b_time rises in proportion to (d_min + cong_dt). CBA sensitivities v56_cong_dt_0, cong_dt_5 and cong_dt_20.

What would replace it

An NTA ERM run of Road Towns' 2050 network: the car journey times on the corridors its towns use, against today's.

Link to this section

Values, Low / Central / High0 / 0 / 0 switch

WhereGlobal inputs, key tunnel_10car_basis

StatusC — convention (a reasonable modelling standard, not a measurement)

Source and basis

v5.6 DEFAULT (23 Oct 2026, Seán; ledger #109, on the v5.5 open call 'platform remainder': 'full'): 0, the full PI estimate, €810m with the pre-design uplift. It counts the step trains' extra cars twice (about €162m with the uplift: the fleet rows buy them as 10-car sets), a conservative allowance by Seán's choice; 1 (€648m, ledger #107) is the sensitivity v56_10car_net and 2 (zero) tunnel_10car_zero. History: Ledger #107 (22 Oct 2026;4 candidate, finding 10): the row's step trains (15 tph) are bought as complete 10-car sets at fleet_set_cost since the v4.4 one fleet basis and ledger #104, so the row's train-lengthening element was counted twice. The remainder (€480m at central) is open for Seán: dart-plus-tunnel/ENGINEERING-SEQUENCE.md puts R09's platforms at 164 m (an 8-car DART, the length of a 10-car DART+), so the platform element may be nil too; tencar_plat_m prices a surface face at €4m. Used only when fleet_one_basis = 1. v5 1, the v4.4 replay basis 0 (inputs/legacy-v44.ts). Sensitivities v54_10car_full (0) and tunnel_10car_zero (2)

Link to this section

Values, Low / Central / High20 / 25 / 25 towns

WhereGlobal inputs, key fill_towns_max

StatusA — ASSUMPTION: no direct source; a sensitivity is run where it matters

Source and basis

Link to this section

Since 24 October Rail Towns builds at most 25 towns at once (20 at the Low band), and its towns keep building after 2050 to 2055; at central its trains are full then. Every option grows after 2050 only as fast as its own towns build.

Version 5.7 (24 Oct). Seán answered the building-speed question (ledger #110). Without a ceiling, Rail Towns needed 31.5 towns building at once to put every home in by 2050. Now it builds at most 25 towns at once (fill_towns_max; 20 at Low), so it has 389,823 homes on rail in 2050. Its towns keep building to 2055 (fill_end), and at central its trains are full then.

A new switch holds every option to its own building pace after 2050 (fill_pace_cap). An option adds no more homes than its towns build at their 2026–2050 rate, and no more than its trains have room for. At central, Road Towns cannot fill its own spare room by 2055. At Low, no option comes near its room. Without the switch, the Low band and the Monte Carlo would put homes on rail that no town could build.

NPV1 is €9.11bn against Road Towns, €1.93bn for Bundle 6 against Road Towns and €7.19bn against Bundle 6. The appraisal still ends in 2068.

v5.7 decision reversed (NPV2)	Rail Towns rail homes	2v1	3v1	2v3
No ceiling and no homes after 2050 (the v5.6 rules)	451,854	€11.88bn	€1.84bn	€10.05bn
No ceiling on the towns at once	451,854	€11.86bn	€1.93bn	€9.93bn
No homes after 2050	389,823	€6.89bn	€1.84bn	€5.05bn
Each option fills its room by 2055 whatever its pace	389,823	€8.93bn	€1.81bn	€7.12bn

Values, Low / Central / High20 / 25 / 25 towns

WhereGlobal inputs, key fill_towns_max

StatusA — ASSUMPTION: no direct source; a sensitivity is run where it matters

Source and basis

Link to this section

Values, Low / Central / High2055 / 2055 / 2055 year

WhereCBA inputs, key fill_end

StatusC — convention (a reasonable modelling standard, not a measurement)

Source and basis

v5.7 DEFAULT (24 Oct 2026, Seán; ledger #110: '25 towns, 2055'): 2055 at every band. Rail Towns builds at most 25 towns at once (fill_towns_max) and its towns keep building to 2055: at central they fill its room by 2055 exactly (2054 falls short), so the cap below does not bind on Rail Towns; it binds on Road Towns, whose towns build 10,258 of its 15,366 homes of room by 2055. With fill_pace_cap 1 (v5.7) each option grows after 2050 only as fast as its towns build, so at Low (20 towns at 300 homes a town a year) Rail Towns adds about 5,800 homes a year to 2055 (its towns' 2026–2050 rate: 81,300 homes over 14 building years; 29,036 by 2055) and stays far below its room. The replay basis keeps 2050. History: v5.1 (14 Oct 2026, Seán: count the value after 2050; ledger #96). At a year after homes_end the increment rises linearly from d_rail in homes_end to the room increment (railcap_pot of the test option less the base option's) in fill_end, and stays there. 2059 = 15,000-home towns at 600 a year from 2035. The extra homes are households after 2050 that both futures house: road-served in the base, rail-served in the test, so only the difference enters. Per-home cost lines (town works, avoided roads with road_phase 1, parking, utilities, denser building, land value capture) scale at their 2026–2050 average per home. With base_timing on (2 v 3) the test and base increments over Road Towns each fill to their own room. v5.1 review (14 Oct 2026): two-sided, so the increment moves to the room increment whether above or below its 2050 level (at central Road Towns has more spare room than Bundle 6, so 3 v 1 falls from 56,371 to 55,155); the fill's homes carry the pair's capital rate per home (÷ d_rail, so a late opening under base_timing no longer inflates them); the scaling residual mid-year is weighted by every completion; the fill's embodied carbon carries the increment's own scaling capex by category (s2 − s1). Until v5.6 the fill was not capped by pace (v5.7: fill_pace_cap): a check flags a fill_end the test option's towns cannot reach (its room less its 2050 rail homes, plus the homes the delay rule holds on roads, against fill_end − homes_end years at t_served ÷ pace_years2; central 2059: 59,692 against 91,800)

Link to this section

Values, Low / Central / High1 / 1 / 1 switch

WhereCBA inputs, key fill_pace_cap

StatusC — convention (a reasonable modelling standard, not a measurement)

Source and basis

v5.7 DEFAULT (24 Oct 2026; ledger #110). With fill_end 2055 at every band (Seán: '25 towns, 2055'), the v5.1 fill would have put homes on rail that the towns cannot build: at Low (20 towns at once, 300 homes a town a year) Rail Towns' room is 524,351 against 141,572 rail homes, 64 years of building; at High Bundle 6 and Road Towns cannot reach their rooms by 2055; and the Monte Carlo's slow-pace draws would all have filled. The cap grows each option by min(need, pace room): need = room − 2050 rail homes (+ the delay-held homes for the test option); pace room = (fill_end − homes_end) × t_served ÷ building years (pace_years2 for Rail Towns, pace_years for Road Towns and Bundle 6). Where both needs fit their pace rooms the result is the v5.1 fill exactly (central with fill_end 2061). At central with fill_end 2055 Road Towns' need does not fit, so the cap moves NPV1 from 8.93 / 1.81 / 7.12 to 9.11 / 1.93 / 7.19. The replay basis keeps 0

Why this value

Ledger #110 (24 Oct 2026). With fill_end 2055 at every band, the v5.1 fill put homes on rail that the towns cannot build: at Low Rail Towns (20 towns at 300 homes a town a year) has 382,779 homes of room left in 2050, about 64 years of building; at High Bundle 6 cannot refill the seats the delay rule leaves free; at central Road Towns cannot reach its own room (15,366 homes against 10,258 its towns build in five years). Each option now grows by the smaller of its need and its towns' pace room. Where every need fits, the result is the v5.1 fill exactly.

What it drives

The increment the homes reach by fill_end (dRailFull) and so every per-home line after 2050. Central NPV1 9.11 / 1.93 / 7.19 with the cap, 8.93 / 1.81 / 7.12 without (Road Towns' room filled whatever its pace).

How it was tested

Sensitivity v57_no_pace_cap; model/fill_pace_cap.test.ts; the CBA check 'ledger #110: with fill_pace_cap on, each option adds after homes_end no more than its towns build'.

What would replace it

A town-by-town phasing (a start year and a build curve for each town), which the model does not have.

Link to this section

Values, Low / Central / High8 / 8 / 8 tph

WhereGlobal inputs, key vent_tph

StatusA — ASSUMPTION: no direct source; a sensitivity is run where it matters

Source and basis

Why this value

What it drives

How it was tested

What would replace it

A ventilation and fire-safety study for the R09 tunnel at 24–26 tph, or a timetable study (Thameslink and Crossrail run 24 tph).

Link to this section

Values, Low / Central / High600 / 1370 / 3000 pphpd

WhereGlobal inputs, key navan_load

StatusD — derived by arithmetic from published figures

Source and basis

Why this value

What it drives

How it was tested

600 / 3,000 sensitivities; navan_load_690, navan_load_1040, navan_load_1370, navan_load_1720 (since v4.4).

What would replace it

IÉ's demand forecast by section (the EPR's CBA), and the park-and-ride space counts.

Link to this section

Values, Low / Central / High0 / 0 / 0 switch

WhereGlobal inputs, key navan_grow

StatusC — convention (a reasonable modelling standard, not a measurement)

Source and basis

Why this value

Irish Rail's Navan forecast (4.3m passenger movements a year) is a 2050 figure, so the line's riders already stand at the model's horizon. Since v5.2 the load is deducted as given.

What it drives

How it was tested

v52_navan_grow (since v5.2) puts the second growth back.

What would replace it

IÉ's demand forecast by section and year (the EPR's CBA), which would say whether its 2050 figure assumes the 2050 population.

Link to this section

Values, Low / Central / High1 / 1 / 1 switch

WhereGlobal inputs, key wicklow_bemu

StatusC — convention (a reasonable modelling standard, not a measurement)

Source and basis

v4.4 (2 Oct 2026, Seán: 'a new plan to bring the DART to Wicklow town … incorporate this into all models'). Iarnród Éireann issued the design-and-build tender for the battery-train charging infrastructure at Wicklow station on 25 Aug 2026 (estimated value €20m ex VAT; prior information notice Sep 2025: €25m): one DART an hour each way Dublin – Wicklow, on overhead power to Greystones and on battery from Greystones to Wicklow, charging on platforms 1 and 2; construction from 2027 and services from 2029, subject to planning and funding approvals; the Railway Order is expected to be lodged with An Coimisiún Pleanála in Q4 2026. Minister for Transport, PQ 15539/26 (26 Feb 2026): the NDP Transport Sectoral Investment Plan funds the extension for around 2029 on the second DART+ fleet order (90 battery-electric carriages), against 'no earlier than 2037' in the GDA Transport Strategy 2022–2042. IÉ Wicklow Capacity Enhancements Pre-Feasibility Report (5 Apr 2024, released under FOI IE_FOI_829): 1 tph battery-electric €56m incl. VAT (Table 11); 2 tph needs two passing loops; the Bray Head single track holds DART+ to 3 tph Bray – Greystones. No homes move with the switch: the model places no towns, committed homes or station infill beyond Bray, and County Wicklow is outside its demand region (Dublin, Meath, Kildare). Note docs/wicklow-dart-2026-10-02.md · v5 DEFAULT (6 Oct 2026, Seán; ledger #93): lo/ce/hi 1/1/1, was 0/0/0 in v4.4 (the replay basis keeps v4.4: model/inputs/legacy-v44.ts)

Link to this section

Values, Low / Central / High2 / 2 / 2 switch

WhereGlobal inputs, key bg_basis

StatusC — convention (a reasonable modelling standard, not a measurement)

Source and basis

v4.4 recheck city-D2b-14-R-bg-1 (confirmed twice, medium). NTA National Rail Census 2024. Table 20 (2024): Northern DART 16,324 + Commuter 7,500 (+ IC 2,636); South Eastern DART 15,989 + Commuter 1,091 (+ IC 683); Kildare Commuter 6,734 (+ IC 11,876); Sligo/Maynooth Commuter 7,508 (+ IC 1,531). Table 21 (08:00–09:00): Northern 3,960 DART + 1,543 Commuter; South Eastern 2,692 DART; Heuston 1,219 Commuter; Sligo 3,092 Commuter. The deduction bg = peak_now × bg_growth is peak-hour DART-class pphpd (peak_now is DART+ commuter/DART capacity), so InterCity riders (64% of the Kildare daily flow) and daily totals are the wrong basis. Table 21 maxima fall on different route sections, so basis 2 is an approximation. Note docs/fisk-2026-09/recheck-fixes.md · v5 DEFAULT (6 Oct 2026, Seán; ledger #93): lo/ce/hi 2/2/2, was 0/0/0 in v4.4 (the replay basis keeps v4.4: model/inputs/legacy-v44.ts)

Why this value

v4.4 (recheck city-D2b-14-R-bg-1, confirmed twice, medium). The model deducts background growth as peak-hour DART-class seats on each leg (bg = peak_now × bg_growth, and peak_now is DART+ commuter/DART capacity), but splits it by leg on Rail Census Table 20 daily flows that include InterCity. InterCity is 64% of the Kildare daily flow and does not use DART-class seats, so leg K takes too much of the deduction and legs N and M too little.

What it drives

Dublin.bg_K, bg_N and bg_M (a switch-gated adjustment; they still sum to 1), and under navan_pool the Maynooth part of leg M (maynooth_bg_frac is read on the same basis); then the background seats deducted on each leg and the seat fit. The incidence depends on which legs bind: on v4.3 it falls on Option 3, on the fisk profile on Option 1.

How it was tested

Sensitivities bg_basis_1 and bg_basis_2 (since v4.4); the sum-to-1 check and the F05 M + V conservation check (checks.ts); model/fisk_recheck-fixes.test.ts (inert at 0; shares and recheck deltas at 1 and 2).

What would replace it

IÉ peak-hour DART and Commuter loads by leg at the Canal Cordon, on one section per leg; then retire the switch.

Link to this section

Values, Low / Central / High1 / 1 / 1 switch

WhereGlobal inputs, key green1_split

StatusC — convention (a reasonable modelling standard, not a measurement)

Source and basis

v4.4 recheck city-D1-11-R-DUB-G1SPLIT (confirmed, medium). tod-scale-economics catchments.site_homes_extended − site_homes over the stations.csv R1 rows, central: Kildare 9,226 (Hazelhatch-Celbridge 6,332 + Heuston West 2,894), Northern 30,452 (Malahide-Drogheda 16,884 + FourNorth 13,568), Maynooth 16,462, South Eastern 1,809, city 2,714; total 60,662 = green1 ce 60k (build_simple_model.py:573/578). The 45/35/20 split (build_simple_model.py:654) credits leg K with Adamstown, but Adamstown and Clonburris are published-capacity rows with no annulus. About 54% of the annulus sits around built-up stations. Note docs/fisk-2026-09/recheck-fixes.md · v5 DEFAULT (6 Oct 2026, Seán; ledger #93): lo/ce/hi 1/1/1, was 0/0/0 in v4.4 (the replay basis keeps v4.4: model/inputs/legacy-v44.ts)

Why this value

v4.4 (recheck city-D1-11-R-DUB-G1SPLIT, confirmed twice, medium). The Option 1 greenfield potential is split 45 / 35 / 20 across legs K, N and M by ASSUMPTION, and the source credits leg K with Hazelhatch, Celbridge and Adamstown. The stations.csv R1 annulus (the ring the green1 total is built from) puts only 15% on the Kildare line, because Adamstown and Clonburris are published-capacity rows with no annulus; the Northern line (Malahide-Drogheda and FourNorth) carries half.

What it drives

Dublin.green1_K, green1_N and green1_M (a switch-gated adjustment of their band values; the total and green1 do not change), then the Option 1 leg town rows, the pace-cap shares and the seat fit on each leg. Option 2 town rows (towns2_*) do not move.

How it was tested

Sensitivity green1_split (since v4.4); the leg-sum check (checks.ts, green1_K + N + M + X + V = green1); model/fisk_recheck-fixes.test.ts (inert at 0; shares sum to the band total at 1).

What would replace it

Per-leg greenfield capacity from the county development plans and LAPs, station by station; then retire the switch.

Link to this section

Values, Low / Central / High2 / 2 / 2 switch

WhereCBA inputs, key time_trips

StatusC — convention (a reasonable modelling standard, not a measurement)

Source and basis

Found 30 Sep 2026 while building the journeys-by-mode charts (desk rail-towns-journeys/): b_time counts 2.4 commuting trips × 0.75 switch × 200 days = 360 rail commuting trips a year per shifted home. The fares line charges 290 rail trips a year (fisk F12d: 266), and the seats are sized on t_town = 0.0722 peak trips, i.e. 0.72 rail trips a weekday or 217 a year at 300 days. A home cannot switch more commuting trips to rail than it makes rail trips. CHOICE of cap; default 0 replays v4.4 · v5 DEFAULT (6 Oct 2026, Seán; ledger #93): lo/ce/hi 2/2/2, was 0/0/0 in v4.4 (the replay basis keeps v4.4: model/inputs/legacy-v44.ts)

Why this value

v4.4 (30 Sep 2026). The model carries three rail trip rates for a rail home. The capital model sizes the seats on t_town: 0.0722 peak-hour trips, i.e. 0.72 rail trips a weekday, 217 a year at 300 days. The fares line charges rail_trips_home: 290 a year (the fisk profile: 266). The time-savings line b_time counts trips_home × switch × days = 2.4 × 0.75 × 200 = 360 switched rail commuting trips a year. A shifted home cannot switch more commuting trips to rail than it makes rail trips, so b_time counts about 1.7 times the capital model's own rail use. Found while building the journeys-by-mode charts on the desk (rail-towns-journeys/).

What it drives

b_time only, and through it PVB1, NPV1, the resident benefit per home (res_ben_home), hvlu_net, b_webs and NPV3. With hvlu_overlap 1 (the Low and Central bands) NPV2 does not move while the HVLU floor is slack: a smaller time saving is matched by a larger land-use residual. At the High band (overlap 0.5) NPV2 moves by half the change in b_time. Rail homes, A and C never move. Inert at 0.

How it was tested

CBA sensitivities time_trips_fares (1) and time_trips_hinge (2); model/time_trips.test.ts; the CBA check 'time-saving trips a year per shifted home'.

What would replace it

A commute-by-mode count for residents of rail-served homes (Census 2022 POWSCAR or the NTA household survey near DART+ stations), which fixes trips_home × switch directly. Then one rail trip rate for the seats, the fares and the time savings.

Link to this section

Values, Low / Central / High1 / 1 / 1 switch

WhereGlobal inputs, key road_sprawl

StatusC — convention (a reasonable modelling standard, not a measurement)

Source and basis

Why this value

What it drives

How it was tested

What would replace it

A network model run of an unconstrained car-led settlement pattern for the GDA (NTA ERM), and Census 2022 POWSCAR commute distances by ring.

Link to this section

Values, Low / Central / High1 / 1 / 1 switch

WhereCBA inputs, key land_method

StatusC — convention (a reasonable modelling standard, not a measurement)

Source and basis

Why this value

What it drives

b_hvlu, so PVB2, NPV2, BCR2 and the public-accounts and funding ratios; NPV1 and every transport line are untouched. Each pair now leads with NPV1 (CBA_HEADLINE_ROWS).

How it was tested

What would replace it

Link to this section

The need bridge: from the Housing Commission's national figure to 619,543 Dublin-region homes to build and 548,254 that load the network.

The number of homes is exogenous and identical in both options, so the model never argues about it; it only has to be sourced. The bridge runs in four steps to the homes to build, then two more (v3.12) to the homes that load the network.

Step	Central	Basis
National need 2024–2050 at 7.25 million people	1.58 million homes	Housing Commission Table 2.1 at 2.2 persons per household
Dublin, Meath and Kildare share	45 per cent, 711,450 homes	Seán's regional forecast sheet, from ESRI RS190's demand share
Less homes completed in 2024 and 2025	31,907	CSO NDQ05 by local authority
Less other committed transit-oriented development	60,000	MetroLink and Luas sites, the same in both options
Homes to build	619,543	the construction count, the same in both options
Less the vacancy allowance, homes with no residents	26,498	5 per cent of the non-replacement increment (Commission p. 34), pro rata
Less in-situ replacement, households already on the network	44,791	half of the replacement of obsolete stock (stock × 0.5 per cent × 27 years); assumption, v3.12
Homes to serve	548,254	the homes that load the railway and the roads

The Low and High cases move the household size (2.4 and 2.0 persons per home), which gives 584,100 and 864,000 gross homes for the region and 406,904 and 715,103 to serve. Those bands are demand scenarios, not confidence intervals. It adds the fair note that the 45 per cent is a demand share rather than a population share, and that both the share and the 60,000 are author assumptions rather than an official forecast.

The reviewer of September 2026 asked for the 6.75 million population column as a coherent Low scenario as well; that is a Stage 2 switch of the rebuild, and until it lands the bands vary household size only. The honesty ledger records one correction that matters here: the Housing Commission never endorsed an annual figure (its majority rejected choosing one), so the model quotes its Table 2.1 matrix and never "60,000 a year" or "90,000 a year".

Where the national total comes from and how the bands are set.

The Report of the Housing Commission (2024) gives a matrix of housing need by population and household size in its Table 2.1. The model reads the largest population column, 7.25 million, from a copy of Seán's regional forecast sheet held in the model's data folder. Household size sets the scenario: 2.4 persons per home in the Low case, 2.2 central and 2.0 in the High case, which gives 1.30, 1.58 and 1.92 million homes nationally. The sheet also holds a 1.9-person column of 2.12 million homes; that is treated as a ceiling and not used as the High case.

Infill homes are apartment-led and use a fixed 2.0 persons per home from the 2016 Census, whatever the scenario.

The Commission's horizon stops at 2050. The one ad hoc test of a 2060 window (v3.4) extrapolated need to 883,043 for the region and is flagged as an extrapolation wherever it is quoted.

Link to this section

60,000 homes around MetroLink and Luas are netted from both options.

Some transit-oriented development is committed on lines the model does not treat as legs: the MetroLink station areas and the Luas corridors. The model takes 60,000 such homes off the total before either option is built, because they are the same in both futures. MetroLink's own committed catchment in the leg table is therefore zero; its spare seats carry two greenfield towns instead.

The figure is a convention, not a count, and it was cross-checked on 4 September 2026 against the Department of Housing and Department of Transport's TOD Dublin Study (May 2023), which inventories fourteen large sites across the four Dublin authorities: roughly 130,000 units across all modes, of which 13,000 to 15,000 are MetroLink sites and 23,000 to 28,000 are Luas-only. The study's heavy-rail sites come to 75,000 to 80,000, about half the model's committed-plus-infill figure, because the model counts Kildare and Meath stations, station infill the study does not survey, and a 2050 horizon. Substituting the study's figures made the rail case slightly stronger, not weaker, because the Road Towns option's lost homes land on tunnel-class roads.

Link to this section

The engine room of the capital model: seats become homes through one hinge.

Every rail-served home in the model is bought with seats. Each step on the investment ladder adds trains per hour on a leg; each train offers 870 spaces; a loading factor and a peak-hour coefficient trim that to about 712 usable spaces per train. Background growth of today's passengers is taken off both options before any new home is seated. What remains is divided by the peak-hour rail trips a new home generates, 0.082 in the central case, to give the homes the leg can serve.

Central case, Dublin	Road Towns	Rail Towns
Seats offered by the ladder (peak hour, peak direction)	18,323	42,054
Less background growth of existing demand	3,792	3,792
Usable seats for new homes	14,531	38,261
MetroLink spare seats, counted separately	6,364	6,364

The hinge is transparent and sourced part by part: the DART+ business case for spaces per train, Vuchic's Urban Transit for the load factor and peak-hour coefficient, the Housing Commission for household size, TCRP Report 128 for trip behaviour.

Three refinements decide where seats bind. Every leg is capped separately, so a full Northern line cannot borrow seats from an empty Maynooth one. The two-track outer sections outside the city are capped by usable train paths, not tunnel seats. And the demand engine scales the trip rate by network, so a home on a weak line uses fewer seats but drives more.

One ceiling is here since v3.10: without the tunnel the Northern line's city end absorbs only 2 of FourNorth's extra trains an hour. Heuston and the loop line carry no ceiling of their own.

Modelling choices referenced here

M2 · Capacity in seats, one hinge

Choice. Seats offered (Ladder × α ÷ PHC) less background growth, divided by peak rail trips per home (t_town), gives homes a corridor can serve.

Alternatives. Codex's single 0.10 passengers per home; a full OD assignment.

Why. Transparent and sourced part by part (Jacobs, TCRP, Commission household size, Vuchic).

Effect. Peak share 8/12% moves Δ A to €1.8bn / €3.6bn.

Status. A on the multipliers Evidence that would replace it. One OD model shared with the CBA's trip rates; IÉ hourly counts.

M5 · Outer two-track sections cap each leg (v3.1)

Choice. Usable outer seats = (paths + Ladder uplifts) × 870 × α ÷ PHC − existing outer demand × (1 + growth); towns must fit core seats, pace AND the outer section.

Alternatives. Core seats only (v1–v3).

Why. The Kildare, Northern and Maynooth outer sections are two-track and shared with intercity, so they bind before the tunnel does.

Effect. The tunnel legs are only 56–77% used in Option 2; Naas four-tracking (+40k homes for €600m) is the cheapest capacity.

Status. A on paths Evidence that would replace it. IÉ timetable path analysis by section.

Why one eight-car train is worth about 712 usable spaces.

The seat count starts from the DART+ South West business case: 20,000 passengers per hour per direction from 23 trains an hour, which is 870 spaces per eight-car train and equals Alstom's rated 551 per five-car set scaled to eight cars. The business case's own 1,200 per train is a legacy planning load of 150 per carriage; wherever its Table 5.7 capacities are used they are converted at 870 over 1,200.

Two haircuts follow Vuchic. A design load factor of 0.9 on the maximum-load section, because trains are never perfectly full along their whole length, and a peak-hour coefficient of 1.1, because the busiest fifteen minutes are busier than the hourly average. Together they leave 870 × 0.9 ÷ 1.1, about 712 usable spaces per train. The model kept the conservative figure; the 1,100 basis appears only as the High value for the Navan branch cap, because Irish Rail's 4,400 passengers an hour implies it.

A Rail Towns rail-factor multiplier exists (Vuchic allows up to 30 per cent more ridership from high-performance rail) and is set to 1.0; the demand engine now carries that effect instead.

Link to this section

Today's passengers will multiply by 2050 whatever is built; their seats are never sold twice.

Existing catchments will generate more peak trips by 2050 whether or not anything on the ladder is built. The model grows today's 22,000 peak-hour, peak-direction passengers by 20, 30 or 45 per cent and removes those seats from both options, by line, before any new home is seated. The line shares come from the NTA's Rail Census 2024 daily inbound flows: Northern 26,460, South Eastern 17,763, Kildare 18,610, Maynooth 9,039.

This replaced the first version's "share of new capacity available for new homes", which the external reviewer rightly called unverifiable. A deduction applied to both options can never favour one by construction, though it costs the Road Towns option more homes because its lines are fuller.

Sensitivity (Δ A is Road Towns minus Rail Towns on state transport; central −€7.9bn): 20 per cent growth gives −€7.9bn, 45 per cent −€7.8bn. The evidence that would replace the assumption is the NTA's regional model loaded sections for 2042, net of the modelled homes.

Modelling choices referenced here

M4 · Background growth deducted from both options

Choice. Existing-catchment peak demand grows 20/30/45% by 2050 and is removed from both options before new homes are seated.

Alternatives. v1's 'share of new capacity for new homes'.

Why. Never favours one option by construction.

Effect. 20/45% → Δ A €1.6bn / €5.4bn; Option 1 loses more homes because its lines are fuller.

Status. A Evidence that would replace it. NTA ERM 2042 loaded sections net of the modelled homes.

Link to this section

The divisor that turns seats into homes, built from five sourced parts.

A new rail-served home is assumed to generate 0.072 peak-hour, peak-direction rail trips in the central case. The figure is a product of five inputs rather than a single guess: 2.2 persons per home, 0.175 daily rail trips per corridor resident, a multiplier of 2.5 for residents within walking distance of a station against the corridor average, a 10 per cent peak-hour share of daily boardings, and a 25 per cent internal-capture discount for trips that stay inside a mixed-use town with 30 per cent commercial land. So 2.2 × 0.175 × 2.5 × 0.10 × 0.75 = 0.072.

Infill homes use 2.0 persons per home, and where an infill station sits inside the inner zone, within about four kilometres of the centre, 70 per cent of its residents' trips are assumed not to load the maximum-load section, because they board inside it.

The peak share is the lever that matters for how many homes a seat buys. At 8 per cent Δ A is −€7.9bn; at 12 per cent it is −€7.8bn. Irish Rail hourly counts would replace the assumption; the Rail Census chart of the hourly profile is a picture, not a table. The reviewer of September 2026 noted that the divisor compounds four assumptions (a catchment trip rate, a station-proximity multiplier, a peak share and internal capture) and that one extra train an hour is worth about 9,900 homes; the CBA's own annual trip figures (360 commuting switches and 290 rail trips per home) describe different trip populations, and a single trip-generation module that feeds both is a Stage 2 switch of the rebuild.

With the demand engine on, this rate is the tunnel-network rate. Each leg and network scales it by the engine's rail-trip factor, so a Kildare town on the committed network generates 0.42 of it.

Link to this section

Towns must fit through the two-track sections outside the city as well as the core.

The Kildare, Northern and Maynooth lines are two-track outside the city and shared with intercity and Enterprise trains. Only so many paths are usable for DART-class trains: 5, 7 and 8 an hour in the central case. A town beyond the four-track section must fit through its outer section as well as the tunnel core, after today's outer-section passengers and their growth are deducted. This came from Gerry Duggan's papers on 4 September 2026 and is why the tunnel's core seats are only partly used in the Rail Towns option.

Two ladder steps change the outer counts. Four-tracking from Hazelhatch to Naas runs the Naas service at the plan's 13 trains an hour on the four-tracked pair (12, 13 or 15), so the Kildare outer section stops being a constraint in the Rail Towns option and the core binds instead; v3.8 made this the service level, replacing a fixed +6 paths. The Northern outer-section package (a passing loop at Skerries and a chord for Enterprise overtaking) adds 30 per cent to the usable paths. FourNorth adds 6 extra Northern peak trains, an assumption in the absence of an Irish Rail service plan.

Naas four-tracking is the cheapest capacity in the model: switching it off drops the Rail Towns option's rail-served homes from 389,823 to 364,808 and Δ A from −€7.9bn to −€7.4bn. The paths themselves (4/6/6 or 6/9/9) do not move the central result, because the core and the pace caps bind first.

The evidence that would replace the path counts is an Irish Rail timetable path analysis by section. The Naas terminal for 13 trains an hour is unpriced.

Modelling choices referenced here

M5 · Outer two-track sections cap each leg (v3.1)

Choice. Usable outer seats = (paths + Ladder uplifts) × 870 × α ÷ PHC − existing outer demand × (1 + growth); towns must fit core seats, pace AND the outer section.

Alternatives. Core seats only (v1–v3).

Why. The Kildare, Northern and Maynooth outer sections are two-track and shared with intercity, so they bind before the tunnel does.

Effect. The tunnel legs are only 56–77% used in Option 2; Naas four-tracking (+40k homes for €600m) is the cheapest capacity.

Status. A on paths Evidence that would replace it. IÉ timetable path analysis by section.

M25 · Naas service level set by the plan on the four-tracked pair (v3.8)

Choice. Inputs naas_4track_tph is the DART frequency to Naas (12 / 13 / 15 tph); the Ladder four-track row supplies (naas_tph − shared paths) × 870 seats on the Kildare outer section, and the Naas extension is fleeted at that frequency through the cycle-time rule (Inputs naas_km).

Alternatives. v3.1–v3.7: four-tracking credited with a fixed +4/6/8 DART paths on top of 5 shared paths, and every extension fleeted at 6 tph.

Why. Seán (7 Sep): four-tracking all the way to Naas was agreed, so the Kildare outer section is not a two-track constraint in Option 2; the engine's own square-root rule (Mohring) says the frequency should follow the towns. The binding constraint on the Naas line becomes the tunnel core.

Effect. Kildare outer usable seats 6,920 → about 8,340 at 13 tph; the Naas-line towns the seats carry rise from 94,483 towards the core limit; the fleet row prices the extra sets.

Status. A (13 tph is the plan; the Naas terminal for 13 tph is unpriced) Evidence that would replace it. A timetable for the four-tracked pair and a Naas turnback design.

Link to this section

Seats are capped per leg, and the tunnel's headroom is shared half and half.

Dublin is modelled as five legs: K, the Kildare and Naas line; N, the Northern line to Drogheda; M, the Maynooth line to Enfield together with the South Eastern line on the surface loop; X, the MetroLink corridor; and V, the Navan branch, added in v3.9 as a sub-leg of M. Committed homes by leg come from the stations file (K 29,176; N 11,355; M 7,469; V 9,000). Infill is mapped station by station to its leg, with the four tunnel-core stations split half to K and half to N.

The tunnel's headroom is scheduled half to the Northern leg and half to the Kildare leg. The loop line it frees goes to the Maynooth leg.

Leg, central	Road Towns: usable seats	Road Towns: town homes fit	Rail Towns: usable seats	Rail Towns: town homes fit, of potential	What binds in Rail Towns
K Kildare / Naas	6,333	4,092	14,689	105,012 of 129,854	delivery pace
N Northern / Drogheda	289	0	11,016	53,193 of 66,218	delivery pace
M Maynooth / Enfield	6,543	14,268	10,027	56,242 of 80,405	delivery pace
X MetroLink	6,364	20,623	6,364	19,200 of 30,000	delivery pace
V Navan branch	0	0	1,159	5,162 of 31,395	delivery pace

The last column is computed: the smallest of the leg's site potential, its delivery-pace cap, the core seats left after committed and infill homes, and the outer-section seats.

In the Road Towns option the binding constraints are K site potential, N core seats, M delivery pace, X delivery pace, V site potential, and the density factors from the demand engine shrink the towns further. Where a Rail Towns leg runs out of core seats, the 30 trains-an-hour step and ten-car trains are the levers that appear in any plan for a larger headline. The reviewer of September 2026 asked for a service diagram showing that the Maynooth-pool split with the Navan branch conserves capacity on each physical section; the branch check on the Results sheet asserts the seat conservation, and a section-by-section diagram is owed.

Link to this section

Committed, common to both options, with spare seats that carry two towns.

MetroLink is committed, so its cost sits in both options and cancels. It has a design capacity of 20,000 passengers per hour per direction against a forecast load of 10,000 from committed development and mode shift, so its spare capacity (6,364 usable spaces after the haircuts) carries two towns of 15,000 homes at Lissenhall and Swords North in both options.

Until v3.4 the Rail Towns option's MetroLink towns carried a ×1.25 accessibility uplift for the tunnel interchange. The demand engine retired it: with the engine on, a MetroLink town scores 0.98 of tunnel-grade accessibility whichever network exists, so the uplift is set to 1.0 when the engine runs. Setting it to 1.0 in the input changes Δ A by +€0.0bn. In particular, the Lissenhall land-readjustment receipts cannot be transferred into the tunnel's case.

Treating MetroLink this way cut the v3.1 gap by about €3bn when it was introduced in v3.2; that is recorded as a fix, not a loss.

Modelling choices referenced here

M7 · MetroLink as a fourth leg, common to both options (v3.2)

Choice. Two towns on the MetroLink corridor in both options (×1.25 in Option 2 for the interchange); MetroLink's own capex is netted.

Alternatives. Crediting the MetroLink towns to Option 2 only.

Why. MetroLink is committed; crediting its towns to one option would be a thumb on the scale.

Effect. Cut the v3.1 gap by ~€3bn when introduced.

Status. A on load and uplift Evidence that would replace it. Railway Order demand tables.

Link to this section

How a 34 km branch at four trains an hour was added, and what it did to the answer.

Irish Rail published the Navan Railway's emerging preferred route on 20 May 2026: a 34 km electrified extension of DART+ West from M3 Parkway with four stations (Dunshaughlin, Kilmessan, Navan Central, Navan North), up to four trains an hour at peak, about 60 minutes from Navan to the city centre, and 4,400 passengers per direction per hour. The NTA's cost estimate is €2bn to €3bn (December 2024); opening is expected around 2036.

The model treats it as committed, in both options, on the FourNorth standard: in the strategy, design under way, funded to Railway Order. Its cost therefore cancels. It becomes leg V, a branch that takes four of DART+ West's twelve trains an hour through Clonsilla, so it gets a third of the Maynooth pool's core seats and background growth rather than any new paths. From that slice come the existing passengers the line captures (1,370 an hour in the central case, from Irish Rail's forecast of 4.3 million a year by 2050; since v5.2 the model deducts the forecast as given, because it already stands at 2050, and no longer grows it by the background growth rate), a committed catchment of 9,000 homes from the Meath development plan, and a branch cap of four trains × 870 spaces after the haircuts. Building years are 13.

The Rail Towns option moves three of its six Enfield-line towns to the branch, keeping twenty towns in all. The demand engine measures the branch against the Maynooth leg's tunnel-grade service, so a 60-minute town at four trains an hour scores 0.48 of tunnel-grade accessibility on the committed network and 0.70 with the tunnel; the loop line freed by the tunnel does almost all of that, because the engine caps waiting at five minutes and frequency barely registers.

The branch is seat-bound: its binding constraint in Rail Towns is delivery pace. After the existing load and the committed homes, 1,159 usable seats remain, enough for 5,162 town homes. So the three Navan towns deliver less than the three Enfield towns they replaced, and the Road Towns option gains the branch's committed catchment. When the line was added in v3.9 the headline fell from 233,375 to 219,288 capacity and Δ A from €7.9bn to €6.9bn; on this version's basis, switching the line off gives capacity 448,492 against 462,654 and Δ A −€8.0bn against −€7.9bn.

What would move it: a 50-minute running time would lift the branch to tunnel-grade accessibility, which is a line-speed question for Irish Rail; single or double track and the cost basis are unknown; and the line's park-and-ride design is not the walk-up town the trip rates assume.

Modelling choices referenced here

M26 · Navan Railway as a sub-leg of the Maynooth leg (v3.9)

Choice. The Navan Railway (IÉ EPR, May 2026) enters the COMMITTED programme in both options. A fifth Dublin leg V is the branch: it takes navan_tph ÷ 12 of the Maynooth pool's core seats and background growth (no new paths through Clonsilla), loses the existing-population demand the line captures (Inputs navan_load, 900 central), has its own committed catchment (committed_V, 9,000), branch cap (navan_tph × 870 × α ÷ PHC less the load), building years (navan_years 13: opening ~2036) and demand-engine factors. Its accessibility is measured against the MAYNOOTH leg's tunnel-grade service (demand_engine REF_LEG), not its own tunnel network: 0.48 on the committed network, 0.70 with the tunnel, so a 60-minute town at 4 tph is not credited as tunnel-grade. Option 2's plan moves three of the six Enfield-line towns to the branch (still 20 towns; Kilcock–Enfield electrification kept for the other three). The engine's network averages exclude V (weight 0) so the CBA parameters and every v3.8 leg factor are unchanged; navan_on = 0 reproduces v3.8.

Alternatives. Keep it out (v3.8, on Duggan's cost warning); treat it as an Option 2 step; a sixth leg with its own tunnel-grade reference (own-network normalisation, the 'dfac 1.0' sensitivity); blend it into leg M (loses the accessibility distinction).

Why. It is in the strategy with the same standing as FourNorth, its EPR fixes the geography and service (34 km, four stations, 4 tph, 60 min), and it changes what the Maynooth leg's towns are and where Option 1's Meath homes sit. Frequency barely registers because the engine caps waiting at 5 minutes; reach (the loop freed by the tunnel) and running time are what move a Navan town.

Effect. See the v3.9 handover addendum and ledger #49 for the central numbers and the sensitivities (OFF, load 600 / 3,000, towns additional, towns in Option 1, 6 tph, catchment 12k, own-network normalisation, 11 years).

Status. A (the line's cost basis, track configuration and the P&R-to-town redesignation are assumptions; the service facts are IÉ's) Evidence that would replace it. IÉ's demand forecast by section and cost basis; single/double track; Meath CDP designations at the station sites; a 50-minute running time would lift the branch to tunnel-grade accessibility (1.13) — a line-speed question for IÉ.

Values, Low / Central / High600 / 1370 / 3000 pphpd

WhereGlobal inputs, key navan_load

StatusD — derived by arithmetic from published figures

Source and basis

Why this value

What it drives

How it was tested

600 / 3,000 sensitivities; navan_load_690, navan_load_1040, navan_load_1370, navan_load_1720 (since v4.4).

What would replace it

IÉ's demand forecast by section (the EPR's CBA), and the park-and-ride space counts.

Link to this section

Values, Low / Central / High0 / 0 / 0 switch

WhereGlobal inputs, key navan_grow

StatusC — convention (a reasonable modelling standard, not a measurement)

Source and basis

Why this value

Irish Rail's Navan forecast (4.3m passenger movements a year) is a 2050 figure, so the line's riders already stand at the model's horizon. Since v5.2 the load is deducted as given.

What it drives

How it was tested

v52_navan_grow (since v5.2) puts the second growth back.

What would replace it

IÉ's demand forecast by section and year (the EPR's CBA), which would say whether its 2050 figure assumes the 2050 population.

Link to this section

What Duggan's answers did to the tunnel's own case, and what the cap is worth on this version.

Through v3.9 the model had no ceiling on how many trains Heuston, Connolly and the loop line can turn round without through-running. The Rail Towns option without the tunnel package assumed the committed core service could still be handled at the terminals. The shared assumption favoured the no-tunnel case, because it let the outer works and the towns deliver most of the homes with nowhere for the extra trains to terminate. The tunnel's own contribution is measured as Rail Towns with the package minus Rail Towns without it, so a terminal ceiling raises the tunnel's value.

Gerry Duggan's written answers of 10 September 2026 closed the Northern-leg half of the question. Connolly's platforms 1 to 4 are not all free: the Enterprise holds one for about 25 minutes and the Sligo service another for 15, which limits how many Drogheda or Dundalk trains can terminate. The Ossory Road flat junction north of Connolly gains nothing from DART+ West or Coastal North; the city-centre resignalling of fifteen years ago already maximised it. So FourNorth delivers trains to a city end that cannot turn them. Heuston is released as a constraint: turn times there can be kept to a few minutes.

v3.10 (10 September) adds one input, the extra Northern-line peak trains the city end can absorb without the tunnel: 0, 2 or 4 an hour, status D, because Duggan gives the mechanism and the band is the model's own. It caps the FourNorth service uplift in Road Towns, and in Rail Towns whenever the tunnel is off; with the tunnel it is uncapped, because Jacobs diverts Northern services to Spencer Dock before Connolly. On its central band the Road Towns option rail-serves 126,158 homes (the Northern leg's offered seats no longer cover its background growth, so its committed and infill homes displace to the roads). When it landed on v3.9 it moved Δ A from €6.9bn to €10.5bn; on this version's basis the sweep is 0 trains an hour −€7.8bn, 3 −€8.1bn, 4 −€8.4bn, 6 −€8.5bn, 8 −€8.5bn, against −€7.9bn at the central 2. Seán chose the central 2 on Duggan's strict reading; the choice favours the tunnel, so the band travels with the number.

Those figures are carried on this page and in the CBA; the v3.11 Kildare band (Duggan's four trains an hour through the Phoenix Park Tunnel against the business case's seven, in both options, so it moves the level and not the gap) is in the Kildare service-plan row. Two items stay open: the Phoenix Park Tunnel count itself (one engineer against the operator's business case; a band, not a central change), and the Connolly overbridge, which Duggan does not recognise and which the business case text does describe.

Modelling choices referenced here

M20 · Option 2 without the tunnel package (v3.4 case)

Choice. A sensitivity that switches off the eight tunnel-package steps, keeping the outer works, town stations and towns; captured for the CBA (CBA_CASE=surface).

Alternatives. Not isolating the tunnel (v3.3).

Why. To see what the tunnel itself buys.

Effect. +78k rail homes without the tunnel vs +165k with it; on a 2050 window the tunnel's own marginal NPV is −€2.2bn, on a 2060 window +€6.8bn.

Status. Method Evidence that would replace it. v3.10 (Duggan, 10 Sep 2026): the city-end ceiling without through-running is now IN the model — Inputs northend_cap_tph caps FourNorth's extra Northern trains in Option 1 and in this no-tunnel case. Still owed: the same test on the Kildare side, where Duggan contests the PBC's PPT 2→7 tph.

M21 · What is not modelled

Choice. No network assignment, no site-level register, no benefits (in this workbook), no schools/health/energy/treatment/flood, no construction disruption. A city-end terminal constraint arrived in v3.10 (Inputs northend_cap_tph) and binds the Northern leg only; Connolly's and Heuston's absolute turnround ceilings are still not modelled.

Alternatives. —

Why. Honesty about scope.

Effect. —

Status. — Evidence that would replace it. Listed in the explainer §7.

Link to this section

Every rail step with its cost band, price year, seats and leg, fixed or scaling with the towns.

The ladder is the model's list of rail works. Each row carries a Low, Central and High cost, the price year it was quoted in, the seats it adds and on which leg, and a toggle for each option. Price years are brought to 2026 with the Society of Chartered Surveyors' tender price index; the tunnel's Jacobs figure is already an out-turn estimate and is indexed at 1.0.

Steps are either fixed, paid in full whether or not the towns fill (the tunnel, its shafts, fleet, depot, traction, signalling, turnbacks, the Naas four-tracking, the Northern package), or scaling, paid in proportion to the greenfield fill (the extensions, their fleet, the town stations, district access).

Rail capital, central, Dublin	Road Towns	Rail Towns
Fixed steps	€8.6bn	€20.8bn
Steps that scale with town fill, at the fill achieved	€0.0bn	€1.1bn
Station and access works for infill	€0.1bn	€0.3bn
Total	€8.7bn	€22.2bn

The committed rows (DART+, FourNorth, MetroLink's towns, Navan) appear in both options and cancel.

Regional ladders for Cork, Limerick and Galway sit on the same sheet with their own rows.

Modelling choices referenced here

M13 · Rail capex: fixed vs scaling with fill

Choice. Ladder steps are fixed (paid in full) or variable (× greenfield fill ratio); price years indexed with the SCSI TPI; the tunnel on Jacobs' out-turn basis (index 1.0).

Alternatives. Codex pays all selected assets in full; rebases the 2021 comparative tunnel cost.

Why. Extensions and stations follow the towns; the tunnel does not.

Effect. Rail Δ €10.0bn undiscounted in v3.4 (Codex €9.65bn).

Status. P/A by step Evidence that would replace it. A QS-normalised estimate ladder on one price basis.

M14 · FourNorth common to both options

Choice. Launched July 2026; its cost and its trains sit in both options.

Alternatives. Codex charged it to the programme.

Why. Committed schemes belong in the do-minimum.

Effect. Would be worth ~€1.7bn of NPV against the programme if charged to it.

Status. P Evidence that would replace it. IÉ service specification for the tph.

R09 and the eight rows that make it a 26 trains-an-hour railway (Jacobs found 20 without the ventilation shafts).

The tunnel R09 runs 7.84 km from Heuston to the Docklands with five stations. Its central cost is €5.5bn (band €5bn to €6bn), taken from Jacobs' 2021 route-options study on the out-turn basis that includes land, VAT, inflation to construction and optimism bias. The All-Island Strategic Rail Review's €3.4bn to €5.3bn at 2021 prices plus 56 per cent optimism bias sits in the same range. This is the largest single cost in the model.

The tunnel's own seats are the loop line it frees, credited to the Maynooth leg; its base eleven trains an hour carry the rerouted services. Eight further rows turn it into a high-capacity railway (the figures are each row's base estimate; since v5 the rows still at pre-design carry a further 35 per cent optimism bias): additional fleet to reach 18 trains an hour (€450m in v4; since v5 priced by the cycle-time rule at one set cost, as every fleet row), a ventilation step (€500m of shafts to reach 26 trains an hour; v5.0 stopped at Jacobs' 20, which needs no intermediate shafts, and v5.1 put the shafts back), ten-car provision (€600m), depot and stabling (€225m), traction power (€150m), in-cab signalling (€225m), west-side turnbacks (€120m) and, since v5.4, an east-side turnback for the step trains that do not run to Malahide (€120m). Together the package is €8.9bn of the €12.2bn of fixed steps that belong to the Rail Towns option only.

A 30 trains-an-hour step (automatic train operation over ETCS Level 2, more turnback capacity) exists with a placeholder cost of €250m and is off by default; Thameslink and Munich's second Stammstrecke are the references that would price it.

The "Rail Towns without the tunnel package" sensitivity switches these eight rows off and keeps everything else.

Link to this section

The cheapest capacity in the model, and the least documented.

Two outer works decide how many towns can reach the tunnel.

Four-tracking the roughly 12 km from Hazelhatch to Sallins and Naas separates DART trains from intercity paths on the Kildare line. The central cost is €600m (band €400m to €900m), with electrification to Naas at €200m on top. From v3.8 the step supplies seats at the plan's service level: 13 DART trains an hour on the four-tracked pair, less the shared paths, times 870. The Kildare outer section then stops being the constraint in the Rail Towns option and the tunnel core takes over. The Naas terminal for that frequency is unpriced.

The Northern outer-section package, a passing loop at Skerries and a chord so Enterprise trains can overtake, lifts the usable DART paths by 30 per cent for €150m. No published figure exists for either piece of work; the costs come from Gerry Duggan's notes on the cost drivers and are status A.

The honesty ledger lists costing both as an open item, because they are the model's cheapest homes: about €600m for some 40,000 homes' worth of seats on the Naas line.

Link to this section

Cork, Limerick and Galway rows, several of them illustrative.

Cork's ladder has the funded first phase of the Cork Area Commuter Rail programme (through-platform at Kent, Glounthaune–Midleton double track, resignalling; €164m), the full programme (electrification Mallow–Cork–Cobh and Midleton, eight stations and fleet; €1.64bn, from the programme's chief executive's figure of about €1.8bn in February 2026) and an illustrative further-corridors row. The full programme's business case is unpublished and the ledger flags it.

Limerick has the committed Colbert station works with Moyross and Parkway access, the Limerick–Limerick Junction electrification with the Limerick–Shannon strategy's rail package and fleet (Jacobs 2022 gave €488m to €561m for its first option), and an illustrative Foynes and Shannon-side row.

Galway has the committed Ceannt station redevelopment (€69.5m) with the Oranmore passing loop, the Galway–Athenry double-tracking and electrification at the All-Island Strategic Rail Review's share, and the Western Rail Corridor north, which the review scored at 0.21 and the model carries with caution.

Link to this section

Which homes take the seats first, how fast towns can be built, and where infill comes from.

Seats are allocated in a fixed order: committed catchments first, station infill second, greenfield towns last. Each group is capped in its own way. Committed homes are capped by seats. Infill is capped by the station-area potential from the densification model. Towns are capped three times over: by core seats, by the outer section, and by delivery pace.

Since version 5.5 (ledger #108) the Rail Towns option adds towns on the seats its trains leave empty, on the Naas, Northern, Enfield and Navan legs, and its counted trains carry 1,088 places as 10-car trains it buys and runs. Since version 5.7 (ledger #110) it builds at most 25 towns at once (20 at the Low band), and the towns keep building after 2050 to 2055 (the CBA's fill_end, at each option's own building pace); at central its trains are full in 2055. In the central case it builds 25.0 towns at once and completes 238,809 town homes by 2050. Road Towns and Bundle 6 keep their programmes; the v5.4 Rail Towns, with its towns limited by pace and 870 places a train, is the sensitivity v55_paced, and the matched comparison, every option filling its trains, is v55_fair_fill (see the version 5.5 section).

Central, Dublin	Road Towns	Rail Towns
Committed homes served	42,392	57,000
Station infill served	44,784	94,014
Town homes served	38,982 of 90,000 potential	238,809 of 300,000
Town homes displaced to the roads	51,018	99,064
Rail-served by 2050	126,158	389,823

The binding constraint by leg in the Rail Towns option is computed each build: K delivery pace, N delivery pace, M delivery pace, X delivery pace, V delivery pace; in the Road Towns option K site potential, N core seats, M delivery pace, X delivery pace, V site potential. Homes displaced from a town do not vanish. The regional total is fixed, so they land in the road tiers at car-suburb density and generate road cost and car trips.

The demand engine scales every one of these steps by network: the town density factor shrinks the potential a site can support, the pace factor slows absorption on weak lines, and an interpolation weight places infill between the Road Towns and Rail Towns densification runs.

Modelling choices referenced here

M3 · Fill order committed → infill → towns

Choice. Committed catchments take seats first (capped by seats), infill next (lightest load, nearest the market), towns last, capped by pace and by the outer sections.

Alternatives. Pro-rata allocation; towns first.

Why. Reflects designation status and market proximity; conservative for the towns.

Effect. Decides which homes are displaced when seats run out.

Status. Convention Evidence that would replace it. A site-level delivery register.

M6 · Delivery pace cap, building years from opening (v3.4)

Choice. Towns × 600 homes per town-year × building years, where the years run from when each town's railway opens (Option 1 19, Option 2 16).

Alternatives. 20 years from 2030 for all towns (v3.3); Codex's dated cohorts at 400/yr from 2036/2040.

Why. No home before its service.

Effect. Additional rail homes 210k → 165k; capacity potential (no cap) reported separately (+213k).

Status. A Evidence that would replace it. Designation and opening dates per town; absorption evidence (Adamstown, Clonburris).

600 homes a year per town, counted from the day its railway opens.

Each town completes 600 homes a year in the central case (300 Low, 1,000 High). The Irish record brackets it: Adamstown and Cherrywood about 230 a year, Clonburris about 700; Leidsche Rijn in the Netherlands reached 1,250. Building years run from when the town's railway opens, the external review's most useful correction: v3.3 had towns completing eight years before the tunnel existed. The Road Towns option's committed-line towns get 19 years from about 2032; the Rail Towns option's towns get 16 from about 2034–35, when the outer works and extensions open ahead of the 2038 tunnel; Navan-branch towns get 13 from about 2036. Both paces are judgements inside the Irish range; the model kept 600 and adopted the opening-date logic.

The engine's pace factor slows absorption on weak networks modestly (elasticity 0, 0.25 or 0.5 on relative accessibility), capped at 1. The reviewer of September 2026 asked what twenty concurrent towns at 600 homes a year, 12,000 a year before infill, would need in sites, assembly, finance, utilities and construction capacity; an aggregate pace cap is a Stage 2 switch of the rebuild, reported against that figure.

Sensitivity: at 300 a year the Rail Towns option rail-serves 269,823 homes and Δ A is −€9.3bn; at 1,000 a year it rail-serves 462,654 and Δ A is −€7.1bn. Pace moves the homes far more than it moves the cost gap, because a slower Rail Towns option also leaves more roads unbuilt in the Road Towns option's favour.

Modelling choices referenced here

M6 · Delivery pace cap, building years from opening (v3.4)

Choice. Towns × 600 homes per town-year × building years, where the years run from when each town's railway opens (Option 1 19, Option 2 16).

Alternatives. 20 years from 2030 for all towns (v3.3); Codex's dated cohorts at 400/yr from 2036/2040.

Why. No home before its service.

Effect. Additional rail homes 210k → 165k; capacity potential (no cap) reported separately (+213k).

Status. A Evidence that would replace it. Designation and opening dates per town; absorption evidence (Adamstown, Clonburris).

Link to this section

Homes within walking distance of existing stations, computed from the built-up footprint.

Infill potential comes from the densification model (model/densification.py in the v3 archive; a station-by-station script the v4 model reads as inputs). For every station it takes the built-up footprint within an 800 m ring from the European Commission's GHSL 2025 grid, applies a turnover share of 12, 18 or 25 per cent over the period, multiplies by the uplift from today's density (22 to 55 homes per hectare by zone) to a station-area target of 60, 80 or 110, and adds 5, 10 or 15 per cent of the unbuilt land as brownfield. Tunnel-core stations get targets of 100, 140 or 180. The catchment is graded: the inner 500 m counts in full, the 500 to 800 m band at 0.5, 0.65 or 0.8. No Census small-area data is used; the existing densities are zone assumptions, and the ledger says so.

The Rail Towns option applies the through-running uplift to every station on the network, which is why its potential is larger: 95,477 against the Road Towns option's 64,309 in the central case. The share of infill at inner-zone stations (16% in the Road Towns option, 30% in the Rail Towns option) feeds the trip rate. With the demand engine on, a network's infill is interpolated between the two runs by where it sits in accessibility. The evidence that would sharpen it is Census small-area density and a survey of station-area sites.

Link to this section

15,000 homes each, sited by leg, at a density the ledger has checked.

A rail town in the model is 15,000 homes within walking distance of a new station. The density calibration was caught and corrected by Seán on 31 August 2026 (ledger #21): the headline is 75 homes per hectare gross at an 800 m radius, about 1.65 times Adamstown's gross density, and at a 1 km radius it is 48 per hectare, which is Adamstown's planned yield. The net-block figure of about 165 per hectare is ordinary Dublin apartment-scheme density once commercial land and the street network are carved out.

The Rail Towns option's twenty towns sit eight on the Naas line, four on the Drogheda line, three on the Enfield line, three on the Navan branch and two on MetroLink. Those are the programme's towns. Since version 5.5 the option adds towns on the seats its trains still leave empty, on the same legs, until the seats are full or the homes to serve run out: 238,809 town homes by 2050 against the programme's 163,200, on 25.0 towns building at once. The Road Towns option carries six committed-line greenfield sites with 1,600 m catchments (27,000 homes on the Kildare line, 21,000 on the Northern, 12,000 on the Maynooth) and the same two MetroLink towns.

The demand engine's density factor scales the homes a site can support. On the committed network a Kildare-line site supports 45% of its tunnel-grade density and a Northern one 62%; the balance goes to the exurb.

The TOD Dublin Study's inventory has Lissenhall at 7,000 homes on 240 hectares, about 30 per hectare, against the model's 15,000 at the land-readjustment model's 120 per hectare. Substituting the study's figure barely moves the answer.

Link to this section

The demand engine: a town on a weak line drives more, rides less and builds lower.

Until v3.4 every rail-served home behaved identically whichever railway served it: the same 0.082 peak rail trips, the same car trips, the same density. Seán's objection on 5 September 2026 was that a dense town on a terminating, low-frequency line cannot behave like one on a through-running network. Its residents drive, and those cars need the roads the roads-led city has to build. The demand engine (model/modules/engine.ts, the one implementation since v4) answers that on the logic of Vuchic's Urban Transit: Operations, Planning and Economics (2005), with weights from Small, Verhoef and Lindsey's The Economics of Urban Transportation (2007).

For each Dublin leg and each of three networks (the committed network; the rail-led network without the tunnel; with it) the engine computes:

door-to-door generalised travel time by rail and by car to three city-centre job clusters;
an accessibility index for each mode, summing jobs weighted by a friction factor that decays with time;
the rail share of peak trips from a diversion curve on the car-to-rail accessibility ratio, calibrated so the tunnel network gives 0.45;
four factors relative to the tunnel network: rail trips per home, strategic car trips per rail home, supportable town density and absorption pace, plus an interpolation weight for infill.

Everything is relative to the tunnel network, which always scores 1. The engine removes demand from weaker networks; it never adds demand to the tunnel. Any uplift above 1 is capped at 1.3, Vuchic's ceiling for high-performance rail.

What it does to the answer: with flat behaviour (engine off) Δ A is −€8.2bn; with the engine it is −€7.9bn (Δ A is Road Towns minus Rail Towns on state transport). The tunnel package's own value in the CBA is +€0.1bn at rung 1 and −€0.7bn at rung 2, the difference between the Rail Towns case with the package and without it; with flat behaviour on v3.4 it was −€2.2bn, because the tunnel-less rail-led city becomes a dense car suburb with a fuller, less useful railway only once behaviour responds. The engine is uncalibrated, which is why its parameters carry status A and bands. The reviewer of September 2026 made two objections that stand: the 0.45 calibration on the tunnel network cancels car accessibility on the main corridors (halving driving times leaves their rail shares unchanged), and the density mechanism supplied most of the v3.9 headline gap. Calibrating on observed station catchments, holding the calibration fixed across the options and testing an equal-density alternative are Stage 2 switches of the rebuild.

Modelling choices referenced here

M22 · Demand engine (v3.5): trips, car trips, density and pace respond to service quality

Choice. Vuchic's demand-forecasting logic per Dublin corridor for three networks (committed; rail-led without the tunnel; with it): door-to-door transit time from his components (§1.3: access, wait h/2 capped at 5 min, on-line, transfer walk + connecting wait + resistance, §4.4); through-running as an integrated line that removes the terminal transfer without dividing frequency (§4.2.4); accessibility a = Σ jobs × friction(time) for transit and car (eq. 10.6); rail share from a diversion curve on the accessibility ratio with captive floors (§10.3, Fig. 10.7), calibrated to 0.45 on the tunnel network; the rail factor caps any uplift at 1.3 (§12.2.4.1). Rail trips per home = share ÷ 0.45; car trips per rail home = (1 − share) ÷ 0.55; pace responds modestly and is capped at 1; density follows accessibility from v3.6 (M23; in v3.5 it was plan-led, §10.4, ε 0.25). Option 2 follows the Ladder tunnel toggle.

Alternatives. Flat rates whichever the network (v3.4); the first draft's PDFH/TAG elasticity form; a full four-step model.

Why. A dense town on a terminating, low-frequency line cannot behave like one on a through-running network. In Vuchic's framework the main effect is modal split, so the engine's first channel is driving; from v3.6 density follows accessibility as well (M23); it removes demand from weaker networks rather than adding it to the tunnel.

Effect. Central: rail-led towns without the tunnel keep a 0.33 rail share against 0.45 (Kildare 0.28), make 0.74× the rail trips and 1.21× the car trips; the committed network 0.40 / 0.89× / 1.09×. See the v3.5 note for homes and costs.

Status. A (Vuchic's method; the service times, node jobs, slope λ 1.5/2/2.5, resistance 3/5/8 min and the 0.45 calibration are stated assumptions) Evidence that would replace it. An NTA ERM run with mode choice for the modelled homes; observed rail shares in Irish station-area TOD (Adamstown, Clongriffin, Donabate).

M23 · Density follows accessibility (v3.6; exponent 1.0 central from v3.7)

Choice. Supportable town density on each Dublin leg = (accessibility relative to the tunnel network) ^ ε, capped at 1, with ε 0.50 (Low) / 1.00 (central) / 1.50 (High) from v3.7 (v3.6: 0.25 / 0.50 / 1.00). Displaced homes are not lost: Dublin's total is fixed, so they land in the road tiers at car-suburb density and generate road cost and car trips.

Alternatives. v3.5's weak channel (ε 0.25: density planned with the railway, Vuchic §10.4, the market responding through driving); v3.6's ε 0.50 (now the Low bound); a step function on apartment viability (the High bound, ε 1.5, stands in for it (a site flips from apartments to houses once the price no longer carries apartment construction).

Why. Seán's reading (5 Sep, late): the homes get built and the residents drive, but at the density the line's land premium supports. Density is what a developer buys when land beside the station is worth more than land further out, and that premium exists only where the station's accessibility is worth paying for (bid-rent: Alonso, Muth, Mills). A terminating line with a change at Heuston earns a smaller premium than a through-running one, so the density that pays at Kilcock or Sallins is lower and the marginal home goes to the exurb where land is cheapest per plot. Irish evidence points the same way: the 2000s commuter belt grew at 20–30 homes/ha in Meath, Laois, Louth and north Kildare while the DART and Kildare line had spare seats; Adamstown, zoned for 10,000 homes on the Kildare line in 2003, has delivered a few thousand beside a line that never got the promised frequency; Cherrywood on the Luas delivered at density; the NPF's compact-growth policy has not stopped exurban growth since 2018 (planning sets a ceiling, the market the floor).

Effect. Central (v3.7) density factors: Kildare towns 0.69× on the committed network and 0.63× without the tunnel; Northern 0.75× / 0.78×; Enfield 0.77× without the tunnel; MetroLink 1. See the v3.6 section of simple-model/demand-engine-2026-09-05.md for homes and costs; the two bounds are run as sensitivities with and without the tunnel package.

Status. A (ε is a judgement bracketed by the two readings; the functional form is the standard bid-rent chain) Evidence that would replace it. A Dublin hedonic station premium (ESRI: Mayor, Lyons, Duffy & Tol 2012, rail in the GDA) chained to a density–price elasticity; delivered densities against service levels at Adamstown, Clongriffin and Cherrywood.

M24 · Out-of-vehicle time weighted in the demand engine (v3.8)

Choice. Access walk, platform wait, transfer walk and connecting wait carry a weight of 1.6 (Low) / 2.0 (central) / 2.5 (High) relative to in-vehicle time in the engine's door-to-door generalised minutes; the car's parking search and walk (8 min) carry the same weight; the transfer resistance (3 / 5 / 8 min) stays as the extra disutility of changing trains. Times passed to the CBA are generalised minutes.

Alternatives. Unweighted minutes (v3.5–v3.7); a separate transfer penalty only.

Why. Small, Verhoef & Lindsey, The Economics of Urban Transportation (2007) §2.6.5: walking and waiting time are valued at 1.6 to 2.0 times in-vehicle time, conventionally 2.0 to 2.5, not counting the context-specific disutility of transfers (Wardman 2001, Mackie et al. 2003). A terminating line that forces a change is therefore worse relative to through-running than face-value minutes say.

Effect. Raises the accessibility gap between the tunnel network and the terminating networks on every leg; see the v3.8 section of the demand-engine note for the factors and results.

Status. P (the weights are the standard appraisal convention; the transfer resistance remains A) Evidence that would replace it. Irish stated- and revealed-preference values for walk, wait and interchange (TII PAG Unit 6.11 / NTA ERM parameters).

Vuchic's components, with out-of-vehicle minutes weighted double.

Travel time is the sum of its parts. For rail: a six-minute access walk to the town station; waiting of half the headway, levelling off at five minutes for longer headways because passengers use the timetable (Vuchic Fig. 1.8); on-line time; and, where the journey needs a change, the transfer walk plus half the connecting headway plus a resistance penalty of 3, 5 or 8 minutes, because changing trains "interrupts travel and requires passengers to orient themselves". Through-running is Vuchic's integrated line: it removes the terminal change for Kildare and Northern passengers without dividing the frequency, which is the whole point of the R09 tunnel.

From v3.8, access, waiting, transfer walking and connecting waits are weighted 1.6, 2.0 or 2.5 times in-vehicle time, following Small, Verhoef and Lindsey §2.6.5, which puts the conventional range at 2.0 to 2.5 with transfers carrying extra disutility on top. The car's eight minutes of parking search and walk carry the same weight. Times passed to the cost–benefit analysis are these generalised minutes.

An example on the central band: a Kildare-line town on the committed network scores 78 generalised minutes to the job clusters against 74 by car; with the tunnel the same town is direct to all three clusters in one seat.

The weights are standard appraisal practice, but they double every change-of-train penalty. The ledger's honest line on v3.8: the widening gap is a statement about how badly a terminating line serves a city with jobs in four directions, and the committed-network Kildare rail share (0.19) should be checked against Census 2022 rail commuting at Adamstown and Hazelhatch before it is quoted.

Modelling choices referenced here

M24 · Out-of-vehicle time weighted in the demand engine (v3.8)

Alternatives. Unweighted minutes (v3.5–v3.7); a separate transfer penalty only.

Effect. Raises the accessibility gap between the tunnel network and the terminating networks on every leg; see the v3.8 section of the demand-engine note for the factors and results.

Link to this section

Jobs at Heuston, the Green and the Docklands, weighted by how long they take to reach.

Accessibility follows Vuchic's index, equation 10.6: the sum over destinations of trip attractions times a friction factor of travel time. The destinations are three employment nodes inside the canal ring, rounded from Census 2022 place-of-work data: West (Heuston and James's) 45,000 jobs, Centre (Stephen's Green, Pearse, Grafton) 160,000, East (Docklands, IFSC, Connolly) 95,000. Vuchic leaves the friction form open; the engine uses the common negative exponential, exp(−0.045 × minutes).

The same index is computed for the car, from peak door-to-door minutes taken from the CBA's congested-road rings (45 to 68 minutes) plus the weighted terminal time. Each leg's rail accessibility is then expressed relative to its tunnel-network value, A_rel.

Leg, central band	Committed network	Rail-led without the tunnel	With the tunnel
K Kildare / Naas	0.45	0.52	1.00
N Northern / Drogheda	0.62	0.68	1.00
M Maynooth / Enfield	1.20	0.68	1.00
X MetroLink	0.98	0.98	1.00
V Navan branch	0.48	0.48	0.70

The Maynooth leg scores above 1 on the committed network because its committed towns are Maynooth and Kilcock, closer in than the Enfield towns the rail-led plan adds. The Navan branch is measured against the Maynooth leg's tunnel-grade service rather than its own, so a 60-minute town at four trains an hour is not credited as tunnel-grade just because the tunnel exists.

Link to this section

Calibrated to 0.45 on the tunnel network, bounded by captive riders and captive drivers.

The rail share of peak trips follows Vuchic's diversion curve (§10.3, Fig. 10.7): share = 0.06 + (0.85 − 0.06) ÷ (1 + k × ratio^λ), where the ratio is car accessibility over rail accessibility, λ is 1.5, 2.0 or 2.5, and the floors are captive riders at 6 per cent and captive drivers at 85 per cent, because the curve never runs from nought to a hundred. The constant k is calibrated on each leg so that the tunnel network gives a rail share of 0.45.

Why 0.45: Census 2022 records 20 to 24 per cent of commuters using rail across whole rail towns such as Donabate and Skerries, most of whose residents live well beyond walking distance of the station. A walkable station area on a through-running line is taken at roughly double.

Leg, central band rail share	Committed network	Without the tunnel	With the tunnel
K Kildare / Naas	0.19	0.23	0.45
N Northern / Drogheda	0.28	0.30	0.45
M Maynooth / Enfield	0.52	0.31	0.45
X MetroLink	0.44	0.44	0.45
V Navan branch	0.31	0.31	0.45

Weighted by town potential, the committed network's average share is 0.33 and the tunnel-less rail-led network's is 0.29.

The slope, the floors and the calibration point are stated judgements. An NTA regional model run with mode choice for the modelled homes would replace them, and observed shares in Irish station-area development (Adamstown, Clongriffin, Donabate) would test them.

Link to this section

The four factors each leg and network feeds into the workbook, all relative to the tunnel.

From the rail share come the factors the capital model reads for every leg and network.

Rail trips per home = share ÷ 0.45, capped at 1.3. A weaker line means fewer rail trips per home, so each seat buys more homes, and a fuller but less useful railway.
Strategic car trips per rail home = (1 − share) ÷ 0.55. This scales the TCRP ratio of 0.51, so a Kildare town on the committed network loads the roads at 1.47 times a tunnel-grade town.
Supportable town density = A_rel to the power ε, capped at 1 (see the next section).
Absorption pace = A_rel to the power 0, 0.25 or 0.5, capped at 1.
Infill weight: where the network sits between the Road Towns and Rail Towns densification runs.

The engine also hands the cost–benefit analysis its travel parameters for each network. On the tunnel network the shifted household saves 15.0 minutes a trip, 75% of its trips switch and it avoids 10,800 car-km a year; on the committed network the figures are 4.3 minutes, 55% and 8,859 km; without the tunnel, 2.3 minutes, 48% and 8,139 km. The surface-case CBA uses these overrides.

Each leg's central factors are set out below.

Central-band factors for the Kildare / Naas leg (leg K) from the demand engine (model/modules/engine.ts, the one implementation). Status A throughout: they are outputs of a stated method, not observations.

Network	Accessibility relative to the tunnel network	Rail share of peak trips (Low to High band)	Rail trips per home	Strategic car trips per rail home	Supportable town density	Absorption pace	Generalised minutes to the job clusters, rail against car
Committed network (Road Towns)	0.45	0.19 (0.28 to 0.12)	0.42	1.47	0.45	0.82	78 against 74
Committed network plus Bundle 6 (Option 3)	0.65	0.29 (0.35 to 0.22)	0.64	1.29	0.65	0.90	69 against 74
Rail Towns without the tunnel	0.52	0.23 (0.30 to 0.15)	0.50	1.41	0.52	0.85	75 against 74
Rail Towns with the tunnel	1.00	0.45 (0.45 to 0.45)	1.00	1.00	1.00	1.00	58 against 74

Every factor is relative to the tunnel network on the same leg, which scores 1 by construction. Rail trips per home = rail share ÷ 0.45, capped at 1.3; car trips per rail home = (1 − share) ÷ 0.55; density and pace are relative accessibility raised to their exponents, capped at 1.

Link to this section

Central-band factors for the Northern / Drogheda leg (leg N) from the demand engine (model/modules/engine.ts, the one implementation). Status A throughout: they are outputs of a stated method, not observations.

Network	Accessibility relative to the tunnel network	Rail share of peak trips (Low to High band)	Rail trips per home	Strategic car trips per rail home	Supportable town density	Absorption pace	Generalised minutes to the job clusters, rail against car
Committed network (Road Towns)	0.62	0.28 (0.34 to 0.20)	0.61	1.32	0.62	0.89	77 against 71
Committed network plus Bundle 6 (Option 3)	0.62	0.28 (0.34 to 0.20)	0.61	1.32	0.62	0.89	77 against 71
Rail Towns without the tunnel	0.68	0.30 (0.36 to 0.23)	0.68	1.26	0.68	0.91	75 against 71
Rail Towns with the tunnel	1.00	0.45 (0.45 to 0.45)	1.00	1.00	1.00	1.00	63 against 71

Link to this section

Central-band factors for the Maynooth / Enfield leg (leg M) from the demand engine (model/modules/engine.ts, the one implementation). Status A throughout: they are outputs of a stated method, not observations.

Network	Accessibility relative to the tunnel network	Rail share of peak trips (Low to High band)	Rail trips per home	Strategic car trips per rail home	Supportable town density	Absorption pace	Generalised minutes to the job clusters, rail against car
Committed network (Road Towns)	1.20	0.38 (0.41 to 0.33)	0.84	1.13	1.00	1.00	80 against 76
Committed network plus Bundle 6 (Option 3)	1.20	0.38 (0.41 to 0.33)	0.84	1.13	1.00	1.00	80 against 76
Rail Towns without the tunnel	0.68	0.31 (0.36 to 0.24)	0.68	1.26	0.68	0.91	93 against 84
Rail Towns with the tunnel	1.00	0.45 (0.45 to 0.45)	1.00	1.00	1.00	1.00	82 against 84

Link to this section

Central-band factors for the MetroLink corridor leg (leg X) from the demand engine (model/modules/engine.ts, the one implementation). Status A throughout: they are outputs of a stated method, not observations.

Network	Accessibility relative to the tunnel network	Rail share of peak trips (Low to High band)	Rail trips per home	Strategic car trips per rail home	Supportable town density	Absorption pace	Generalised minutes to the job clusters, rail against car
Committed network (Road Towns)	0.98	0.44 (0.44 to 0.45)	0.99	1.01	0.98	1.00	53 against 59
Committed network plus Bundle 6 (Option 3)	0.98	0.44 (0.44 to 0.45)	0.99	1.01	0.98	1.00	53 against 59
Rail Towns without the tunnel	0.98	0.44 (0.44 to 0.45)	0.99	1.01	0.98	1.00	53 against 59
Rail Towns with the tunnel	1.00	0.45 (0.45 to 0.45)	1.00	1.00	1.00	1.00	52 against 59

Link to this section

Central-band factors for the Navan branch leg (leg V) from the demand engine (model/modules/engine.ts, the one implementation). Status A throughout: they are outputs of a stated method, not observations.

Network	Accessibility relative to the tunnel network	Rail share of peak trips (Low to High band)	Rail trips per home	Strategic car trips per rail home	Supportable town density	Absorption pace	Generalised minutes to the job clusters, rail against car
Committed network (Road Towns)	0.48	0.31 (0.36 to 0.24)	0.68	1.26	0.48	0.83	101 against 92
Committed network plus Bundle 6 (Option 3)	0.48	0.31 (0.36 to 0.24)	0.68	1.26	0.48	0.83	101 against 92
Rail Towns without the tunnel	0.48	0.31 (0.36 to 0.24)	0.68	1.26	0.48	0.83	101 against 92
Rail Towns with the tunnel	0.70	0.45 (0.45 to 0.45)	1.00	1.00	0.70	0.91	90 against 92

The Navan branch branch is measured against the Maynooth / Enfield leg's tunnel-grade service rather than its own network, so its factors with the tunnel are below 1. Rail trips per home = rail share ÷ 0.45, capped at 1.3; car trips per rail home = (1 − share) ÷ 0.55; density and pace are relative accessibility raised to their exponents, capped at 1.

Link to this section

The bid-rent argument, the Irish evidence and an exponent that is a judgement.

Vuchic treats density as planned with the transit system (Stockholm's suburbs) and puts the market's first response to a weak line in mode share. v3.5 followed him and gave density only a weak elasticity of 0.25. Seán's reading on 5 September added the supply side: the homes still get built and their residents still drive, but at the density the line's land premium supports. Density is what a developer buys when land beside the station is worth more than land further out, and that premium exists only where the station's accessibility is worth paying for. This is the standard bid-rent chain (Alonso, Muth, Mills). A terminating line with a change at Heuston earns a smaller premium than a through-running one, so the density that pays at Kilcock or Sallins is lower and the marginal home moves to the exurb, where land is cheapest per plot.

Irish evidence runs the same way. The 2000s commuter belt grew at 20 to 30 homes per hectare in Meath, Laois, Louth and north Kildare while the DART and the Kildare line had spare seats. Adamstown, zoned for 10,000 homes on the Kildare line in 2003, has delivered a few thousand beside a line that never got its promised frequency. Cherrywood on the Luas delivered at density. The National Planning Framework's compact-growth policy has not stopped exurban growth since 2018: planning sets a ceiling, the market sets the floor.

The rule: supportable density = A_rel^ε, capped at 1 because a 15,000-home site does not grow with better access. The exponent moved twice. v3.6 set it at 0.25, 0.50, 1.00. v3.7 (7 September) set the central case to 1.00, density proportional to accessibility, on Seán's argument that in a city with jobs in the centre, north, south and west a terminating line cannot be relied on; Low is 0.50 and High is 1.50, standing in for a step effect in which a site flips from apartments to houses once the price no longer carries apartment construction.

Central density factors: Kildare towns 0.45 on the committed network and 0.52 without the tunnel; Northern 0.62 and 0.68; Enfield 0.68 without the tunnel; MetroLink 0.98; the Navan branch 0.48 and 0.70 with the tunnel. The exponent moves Δ A between −€7.9bn (0.25) and −€7.9bn (1.5).

What would replace it: a Dublin hedonic station premium (Mayor, Lyons, Duffy and Tol, 2012) chained to a density–price elasticity, and delivered densities at Adamstown, Clongriffin and Cherrywood set against the service each got.

Modelling choices referenced here

M23 · Density follows accessibility (v3.6; exponent 1.0 central from v3.7)

Link to this section

The timetable assumptions the engine runs on, all status A.

The engine's inputs mirror the workbook's service plan: trains per hour at the town station, on-line minutes to the arrival node, the transfer walk and connecting headway where a journey changes, and car minutes to each node.

Leg	Committed network	Rail-led without the tunnel	With the tunnel
K Kildare / Naas	5 tph to Heuston (32 min); Centre and Docklands by Luas Red or on foot	13 tph on the four-tracked pair, terminating at Heuston (34 min)	13 tph through R09: Heuston, Stephen's Green, Spencer Dock in one seat (34 / 40 / 46 min)
N Northern / Drogheda	7 tph to Connolly (38 min); Centre by walk or Luas; Heuston by Luas Red	9 tph, same reach	9 tph through R09 (38 / 46 / 54 min)
M Maynooth / Enfield	8 tph to Docklands / Connolly (42 min)	6 tph Enfield extension (52 min)	6 tph, loop freed: Pearse in one seat (52 / 58 min); Heuston still a change
X MetroLink	20 tph to Tara Street (28 min); Docklands on foot	as committed	cross-platform change to the tunnel at Tara Street
V Navan branch	4 tph to Connolly / Spencer Dock (60 min)	as committed	4 tph, loop freed: Pearse in one seat (60 / 66 min)

Car door-to-door minutes in the peak run from 40 to 78 depending on leg and node, from the CBA's congested-road rings, plus 8 minutes of parking search and walk. The Navan branch's car times are an assumption: about eight minutes beyond a Maynooth-leg town by the M3 and M50.

None of these is an observation. They are the plan's service levels and timetable-anchored minutes, and the NTA regional model would replace them.

Link to this section

Strategic roads as scarcity tiers, then distributors, parking and utilities, all per home.

Every home the railway cannot carry is a car-suburb home and is served by five things the model prices separately: strategic road capacity, distributor roads, destination parking, utilities and connections, and the operating costs of roads and rail. Rail-served homes are priced on the same five lines at lower rates; nothing is charged to one typology and waived for the other.

Central, Dublin	Road Towns	Rail Towns
Strategic roads (scarcity tiers)	€9.0bn	€5.9bn
Distributor and link roads	€5.8bn	€3.2bn
Destination parking (private)	€12.9bn	€9.1bn
Utilities and connections	€12.0bn	€10.2bn
Rail and road operating costs inside the window	€2.6bn	€3.4bn

Strategic roads are the decisive line and the least documented. The model does not assign traffic to a network. It stacks car-home equivalents cumulatively against Dublin's road network and charges each tranche at the cost of the marginal road it would need, from spare headroom through rural dual carriageway, orbital and urban-edge insertion to Port Tunnel class once the city is inside an M50 that TII says cannot be widened. The tier rates are derived from sourced parts; the tier bounds, where one class of road gives way to the next, are assumptions, and they decide the state-transport answer. On this version's basis the Road Towns option stops -1,416 car-home equivalents short of the tunnel-class tier, so the strategic-road gap is €3.1bn and the state-transport gap Road Towns €7.9bn cheaper; the reviewer of September 2026 showed that a 7 per cent move in that one bound was enough to remove the v3.9 saving, and the switching-value table below now reports it by input.

3). The model's position: a rural dual-carriageway rate cannot price capacity inside an M50 that cannot be widened; either the road is built, and it costs what insertion or tunnelling costs, or it is not, and congestion and the switchers' time savings rise. Its own break-even is €84m per lane-kilometre.

Modelling choices referenced here

M8 · Roads as scarcity tiers on car-home equivalents

Choice. Cumulative car-home equivalents (road homes + rail homes × 0.51) climb five tiers with derived €/equivalent rates (headroom → rural DC → orbital → urban-edge → tunnel class); bounds per city.

Alternatives. Codex: one blended €5m per lane-km with 50% of strategic travel needing no new capacity; a network assignment.

Why. Dublin's marginal road capacity is not a rural dual carriageway; TII says the M50 cannot be widened. A blended rate hides the scarcity.

Effect. Tier bounds ×2 → Δ A −€5.0bn; ×0.5 → +€11.6bn. THE decider.

Status. A on bounds; rates derived from sourced components Evidence that would replace it. NTA ERM select-link runs; TII radial cordon capacities.

M9 · Rail homes carry road cost (v3.4: TCRP ratio only)

Choice. A rail home loads the strategic network at 0.51 of a car home (TCRP 128).

Alternatives. v3.3's 0.38 (with a further internal-capture discount); v1's unsourced 0.21.

Why. The TCRP ratio is measured at the dwelling and already contains local trips; the strategic share applies to both typologies.

Effect. 0.40/0.60 → Δ A €0.3bn / €5.1bn (Option 1's extra equivalents fall in the tunnel-class tier).

Status. P Evidence that would replace it. Irish TOD trip surveys (Adamstown, Clongriffin).

M10 · Parking as a resource cost, discounted once (v3.4)

Choice. Destination spaces per car home from cars × per-car spaces × sharing (1.32); rail homes = car-home spaces × TCRP TOD ratio 0.5 (0.66). Buildcost rates plus RZLP land.

Alternatives. v3.3's two discounts (0.40 spaces); Codex's 1.3 / 0.8 at €20k.

Why. TCRP's per-dwelling finding already embodies car-shedding.

Effect. Δ parking €3.6bn → €2.0bn.

Status. P on rates and ratio; A on mix Evidence that would replace it. An Irish destination-parking inventory per home.

M11 · Utilities: connections equal, local networks by street-metres, trunk mostly per household (v3.4)

Choice. Uisce/ESB charges per home on both; local buried networks by street-metres × €1.1k/m; trunk allowance €6.7k with a 10% conveyance saving for rail towns.

Alternatives. v3.3's 60% trunk saving; Codex's €9.5k/€5k local and 6% trunk.

Why. Treatment and source capacity is per household; only pipes are shorter in a compact town.

Effect. Δ utilities €2.1bn → €1.1bn.

Status. A Evidence that would replace it. Uisce capital-programme split; quotable-rates schedule.

M12 · Distributor roads on both typologies (v3)

Choice. LIHAF-class €12.3k per car-suburb home; rail towns scaled by street-metres.

Alternatives. Charging distributors to car suburbs only (v1).

Why. Symmetric boundary.

Effect. Δ distributors €1.5bn.

Status. P (LIHAF rate) Evidence that would replace it. Matched-layout cost comparison.

Five tiers, each dearer than the last, on cumulative car-home equivalents.

Homes are converted to car-home equivalents: a car-suburb home counts 1, a rail-served home 0.51 (scaled by the engine's car-trip factor). The equivalents are stacked cumulatively and charged in five tiers.

Tier	Marginal road it stands for	Bound (cumulative equivalents)	Rate per equivalent	Road Towns take	Rail Towns take
1	Existing headroom, junction allowance only	to 50,000	€5.0k	50,000	50,000
2	Rural dual carriageway	to 150,000	€9.3k	100,000	100,000
3	Outer orbital (Leinster Orbital class) with M50 junction rebuilds	to 300,000	€23.6k	150,000	150,000
4	Urban-edge insertion (M28 class)	to 500,000	€21.4k	200,000	57,241
5	Beyond the urban edge: more urban-edge road further out (urban tunnel class until v5.2)	open-ended	€21.4k	1,416	0

The Road Towns option reaches 501,416 equivalents and pays €0.0bn in the top tier; the Rail Towns option stops at 357,241. The strategic-road gap is €3.1bn. Until v3.11 the Road Towns option sat 53,186 equivalents inside the top tier and the gap was €15.4bn; the v3.12 demand reconciliation took it out.

The regional cities use the same structure with bounds scaled by population and the top tier priced at the M28 urban-edge rate, because only Dublin's ring is at capacity. Since v5.3 (ledger #100) Dublin's top tier is priced the same way: no urban road tunnels, more urban-edge road further out. The car homes beyond the bound then drive further, and the cost–benefit analysis counts those longer commutes, from the year the region's car homes reach the bound's ramp.

The alternative was a network assignment, which the model does not have, or a single blended rate, which hides the scarcity. What would replace the tiers: NTA regional model select-link runs on the radials and TII cordon capacities, corridor by corridor.

Modelling choices referenced here

M8 · Roads as scarcity tiers on car-home equivalents

Alternatives. Codex: one blended €5m per lane-km with 50% of strategic travel needing no new capacity; a network assignment.

Why. Dublin's marginal road capacity is not a rural dual carriageway; TII says the M50 cannot be widened. A blended rate hides the scarcity.

Effect. Tier bounds ×2 → Δ A −€5.0bn; ×0.5 → +€11.6bn. THE decider.

Status. A on bounds; rates derived from sourced components Evidence that would replace it. NTA ERM select-link runs; TII radial cordon capacities.

Link to this section

How many kilometres of new strategic road a car home needs.

The rate for each tier is derived rather than picked. A car-dependent home generates 0.67 vehicle departures in the morning peak hour at its site access (TRICS outputs in Irish traffic assessments). Forty per cent of those reach the strategic network of motorways and national roads (an assumption; 30 to 50 per cent in the bands). Each strategic trip needs 12 km of new capacity (8 to 16). The 1,100 vehicles an hour a lane (TII TD 9/07, level of service D) is a two-way design flow spread over four lanes, so a new four-lane route carries 4,400 an hour both ways; 60% of that runs in the peak direction, 2,640 vehicles an hour over two lanes, or 1,320 a lane, in line with TA 79/99 UAP3. Holding that level of service would take one new route-kilometre for about 820 car homes: about 550 km for Road Towns.

Since v5.4 (ledger #105) the options build 27.3% of that road. Seán set 150 km of new motorway for Road Towns, an outer ring from Naas to the M1 and two radials, and the rest of its traffic runs on the existing network. So a car home is charged 0.67 × 0.4 × 12 × 0.273 ÷ (1,100 × 4 × 0.60) = 0.00033 route-kilometres, one for about 3,000 car homes. The road not built shows up as congestion: less of the decongestion credit falls on new capacity, so only that share of its overlap deduction comes off. The share is Dublin's: Cork, Limerick and Galway build all the road their physics ask for (ledger #107).

The model prices the 150 km at its scarcity tiers, not at the map's mix of roads. Road Towns' built kilometres fall 33 km in the rural tier, 50 km orbital and 67 km urban-edge, with 16 interchange-equivalents: €9.0bn. The map's ring and two radials at the same rates would cost €5.8bn; the sensitivity road_map_mix prices every option that way. Car journey times are today's workbook rings and do not rise with the 150 km network. Its congestion is priced only through the €0.25 marginal external cost on the car-kilometres Rail Towns avoids. The longer drives of car homes beyond the urban edge are counted at sprawl speed (ledger #100); the slower speeds a congested network imposes on every Road Towns driver are not in the headline. An NTA regional model run would replace both.

The two road-physics assumptions are the second-largest lever after the tier bounds: a strategic share of 0.3 or 0.5 moves Δ A to −€8.5bn or −€7.3bn; 8 or 16 km per trip moves it to −€8.7bn or −€7.1bn. NTA select-link data on the M7, N4, M3 and M1 radials by home location would replace both.

Link to this section

The sourced parts each tier rate is built from.

Each tier's rate per equivalent = route-km per home × (€ per route-km + land) + junction allowance, all × (1 + optimism bias).

The road costs per route-kilometre come from out-turns and studies: rural dual carriageway €12.5m (the Department of Transport's July 2025 submission to the Oireachtas gives N4 €9.2m, N5 €10.6m, N22 €14.9m per km); the outer orbital €30m (the 2007 Leinster Orbital Route feasibility study, indexed); urban-edge insertion €36.5m (the M28 Cork to Ringaskiddy projection); urban tunnel €245m (the Dublin Port Tunnel's €752m for 4.5 km in 2006, indexed by 1.47). The M50's €1bn upgrade and TII's statement in December 2025 that it cannot be widened or double-decked anchor the top tier's existence.

Junctions are an allowance rather than named schemes: one €200m interchange (the Dunkettle out-turn) per 40,000 equivalents. Land is 7.8 hectares per route-kilometre at the CSO's residentially zoned land prices: €0.03m per hectare rural, €1.59m at the urban edge. Optimism bias is 35 per cent, between TII's reference-class 26 per cent and the UK's 46 per cent; the rail figures carry their own and the bias is never stacked twice.

Until v5.2 the top tier's rate, the urban tunnel's, was the model's least-sourced number. Since v5.3 Dublin prices no urban tunnel, so that rate no longer moves the answer (the Road Towns option has 1,416 equivalents in the top tier); the sprawl travel beyond the bound replaces it.

Link to this section

Where one class of road gives way to the next is an assumption, and it decides the state-transport answer.

The bounds say how many cumulative car-home equivalents Dublin's network can absorb at each class of road: 50,000 of existing headroom, then up to 150,000 with rural dual carriageways, 300,000 with the orbital and rebuilt M50 junctions, 500,000 with urban-edge insertion, and tunnel class beyond that. They carry status A because nothing published counts the region's radial capacity that way. Lane physics sit beside them as a sanity check: 100,000 car homes would need about 122 new route-kilometres at level of service D, of which the options build about 33.

This is the single input that decides the state-transport comparison. At the central bounds Δ A is −€7.9bn; doubling every bound gives −€7.6bn and halving them −€7.9bn. The Road Towns option's car-home equivalents (501,416) sit -1,416 below the tier-4 bound, so the sign of the state-transport gap turns on where that bound lies. The reviewer of September 2026 asked for the single-threshold break-even to be published and for the staircase to be replaced by identifiable bottlenecks and schemes; a continuous or scheme-based road-cost module is a Stage 2 switch of the rebuild.

The ledger has said this in every version, and the one-pager's footer carries it. What would replace the bounds is a corridor-by-corridor capacity count: TII's radial cordons and TD 9 lane capacities set against the NTA regional model's 2042 loads.

Link to this section

A rail-served home loads the strategic network at 0.51 of a car home.

A rail-served home is not car-free. TCRP Report 128 measured household vehicle trips in transit-oriented development at 44 per cent below suburban rates in the morning peak, so a rail home counts as 0.51 of a car home in the road tiers (0.40 to 0.60 in the bands). v3.3 had applied a further "internal capture" discount on top of this; the TCRP ratio is measured at the dwelling and already contains local trips, so v3.4 removed the double count, lifting the equivalence from 0.38 to 0.51.

With the demand engine on, the 0.51 is scaled by each leg's car-trip factor, so a Kildare town on the committed network counts as 0.75 of a car home and a tunnel-grade town as 0.51.

Sensitivity: at 0.40 Δ A is −€7.3bn; at 0.60 it is −€8.3bn. Irish surveys of trip generation at Adamstown and Clongriffin would replace the American ratio.

Modelling choices referenced here

M9 · Rail homes carry road cost (v3.4: TCRP ratio only)

Choice. A rail home loads the strategic network at 0.51 of a car home (TCRP 128).

Alternatives. v3.3's 0.38 (with a further internal-capture discount); v1's unsourced 0.21.

Why. The TCRP ratio is measured at the dwelling and already contains local trips; the strategic share applies to both typologies.

Effect. 0.40/0.60 → Δ A €0.3bn / €5.1bn (Option 1's extra equivalents fall in the tunnel-class tier).

Status. P Evidence that would replace it. Irish TOD trip surveys (Adamstown, Clongriffin).

Link to this section

Local roads at the LIHAF rate, charged to car suburbs and rail towns alike.

Distributor and link roads are charged to every home. A car-suburb home carries €12,300, the Comptroller and Auditor General's revised per-home figure for the Local Infrastructure Housing Activation Fund (its committed figure was €7,300 and its per-delivered-home figure €17,600). A rail-town home is scaled by street-metres per home, 2.6 against 8.5, because compact layouts need fewer metres of road per dwelling. Charging both typologies keeps the boundary symmetric; the first version charged car suburbs only.

At the committed LIHAF rate of €7,300 Δ A moves from −€7.9bn to −€8.9bn. A matched-layout cost comparison would replace the street-metre scaling.

Modelling choices referenced here

M12 · Distributor roads on both typologies (v3)

Choice. LIHAF-class €12.3k per car-suburb home; rail towns scaled by street-metres.

Alternatives. Charging distributors to car suburbs only (v1).

Why. Symmetric boundary.

Effect. Δ distributors €1.5bn.

Status. P (LIHAF rate) Evidence that would replace it. Matched-layout cost comparison.

Link to this section

1.32 spaces per car-suburb home, half that for a rail home, at Irish construction rates plus land.

A car-suburb home needs destination parking at work, shops, schools and leisure: cars per home (1.65) times spaces per car (0.45 at workplaces and 0.55 elsewhere, from the Dutch national parking inventory) times a shared-use factor of 0.8 for complementary peaks (Urban Land Institute) gives 1.32 spaces. A rail-served home needs 50% of that, 0.66, transferred from TCRP 128's finding on residential parking; the reviewer of September 2026 pointed out that the report does not measure destination parking, so the 50 per cent carries status A pending destination arrival and occupancy data. v3.3 had discounted rail homes twice, by cars per home and by the TOD ratio, giving 0.40 spaces; v3.4 discounts once.

Spaces are priced at Buildcost 2024 construction rates (€5,250 surface, €22,000 multi-storey, €60,000 underground) in a 60/30/10 mix, plus land at 27.5 m² for a surface space and 5.5 m² for a multi-storey one at Dublin's zoned-land price of €1.59m per hectare. The result is about €24.6k per car-suburb home and €12.3k per rail home. The regional cities scale the land component only.

Parking is a private cost, so it sits in subtotal B rather than A. Both models need an Irish inventory of destination spaces per home; none exists.

Modelling choices referenced here

M10 · Parking as a resource cost, discounted once (v3.4)

Choice. Destination spaces per car home from cars × per-car spaces × sharing (1.32); rail homes = car-home spaces × TCRP TOD ratio 0.5 (0.66). Buildcost rates plus RZLP land.

Alternatives. v3.3's two discounts (0.40 spaces); Codex's 1.3 / 0.8 at €20k.

Why. TCRP's per-dwelling finding already embodies car-shedding.

Effect. Δ parking €3.6bn → €2.0bn.

Status. P on rates and ratio; A on mix Evidence that would replace it. An Irish destination-parking inventory per home.

Link to this section

Equal connection charges, shorter pipes in a compact town, and treatment capacity that is per household.

Utilities are built up in three layers. Connection charges are the same for every home: Uisce Éireann's standard water and wastewater connection (about €5,100, from the CRU's 2018 policy, unchanged in 2023) and ESB Networks' housing-scheme charge (€1,050). Local buried networks (water, foul, storm, low-voltage electricity, telecoms) are priced by serviced street-metres, 8.5 for a car suburb and 2.6 for a rail town, at €1,100 a metre, with a 25 per cent premium for the denser working conditions of a rail town. The trunk allowance for growth in water and wastewater capacity is €6,700 a home, the Government's €2bn ring-fence divided by 300,000 homes, and rail towns pay 90 per cent of it, because treatment and source capacity is per household and only the pipes are shorter. v3.3 had given rail towns a 60 per cent saving; v3.4 corrected it.

The result is about €22.2k per car-suburb home and €15.8k per rail home. The honesty ledger records that this layer shrank when the evidence arrived: the first estimate of the local-network rate was corrected down from €1.5–2.5k a metre to €0.8–1.5k on Uisce's published excess-length charges. Uisce's capital-programme split and a quotable-rates schedule would replace the assumptions.

Modelling choices referenced here

M11 · Utilities: connections equal, local networks by street-metres, trunk mostly per household (v3.4)

Choice. Uisce/ESB charges per home on both; local buried networks by street-metres × €1.1k/m; trunk allowance €6.7k with a 10% conveyance saving for rail towns.

Alternatives. v3.3's 60% trunk saving; Codex's €9.5k/€5k local and 6% trunk.

Why. Treatment and source capacity is per household; only pipes are shorter in a compact town.

Effect. Δ utilities €2.1bn → €1.1bn.

Status. A Evidence that would replace it. Uisce capital-programme split; quotable-rates schedule.

Link to this section

Rail O&M per seat and road renewals, shown beside the capital subtotals.

The capital model shows operating costs beside its subtotals so that a reader can see the whole-life direction without waiting for the cost–benefit analysis. Rail operation and maintenance is €4,200 per usable seat-year, from the DART+ business case's incremental operating cost of €41.1m a year for 13,920 seats, indexed; it runs for the 15 years the new seats operate inside the window. Road renewals are 0.8 per cent of road capital a year over an average 12.5 years of liability inside the window.

In the central case, rail O&M is €1.2bn in the Road Towns option and €2.5bn in the Rail Towns option; road renewals are €1.5bn and €0.9bn.

Link to this section

Three subtotals, a quick present value, the incidence rows and land value capture as a transfer.

The Model sheet adds the lines into three subtotals, each wider than the last, so that no single "all infrastructure" figure can be quoted without saying what it contains.

A, state transport: rail capital plus strategic roads plus distributor roads.
B: A plus destination parking, a private cost.
C, modelled transport, parking and selected utilities: B plus connections, local networks and trunk capacity.

Central, Dublin	Road Towns	Rail Towns	Difference (Road Towns minus Rail Towns)
A, state transport	€23.5bn	€31.4bn	−€7.9bn
B	€36.4bn	€40.5bn	−€4.1bn
C	€48.4bn	€50.7bn	−€2.3bn
C per home built	€78,200	€81,900
Present value of A at 4 per cent	€16.3bn	€23.0bn	−€6.7bn
Land value capture receipts	€0.8bn	€4.8bn
A net of land value capture	€22.7bn	€26.6bn	−€3.9bn

C excludes schools, health, energy, treatment plants, flood defence and land assembly. On C, Road Towns €2.3bn cheaper; on state transport alone, Road Towns €7.9bn cheaper; the difference between the two, €5.6bn, is the density dividend in parking, utilities and distributors. Until v3.11 the rail-led city was also cheaper on state transport; the v3.12 demand reconciliation removed that saving, and the ledger's honest line is that the rail-led city's case now rests on parking, utilities and the CBA's benefits, not on avoided roads.

Present value here is a quick view at DPER's 4 per cent real rate, with rail on a package profile and everything else spread with the homes; the cost–benefit analysis is the appraisal. Land value capture is a transfer from landowners to the State: it changes who pays, so it appears only in the funding rows and never in A, B, C or the CBA's net present value.

Modelling choices referenced here

M15 · Subtotals A / B / C and incidence

Choice. A state transport; B + private parking; C + connections, local networks, trunk. Public / developer / private incidence rows.

Alternatives. One 'all infrastructure' figure.

Why. C is not all infrastructure (schools, health, energy, treatment, flood, land assembly excluded); saying so avoids overclaiming.

Effect. Δ A €2.9bn vs Δ C €6.0bn in v3.4.

Status. Convention Evidence that would replace it. —

M16 · Present value and whole-life views

Choice. PV at DPER 4% with rail on a package profile and everything else with the homes; whole-life A adds O&M and renewals inside the window.

Alternatives. The CBA's annual discounting (which supersedes this for appraisal).

Why. A quick view inside the capital model; the CBA is the appraisal.

Effect. PV Δ A +€0.7bn (v3.3 −€0.5bn).

Status. Convention Evidence that would replace it. —

M17 · Land value capture as a transfer (v3.3)

Choice. Receipts per town home enter only the funding rows (A net of LVC); A/B/C and the CBA NPV are unchanged.

Alternatives. Counting LVC as a benefit (double counts the location value already in the land-use line).

Why. A transfer from landowners to the state is not a resource saving.

Effect. Δ A net of LVC €5.6bn vs Δ A €2.9bn.

Status. A (the €20k is provisional, ledger #35b) Evidence that would replace it. A named capture mechanism and its own costs.

A quick 4 per cent view; the CBA is the appraisal.

The capital model discounts its subtotals at 4 per cent real, the Department of Public Expenditure's test rate from May 2026. Rail packages are spread uniformly over 11 years from 2028; roads, parking and utilities are spread uniformly with the homes from 2026 to 2050. A whole-life view adds the operating costs and renewals inside the window.

This is deliberately a quick view. The cost–benefit analysis discounts year by year with residual values, phases the rail package on an S-curve and the homes from 2035, and is the appraisal figure. In this version the capital model's present value of A is €16.3bn against €23.0bn, €6.7bn in Road Towns' favour; in v3.3 the same difference was −€0.5bn and in v3.9 €3.3bn the other way, which is one measure of what the demand engine, the later corrections and the demand reconciliation did.

Modelling choices referenced here

M16 · Present value and whole-life views

Choice. PV at DPER 4% with rail on a package profile and everything else with the homes; whole-life A adds O&M and renewals inside the window.

Alternatives. The CBA's annual discounting (which supersedes this for appraisal).

Why. A quick view inside the capital model; the CBA is the appraisal.

Effect. PV Δ A +€0.7bn (v3.3 −€0.5bn).

Status. Convention Evidence that would replace it. —

Link to this section

€20,000 a home from Seán's land-readjustment model, provisional, and never a benefit.

Seán's land-readjustment model suggested that a pooling scheme in the new rail-town areas could capture about €20,000 per home for the State. The workbooks treat that as a transfer from landowners to the State: never a resource saving, never a benefit. The location value of a rail home is already counted once, in the CBA's land-use line, so counting the capture as well would double count. The economic results (A, B, C and every net present value) are therefore unchanged by it; only the funding rows move.

A scope switch decides which homes are in: by default, greenfield rail towns in both options, because the instrument is available to whichever programme designates towns. In the central case that is 38,982 town homes in the Road Towns option (€0.8bn) and 238,809 in the Rail Towns option (€4.8bn), so A net of capture is €22.7bn against €26.6bn, a gap of −€3.9bn, and the present-value gap −€4.1bn.

The figure is provisional. The 4 September fisk of the readjustment model (ledger #35b) found that its base case yields no cash surplus for rail, that €20,000 a home would need a residual share above the scheme's own ceiling, and that the nearest real mechanism is the Land Value Sharing levy at the Bill's 25 per cent, which would pay the council, not the railway. Until Seán names the mechanism, every figure that rests on the €20,000 is quoted as provisional.

Modelling choices referenced here

M17 · Land value capture as a transfer (v3.3)

Choice. Receipts per town home enter only the funding rows (A net of LVC); A/B/C and the CBA NPV are unchanged.

Alternatives. Counting LVC as a benefit (double counts the location value already in the land-use line).

Why. A transfer from landowners to the state is not a resource saving.

Effect. Δ A net of LVC €5.6bn vs Δ A €2.9bn.

Status. A (the €20k is provisional, ledger #35b) Evidence that would replace it. A named capture mechanism and its own costs.

Link to this section

Dublin in full, the regions in brief, and the Low scenario beside them.

Dublin, Meath and Kildare, central	Road Towns	Rail Towns
Homes to build	619,543	619,543
Homes to serve (loading the network)	548,254	548,254
Rail-served by 2050	126,158	389,823
Capacity once the works operate	141,525	462,654
Road-served by 2050	422,096	158,431
Rail capital	€8.7bn	€22.2bn
Strategic roads	€9.0bn	€5.9bn
A, state transport	€23.5bn	€31.4bn
C, modelled total	€48.4bn	€50.7bn
C per home built	€78,200	€81,900

Region, central	Homes	Rail-served, Road Towns / Rail Towns	Δ A	Δ C
Cork	151,456	10,047 / 52,960	+€0.1bn	+€0.9bn
Limerick	48,696	3,723 / 25,053	−€0.0bn	+€0.4bn
Galway	71,462	4,767 / 26,751	+€0.1bn	+€0.5bn

In the Low scenario Dublin has 406,904 homes to serve, rail-serves 64,593 against 141,572, and on state transport Road Towns €8.8bn cheaper (A €13.4bn against €22.2bn). Every version to v3.11 reversed in the Low case; in this version the central case is itself negative on state transport, so the Low case deepens it, and the one-pager's footer carries the line.

The version-history section reports the gap lineage across versions, so that a reader can see how much of the current gap arrived with each modelling change.

Link to this section

What the rail-led city is worth: Rail Towns minus Road Towns under DPER conventions.

The capital model says what each city costs. The cost–benefit analysis (model/modules/cba.ts) says what the rail-led city is worth, as the Rail Towns option minus the Road Towns option, in present value at the Department of Public Expenditure's 4 per cent real rate over 2026 to 2068. It reads the capital model's central quantities (the shifted homes, the rail capital, the roads, parking and utilities avoided, the extra seats) and adds the benefits to the households that are rail-served in one future and road-served in the other.

Central, Dublin, present value
Rail capital, fixed and scaling, with renewals and residuals	€10.05bn
Roads, parking and utilities avoided, with renewals and residuals	−€7.40bn
Extra housing construction cost, operating costs, programme body, public-funds premium	€7.87bn
Present value of costs	€10.52bn
Time savings, decongestion, vehicle operating costs, safety, carbon, air, health	€19.63bn
Higher-value land use, net of overlap	€0.00bn
Agglomeration	€2.20bn
Net present value: transport only / with land use / with agglomeration	€9.1bn / €9.1bn / €11.3bn

Since v5.3 the headline is rung 1, the transport-only net present value, €9.1bn; land use follows as its own step. "Transport only" names the benefit side: rung 1 sets the transport benefits against all resource costs, with the roads, parking and utilities avoided and the denser-building cost on the cost side. The present value of costs is €10.5bn: the extra rail spending exceeds what it avoids, so a conventional benefit-cost ratio can be formed on it. The economic ratios are 1.87 at rung 1, 1.87 with land use and 2.08 with agglomeration. Readers also expect a ratio per euro of rail investment, so the results report funding-view ratios on the gross rail investment (€11.5bn): 2.1 on transport benefits, 2.4 with land use and all the avoided infrastructure counted, labelled as such wherever quoted. (The v3.9 page reported costs of −€3.9bn, "dominant", and a funding-view "BCR 2.7"; the v3.12 demand reconciliation turned the road saving from €13.6bn of present value into €3.9bn and the cost side positive.)

The State's net cash position is €8.2bn worse under the Rail Towns option in present value, €5.7bn worse with land value capture. The Low and High parameter bands give −€2.3bn and +€17.7bn at rung 1; 2 and 6 per cent discount rates give €21.8bn and €2.6bn.

The method follows the NSW and Australian Transport Assessment guidance for land-use-aware appraisal: benefits are reported on a ladder of three rungs, avoided sprawl infrastructure is a cost saving and never a benefit, and land value uplift is never added to user benefits or agglomeration.

Modelling choices referenced here

C1 · Incremental, two futures, same homes

Choice. Every line is Option 2 minus Option 1 for the same regional need; the 'additional rail homes' are homes served by rail in one future and roads in the other.

Alternatives. A project CBA of the tunnel on a fixed land-use map (Jacobs 2021, BCR 0.71); an open-region reading with extra homes.

Why. TAF's fixed-map method cannot see what the homes need or are worth; the open-region reading is unsupported.

Effect. Frame for all results.

Status. Frame Evidence that would replace it. —

C3 · NSW / ATAP O8 reporting ladder

Choice. Rung 1 transport; rung 2 + higher-value land use net of overlap; rung 3 + agglomeration. Avoided sprawl infrastructure is a cost saving, never a benefit.

Alternatives. One headline with everything in.

Why. Shows what each layer of the case rests on and prevents double counting.

Effect. v4.3 central NPV1 / NPV2 / NPV3: 2v1 €5.37bn / €6.58bn / €7.64bn; 3v1 €3.29bn / €3.62bn / €3.95bn; 2v3 €1.59bn / €2.44bn / €3.19bn.

Status. Method Evidence that would replace it. —

C12 · Funding-view ratios beside the economic BCR

Choice. v4.4 (fisk F19): each pair leads with NPV2 and BCR2 = PVB2 ÷ PVC, which reconciles to NPV2; where PVC ≤ 0 (3v1) the public-accounts ratio BCR2_pa is quoted. The funding ratios on the gross rail investment (BCRg, savings credited as benefits) and on the public investment net of LVC (BCRn) are kept as labelled secondary rows. They were the lead ratios in v3.4, when PVC ≈ 0.

Alternatives. Leading with BCRg (the v3.4–v4.3 practice); a TAF-convention ratio only.

Why. BCRg does not reconcile to NPV: k_house and the SPPF sit on neither side. A DPER reader reconciles any ratio to the NPV first.

Effect. 2v1 central v4.3: BCR2 2.38, BCR2_pa 2.08; funding ratios BCRg2 2.17 / BCRn2 2.60 (2.40 with bcrn_net_lvc = 1).

Status. Convention Evidence that would replace it. —

The appraisal frame: horizon, discounting, residuals and price bases.

Every line is the Rail Towns option minus the Road Towns option for the same regional need; the "additional rail homes" are homes served by rail in one future and by roads in the other. The frame is the Department of Public Expenditure's: 4 per cent real, constant 2026 prices, a horizon of construction plus 30 years from the 2038 opening (2026 to 2068), and straight-line residual values at 2068 by asset class, with the tunnel civils on a 100-year life, rolling stock 35 and other works 40. Parameters quoted in 2011, 2016 or 2018 prices are indexed with the CPI.

The tunnel opens in 2038 in the central case (2036 to 2040), from a Railway Order around 2029 and a seven-year build. The first additional town homes complete three years before that, because the outer works and the Naas four-tracking open ahead of the tunnel. The two are roughly equivalent; the discounting alone is worth about €0.1bn. A 2078 horizon adds €1.4bn; a 2060 delivery window with a 2078 horizon was tested once on v3.4 and raised the transport-only net present value by about a third.

Modelling choices referenced here

C1 · Incremental, two futures, same homes

Choice. Every line is Option 2 minus Option 1 for the same regional need; the 'additional rail homes' are homes served by rail in one future and roads in the other.

Alternatives. A project CBA of the tunnel on a fixed land-use map (Jacobs 2021, BCR 0.71); an open-region reading with extra homes.

Why. TAF's fixed-map method cannot see what the homes need or are worth; the open-region reading is unsupported.

Effect. Frame for all results.

Status. Frame Evidence that would replace it. —

C2 · DPER frame: 4% real, 2026 prices, 2026–2068, residuals

Choice. Construction plus 30 years from the 2038 opening; straight-line residuals at 2068 by asset class.

Alternatives. Codex: 2087 with full replacement at life and no residual; NSW 5%/3%/7% conventions.

Why. Irish guidance.

Effect. v4.3 central 2v1: 2% / 6% → NPV1 €11.5bn / €2.2bn (from €5.37bn); a 2078 horizon raises NPV1 to €6.68bn (+€1.31bn). The v3.4 ad hoc test of a 2060 delivery window with 2078 (NPV1 €17.1bn) has not been rerun.

Status. P Evidence that would replace it. —

Link to this section

Three rungs of benefit, avoided infrastructure on the cost side, and the funding-view ratios.

Benefits are reported on a ladder so that a reader can see what each layer of the case rests on. Rung 1 is transport: time savings, decongestion, vehicle operating costs, safety, carbon, air quality and noise, and health. Rung 2 adds higher-value land use, net of the benefits already counted. Rung 3 adds agglomeration. Avoided sprawl infrastructure (roads, parking, utilities) is a cost saving on the cost side of the ledger, never a benefit; that is the NSW and ATAP O8 convention and it prevents double counting.

The bridge from rung to rung, at central (ledger #107, Codex v5.4 review, finding 14):

Rung bridge, central	Present value
Transport benefits, PVB1	€19.63bn
Transport costs, PVC_tr: rail capital, roads avoided, operating costs, delivery body, renewals, residuals, public-funds premium	€10.20bn
Parking avoided, utilities avoided, denser-building cost	−€2.32bn, −€1.09bn, €3.73bn
Present value of costs, PVC, all of the above	€10.52bn
NPV1 = PVB1 − PVC	€9.11bn
Higher-value land use, net of overlap	€0.00bn
NPV2	€9.11bn
Agglomeration	€2.20bn
NPV3	€11.31bn

The present value of costs is €10.5bn: the extra rail spending exceeds what it avoids, so a conventional benefit-cost ratio can be formed on it. The economic ratios (benefits over incremental cost) are 1.87, 1.87 and 2.08 for the three rungs, and they are meaningful only while the cost side is clearly positive; the results say "dominant" whenever it is not. Two funding conventions are reported beside them. Ratios on the gross rail investment (present value €11.5bn), with the savings credited as benefits: 2.05 on transport benefits alone, 2.35 with all savings, 2.35 at rung 2 with all savings, 2.54 at rung 3. Ratios on the public investment net of land value capture (€9.0bn): 2.34, 2.72 and 2.96.

The rebuilt one-pager's key-numbers grid leads with the rung-2 gross ratio, 2.4, labelled as benefits plus avoided roads, parking and utilities per euro of gross rail investment, and says beside it what the economic ratio is.

Modelling choices referenced here

C3 · NSW / ATAP O8 reporting ladder

Choice. Rung 1 transport; rung 2 + higher-value land use net of overlap; rung 3 + agglomeration. Avoided sprawl infrastructure is a cost saving, never a benefit.

Alternatives. One headline with everything in.

Why. Shows what each layer of the case rests on and prevents double counting.

Effect. v4.3 central NPV1 / NPV2 / NPV3: 2v1 €5.37bn / €6.58bn / €7.64bn; 3v1 €3.29bn / €3.62bn / €3.95bn; 2v3 €1.59bn / €2.44bn / €3.19bn.

Status. Method Evidence that would replace it. —

C12 · Funding-view ratios beside the economic BCR

Alternatives. Leading with BCRg (the v3.4–v4.3 practice); a TAF-convention ratio only.

Why. BCRg does not reconcile to NPV: k_house and the SPPF sit on neither side. A DPER reader reconciles any ratio to the NPV first.

Effect. 2v1 central v4.3: BCR2 2.38, BCR2_pa 2.08; funding ratios BCRg2 2.17 / BCRn2 2.60 (2.40 with bcrn_net_lvc = 1).

Status. Convention Evidence that would replace it. —

Link to this section

263,664 households, 2.4 trips a day, 200 days, three-quarters switching, 15 minutes each.

The households in the increment are the 263,664 homes that are rail-served in the Rail Towns option and road-served in the Road Towns option (389,823 less 126,158). Each makes 2.4 commuting trips per weekday (about 1.2 workers per household in the commuter belt, both directions) on 200 days a year; the days fell from 230 in v3.3 on the CSO's finding that 35 per cent of workers usually or sometimes work from home. Three-quarters of those trips switch from car to rail and realise a time saving (since v5 the count is capped at the rail trip rate the seats are sized on, 245 a year per shifted home in this appraisal: Inputs time_trips 2); the rest were already on a bus or would not have driven, a haircut adopted in v3.4. The saving is 15 minutes a trip, from the detailed workbook's access rings (rail 36 to 42 minutes door to door against 45 to 68 by congested road), valued at €18 an hour from the 2016 Common Appraisal Framework's commuting value indexed by CPI, with no growth applied. Those road times are today's rings. They do not rise with Road Towns' 150 km network (ledger #107, Codex v5.4 review, finding 7): that network's congestion enters only through the marginal external cost on the car-kilometres Rail Towns avoids, in the next section. The slower speeds it imposes on every Road Towns driver are not in the headline. An NTA regional model run would replace both.

Each shifted home also avoids about 11,000 car-kilometres a year (10,800 since v5, scaled by each option's demand engine): a car-dependent household with 1.65 cars driving 12,000 km each against a rail-served one with 1.0 car driving 9,000, consistent with TCRP's 44 per cent reduction in vehicle trips. This is a household-level difference that captures second-car shedding and non-work travel, which a commute-only figure would miss. Rail trips per home are 266 a year for the fare line; the reviewer of September 2026 noted that this and the 360 annual car-to-rail commuting switches (2.4 × 200 × 0.75) describe different trip populations. Since v5 the time savings use the seat-sizing rate (245 a year) while the fare line still uses 266, so two rates remain; a single trip-generation module feeding capacity, benefits, fares and carbon is still a Stage 2 switch of the rebuild.

Rung 1 line, central	Present value
Time savings	€5.1bn
Decongestion	€6.8bn
Vehicle operating costs	€3.5bn

With the surface case, the demand engine overrides the minutes, switch share and kilometres for the tunnel-less network (2.3 minutes, 48%, 8,139 km), which is why the tunnel's own contribution is positive.

Modelling choices referenced here

C6 · Travel behaviour of shifted households

Choice. 2.4 commuting trips per home-day × 200 days × 75% switch × 15 min × €18/h; 11,000 car-km a year avoided at household level.

Alternatives. Codex: 1.2 commuters × 180 days × 65% × 10 min × €15.3/h; 5,900 car-km (commute only).

Why. Each parameter sourced or banded; the car-km figure is a household-level empirical difference.

Effect. Time €2.4bn, decongestion €2.5bn, VOC €2.0bn PV.

Status. A/D Evidence that would replace it. An NTA ERM run for the modelled homes; CSO NTS household car-km.

Link to this section

Per-kilometre externalities transferred from UK and Irish appraisal values.

The avoided car-kilometres carry five per-kilometre values. The marginal external cost of congestion is €0.25 a kilometre since v5.4 (ledger #106): with 150 km of new road, Road Towns' traffic runs on a congested network. Today's network gives €0.15, weighted from the UK's TAG A5.4 schedule for a Dublin commuter-belt trip that mixes rural motorway, urban A-roads and the M50; no Irish marginal-cost schedule exists, and the Department of Transport's aggregate figure for aggravated congestion in the Greater Dublin Area supports the order of magnitude. Vehicle operating costs saved are €0.12 a kilometre from the TII appraisal guidance's fuel and non-fuel rates indexed and adjusted for electric vehicles. Collision costs avoided are €0.028 a kilometre from the RSA and CSO casualty rates at the guidance's casualty values. Air quality and noise are €0.007 a kilometre and fall as the fleet electrifies. Health is €220 per shifted home a year from the World Health Organisation's HEAT method as used in the NTA's health module, discounted for those who already walk.

The marginal external cost is what a car-kilometre costs other road users, their time and vehicle costs (TAG A5.4). It does not overlap the switchers' own time savings or vehicle operating costs: those lines count different people. One overlap rule applies (ledger #107, Codex v5.4 review, finding 8). If Road Towns built all the road the trip physics ask for, 27 per cent of the decongestion credit would fall on the new links (cong_overlap, status A); times the 27.3% built, 7.4 per cent comes off, leaving about €0.23 a kilometre. The split is an assumption pending an NTA regional model select-link run; the Dnes review of 5 October gives TAG A5.4's bands for an unexpanded network, not the split. The 150 km network's own congestion enters the appraisal here and nowhere else: the car journey times are today's rings, and Road Towns' drivers' slower trips are not priced.

Line, central	Present value
Safety	€0.82bn
Air quality and noise	€0.20bn
Health	€0.58bn

Congestion is the lever: at €0.15, the v5.3 value, the transport-only net present value is €6.4bn; at €0.35 it is €11.8bn. An NTA regional model marginal-cost run would replace the transfer.

Modelling choices referenced here

C7 · Marginal congestion cost transferred from UK TAG

Choice. €0.25/km since v5.4 (ledger #106: Road Towns' traffic runs on a congested network with 150 km of new road); €0.15/km weighted for a Dublin commuter-belt trip on today's network was the v5.3 value.

Alternatives. Codex €0.08; an Irish marginal-cost schedule (none exists).

Why. Best available; the DoT aggregate congestion cost supports the order of magnitude.

Effect. v4.3 central 2v1: €0.08 / €0.25 → NPV1 €4.00bn / €7.34bn; NPV2 €5.20bn / €8.54bn (v3.4: €7.9bn / €10.6bn).

Status. A Evidence that would replace it. NTA ERM marginal-cost run.

Values, Low / Central / High0.15 / 0.25 / 0.35 €/km

WhereCBA inputs, key mec_cong

StatusC — convention (a reasonable modelling standard, not a measurement)

Source and basis

Why this value

What it drives

Decongestion (v4.3 central 2v1: €2.95bn PV).

How it was tested

v4.3 central 2v1: €0.08 / €0.25 → NPV1 €4.00bn / €7.34bn; NPV2 €5.20bn / €8.54bn.

What would replace it

An NTA ERM marginal-cost run on the radials.

History

Link to this section

DPER's shadow price on a physical carbon account: construction and operation of roads and rail in both options (v3.13).

Carbon is valued at the Department of Public Expenditure's shadow price schedule from its March 2024 technical guidance: €349 a tonne in 2026 rising to €890 in 2050 and €1,083 by 2055, in 2018 prices and indexed, with no split between traded and non-traded sectors and values held flat after 2055. Since v3.13 (11 September 2026) the two monetised lines take their tonnes from a physical carbon account of both options over the appraisal period: construction carbon for rail by ladder category (tonnes per €m: tunnel 70, four-tracking 120, electrification, fleet, stations, systems) and for roads by tier route-kilometre (rural dual carriageway 4,000 t/km, tunnel class 60,000 t/km) plus junctions and distributors per €m; operating carbon from car-kilometres at a tailpipe factor that falls from 0.13 kg a kilometre in 2026 to 0.03 in 2050, and from rail traction at 4,500 kWh per offered seat-year on a grid factor that falls to 0.08 kg/kWh by 2035. Over the 30-year reporting window the Road Towns option emits 13.81 Mt and the Rail Towns option 12.26 Mt, a difference of -1.55 Mt; the cumulative difference turns in the Rail Towns option's favour in 2043, the carbon payback year, and comes to -5.9 tonnes per additional rail-served home.

Line, central	Present value
Operational carbon	€1.13bn
Embodied carbon, net	−€0.00bn

Every factor is an assumption pending the MetroLink and DART+ environmental reports and TII's carbon tool; the v3.4 lump convention (450,000 tonnes for the rail increment, 150 tonnes per €m of road) is kept as a switch.

Modelling choices referenced here

C13 · Carbon: shadow price

Choice. DPER shadow price (€349/t 2026 → €890/t 2050, €2018, indexed to 2026; flat after 2055). From v3.13 the two monetised lines take the physical increment from the Carbon sheet (C17).

Alternatives. ETS split; the v3.4 lump lines (carbon_link = 0).

Why. DPER's single schedule; one physical account for tonnes and euros.

Effect. Carbon lines PV: see the Results carbon block and the build log.

Status. P Evidence that would replace it. —

Link to this section

A matched rail-access premium of €26,180 a home, less the €45,197 of transport benefits already counted, leaves €0.

A home near a station sells for more than the same home further away. Since v5.3 the model values that gain on matched evidence (land_method 1): buyers in Dublin pay about 7% more for a home within 500 metres of a train station, other things equal (Mayor, Lyons, Duffy and Tol, 2012; the band runs from 4.2 to 10 per cent). On a €440,000 home, less a 15 per cent normal developer return, that is €26,180 a home.

The NSW guidance (TPG23-08) and ATAP O8 say that ordinary land-value uplift is the capitalisation of the user benefits already counted in rung 1 and is not additional, and Mayor et al. attribute the premium to shorter commutes. So the model deducts the transport benefits each resident already gets, net of the fares they pay (time, vehicle operating costs and health per discounted home, less fares: €45,197; the longer drives of sprawl are left out, because no station premium prices them) and counts only the remainder: €0 a home, €0.00bn in present value. At central the premium is below Rail Towns' residents' net benefits, so against Road Towns land use adds nothing; Bundle 6's residents count fewer transport benefits, so Bundle 6 keeps €0.00bn of land value against Road Towns. That is the outcome of the deduction rule, not proof that the premium is wholly transport benefit: the evidence is for existing Dublin homes sold in 2001 to 2006, not for new homes in the outer counties.

Until v5.2 the margin was a €65,000 gap between county median prices, which mix dwelling type, size, schools and distance to Dublin; on that method rung 2 is €10.8bn. At a 10 per cent premium it is €9.1bn; with half the overlap deducted, €9.7bn. A dependent-development method, tried on 22 October, was withdrawn: the model houses the same households in every option, so the land a change of use creates arises in Road Towns too.

Modelling choices referenced here

C4 · Higher-value land use = gross margin less capitalised benefits (v3.4)

Choice. Gross location margin (€65k, net of normal developer return) less the resident-accruing benefits already counted (time + VOC + health per discounted home, €53.6k), floored at zero → €11.4k.

Alternatives. v3.3's €55k price margin counted in full; Codex's full residual-valuation build-up (sale value − production cost − return, less a picked €50k overlap and €10k displacement).

Why. NSW TPG23-08 / ATAP O8: ordinary uplift is capitalised user benefit and not additional. Our own resident benefits are computed, so the deduction is derived, not picked.

Effect. Rung 2 line €6.0bn → €1.0bn. At 100% overlap rung 2 is insensitive to time-saving assumptions.

Status. A (margin) / D (overlap) Evidence that would replace it. Matched residual valuations by location.

Values, Low / Central / High1 / 1 / 1 switch

WhereCBA inputs, key land_method

StatusC — convention (a reasonable modelling standard, not a measurement)

Source and basis

Why this value

What it drives

b_hvlu, so PVB2, NPV2, BCR2 and the public-accounts and funding ratios; NPV1 and every transport line are untouched. Each pair now leads with NPV1 (CBA_HEADLINE_ROWS).

How it was tested

What would replace it

Link to this section

Denser built form costs more to build: €20,000 a home, charged once, sign uncertain.

A denser built form at matched floor area and quality is a real resource cost, and from v3.4 it is charged once in the core account at €20,000 per shifted home (band €40,000 to zero). The reasoning: the Society of Chartered Surveyors' cost studies put a mid-rise apartment's hard cost €80,000 to €100,000 above a three-bed semi's, while terraces and duplexes cost about the same; the increment is roughly 15 per cent apartment-led station infill and 85 per cent rail-town perimeter blocks of houses, duplexes and low-rise at about 48 homes a hectare gross, which weights to about €20,000.

The line costs €3.7bn in present value. At zero the transport-only net present value is €9.1bn; at €40,000 it is €9.1bn. The sign is genuinely uncertain, and a quantity surveyor's comparison of matched dwellings by built form is owed.

Modelling choices referenced here

C5 · Extra housing construction cost in core (v3.4)

Choice. €20k per shifted home (band €40k–€0) charged once at completion, in the economic views but not public cash.

Alternatives. Netting it inside the land-use margin (v3.3); Codex's €30k.

Why. A real resource cost of the denser built form; the NSW bridge logic charges it once.

Effect. −€1.75bn PV.

Status. A Evidence that would replace it. QS comparison of matched dwellings.

Link to this section

Thirteen per cent of transport user benefits, on the 2015 DART business case's precedent, never headlined.

Wider economic benefits are carried as rung 3, as a fixed share of the transport user benefits: 13 per cent central (10 to 20), from the 2015 DART Expansion Revised Business Case, which found agglomeration of 12.5 per cent on DART Underground and 14 per cent on the whole programme by the UK effective-density method. The 2010 business case's 67 per cent is retired and never used. The line is €2.20bn in present value, about 10% of rung-3 benefits; nothing headlined depends on it, and the one-pager frames it as incomes (about €300 a year per new household).

Three rules travel with it, recorded in the agglomeration note of 10 September 2026 for anyone building a second scheme that must add up with this one: no land value uplift on top of agglomeration; static or dynamic clustering, never both; and agglomeration follows a real accessibility or density change, so homes on their own do not earn it under this method. The matched premium of rung 2 is net of the residents' own user benefits only; no agglomeration overlap is deducted from it (ledger #107, Codex v5.4 review, finding 15). At central that exposure is zero against Road Towns (€0.00bn) and at most €0.00bn on Bundle 6 against Road Towns and Rail Towns against Bundle 6; it is larger in the High band, where half the overlap is deducted.

The older detailed workbook holds an indicative effective-density proxy (elasticity 0.046, a 45-minute contour of one million residents, €105,000 of GVA per worker) that is labelled coarse and not in the CBA ladder. The honest upgrade would be a distance-decayed, sector-weighted effective density using the 2015 business case's four elasticities and CSO employment by sector, which would also let the model replace the 13 per cent precedent with a calculation of its own.

Link to this section

Rail capital fixed and scaling, the avoided infrastructure, operating costs, renewals and the delivery body.

The cost side reads the capital model's central quantities. Fixed rail capital is the Rail Towns option's fixed steps less the Road Towns option's, €12.2bn undiscounted, of which 54% is 100-year civil engineering and 18% rolling stock; it is spent on an S-curve over the nine years before opening. Scaling rail capital (extensions, their fleet, stations, access) is €1.4bn and follows the completions. Roads avoided are the tiers plus distributors, €5.6bn undiscounted; parking avoided €3.8bn; utilities avoided €1.8bn.

Operating costs are €4,200 per offered seat-year on 21,724 additional seats from the DART+ business case's incremental figure. Renewals run at 1.2 per cent of rail capital a year from opening and 0.8 per cent of road capital from construction. A programme delivery body for designation, land pooling and the planning of twenty towns costs €20m a year from 2027 to 2050, added in v3.4. Optimism bias on rail is zero in the central case because the tunnel figure is Jacobs' out-turn estimate; the High band applies TII's 37 per cent, which takes the transport-only net present value to €6.2bn.

Cost line, central	Present value
Fixed rail capital	€9.18bn
Scaling rail capital	€0.80bn
Roads avoided	−€3.90bn
Parking avoided	−€2.32bn
Utilities avoided	−€1.09bn
Extra housing cost	€3.73bn
Rail operating costs	€1.12bn
Delivery body	€0.34bn
Renewals, rail and road	€1.30bn and −€0.54bn
Residuals, net	−€0.78bn
Public-funds premium	€2.69bn
Present value of costs	€10.52bn

Link to this section

1.3 on net public spending, so nearly a wash where the avoided roads are public too.

The Public Spending Code and TII's appraisal guidance apply a premium of 30 per cent to Exchequer expenditure. From v3.4 the model shows it explicitly, applied to net public spending: rail, strategic roads, trunk utilities, operating costs, renewals and the programme body, whether spent or avoided, with residuals excluded. Because the roads the rail-led city avoids are public too, the premium nearly nets out; in the central case it is a charge of €2.7bn in present value. Setting it to 1.0 gives €11.8bn at rung 1.

16 factor-cost-to-market-price conversion. That is a standard-form convention, but it penalises whichever option's savings happen to be private. The model shows the premium and not the 1.16 uplift, which would favour the Rail Towns option here, and flags the choice.

Modelling choices referenced here

C8 · Shadow price of public funds on net public spending (v3.4)

Choice. 1.3 on net public spending (spending and avoided spending), residuals excluded.

Alternatives. Not applying it (v3.3); Codex's gross-spend convention with private savings excluded.

Why. DPER/TII convention, shown explicitly. Symmetric application means it is small here and a credit where net public spending is negative.

Effect. v4.3 central: +€1.55bn PV cost for 2v1, +€0.07bn for 3v1, +€1.64bn for 2v3; +€0.31bn in the surface case (tunnel steps off). It was +€0.29bn in v3.4.

Status. P Evidence that would replace it. Appraisal-authority agreement on the base.

Link to this section

Transfers change who pays, not the resource cost, so they live in the fiscal rows only.

Fares (€2.20 a rail trip in the 90-minute fare class) and land value capture (€20,000 per home in the new rail-town areas, from Seán's land-readjustment model, provisional) are transfers. They appear in the fiscal and funding rows and never in the economic net present value; the build script asserts that every rung is identical with capture at zero and at €20,000.

In the central case the fare line is −€1.55bn to the State (negative is a receipt), land value capture −€2.46bn, and the State's net cash position is €8.2bn worse under the Rail Towns option before capture and €5.7bn worse with it. The scope switch mirrors the capital model's: greenfield rail towns in both options by default, so only the 199,827 additional town homes count; the other settings credit all of the Rail Towns option's towns or all shifted homes.

The 4 September fisk of the readjustment model found that the €20,000 is not an output of it, so every figure that rests on the capture is quoted as provisional until Seán names the mechanism.

Modelling choices referenced here

C11 · Fares and LVC as transfers

Choice. Fares and land value capture appear only in the fiscal and funding rows; the economic NPV is asserted identical with LVC at €0 and €20k.

Alternatives. Counting LVC as a benefit (double counts the land-use line).

Why. Transfers change who pays, not the resource cost.

Effect. Net public cash −€0.4bn before LVC, −€1.8bn with it.

Status. Method Evidence that would replace it. A named capture mechanism (ledger #35b).

Link to this section

Bands, one-factor tests and switching values: the handful of inputs that move the result.

Every input carries Low, Central and High values, and the build evaluates all three scenarios, 293 one-factor tests on the capital model and 114 on the cost–benefit analysis, and switching values by input, in under a second. The bands are the model's cost-and-demand bands, not confidence intervals. Δ A below is Road Towns minus Rail Towns on state transport: positive favours the rail-led option.

Seven things decide the answer, in order.

Road tier bounds. Central Δ A is −€7.9bn; doubling the bounds gives −€7.6bn, halving them −€7.9bn. Status A, and the Road Towns option sits -1,416 equivalents below the tier-4 bound, so this one input sets the sign.
The demand basis. The v3.12 reconciliation of the gross requirement to net transport demand: on the v3.11 basis, with every home loading the network, Δ A is −€8.0bn; with no in-situ replacement −€7.9bn; with all replacement in situ −€7.8bn.
Road physics. The strategic share of trips (0.3 or 0.5) moves Δ A to −€8.5bn or −€7.3bn; the kilometres of new capacity per trip (8 or 16) to −€8.7bn or −€7.1bn.
The Low scenario: Dublin Δ A −€8.8bn, Δ C −€7.7bn. It must be said whenever the central case is quoted.
The demand engine. Off, Δ A is −€8.2bn; at its Low band −€8.1bn; at its High band −€7.7bn.
The density exponent moves Δ A between −€7.9bn (0.25) and −€7.9bn (1.5).
Terminal capacity without through-running (v3.10): with the city end absorbing none of FourNorth's trains Δ A is −€7.8bn; at 4 an hour −€8.4bn; at 8 −€8.5bn.
The appraisal window. On a 2050 window the outer works carry much of the case; on a 2060 window (tested once, on v3.4) the tunnel does. The horizon is a decision, not a detail.

In the cost–benefit analysis the switching values are now computed by changing an input and re-evaluating everything it touches (the v3.x table scaled one output line at a time and mislabelled the rung; the reviewer of September 2026 found both errors). The tunnel package's construction cost would have to be ×2.00 (16,493) to take rung 1 to zero and ×2.00 (16,493) for rung 2; a 40 per cent smaller road saving leaves rung 1 at €7.2bn.

Modelling choices referenced here

M18 · Three scenarios and one-factor sensitivities

Choice. Low / Central / High bands on every input (Inputs!C2) plus 47 one-factor tests written to outputs/results.json.

Alternatives. Monte Carlo; a single central case.

Why. Shows which assumptions decide the answer; the Low scenario reverses the ranking in every region.

Effect. See the Sources sheet status column and the explainer.

Status. Method Evidence that would replace it. —

C14 · Sensitivities, bands and switching values

Choice. Low/Central/High parameter bands, 2%/6%, 29 one-factor cases incl. 'v3.3 conventions', switching values on every line.

Alternatives. Monte Carlo.

Why. Shows which assumptions decide the answer.

Effect. No single line takes rung 1 negative; the roads-bounds ×2 analogue leaves €4.0bn.

Status. Method Evidence that would replace it. —

The inputs that decide the state-transport comparison, with their effects on this version.

One-factor test on the capital model	Δ A	Δ C
Central case	−€7.9bn	−€2.3bn
Dublin tier bounds ×2 (100 / 300 / 600 / 1,000k)	−€7.6bn	−€2.1bn
Dublin tier bounds ×0.5 (25 / 75 / 150 / 250k)	−€7.9bn	−€2.3bn
v3.11 demand basis (every home loads the network)	−€8.0bn	−€2.7bn
Strategic share of trips 0.30 / 0.50	−€8.5bn / −€7.3bn	−€2.9bn / −€1.7bn
Kilometres of new capacity per trip 8 / 16	−€8.7bn / −€7.1bn	−€3.1bn / −€1.5bn
Tunnel-class rate €200m / €350m per km	−€7.9bn / −€7.9bn	−€2.3bn / −€2.3bn
Road optimism bias 26 per cent	−€8.0bn	−€2.5bn
TOD car-trip ratio 0.40 / 0.60	−€7.3bn / −€8.3bn	−€1.7bn / −€2.8bn
v3.3 equivalence 0.38 (the removed double discount)	−€7.2bn	−€1.7bn

Every one of these is on the road side or the demand basis. The largest rail-side movers are switching the Naas four-tracking off (−€7.4bn) and FourNorth at 4 trains an hour (−€7.8bn). The road inputs are also the least sourced: the bounds, the strategic share and the kilometres per trip are all status A.

The evidence that would settle them is the NTA regional model's select-link data on the M7, N4, M3 and M1 radials by home location, and a corridor-by-corridor capacity count from TII's cordons and TD 9 lane capacities.

Link to this section

The Rail Towns option with the eight tunnel rows switched off, run through the CBA with the engine's no-tunnel parameters.

To see what the tunnel itself buys, a sensitivity switches off the eight tunnel-package rows (the tunnel, its fleet, shafts, ten-car works, depot, traction, signalling and turnbacks) and keeps the outer works, extensions, stations and towns. The demand engine supplies the tunnel-less network's factors. The capital model's result is captured for the CBA as CBA_CASE=surface.

Central, Dublin	Rail Towns with the tunnel package	Without it	Tunnel package alone
Rail-served by 2050	389,823	318,815	318,815 against 389,823
Capacity	462,654	412,581	+50,073
A, state transport	€31.4bn	€23.5bn
Δ A against Road Towns	−€7.9bn	−€0.0bn
Net present value, rung 1 / rung 2	€9.1bn / €9.1bn	€9.0bn / €9.8bn	+€0.1bn / −€0.7bn
Present value of costs	€10.52bn	€0.31bn	+€10.2bn
State's net cash position	€8.2bn worse	€1.0bn worse	+€7.2bn

The tunnel-less rail-led city is a dense car suburb with a fuller, less useful railway: more homes fit per seat because each uses fewer, but its strategic road bill is €1.9bn higher than the tunnel city's.

Two caveats carry every quotation of these numbers: the terminal question (the no-tunnel case assumes the committed core service can still reach the city without through-running) and the exclusions (existing-user benefits, post-2050 capacity). The history of this figure is the history of the model: −€2.2bn on v3.4 with flat behaviour, +€3.0bn on v3.5, +€4.3bn on v3.6, +€6.4bn on v3.7, +€10.4bn on v3.8, +€7.3bn on v3.9 with the Navan line.

Modelling choices referenced here

M20 · Option 2 without the tunnel package (v3.4 case)

Choice. A sensitivity that switches off the eight tunnel-package steps, keeping the outer works, town stations and towns; captured for the CBA (CBA_CASE=surface).

Alternatives. Not isolating the tunnel (v3.3).

Why. To see what the tunnel itself buys.

Effect. +78k rail homes without the tunnel vs +165k with it; on a 2050 window the tunnel's own marginal NPV is −€2.2bn, on a 2060 window +€6.8bn.

C15 · Option 2 without the tunnel package (CBA_CASE=surface) and window tests

Choice. The same workbook on the capital model's no-tunnel case; ad hoc 2060-window runs via CBA_OVERRIDES / CBA_HOMES_END / CBA_Y1.

Alternatives. Not isolating the tunnel.

Why. To see what the tunnel itself buys.

Effect. No-tunnel NPV1 €11.2bn on 78k homes (2050) → tunnel alone −€2.2bn; on a 2060 window the tunnel alone +€6.8bn.

Status. Method Evidence that would replace it. v3.10: the terminal-capacity ceiling without through-running is now applied here too (northend_cap_tph), so this case no longer credits FourNorth trains the city end cannot turn.

Link to this section

How far each input would have to move, with everything it touches re-evaluated, to take each rung to zero.

The v3.x table scaled one output line at a time and mislabelled the rung (the reviewer of September 2026 found both errors). This table changes an input and re-evaluates everything downstream of it, by bisection between ×0.05 and ×8 of the central value.

Modelling choices referenced here

C14 · Sensitivities, bands and switching values

Choice. Low/Central/High parameter bands, 2%/6%, 29 one-factor cases incl. 'v3.3 conventions', switching values on every line.

Alternatives. Monte Carlo.

Why. Shows which assumptions decide the answer.

Effect. No single line takes rung 1 negative; the roads-bounds ×2 analogue leaves €4.0bn.

Status. Method Evidence that would replace it. —

Link to this section

Each CBA input moved alone on the central case; the Bundle 6 cost cases shown for Bundle 6 against Road Towns too.

One-factor test on the CBA (Rail Towns v Road Towns unless stated)	Net present value, rung 1
Central	€9.1bn
Road saving −40 per cent (the analogue of doubling the tier bounds)	+€7.2bn, costs €12.44bn
Road saving +40 per cent	€11.0bn
Rail optimism bias +37 per cent	+€6.2bn
Rail optimism bias +37 per cent, Bundle 6 v Road Towns	+€1.5bn (central €1.9bn)
Bundle 6 works at twice the 2018 estimates, Bundle 6 v Road Towns	+€1.2bn
Decongestion overlap deduction held at 27 per cent with 150 km built	+€7.7bn
Extra housing cost €40,000 / zero	+€9.1bn / €9.1bn
Congestion €0.15 / €0.35 per km	+€6.4bn / +€11.8bn
Minutes saved 10 / 20	+€8.1bn / €10.1bn
Tunnel opens 2036 / 2040	+€9.1bn / +€6.8bn
v3.3 conventions (230 days, full switch, no housing cost, no overlap, no premium, no body)	€12.2bn, rung 2 €16.5bn
Discount rate 2 / 6 per cent	€21.8bn / +€2.6bn
Low / High parameter band	−€2.3bn / +€17.7bn

What takes the result to zero: the road side of the capital model, the demand basis and the Low band.

Modelling choices referenced here

C14 · Sensitivities, bands and switching values

Choice. Low/Central/High parameter bands, 2%/6%, 29 one-factor cases incl. 'v3.3 conventions', switching values on every line.

Alternatives. Monte Carlo.

Why. Shows which assumptions decide the answer.

Effect. No single line takes rung 1 negative; the roads-bounds ×2 analogue leaves €4.0bn.

Status. Method Evidence that would replace it. —

Link to this section

Status letters on every input, a ledger of every load-bearing claim, an adversarial rebuild, and rules for quoting.

Seán's standing instruction to the model is absolute honesty: "don't give me what I want if the numbers aren't there". Five mechanisms enforce it.

A source for every input. The model's register has 572 rows for the capital model, one per input or ladder step, each with its Low, Central and High values, its basis and a status letter; the CBA has 184. Every input and every modelling choice also carries a commentary entry saying what it is, why this value, what it drives, how it was tested and what would replace it; those entries are the lowest level of this page.
A Methods register, M1 to M27 and C1 to C17, recording each modelling choice, the alternatives rejected, the reason, the effect on the answer and the evidence that would replace the judgement. The labels on this page's headings are those items, and each heading opens them.
The honesty ledger, 117 numbered items, one for every load-bearing claim or change, each with a status (verified, corrected, dropped, pending) and an honest line in bold that says what the change means for the case, including when it weakens it.
External review. A ChatGPT review of v1 in early September, an independent from-scratch rebuild on 5 September, and an analytical review in September 2026 whose nine findings set the agenda for Stage 2 of the v4 rebuild; each was sorted into adopted, rejected with reasons, and things the model still owes.
Version discipline. Since v4 the model is code under version control, and a replay test asserts on every change that, on the frozen v3.12 inputs, it still reproduces the v3.12 and v3.13 outputs to one part in a million; every structural change since is a named switch with a recorded delta, and switches in the inputs reproduce earlier versions (navan_on = 0 gives v3.8; the engine off gives v3.4; the v3.11 demand basis is a sensitivity).

The rule for quoting: read the ledger's tail and the version stamp before quoting anything; quote the completions and the capacity by name, and the programme's capacity separately from the tunnel package's; say what the Low scenario does; and never quote a superseded headline (233,000, €14.5bn, €2.90, 210,000, €20.9bn, or the v3.9 page's 219,000, 158,000, €6.9bn and BCR 2.7).

P, D, C, A, U: what they mean and how the register is distributed.

Letter	Meaning
P	Published figure, cited
D	Derived by arithmetic from published figures
C	Convention: a reasonable modelling standard, not a measurement
A	Assumption: no direct source; the band is widened and a sensitivity is run where it matters
U	Unverified placeholder

Of the capital register's 572 rows, 78 are P, 116 D, 113 C, 237 A and 4 U. The As are concentrated in three places: the road tiers (the bounds, the strategic share, the kilometres per trip), the demand engine's factor rows (which are outputs of a stated method rather than observations), and the outer-section service assumptions; since v4 the destination-parking factor is A as well, on the reviewer's point that TCRP 128 does not measure destination parking. The Navan Railway's cost is the nearest thing to a U: its price basis is unstated, and its commentary says so. The status column is what a reviewer should read first, and the ledger's list of what would replace each assumption is the model's research agenda.

Link to this section

What each version changed and what it did to the gap.

Version, date	Change	Δ A	Δ C	NPV rung 1
v1, 3 Sep	Simple four-city model: ladder, fill order, road tiers on car-home equivalents	€11.3bn	€21.2bn
Vuchic update, 3 Sep	Load factor and peak-hour coefficient; per-leg caps; fleet by cycle time; graded catchments	€7.5bn	€15.4bn
v3, 4 Sep	Response to the ChatGPT review: Sources sheet, need bridge, background growth, derived tier rates	€5.5bn	€11.7bn
v3.1, 4 Sep	Outer two-track sections cap each leg (Duggan)	€4.2bn	€9.9bn
v3.2, 4 Sep	MetroLink as a fourth leg, common to both options	€1.1bn	€6.8bn
v3.3, 4 Sep	Land value capture as a transfer; the CBA built	€1.1bn	€6.8bn	€14.9bn
v3.4, 5 Sep	The external review's six findings; the tunnel isolated; capacity beside completions	€2.9bn	€6.0bn	€9.0bn
v3.5, 5 Sep	Demand engine on Vuchic's basis	€4.0bn	€7.1bn	€9.7bn
v3.6, 5 Sep	Density follows accessibility, exponent 0.5	€4.2bn	€7.4bn	€10.1bn
v3.7, 7 Sep	Exponent 1.0 central	€4.7bn	€7.9bn	€10.8bn
v3.8, 7 Sep	Out-of-vehicle time weighted; Naas at the plan's 13 tph	€7.9bn	€11.4bn	€14.5bn
v3.9, 7 Sep	Navan Railway in the committed programme	€6.9bn	€10.2bn	€12.8bn
v3.10, 10 Sep	City-end cap on the Northern leg without the tunnel (Duggan's answers)	€10.5bn	€14.5bn
v3.11, 10 Sep	Kildare service-plan increment as a band (Duggan's Phoenix Park Tunnel reading); CBA rebuilt	€10.5bn	€14.5bn	€18.1bn
v3.12, 11 Sep	Gross requirement reconciled to net transport demand (vacancy allowance, in-situ replacement); the Road Towns option leaves the tunnel-class tier	−€2.2bn	€1.9bn	€5.6bn
v3.13, 11 Sep	Carbon module for both options; the monetised lines follow it	−€2.2bn	€1.9bn	€5.4bn
v4.0–v4.2, 11–22 Sep	Rebuild in code, exact to one part in a million; Option 3, Bundle 6; the stress test	−€2.2bn	€1.9bn	€5.4bn
v4.3–v4.4, 27–29 Sep	The IAE / RIAI switches and the September fisk, each off by default	−€2.2bn	+€1.9bn	€5.4bn
v5, 6 Oct	The decided basis: the fisk profile with the road curve, 20 trains an hour, the trip cap, the Wicklow DART	+€4.0bn	+€8.0bn	€7.7bn
v5.1, 14 Oct	Ventilation shafts back (26 trains an hour)	+€2.9bn	+€6.9bn	€6.2bn
v5.2, 20 Oct	The Navan forecast counted once	+€3.1bn	+€7.1bn	€6.4bn
v5.3, 22 Oct	No urban road tunnels; land on a matched premium; the Codex fixes	−€7.9bn	−€2.3bn	€9.1bn

The gap fell through the first six versions as corrections landed against the rail case, rose again when behaviour was allowed to respond to service and when the city end was capped, and fell below zero when the gross requirement was reconciled to net demand. All three movements are recorded as they happened; the ledger's line on v3.4 was that the programme still cleared but the tunnel's own case rested on what happens after 2050, its line on v3.9 was that a four-trains-an-hour branch takes 14,000 off the headline, and its line on v3.12 is that the rail-led city no longer saves the State money on transport and its case rests on parking, utilities and the CBA's benefits.

Link to this section

What would replace the judgements, and what is still owed.

Road tier bounds and the two road-physics inputs: NTA regional model select-link data on the radials and TII cordon capacities, and a scheme-based road-cost curve to replace the staircase (reviewer finding 1). Decides the state-transport comparison; the single-threshold break-even is now published in the switching-value table.
The demand engine's calibration on observed station catchments (Census 2022 rail shares within 800 m of the Kildare, Northern and Maynooth stations), held fixed across the options, with an equal-density test (reviewer finding 2). Until then an NTA regional model run with mode choice for the modelled homes would replace the engine's service times, node jobs, slope and calibration.
Coherent end-to-end scenarios, with the 6.75 million population column as a Low case, and a phasing switch that ties the avoided roads to the homes that need them (reviewer findings 6 and 8). Terminal capacity from Duggan's answers is modelled; the Phoenix Park Tunnel count and the Connolly overbridge stay open.
The committed-network Kildare rail share (0.19) against Census 2022 rail commuting at Adamstown and Hazelhatch, owed since v3.8.
A Dublin hedonic station premium chained to a density–price elasticity, and delivered densities at Adamstown, Clongriffin and Cherrywood against the service each got, to replace the exponent.
The Navan Railway's track configuration, cost basis and DART+ West path split (Irish Rail); a 50-minute running time would make the branch tunnel-grade.
A quantity surveyor's comparison of matched dwellings by built form, for the sign of the extra housing cost.
Existing-user benefits from Jacobs on a consistent price base.
CSO National Travel Survey household car-kilometres by area type; an Irish destination-parking inventory per home.
The density-bound CBA runs, the TOD-study scaled rows and the 2060 window, re-run on the current basis.
A named land-value-capture mechanism, so the €20,000 stops being provisional.
The external appraisal of 5 September, recorded in the ledger; the optimised non-tunnel alternative (surface rail, interchanges, BusConnects, parking restraint, road pricing), existing-user benefits and construction disruption from the September 2026 review, each a Stage 2 switch of the rebuild.
The 150 km network's congestion (v5.4; ledger #107, Codex v5.4 review, finding 7). The car journey times are today's workbook rings and do not rise with the network; its congestion is priced only through the €0.25 marginal external cost on the car-kilometres Rail Towns avoids, and the slower trips of Road Towns' own drivers are not in the headline. An NTA regional model run would replace both. A select-link run would also set the share of the decongestion credit that falls on the new links, and whether the 150 km is priced at the tiers (€9.0bn) or at the map's mix (€5.8bn).

The Rail Towns case is one-sided by decision (v5.5, ledger #108): only Rail Towns fills its trains and runs 10-car trains. Every option filling gives €1.09bn against Road Towns; at the Low band Rail Towns builds 84 towns at once. A ceiling at Low is Seán's call.

Link to this section

Where everything lives, and the order the scripts run in.

Since v4 the model is one TypeScript repository, rail-towns, under version control. model/inputs/ holds every input with its bands, status, source and commentary; model/modules/ holds the arithmetic (need, ladder, capacity by leg, the demand engine, roads, costs, the CBA and the carbon account, the checks); model/cities/ holds the ladder and the Dublin map; model/sensitivities.ts the one-factor cases; export/ writes results.json, cba_results.json, the manifest, the workbook and the chart data; explorer/ is the interactive page; anatomy/ is this site; ledger/honesty-ledger.md is the ledger. The v3 workbook model, its explorer and its site are archived in the desk repository under tod-scale-economics/.

Rebuild order:

npm test          # the replay gate: v3.12 / v3.13 reproduced at 1e-6, the explorer parity, the invariants
npm run build     # results, manifest, workbook, charts, explorer, this site → dist/

This page is built by anatomy/build.ts, which resolves every figure in the curated text (anatomy/content/*.md) from the model at build time, generates a leaf for every input from the registry and the commentary, and refuses to build if a figure cannot be resolved, a section exceeds 500 words, or a superseded literal appears in the prose.

Link to this section

Input	Central	Rung 1 (transport only) reaches zero at	Rung 2 (with land use) reaches zero at
Tunnel package construction cost (eight steps, €m)	8,242	×2.00 (16,493)	×2.00 (16,493)
Dublin road tier bounds (all four together; b4 shown)	500,000	none between ×0.05 and ×8 of the central value	none between ×0.05 and ×8 of the central value
Urban tunnel road class rate (€m/km)	245	none between ×0.05 and ×8 of the central value	none between ×0.05 and ×8 of the central value
Route-km of new capacity per strategic trip	12	none between ×0.05 and ×8 of the central value	none between ×0.05 and ×8 of the central value
Strategic share of car departures	0.40	none between ×0.05 and ×8 of the central value	none between ×0.05 and ×8 of the central value
TOD car-trip ratio	0.51	×3.02 (1.539)	×3.02 (1.539)
Homes per town per year	600	×0.50 (301)	×0.50 (301)
Background growth of existing demand	0.20	×6.72 (1.344)	×6.72 (1.344)
Land-use margin per home (€)	65,000	none between ×0.05 and ×8 of the central value	none between ×0.05 and ×8 of the central value
Extra housing cost per home (€)	20,000	none between ×0.05 and ×8 of the central value	none between ×0.05 and ×8 of the central value
Minutes saved per trip	25	none between ×0.05 and ×8 of the central value	none between ×0.05 and ×8 of the central value
Commuting days a year	200	none between ×0.05 and ×8 of the central value	none between ×0.05 and ×8 of the central value
Marginal external cost of congestion (€/km)	0.25	none between ×0.05 and ×8 of the central value	none between ×0.05 and ×8 of the central value
Value of time (€/h)	18	none between ×0.05 and ×8 of the central value	none between ×0.05 and ×8 of the central value
Discount rate	0.04	×1.85 (0.074)	×1.85 (0.074)
Rail O&M per seat-year (€k)	4.2	×7.28 (30.586)	×7.28 (30.586)

What the model is, and how to read this page

Two cities, the same homesM1C1324 words5 sections below

The Road Towns option: the committed railway, then roadsM14M7290 words

The Rail Towns option: the investment ladder and twenty townsM13432 words

Capacity and completions are different numbersM19255 words

What is deliberately outside the modelM21C16225 words

Cork, Limerick and Galway332 words3 sections below

Cork inputs37 words20 sections below

Housing need 2024–2050 (Housing Commission Table 2.1 at 7.25m population; low/central/high = 2.4/2.2/2.0 persons per household)P175 words

Less: homes completed in 2024 and 2025 (the Commission period starts in 2024)P73 words

Homes served by other committed TOD (netted out of both options)C134 words

Committed rail catchment homes (greenfield / SDZ at committed stations)D89 words

Infill potential at existing stations — Option 1D79 words

Infill potential at existing stations — Option 2 (network-wide uplift)D80 words

Share of Option 1 infill at INNER-zone stations (< 4 km from the centre)D86 words

Share of Option 2 infill at INNER-zone stationsD51 words

Further greenfield TOD potential on committed lines — Option 1D74 words

Greenfield rail-town potential — Option 2U152 words

Number of town-scale sites carrying the Option 1 greenfield potential (for the pace cap)A76 words

Number of rail towns carrying the Option 2 potential (for the pace cap)C69 words

Existing peak-hour, peak-direction rail demand into the centre (2024)D106 words

Roads tier 1 upper bound (cumulative car-home equivalents)A158 words

Roads tier 2 upper bound (cumulative car-home equivalents)A68 words

Roads tier 3 upper bound (cumulative car-home equivalents)A65 words

Roads tier 4 upper bound (cumulative car-home equivalents)A88 words

Roads tier 5 upper bound (cumulative car-home equivalents)A56 words

Tier 5 environment: 1 = urban tunnel / parallel-M50 class (Dublin), 0 = urban-edge insertion class (regional cities)C110 words

Parking LAND factor (1.0 = Dublin RZLP values; applies to the land component only)D86 words

Limerick inputs37 words20 sections below

Housing need 2024–2050 (Housing Commission Table 2.1 at 7.25m population; low/central/high = 2.4/2.2/2.0 persons per household)P175 words

Less: homes completed in 2024 and 2025 (the Commission period starts in 2024)P73 words

Homes served by other committed TOD (netted out of both options)C136 words

Committed rail catchment homes (greenfield / SDZ at committed stations)D84 words

Infill potential at existing stations — Option 1D79 words

Infill potential at existing stations — Option 2 (network-wide uplift)D80 words

Share of Option 1 infill at INNER-zone stations (< 4 km from the centre)D86 words

Share of Option 2 infill at INNER-zone stationsD51 words

Further greenfield TOD potential on committed lines — Option 1D73 words

Greenfield rail-town potential — Option 2U145 words

Number of town-scale sites carrying the Option 1 greenfield potential (for the pace cap)A76 words

Number of rail towns carrying the Option 2 potential (for the pace cap)C69 words

Existing peak-hour, peak-direction rail demand into the centre (2024)D101 words

Roads tier 1 upper bound (cumulative car-home equivalents)A148 words

Roads tier 2 upper bound (cumulative car-home equivalents)A68 words

Roads tier 3 upper bound (cumulative car-home equivalents)A65 words

Roads tier 4 upper bound (cumulative car-home equivalents)A88 words

Roads tier 5 upper bound (cumulative car-home equivalents)A56 words

Tier 5 environment: 1 = urban tunnel / parallel-M50 class (Dublin), 0 = urban-edge insertion class (regional cities)C110 words

Parking LAND factor (1.0 = Dublin RZLP values; applies to the land component only)D86 words

Galway inputs37 words20 sections below

Housing need 2024–2050 (Housing Commission Table 2.1 at 7.25m population; low/central/high = 2.4/2.2/2.0 persons per household)P175 words

Less: homes completed in 2024 and 2025 (the Commission period starts in 2024)P73 words

Homes served by other committed TOD (netted out of both options)C136 words

Committed rail catchment homes (greenfield / SDZ at committed stations)D85 words

Infill potential at existing stations — Option 1D79 words

Infill potential at existing stations — Option 2 (network-wide uplift)D80 words

Share of Option 1 infill at INNER-zone stations (< 4 km from the centre)D86 words

Share of Option 2 infill at INNER-zone stationsD51 words

Further greenfield TOD potential on committed lines — Option 1D73 words

Greenfield rail-town potential — Option 2U131 words

Number of town-scale sites carrying the Option 1 greenfield potential (for the pace cap)A76 words

Number of rail towns carrying the Option 2 potential (for the pace cap)C69 words

Existing peak-hour, peak-direction rail demand into the centre (2024)D101 words

Roads tier 1 upper bound (cumulative car-home equivalents)A156 words

Roads tier 2 upper bound (cumulative car-home equivalents)A68 words

Roads tier 3 upper bound (cumulative car-home equivalents)A65 words

Roads tier 4 upper bound (cumulative car-home equivalents)A88 words

Roads tier 5 upper bound (cumulative car-home equivalents)A56 words

Tier 5 environment: 1 = urban tunnel / parallel-M50 class (Dublin), 0 = urban-edge insertion class (regional cities)C110 words

Parking LAND factor (1.0 = Dublin RZLP values; applies to the land component only)D86 words

A third city: Bundle 6242 words4 sections below

What the 2018 study found346 words

What the model takes from it463 words12 sections below

City end WITH the Bundle 6 works and WITHOUT a tunnel: extra Northern-line peak trains Connolly (four through platforms), Spencer Dock (18 tph terminating) and the Loop Line (18 tph) absorb (Option 3; caps FourNorth's uplift as northend_cap_tph does in Option 1)D285 words5 sections below

Why this value: City end WITH the Bundle 6 works and WITHOUT a tunnel: extra Northern-line peak trains Connolly (four through platforms), Spencer Dock (18 tph terminating) and the Loop Line (18 tph) absorb (Option 3; caps FourNorth's uplift as northend_cap_tph does in Option 1)D123 words

What it drives: City end WITH the Bundle 6 works and WITHOUT a tunnel: extra Northern-line peak trains Connolly (four through platforms), Spencer Dock (18 tph terminating) and the Loop Line (18 tph) absorb (Option 3; caps FourNorth's uplift as northend_cap_tph does in Option 1)D87 words

How it was tested: City end WITH the Bundle 6 works and WITHOUT a tunnel: extra Northern-line peak trains Connolly (four through platforms), Spencer Dock (18 tph terminating) and the Loop Line (18 tph) absorb (Option 3; caps FourNorth's uplift as northend_cap_tph does in Option 1)D71 words

What would replace it: City end WITH the Bundle 6 works and WITHOUT a tunnel: extra Northern-line peak trains Connolly (four through platforms), Spencer Dock (18 tph terminating) and the Loop Line (18 tph) absorb (Option 3; caps FourNorth's uplift as northend_cap_tph does in Option 1)D84 words

History: City end WITH the Bundle 6 works and WITHOUT a tunnel: extra Northern-line peak trains Connolly (four through platforms), Spencer Dock (18 tph terminating) and the Loop Line (18 tph) absorb (Option 3; caps FourNorth's uplift as northend_cap_tph does in Option 1)D78 words

Bundle 6: extra Kildare-line DART paths an hour through the Phoenix Park Tunnel to Docklands, over the DART+ baseline (7 tph)D294 words