Rail fares freeze: the engine ships both legs of the counterfactual (#105)

[vahid-ahmadi]

Closes #105. Branches from main — independent of the other UK PRs.

Same shape as #102: a reform specification with every path resolved against a real policyengine-uk 2.89.2, and no value claims staged.

Why this lane is cheap

The 2026 regulated fares freeze is already in the certified baseline, so the scored object is the counterfactual — what fares would have done under prior law. Unusually, the engine carries that leg too:

year	gov.dft.rail.fare_index	gov.dft.rail.prior_law_fare_index	gap
2026	1.217	1.288	+5.83%
2027	1.268	1.342	+5.84%
2028	1.318	1.394	+5.77%

#13 and #67 both had to construct an indexed counterfactual by hand — the axis #67 found undecomposed in OBR's own re-estimates. Here it is a parameter. The reform is one line: fare_index := prior_law_fare_index.

It reaches an outcome, which is what makes it a lane rather than a curiosity:

rail_usage × fare_index → rail_subsidy_spending → dft_subsidy_spending

That factorisation is deliberate. rail_subsidy_spending's own docstring says it exists "allowing reforms to modify fare prices independently of usage quantity."

What the validator refuses

Each of these is a way the spec could be wrong while still looking right:

• A counterfactual that is not a change — it would score nothing and read as agreement. Checked per year, so a freeze that lapses mid-horizon can't pass on 2026 alone.
• Swapped legs — prior law must sit above the freeze. Reversed, the spec scores a fare cut and still looks plausible.
• Staged value claims — DfT's saving figures are DfT's. The Ingest the two harvested-but-unused UK think-tank families: IFS (268) + Resolution Foundation (71) #86/Ingest the two harvested-but-unused UK think-tank families: IFS + Resolution Foundation (#86) #91 re-publication rule applies: they belong in their own harvest lane, read from primary documents, never copied from reporting into a spec.
• Unresolved paths — --resolve reads every path and head variable out of a real engine. This repo has published two false engine gaps from paths that were never resolved (Autumn Budget 2026 scoreability registry: ready before 28 October (#96) #100, Baseline integrity: does the certified world carry the law legislated into its future? (#99) #101).

Scope, named up front

Regulated fares are a minority of tickets sold in England, and DfT reports savings per passenger where the engine's quantities are per household. Both are recorded as named divergence axes (#59) before any comparison runs — unnamed, either gap gets attributed to the engine.

Not computed here

This PR specifies and resolves; it computes no counterpart. UK counterpart compute runs through #51, which isn't merged. Publishing a number before that lands would bypass the compute contract.

Gates

275 tests (12 new), ruff format --check clean, two build_db runs agree on content_hash, no drift.

#106 is the other half of the pair — bus, which is not expressible.