Bus fare cap: an earned gap, and the near-miss that nearly made it false (#106)

[vahid-ahmadi]

Closes #106. Branches from main — independent of the other UK PRs.

The £2/£3 bus fare cap cannot be scored at the certified pin. The verdict matters less than how it was reached, because my first pass at it was wrong — and it was wrong in exactly the way #100 and #101 were wrong.

The near-miss

That first pass searched the parameter tree, found gov.dft has only rail and spending, and concluded bus was absent from the model.

It isn't. bus_subsidy_spending is a variable, and a parameter-only search misses it:

dft_subsidy_spending   adds = [rail_subsidy_spending, bus_subsidy_spending]

Bus subsidy is a household-level quantity and its incidence is modelled. Caught before publishing this time, rather than in review afterwards.

The gap that survives

Narrower and more useful than "PolicyEngine doesn't do buses":

	rail	bus
quantity	rail_usage	—
price lever	gov.dft.rail.fare_index	—
relationship	rail_usage × fare_index	flat input
reform possible	yes	no

bus_subsidy_spending is a pure input — no formula, no parameter behind it. So the gap is a missing factorisation, not a missing subject area. That makes the upstream ask small and precedented: give bus what rail already has (bus_usage, gov.dft.bus.fare_index, and the product between them). It would make every future fare-cap question scoreable, not just this one.

The validator enforces the lesson, not the conclusion

--resolve can't catch a false gap — a not_expressible entry carries no path to resolve. So:

• A gap claim needs a name search covering variables as well as parameters. --probe re-runs both against a real engine and refuses drift.
• The registry must record what the engine does have. Omitting the present-but-unusable machinery overstates the gap.
• has_formula must be False. That single fact is what the verdict rests on — if it changes upstream, the probe fails loudly rather than leaving a stale opinion in the repo.
• The nearest lever (gov.contrib.policyengine.economy.transport, "raise transport expenses by this percentage", currently 0) must be named and ruled out. It moves all transport, isn't bus-specific, and isn't a fare cap — using it would model something else and label it as the measure. Unruled-out, a reader finds it and assumes nobody looked.
• No staged value claims — the DfT-commissioned Frontier Economics/SYSTRA evaluation figures belong to their originators (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).

One honest detail pinned in the tests: the search pattern's bus matches gov.hmrc.business_rates. Substring false positive, not a bus lever, recorded as such so nobody mistakes it later.

Why hold it at all

The £3 cap runs to March 2027, so it's live for the 28 October Budget, and DfT has a published independent evaluation of the £2 cap. The claims are worth holding even while the counterpart isn't computable — that's what a mode-1 lane with a named gap is for.

Gates

277 tests (14 new), ruff format --check clean, two build_db runs agree on content_hash, no drift.

#107 is the scoreable half of the pair.