Common ScoreResult schema across scoring adapters + OBR microsim static-costing bridge

[vahid-ahmadi]

Closes #9
Closes #10

ScoreResult schema (#10)

One pydantic ScoreResult for every scoring entry point, emitted per the module's plain-dicts-out convention and additively: og_score_reform, the new obr_score_reform, and pe_population_impact keep their existing payloads unchanged and gain a "score" key containing:

• model (id) + model_class (microsim / semi-structural / olg-ge), country, reform, horizon (steady-state / quarterly window … / annual …)
• quantities — gdp, consumption, investment, government, revenue, debt, each a ScoreQuantity with level_bn / delta_bn / delta_pct plus required units and basis; a model fills only what it genuinely produces (the static microsim fills revenue only, and says why in caveats)
• assumptions, caveats, optional uncertainty, optional distributional block (deciles / winners / losers — microsim only for now)

Payoff wired: macromod compare --reform '…' --models microsim,obr (or og) renders one table from N ScoreResults; --json emits the list. The table footer states explicitly that steady-state and budget-window deltas are each model's own concept.

OBR bridge (#9)

score_reform(model="obr") / obr_score_reform: static costing in → second-round effects out.

1. pe_population_impact(country="uk", reform, year=y) for each year in [start_year, start_year+years) → annual budgetary_impact_bn path (baseline sims cached per year).
2. obr_costing_to_shock (pure arithmetic, unit-tested with no heavy deps): annual £bn → quarterly £m, flat within each year (declared), sign-corrected (revenue raised ⇒ disposable income falls).
3. Injected via run_reform(var="HHDI", shock=[…]) using the upstream path-shock interface (Path-valued shocks and an add-factor injection point for externally costed reforms obr-macroeconomic-model#9's path-valued shocks are on obr main). HHDI is the demand-side incidence channel: HHDI → RHHDI → CONS → GDPM.

Declared boundaries — in the result payload (caveats), not just docstrings:

• run_reform exogenises HHDI identically in baseline and shocked runs, so the delta isolates the shock, at the stated cost that the economy's feedback onto disposable income itself is not recycled (this is a plain exogenous path, not upstream's proposed held add-factor injection point, which is not implemented there yet).
• Corporation tax is not household-borne in the microsim: refused with a ValueError pointing at obr_shock(var="TCPRO"), before any heavy import.
• Demand-side incidence only; supply-side channels belong to the OG member.
• The revenue quantity in the OBR score block echoes the static costing and its basis says it is the bridge INPUT, not an emulator output.

Also

• score_reform gains a microsim model so the compare table can include the static costing itself; country=us works there.
• MCP score_reform docstring/params updated (years, dataset); CLI score gains --years/--dataset and renders the common block for non-OG models.
• Tests: fast unit tests for the costing→shock translation, schema converters, bridge wiring (both heavy deps stubbed), corp-tax boundary, and compare; the real UK pipeline test is slow-marked and skips cleanly without HUGGING_FACE_TOKEN / obr_macro. Fast suite: 60 passed, 35 skipped (the pre-existing test_summary_parses needs boe_var or MACROMOD_BOE_VAR_REPO, unrelated to this change and green with either).

Deliberately deferred: the site/docs page (#10 mentions it; this PR is the integration layer), any uncertainty band producer (SVAR is not a reform-scoring adapter yet), and the upstream held-add-factor injection point.

🤖 Generated with Claude Code

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
macromod	Ready	Preview, Comment	Jul 17, 2026 10:17am