First calibrated rowwise UK local candidate (#495 increment 6) #762

Parent: #495 (workstream G of the UK-migration epic #665). Depends on #708, the local target surface compile, the fence-enforcement issue, and increment 5's K ruling.

Why

This is the WS-G milestone: the first UK local artifact whose weights are solved under declared doctrine against targets compiled from Chronicle facts, on a spine microcosm built itself.

Scope

1. One uniform doctrine solve at the adjudicated K via solve_uk_rowwise_weights_under_doctrine: national families and every area × metric row in one operator, uniform target weights, one declared cap and stretch bound, no per-target knobs. A miss is support work or target work, never a knob.
2. Weight-kind transition recorded — importance → locally calibrated — with exactly one appended mass-change record carrying the canonical rowwise_calibration_mass_reason text.
3. The rowwise release contract from the epic, in the manifest: identity chain (spine sha → source-stage shas), geography block (ladder sha, per-layer vintages, 650/650, gate pass), weights block (exact mass conservation, weight-kind chain, carried mass log, declared stretch bound), solve block (loss shape, cap, ratio bound, epochs, seed), per-family fit, past-cap census, per-area support with distinct-source counts.
4. Adjudicate both doctrine constants against measured fit — the loss cap (10.0) and UK_LOCAL_MAX_WEIGHT_RATIO (100.0, against the US design bound of 5.0 and the exact-k ladder's 20.0). This is the revision path max_weight_ratio anchors differently per arm: 5x vs design (dense) but ~25x effective (sparse refit re-anchors) — declare one stretch contract #493 named and the reason Loss shape: flat cap at 1.0 makes past-cap rows free dumping grounds (17 rows pushed out in Build N); bounded-tail experiment + past-cap census diagnostic #492's past-cap census exists; confirm or revise, and record the measurement in local_doctrine.py and its pinned test.
5. Score against the incumbent on a frozen local register — constituency and LA estimates versus uk-data's wide-format weights — plus weakest-family and weakest-area tables.
6. Private publish only. populace-publish-release --repo-id policyengine/populace-uk-private --no-latest, immutable tag, dataset_role: non_default_local_area. Aggregate diagnostics are committable; household-grain rows never leave the private repo.

Out of scope

• The full credibility-gate port (Port UK local target surface and credibility gates from uk-data #147) beyond what makes this candidate reviewable.
• The published local variant (Publish a Populace UK local variant with rowwise H5 and long weights #146) and loader consumption (workstream I).

Acceptance

• A calibrated rowwise candidate exists with the full manifest above, every rung in Logbook.
• Constituency and LA estimates match or beat the incumbent on the frozen local register.
• Both doctrine constants are either confirmed or revised, with the measurement recorded where the constant lives.
• The artifact is loadable as a variant="local" entry and never moves latest.json.

🤖 Generated with Claude Code