Repository navigation
Assemble the microcosm-built UK spine: whole-spine parity and swap acceptance #686
Description
Rescoped 2026-08-19 — the parity vintage is FRS 2024-25. Per the vintage decision recorded on #723, the frozen incumbent reference is now
enhanced_frs_2024_25(survey/base 2024, calibration year 2025, the v1.56.x production line), with the parity instruments, record-count identity, and integrity baselines re-frozen at that vintage by #723. Read the vintage-specific lines below accordingly. The certified 2023 H5 keeps exactly one role here: the control build input for swap acceptance. #630 has since been resolved, satisfying that dependency. Sequence: #681/#682 → #723 (retarget) → #684 → this child. The 2023-vintage build itself is deferred to #687.
Workstream E10 of the UK-migration master epic #665; child of #145. The workstream's exit: the microcosm-built spine replaces the certified incumbent H5 in the national build.
Scope
- One driver path running the full ingest+imputation
StagePlanat the scale ladder's rungs (smoke/dev/full), every attempt — pass, fail, or abort — landing a Logbook row. - Whole-spine parity: candidate extraction (extraction-only mode added to
tools/build_uk_efrs_parity_reference.py) diffed against the committed 145-column reference — per-column nonzero-share deltas, entity counts satisfying the record-count identity, missing/extra columns through a signed register. Weighted totals across the 131-column licensed register via--emit-weighted-totalson both sides (reference sha-verified; deltas published SDC-safely). New-spine baselines recorded with the weighted-integrity instrument — these become the candidate thresholds for arminguk_input_mass_parityanduk_qrf_tail_concentration. - Swap acceptance: a control build (certified H5) and a candidate build (microcosm spine at the grain the joint E/D adjudication settles;
verify_certified_uk_candidateextended with a second certified identity — never a bypass).tools/compare_uk_h5_payload.pywith a--structure-onlyverdict mode (same surface; values differ only where signed); the full battery in release-candidate posture;macro_realismdeclared for the assembled spine. - Signed-differences register (one committed JSON): every intentional deviation — the CGT weight mechanism, the defect fixes, the RNG-stream class, the QRF-implementation class, net-new columns — with id, magnitude evidence, adjudicator, and date. Anything differing that is not in the register is a defect.
- Calibration-seam rehearsal: the rowwise candidate driver consumes the candidate staging H5, calibrates through the front door, and the battery passes there too.
Acceptance
The WS-E epic bar (#145): the microcosm-built spine replaces the certified incumbent H5 in the national build; eFRS parity fixtures beat or match the incumbent on the frozen register with differences signed; determinism covers every stochastic assignment on the live path. Final calibrated sign-off additionally requires WS-D (#623) landed and #630 adjudicated.
Depends on: E2–E9; the joint E/D spine-grain adjudication; #630 for the release-candidate posture.
Activity
Regenerate the parity instruments from enhanced_frs_2024_25
Move the gate-side reference pins to the 2024-25 line
Adversarial-review dispositions: dual source pins on the HMRC family …
Address the #733 automated review: single failure handler, mass-log c…
Regenerate the parity instruments from enhanced_frs_2024_25
Move the gate-side reference pins to the 2024-25 line
Adversarial-review dispositions: dual source pins on the HMRC family …
Address the #733 automated review: single failure handler, mass-log c…
juaristi22 commentedon Aug 22, 2026
WS-D input for this issue's swap decision, from #743 (draft): the armed-run runbook (
docs/uk-national-calibration-runbook-623.md— command shape, evidence-dir layout, unblock conditions) andtools/score_uk_national_candidate.py(#578 rule 1: both artifacts rescored on one frozen register, June-schemascore_vs_enhanced_frsblock). Final calibrated sign-off here still requires the held #623 run to execute; its score receipt is this issue's swap-decision input. Grain: the #688 staging-input half is now adjudicated pre-clone for the national default (see the ruling recorded there) — the strata-cap re-derivation (100,000 → 10,000) is measured at run time; arming stays with this issue.Add the UK eFRS parity re-pin tool and the identity-mirror lockstep test
Re-pin the UK enhanced-FRS parity reference to the incumbent's 1.56.16
Record the #686 re-pin and Scottish-water measurement receipts
Pin the new country-package resource in the resource-list assertions
Re-pin the UK enhanced-FRS parity reference to the incumbent's 1.56.16
Record the #686 re-pin and Scottish-water measurement receipts
Pin the new country-package resource in the resource-list assertions
Merge pull request #747 from PolicyEngine/uk-spine-assembly-686
Add the UK eFRS parity re-pin tool and the identity-mirror lockstep test