Remove the UK QRF tail-concentration gate (#1276)

changelog.d/1124-uk-qrf-tail-surface.changed.md

@@ -1,3 +1,3 @@
 The UK QRF tail-concentration gate no longer checks the report columns `spi_benefit_coherence` (microcosm#1100) replaces on every SPI-channel row. The stage zeroes the income-tested legacy reports (contributory and income-based JSA, Income Support, Working and Child Tax Credit, income-related ESA, Severe Disablement Allowance) and restores IIDB, AFCS and bereavement support from the FRS twin. After it these columns hold only FRS respondents' own reports, so the donor-tail broadcast the gate looks for (#462) cannot occur in them; their concentration is that of a small population of genuine reporters and their clone copies. The surface reads the superseded columns from the stage's declared operations and lists them in its details as `superseded_by_spi_benefit_coherence`.
 
-`hmrc_spi_other_income` and `hmrc_spi_taxable_termination_pay` stay on the surface: they are real stage-1 draws. They failed national release-cut certification as genuinely concentrated instruments, rare on the SPI tape, whose top 100 records carry nearly their whole mass whether counted by row or by independent record; the CGT clones and SPI copies also take them past the carrier floor. They are deferred behind national-line reviewed exclusions, time-boxed to 2027-01-14, until the gate is re-armed on independent draws (microcosm#1277).
+`hmrc_spi_other_income` and `hmrc_spi_taxable_termination_pay` stay on the surface: they are real stage-1 draws. They failed national release-cut certification as genuinely concentrated instruments, rare on the SPI tape, whose top 100 records carry nearly their whole mass whether counted by row or by independent record; the CGT clones and SPI copies also take them past the carrier floor. They were deferred behind national-line reviewed exclusions until microcosm#1281 removed the gate and its exclusion register; the check that replaces it is tracked in microcosm#1276.

changelog.d/1276-remove-uk-qrf-tail-gate.removed.md

@@ -0,0 +1 @@
+The UK national release-cut battery no longer runs the QRF tail-concentration gate (`uk_qrf_tail_concentration`), and its exclusion register (`qrf_tail_reviewed_exclusions.json`) is gone. On the UK surface the gate's top-100 share could only bite on a column with fewer than 100 independent draws whose CGT-clone copies took it past the 104-row floor, so its verdict measured how little mass those copies carry rather than whether an imputation broadcast a donor tail; and at the armed 0.99947 threshold the #462 incident it was built for (89% of a column's mass in 100 records) would have passed. The data contract drops the gate's pins, detail schema and validation, and re-pins the release-cut part and full-battery digests. `uk_qrf_tail_concentration_columns` and the weighted-integrity baselines tool stay as the measurement surface for the replacement microcosm#1276 asks for: a source-fidelity check of each imputed column against its donor.

docs/gate-battery-contract.md

@@ -128,7 +128,6 @@ Canonical artifact keys:
 | `parity_evidence` | the UK national build | `export_surface`, `target_surface`, `target_fit` |
 | `input_mass_reference` | the UK national build, when the licensed sidecar is supplied | `input_mass_parity` (UK override) |
 | `reviewed_input_mass_exclusions` (optional override) | the UK national build | `input_mass_parity` (UK override) |
-| `reviewed_qrf_tail_exclusions` (optional override) | the UK national build | `tail_concentration` (UK override) |
 | `uk_area_support_summary`, `exclusions_evaluated_on` | the UK local rowwise build | `area_support` (UK binding; reviewed exclusions carry measured shortfalls and stale entries fail) |
 | `cgt_projection` | the UK calibration seam: read from the installed engine's growth path and exempt amounts, or stated from the manifest pins when no engine is installed (the receipt's `projection_engine` names the source, and the release certifier refuses a seam part whose fence read the pins) | `cgt_projection_entrants` (UK binding; the count is a cumulative stock by year, the bound is the vendored HMRC thin-band taxpayer count as a plausibility ceiling, and the growth path and exempt amount are pinned in the manifest and drift-checked, microcosm#970) |
 
@@ -139,8 +138,9 @@ reference implementation of a country registry. Its bindings adapt the
 reviewed policy objects from the frozen declared parameters, hold
 runtime-supplied references to the spec-declared pins, declare the
 parameter vocabulary each evaluator can route (so a stray key is refused
-at arm time, not discovered mid-battery), and re-mint exactly
-two UK-flavored result names onto the shared vocabulary. Every verdict
+at arm time, not discovered mid-battery), and re-mint the
+UK-flavored result names of the legacy evaluators onto the shared
+vocabulary. Every verdict
 is computed by the existing UK gate functions — a binding adapts
 evidence; it never re-implements a comparison. The differential test
 (`tests/test_uk_battery_bindings.py`) holds the battery's verdicts equal

docs/uk-national-release-assembly-runbook-806.md

@@ -104,6 +104,26 @@ With the default paths, this writes
 `microcosm_uk_2024_25.release_certification.json` next to the candidate. Continue
 only when the certification says `shippable: true`.
 
+**Interim check until microcosm#1276 lands.** microcosm#1281 removed the QRF
+tail-concentration gate, and nothing release-blocking yet checks the SPI
+stage-1 columns (`hmrc_spi_*` and the other stage-1 outputs). Before assembly,
+measure the candidate and the currently promoted cut together:
+
+```bash
+uv run --no-sync python tools/measure_uk_weighted_integrity_baselines.py \
+  --h5 candidate=<candidate-dir>/microcosm_uk_2024_25.h5 \
+  --h5 previous=<promoted-cut-h5> \
+  --output <candidate-dir>/stage1_comparison.json
+```
+
+Compare each stage-1 column's weighted total (`input_mass_totals`) and carrier
+count (`qrf_tail`) between the two, and explain any column that moves by more
+than half in either direction before continuing. microcosm#1256 cut
+`hmrc_spi_incapacity_benefit_income` by about six-sevenths, and only this kind
+of comparison would have shown it. The file is licensed output: keep it
+local, out of the release directory and out of any public text. The tool
+suppresses carrier counts below 10.
+
 ## 3. Assemble the release directory
 
 Assembly verifies the complete hash join before writing, mints the calibration

packages/microcosm-build/src/microcosm/build/uk/country_package.json

@@ -242,11 +242,6 @@
       "kind": "legacy_json",
       "schema_id": "legacy_json"
     },
-    {
-      "path": "qrf_tail_reviewed_exclusions.json",
-      "kind": "legacy_json",
-      "schema_id": "legacy_json"
-    },
     {
       "path": "release_input_coverage_manifest.json",
       "kind": "legacy_json",

packages/microcosm-build/src/microcosm/build/uk/gates.json

@@ -1247,19 +1247,6 @@
       },
       "notes": "Weighted input-mass parity against the active named reference in the registry, with reviewed exclusions scoped per reference so a channel-blind reference exclusion does not suppress comparison against a future channel-aware reference. The tolerance is the worst surviving measured drift at the reviewed floor, exact, with no headroom; the reviewed per-reference exclusions (charitable_investment_gifts, signed at +15,168.8% on the post-#710 surface; owned_land, signed for seed variance) were retired on 2026-09-22 when the first release-cut certification measured both inside the fence (-99.0% and -73.5% against the enhanced FRS reference, microcosm#967); the register is empty and the gate's current failures are the two genuine breaches it names. The 0.0 floor excludes only arithmetically meaningless zero-mass denominators. Totals stay uncommitted under the UKDS licence, supplied per-run and pinned by totals_sha256; missing sidecar evidence records evidence_absent. Totals keys are flat frame column names since the #611 A4 dedup; that digest re-freeze was keys-only against the #610 values (identity unchanged, values byte-identical). Sensitivity floor, recorded rather than silent (microcosm#706 review): the tolerance is the worst surviving measured drift, and that edge is currently set by a column whose deviation is explained (vintage plus SPI composition, microcosm#703's head-to-head receipt), so at this value the gate is a gross mass-loss/explosion fence rather than a regression detector — the compared population sits at median 0.24 and p90 0.63. Tightening runs through this register: each explained top drift that earns a receipt lets the tolerance re-mint at the worst unexplained survivor, with the full distribution and the exclusions-versus-tolerance tradeoff in experiments/630-uk-gate-adjudication-receipts.md."
     },
-    {
-      "id": "uk_qrf_tail_concentration",
-      "gate": "tail_concentration",
-      "phase": "terminal",
-      "criticality": "release_blocking",
-      "parameters": {
-        "reviewed_exclusions_resource": "qrf_tail_reviewed_exclusions.json",
-        "top_k": 100,
-        "max_top_share": 0.9994670564654868,
-        "min_nonzero_records": 104
-      },
-      "notes": "Weighted tail-concentration audit of the QRF-imputed output columns derived from the HMRC source manifest. Armed from the #686 L3 baselines per #757 B4, each threshold recording the run that measured it: the 47-column qrf_tail grids of uk_weighted_integrity_baselines_686.json (sha256 8b3f6c4e9c522445bf3e05d3451ee8557f4e14d25524ceac3af8bfe05401a146, licensed acceptance dir 686-spine-swap), measured on spine-a.h5 (sha256 a65f2132736d1dcecb91709a513a0d54a5887635ea03760629cc888d188ee306, 113,649 person rows, design weights) for the #609/#578 threshold adjudication. top_k 100 is the measurement grid anchor [10, 100, 500, 1000]; max_top_share is the exact measured maximum over the checked surface (hmrc_spi_other_social_security_income, 104 carriers), no headroom; min_nonzero_records is the thinnest measured column above the grid anchor (the same column) - the policy domain requires min_nonzero_records > top_k, so the three saturated sub-anchor columns (hmrc_spi_taxable_termination_pay 12, charitable_investment_gifts 22, sda_reported 24 carriers, each top-100 share 1.0) sit below it and go thin visibly in the signed details on every run, never a silent pass. The baselines are design-weight measurements and the terminal gate runs on the calibrated release frame; a calibrated-weight breach names its column and is a finding, not noise. The 12 household_qrf_tail grids (was_wealth surface) are measured in the same baselines file and not yet armed - a household-surface gate entry is a declared follow-up. Reviewed exclusions live in the named register resource of this package."
-    },
     {
       "id": "uk_local_geography_ladder_post_calibration",
       "gate": "spine_agreement",

packages/microcosm-build/src/microcosm/build/uk_runtime/__init__.py

@@ -577,14 +577,12 @@
 from microcosm.build.uk_runtime.weighted_integrity import (
     UKInputMassParityPolicy,
     UKInputMassReference,
-    UKQRFTailConcentrationPolicy,
     load_uk_input_mass_reference,
     load_uk_local_area_support_exclusion_register,
     load_uk_reviewed_exclusion_register,
     uk_input_mass_parity_gate,
     uk_input_mass_totals,
     uk_qrf_tail_concentration_columns,
-    uk_qrf_tail_concentration_gate,
 )
 
 __all__ = [
@@ -1025,7 +1023,6 @@
     "UK_EDUCATION_GRANT_CAPACITY_PREDICTORS",
     "UK_LEGACY_PROXY_PREDICTORS",
     "UK_YEAR_RULES",
-    "UKQRFTailConcentrationPolicy",
     "UKZeroWeightStratumDeclaration",
     "load_uk_frs_release",
     "load_uk_input_mass_reference",
@@ -1036,7 +1033,6 @@
     "uk_input_mass_parity_gate",
     "uk_input_mass_totals",
     "uk_qrf_tail_concentration_columns",
-    "uk_qrf_tail_concentration_gate",
     "uk_target_fit_gate",
     "uk_target_surface_gate",
     "uk_weight_ess_gate",