Skip to content

UK dense: bus floor to 1.15%, four renewals and build 1r's gate-only sign-off (#1207, #1301) - #1640

Merged
juaristi22 merged 4 commits into
mainfrom
uk-dense-bus-floor-21-families
Nov 16, 2026
Merged

juaristi22 merged 4 commits into
mainfrom
uk-dense-bus-floor-21-families

Conversation

@juaristi22

@juaristi22 juaristi22 commented Nov 16, 2026 •

Copy link
Copy Markdown
Collaborator

Summary

Three commits on 3a7067fa2 (main on 15 November) carry the first UK dense build at main's head on Chronicle feed 4ce631f5 to an assembled release:

  1. 79601a89c lowers the dense solve's loss-share floor for DfT's bus journeys from 1.25% to 1.15% (UK consumption-tax levels: open items after #1167 #1207).
  2. e1f70ddd3 renews four of the local line's gate-only national target-fit entries that expired on 16 November (UK local-area release: publish a K=15, 150,000-household dense dataset in at most two builds #1301).
  3. 3dbf54e19 is the gate-only sign-off drafted by draft_uk_dense_signoff.py from dense build 1r, applied unchanged (UK local-area release: publish a K=15, 150,000-household dense dataset in at most two builds #1301).

The changes are a constant, register data and test pins. No solve or gate logic changes.

Built and assembled from this head

  • Dense build 2 (attempt uk-local-candidate-f100-s42-20261116T102715Z-e2745e59) ran at 3dbf54e19 with gate clock 2026-11-16 and --require-solve-reuse.
  • draft_uk_dense_signoff.py --verify against build 1r (attempt uk-local-candidate-f100-s42-20261116T010526Z-602766ae): the solve was reused, 0 rows differ and nothing failed.
  • Pre-flight passed. The release was assembled on 2026-11-16 and is shippable: 7 of 7 local gates passed. The diagnostic uk_high_value_homes_support failed, as it did in build 1r. It does not block the release.
  • Both builds were published for inspection on 2026-11-16, the national one being uk-frs-calibration-attempt-20261115T104657Z-fcfac709. They are listed in their releases' folders but not current: the national release stays at eeb6093e and the dense release at 38605444. Promotion is a separate step.
  • An earlier build 2 attempt at this head was stopped after 98 seconds because of a local launch fault. Its Logbook row (f8debfe4e5a6eed6cdc7bdb26d0cc0c0e6b2942f419fd0cac1ac24e810811ffb, disposition discarded) comes before the build above in the uk/local chain.

The dense release records 3dbf54e19. Main (0402f5628) is merged in at 65606b552, with a merge commit, not a rebase, so that 3dbf54e19 stays reachable. Keep updating the branch with merges. The merge had no conflicts and no dependency changes, and 1,109 tests pass after it: the three changed test files and the engine-free register and gate tests.

Bus-journeys floor (#1207)

At 3a7067fa2, the floor refused the first dense build before its solve: dft_bus_journeys 1.1905% < 1.25%.

The cause is 06eeb2835, which binds England's PIP cases by age. It made dwp_pip the twenty-first family at the dense surface's nation and region grain. Under nation_grain_family_equal_sqrt_count, the bus journeys' share fell from 0.25/20 to 0.25/21. The national line has no floors, so the national arm could not show this.

The floor remains a check on the rule, never a change to its arithmetic (the 2026-10-14 ruling). At 1.15% it admits twenty-one families and refuses a twenty-second, so the next family bound at that grain comes back for a ruling. A new test, test_the_bus_floor_admits_21_region_families_and_refuses_22, pins this on the doctrine's own rule. The doctrine pin and the constant's note change with it. Energy's 2.5% floor is unchanged.

Four national renewals (#1301)

These are HMRC's dividend income bands from 20,000 to 50,000 (three entries) and its private pension income band from 12,570 to 15,000.

They expired on 16 November, and build 1r breaches all four. The sign-off would have kept them at that expiry beside approvals of the same day. test_uk_national_graph's review invariant refuses that: the latest approval must come before the earliest expiry. The gate clock cannot be set in the future either.

Each renewal is built with the drafting tool's own fit_reason and Window.entry. It cites build 1r's attempt and full gate report, and uses the sign-off's adjudication. Each is approved 2026-11-16, expires 2026-12-14, and applies to the local line only.

Gate-only sign-off from dense build 1r (#1301)

Build-1 attempt uk-local-candidate-f100-s42-20261116T010526Z-602766ae, full gate report sha256 405033ef697fda5e04ca08a28dba5cf59e44c56d7918738a0ee4164e76c3bbaf, gates evaluated on 2026-11-15.

Every signed row stays bound in the solve. Each entry is scoped to the local line and suppresses only that line's release fence. A build in which a signed row passes fails its entry as stale.

  • National target-fit rows (uk_target_fit): 3 added, 13 kept. The four renewed entries are among the kept.
  • Local target-fit cells (uk_local_target_fit): 12 added, 22 kept, 9 retired (entries build 2 would fail as stale).
  • Areas below the support floor (uk_local_area_support): 6 kept.

No entry was rescoped or left for review.

National target-fit rows: 1 stale-risk and 0 near-bound. Stale-risk counts signed rows within 0.01 above the bound, which build 2 fails as stale if it does not reproduce build 1 exactly. Near-bound counts unsigned rows within the same margin below it. Build 2 reproduced build 1r exactly. The local registers' counts are not printed here.

Window. Approved 2026-11-16 by juaristi22. The adjudication reads: "María's approval of the gate-only sign-off for the pair built from microcosm main on 2026-11-15, with the bus-floor ruling of 2026-11-16 (microcosm#1301, microcosm#1207)". Fit entries expire 2026-12-14 and area entries 2027-02-16.

Assembly deadline: 2026-11-16. Eleven local entries expire then, including uk_local_target_fit:age_structure/E07000008/age/20_30, uk_local_target_fit:age_structure/E14001131/age/20_30 and uk_local_target_fit:age_structure/E14001149/age/20_30. Build 2, its --verify and the assembly all ran on that day. Assembly and upload refuse a measure exclusion past its expiry.

Tests

uv run --no-sync pytest packages/microcosm-build/tests/engine_free/uk/test_uk_local_doctrine.py packages/microcosm-build/tests/engine_free/uk/test_uk_target_weights.py
uv run --no-sync pytest packages/microcosm-build/tests/engine_free/uk/test_uk_terminal_gates.py

Locally, at the commit that changed them:

  • at 79601a89c, the three target-weight test files: 30 passed;
  • at e1f70ddd3, the register and national-graph tests: 689 passed;
  • at 3dbf54e19, the terminal-gate and national-graph tests: 90 passed.

Changelog fragments: 1207-uk-dense-bus-floor-21-families.changed.md, 1301-uk-dense-renew-four-national-rows-20261116.changed.md, 1301-uk-dense-signoff-2026-11-16.changed.md.

Disclosure: this body carries:

  • gate ids and verdicts, and attempt ids, report digests and a Logbook row digest;
  • register entry counts, and the national register's stale-risk and near-bound counts;
  • the reviewer, the adjudication and policy dates;
  • the names of the renewed entries and of the entries that set the assembly deadline;
  • the floor's arithmetic, and test paths.

Per-row errors, the local registers' stale-risk and near-bound counts, and per-area support figures stay in the private review file, outside every git work tree.

🤖 Generated with Claude Code

juaristi22 and others added 4 commits November 16, 2026 01:02
María's ruling of 2026-11-16 ("Lower the floor"), after the first dense
build at 3a7067f was refused before its solve: dft_bus_journeys 1.1905%
< 1.25%. Binding England's PIP cases by age (06eeb28, her call 7 of
2026-11-11) made dwp_pip the twenty-first family at the dense surface's
nation and region grain, so the bus journeys' share under
nation_grain_family_equal_sqrt_count fell from 0.25/20 to 0.25/21. The
national line has no floors, so its PIP arm could not show it.

The floor stays a check on the rule, never a change to its arithmetic
(the 2026-10-14 ruling): 1.15% admits twenty-one families and refuses a
twenty-second. A new test pins that on the doctrine's own rule; the
doctrine pin and the constant's note move with it. Energy's 2.5% floor is
unchanged.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
María's approval of 2026-11-16 ("Yes, go ahead"). HMRC's dividend income
bands from 20,000 to 50,000 and its private pension band from 12,570 to
15,000 expired that day and the dense build 1r of that day still breaches
all four, so its gate-only sign-off would keep them at that expiry beside
approvals of the same day: test_uk_national_graph's review invariant (the
latest approval before the earliest expiry) refuses that, and a future
gate clock is refused too.

The renewals are built with draft_uk_dense_signoff.py's own fit_reason and
Window.entry, citing build 1r's attempt and full gate report, with the
sign-off's adjudication, approved 2026-11-16, expiring 2026-12-14, on the
local line only. The sign-off is drafted on top of this commit.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…m#1301)

María's approval of 2026-11-16 ("Approved, run build 2") of the draft from
draft_uk_dense_signoff.py on build 1r (attempt
uk-local-candidate-f100-s42-20261116T010526Z-602766ae), applied unchanged:
3 national target-fit rows and 12 local target-fit cells added, 9 local
entries build 2 would fail as stale retired, the rest kept; the terminal
gate test's key-set pins move with them. Every signed row stays bound in the
solve. Build 2's gate clock is 2026-11-16, and the assembly deadline is
2026-11-16.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A merge, not a rebase: the dense release published on 2026-11-16 records
3dbf54e, which must stay reachable from this branch.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

@vahid-ahmadi vahid-ahmadi left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I checked the three commits on 3a7067fa2 (git diff 3a7067fa2 3dbf54e19):

  • The floor arithmetic: 0.25/21 = 1.19% meets 1.15%, and 0.25/22 = 1.14% fails it. The new test pins both.
  • The register counts: the national register has 16 entries (3 added, 13 kept) and the local one 34 (12 added, 22 kept). The four renewals are approved 2026-11-16, expire 2026-12-14 and apply to the local line only.

Three points:

  1. Please land this with a merge commit, not a squash.
    • The dense release, and the certificate in PolicyEngine/policyengine.py#759, record 3dbf54e19. The body says to keep updating with merges but not how to land it, and Microcosm allows squash and rebase merges.
    • A squash leaves 3dbf54e19 off main. It also means main's solve differs from what built the release, since 65606b552 brings in main's calibrate/solve.py changes.
    • Could the body say so, so whoever merges sees it?
  2. 11 local target-fit entries expire on 2026-11-16, so they'll already be expired on main when this merges.
    • These are the entries approved 2026-10-19, including the three age_structure/.../20_30 entries the body names.
    • The four national entries in the same position were renewed, because the national graph's invariant would refuse them. The local register has no such check, so these stayed as they were.
    • If build 1r still breaches any of them, the next local build from main fails them as expired. Could they be renewed here on the same basis, or could the body note that the next local cut must re-sign them?
  3. The "Built and assembled" section says both builds are published but not current, with the releases staying at eeb6093e and 38605444. policyengine.py#759 says both were made current at 12:48 and 12:49 UTC on 16 Nov. Worth updating so the two PRs agree.

Review drafted with AI assistance and checked against the code at 3dbf54e and 65606b5.

@juaristi22
juaristi22 merged commit f1f5f73 into main Nov 16, 2026
12 checks passed
MaxGhenis added a commit that referenced this pull request Nov 16, 2026
Main brought five first-parent merges since 14797c6: #1640 (UK dense
bus floor for 21 families), #1642 (Belgium's ten gate entries in the gate
contract), #1643 (Belgium's Axiom engine pinned at 753ce7d9), #1608
(promotions notify collector; emitter-interface floor 0.16.0) and #1614
(LWI's US county facts pin). Main changed nothing under
packages/microcosm-fit or packages/microcosm-graph.

Two pin files conflicted, both through #1614; each block took main's text
plus this branch's own comment lines, placed after main's sentence:
- docs/evidence/spec-engine/us-f0-coverage.json: two spec_sha256 blocks,
  main's side (the branch has no comment there).
- packages/microcosm-build/tests/engine_workflow/us/test_us_multispine_pool_tool.py:
  main's #922 county-facts comment, then the branch's #1543 line, then
  spec_sha256.
No code or test file other than these two pins conflicted, and no other
file was changed by both sides.

The tools then wrote every value on the merged tree:
- tools/repin_us_spec_engine_contracts.py: the multispine pool
  spec_sha256 9ad067de (tools/generate_us_bundle_from_constants.py --check
  computes the same digest). It left the branch's EXPECTED_HASHES seed_map
  d5898b93 and seed_protocol 206be396 and the loader golden b7c843a1 as
  they were, and main's field-usage and coverage-tool pins as main has them.
- tools/spec_engine_coverage.py: us-f0-coverage.json's two spec_sha256
  lines (42401/42401 configuration fields; 41/41 inventory checks).
- fit.qrf/pins.json keeps the branch's values: the merged fit sources equal
  the branch's and the kernel's implementation hash reproduces, so
  tools/graph_parity_repin.py was not run.
The repinner's quote style was then reformatted with ruff format.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants