Skip to content

Pin Belgium's Axiom engine at 753ce7d9 and refresh its input surface - #1643

Merged
MaxGhenis merged 1 commit into
mainfrom
be/axiom-surface-refresh
Nov 16, 2026
Merged

MaxGhenis merged 1 commit into
mainfrom
be/axiom-surface-refresh

Conversation

@MaxGhenis

@MaxGhenis MaxGhenis commented Nov 16, 2026 •

Copy link
Copy Markdown
Contributor

Ready. Round 1 of the Subfleet hard-tier review (20261116-095720-be-axiom-pin-r1) approved 0f628e7 with no findings, so the review has finished. It reproduced the Belgian surface, both concept-coverage outputs and both country-spec goldens, and checked every engine citation at 753ce7d9. The NZ hub OK'd the four shared files below at 0f628e7. Waiting on CI.

Belgium's Axiom input surface moves to the pins its bindings already carry for RuleSpec and to New Zealand's engine pin, and Belgium's bindings now pin that engine. This is the follow-up that #1602's engine.status names.

What moves

Was Now
be.json rulespec.commit b105e2b3 632e1da1 (the bindings' RuleSpec pin, rulespec-be main's head)
be.json engine.commit 04315d94 753ce7d9 (NZ's pin)
be/axiom_rules_bindings.json engine.commit, wheel_sha256 null 753ce7d9, 054c9196… (the wheel the NZ hub built at 753ce7d9)

Nothing in the surface changes but the two pins. All 116 modules keep their bytes, inputs, status and canonical names: the five rulespec-be country trees are identical at b105e2b3 and 632e1da1. So PR 4a's closure counts (277 roots: 39 encoded, 238 defaulted) and every default stand.

How it was generated. tools/refresh_axiom_input_surface.py --country be in dense mode, with the NZ hub's 753ce7d9 CLI and dense extension. The engine directory matches the commit file for file (336 files), and the wheel's sha256 is the one nz/axiom_rules_bindings.json pins. Two checks back the setup: the same run with --check --country nz reproduces main's nz.json byte for byte, and the Belgian run's differential found 0 input, status or catalog mismatches between the dense surface and the CLI walk (147 of 156 entity programs dense-compiled; 7 modules dense-unsupported, as before).

What follows from it

  • The closure's engine evidence, re-cited at 753ce7d9. Every claim but one holds with new line numbers. The one that changes: at 04315d94 a RuleSpec document had no inputs section; at 753ce7d9 it has one (src/rulespec.rs:664-668). The block still gives no input a default: the engine keeps it only to check only_related and selected-year reads and to read elections. None of the twelve bound modules, nor any other surface module, has an inputs key. The evidence now also cites where the first row error fails the whole request (src/dense.rs:1539, python-ext/src/lib.rs:404-409), as NZ's does. The four engine_source_pins move to 753ce7d9.
  • The bindings' notes: period_note cites parse_period_kind (python-ext/src/lib.rs:985-992) and views_note cites parse_dtype (src/formula.rs:1419-1430) at 753ce7d9. Both behave as at 04315d94: a custom period kind keeps its name, and a Count dtype still falls through to Decimal. The RuleSpec verification note no longer needs its ancestor check, since the surface is now generated at the pin; verified_on is 2026-11-16.
  • The concept mapping's engine_version is the RuleSpec commit it was reviewed against, and test_axiom_concept_mapping.py requires it to equal the surface's, so it moves to 632e1da1 (not to the engine commit; NZ's mapping likewise records its surface's RuleSpec commit). Five unmapped reasons say "rulespec-be at 632e1da1"; their claims rest on module text that is identical at both commits. The concept-coverage golden and docs/concept-coverage/axiom-be.md are regenerated with tools/refresh_concept_coverage.py.
  • Tests. test_the_engine_pin_is_the_engine_that_generated_the_surface replaces the test that required null pins: the pin must equal the surface's and the closure's engine commit, and the wheel digest must be a sha256. test_belgium_still_refuses_to_activate drops the two pin gaps and now asserts no axiom_rules_bindings gap; the benefit-unit rule and the transport graph remain gaps, so Belgium stays inactive.
  • The country-spec golden moves by two resource hashes (axiom_input_closure.json, axiom_rules_bindings.json) and the fingerprint.

Shared files

The surface README and the refresh tool's docstring now say both snapshots record 753ce7d9, give Belgium's new RuleSpec commit, and build the extension with --locked. The note that explained the missing --locked at 04315d94 is kept as one parenthesis, since no snapshot records that engine now. The NZ hub's record of its 054c9196 wheel names maturin build --release --locked. Two shared test files change only for Belgium: a comment in test_axiom_concept_mapping.py and test_belgium_still_refuses_to_activate in test_be_target_references.py. _separation_violations, which NZ loads by path, is untouched. The NZ hub checked these four files' hunks and blobs at 0f628e7 and OK'd them; its OK covers those files at that head.

Verification

At 0f628e7 (on main 9504278), one file per process with -n 0:

  • test_axiom_concept_mapping.py: 15 passed.
  • test_be_axiom_input_closure.py: 20 passed.
  • test_be_target_references.py: 156 passed, 32 skipped (the Chronicle feed audit; BE_CHRONICLE_ARTIFACT unset).
  • test_transport_holdout_separation.py: 51 passed.
  • test_concept_mapping.py, test_country_spec.py and test_refresh_axiom_input_surface.py: all passed.
  • tools/refresh_concept_coverage.py --engine axiom-be --engine axiom-nz --check and tools/refresh_axiom_input_surface.py --country be --check at the two pins: up to date.
  • ruff check and ruff format --check (0.15.16, the lockfile's version) pass on the four changed Python files.

In a run by the reader that mapped the change (below), with only be.json changed, three tests failed: the mapping's RuleSpec commit test and two closure tests (the mapping and surface pins, and the engine-source pins). The rest of the change answers them.

Two independent readers drafted the re-citations from the 753ce7d9 source, and two more checked each draft, one for the cited lines and one for the substance. Their two corrections are taken: the RulesDocument span is narrowed to the inputs field, and "can't declare a default" becomes "gives no input a default", since an inert default key is accepted, not refused. A third reader mapped every file the move touches. No kernel identity or gate-battery binding source hashes these JSON files. The builder source digest (code_identity.py) does change, since it covers all packaged JSON, but it is computed live and has no committed value.

No engine was built here, and nothing runs the bindings yet: Belgium has no transport graph (PR 4b).

🤖 Generated with Claude Code

Belgium's input surface (be.json) is regenerated at rulespec-be 632e1da1,
the bindings' RuleSpec pin, and axiom-rules-engine 753ce7d9, New Zealand's
engine pin. The refresh used the NZ hub's 753ce7d9 CLI and dense extension
(the engine tree matches the commit file for file; wheel sha256
054c9196...). The same setup regenerates main's nz.json byte for byte. No
module's bytes, inputs, status or canonical names move, only
rulespec.commit and engine.commit: the five rulespec-be country trees are
identical at b105e2b3 and 632e1da1.

- axiom_rules_bindings.json pins engine commit 753ce7d9 and the wheel's
  sha256, as nz/axiom_rules_bindings.json does. The surface is now
  generated at the RuleSpec pin, so the verification note drops its
  ancestor check, and verified_on moves to 2026-11-16. The period and
  views notes cite parse_period_kind and parse_dtype at 753ce7d9.
- axiom_input_closure.json records the new surface pins and the mapping's
  new RuleSpec commit, and engine_optional_evidence is re-cited at
  753ce7d9. One sentence changes in substance: at 753ce7d9 a RuleSpec
  document has an inputs block, which gives no input a default (the
  engine keeps it only for only_related and selected-year checks and
  for elections), and none of the 116 surface modules has one. The
  evidence also says that the first row error fails the whole request.
  The four engine source pins move to 753ce7d9.
- axiom_concept_mappings/be.json takes the surface's RuleSpec commit as
  its engine_version, as test_axiom_concept_mapping.py requires, and five
  unmapped reasons now say "at 632e1da1". The concept-coverage golden and
  doc are regenerated.
- The engine-pin test now checks that the pin is the surface's and the
  closure's engine. Belgium's activation test no longer expects the two
  pin gaps; a benefit-unit rule and a transport graph remain. The
  country-spec golden moves by two resource hashes and the fingerprint.
- The surface README and the refresh tool's docstring say both snapshots
  record 753ce7d9, build the extension with --locked, and give Belgium's
  new RuleSpec commit.

Belgium stays inactive.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@MaxGhenis
MaxGhenis marked this pull request as ready for review November 16, 2026 15:31
@MaxGhenis
MaxGhenis merged commit 5f18174 into main Nov 16, 2026
12 checks passed
@MaxGhenis
MaxGhenis deleted the be/axiom-surface-refresh branch November 16, 2026 18:32
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.

1 participant