Skip to content

Measured compile: a unit variant is bound through its parent coproduct (std.measure census 147 -> 0) - #12132

Merged
gunbai-bot[bot] merged 5 commits into
mainfrom
session/bright-badger-206-unit-variant-binding
Sep 23, 2026
Merged

Measured compile: a unit variant is bound through its parent coproduct (std.measure census 147 -> 0)#12132
gunbai-bot[bot] merged 5 commits into
mainfrom
session/bright-badger-206-unit-variant-binding

Conversation

@gunbai-bot

@gunbai-bot gunbai-bot Bot commented Sep 23, 2026 •
edited
Loading

Copy link
Copy Markdown
Contributor

Defect

Two routes of one compiler answered one question in opposite ways. The question: does std.measure's Measure<Memory, One, Nat> resolve, where Memory and One are std.measure's own Quantity and scale unit variants passed as phantom type arguments?

  • Unmeasured resolve: admits them. The resolver's exact-one judgment lookup_unit_variant_phantom_type resolves the leaves.
  • Measured compile (the one compile_dag_diagnostic_census arms, gunbc.type_ref_hit_ne_bind_measure): charged every such leaf with a blocking UnresolvedType, because type_ref_measure_binding_authority had no arm for a zero-payload variant.

As a result, a census of a source that imports std.measure and nothing else reported 147 blocking rows on main. Every census whose closure reaches std.measure read red for a reason that isn't its subject. std.decimal reaches it through std.bytes → std.integer. That's how #12000's stack (cas_slot_keys importing std.decimal) turned test.claim.file_hold_plan_refusal_probe_witness_test the_harness_runs_and_a_clean_source_over_the_hold_store_is_clean red. It bisects to 9945e92, and its parent is clean.

Which side is right

The measure is wrong. A unit variant has no binding of its own: its binding is its declaring coproduct's. The resolver already owns the judgment of when a bare leaf names exactly one visible unit variant, so the measure must consume that judgment, not restate it.

Change

  • One judgment, relocated: lookup_unit_variant_phantom_type and UnitVariantPhantomLookup move from v1.compiler.infer_resolve to v1.compiler.infer_env, beside the unit_variant_index they judge. The move is needed so the measure can consume them without an import cycle. The resolver and the Rust emitter keep consuming the same function.
  • Measure arm: type_ref_measure_binding_authority gains an inline arm. A bare leaf is bound iff lookup_unit_variant_phantom_type answers UnitVariantPhantomPresent, meaning exactly one visible contribution with count 1. Zero contributions, two parents, one parent declaring it twice, a shadow that empties the key, and an unobserved index all answer unbound. No new pub fn is emitted, so the pub set moves and does not grow. No exemption is added, and the measure is armed exactly as before.
  • Superseded: the first cut's key-presence helper type_ref_unit_variant_parent_is_bound is gone. Key presence is not the resolver's contract: the white-box tests below fail under it.
  • Admission: the v1 purpose-test admission (gunbc.v1_maintenance_standing v1_seed_standing, instance DefectRepairDiagnosticsMayMove) is recorded as a // annotation beside the arm it admits (DESIGN §4c). No String data row is added.
  • Authority note: the note in gunbc.type_ref_hit_ne_bind_measure is updated to the three-arm authority.
  • Mirrors: regenerated with claim_executor --required-regen and installed only by three-way merge against a clean-main candidate. Main carries about 31 files of pre-existing mirror drift, which this PR does not take. compiler_tests.rs got its second generation after its generator mirror was installed.

Evidence

Census answer controls: six claims in test.claim.unit_variant_binding_authority_witness_test, all passing on this head (6/6, re-run in the sweep below).

Claim Fixed
a_source_importing_only_std_measure_is_clean_under_the_measured_compile (147 → 0) PASS
a_variant_of_an_imported_coproduct_is_bound_as_a_type_argument PASS
a_variant_whose_parent_is_not_bound_is_still_charged_by_the_measure PASS
a_variant_declared_by_two_visible_coproducts_is_refused_as_ambiguous PASS
a_variant_declared_twice_by_one_coproduct_is_refused PASS
a_variant_whose_only_parent_is_shadowed_is_refused PASS

The last three are answer controls, not discriminators of the measure: the resolver or emitter refuses those leaves before the measure decides.

Direct discriminating boundary: the white-box test generator ct_measure_unit_variant_exact_one_test (in v1.compiler.compiler_tests_rust) emits five Rust tests. Each drives type_ref_measure_binding_authority over an otherwise-empty TypeEnv, so only the contribution map decides:

  • one contribution → bound
  • two parents → unbound
  • count 2 → unbound
  • emptied key → unbound
  • unobserved index → unbound

All five pass. Replacing the judgment with bare key presence fails exactly the two-parent, count-2 and emptied-key tests. These white-box results and the mutation were run by the previous owner and accepted in review 5291522360; I did not re-run them.

Neighbouring witnesses: phantom_marker_type_argument_identity_witness_test passes 8/8, per the previous owner's run (it does not call the census, so it is outside the sweep).

Regression sweep: every witness file that calls compile_dag_diagnostic_census or compile_census_probe was run with claim_batch built from this head and with an unfixed claim_batch built from the branch's merge-base, 034aef0. The table and verdicts are in the PR comments: no comparable file differs from base, and the one red file fails identically on base.

Exact-head CI on 038e3ac11dd: clippy, compiler, emit-build, floor and witnesses all pass.

Recorded, not changed

Arm (1) treats a coproduct name reached only transitively as bound: it draws only the advisory UnlistedImportUse. The variant arm is stricter than that on the same fixture. This is the existing grain of arm (1), and this PR doesn't touch it.

🤖 Generated with Claude Code

gunbc-ci-auto-heal and others added 3 commits September 23, 2026 06:31
gunbc-ci-auto-heal @claude
Measured compile: a unit variant is bound through its parent coproduc…
017eb98 
…t, so std.measure's phantom type arguments stop reading as missing imports

The measure compile_dag_diagnostic_census arms (gunbc.type_ref_hit_ne_bind_measure) charged every
bare leaf without a binding of its own to a missing import. A zero-payload variant never has one --
its binding is its declaring coproduct's -- so std.measure's Measure<Memory, One, Nat> produced 147
blocking UnresolvedType rows in a census of a source importing std.measure alone, while the
unmeasured resolve admitted the same leaves: two routes of one compiler answering one question
opposite ways. type_ref_measure_binding_authority gains the arm the resolver's unit_variant_index
already answers (built only from bindings visible to the env); an unobserved index stays strict.

Mirror regenerated with --required-regen; only this change's hunk is installed (the candidate also
carries 31 files of pre-existing main drift, not taken here).

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
gunbc-ci-auto-heal
Merge remote-tracking branch 'origin/main' into session/bright-badger…
206d90a 
…-206-unit-variant-binding-2
gunbc-ci-auto-heal @claude
Admission note names its PR (#12132)
176d391 
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
briansrls
briansrls requested changes Sep 23, 2026
View reviewed changes

@briansrls briansrls 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.

SOURCE HOLD at GitHub's current exact head 176d391ee26ca2521464f51ff8a48570666ced32.

(The request named 36529891527…; that is not #12132's head. I bound this review to the current PR head reported by GitHub.)

The diagnosis is correct: the measured compile must not call std.measure's visible zero-payload variants unresolved when the ordinary resolver binds them through their parent coproduct. Keeping the change measurement-only, leaving an unobserved index strict, and regenerating only the corresponding stage0 hunk are the right boundaries. Two source walls remain.

1. An index key is not the same fact as one resolvable visible unit variant

type_ref_unit_variant_parent_is_bound currently returns true for any map_get(env.unit_variant_index, name) hit. The ordinary resolver's canonical lookup_unit_variant_phantom_type is stricter: it sums every contribution and resolves only when the total is exactly one; zero or several refuse. Its source explicitly states that duplicate names within one coproduct and multiple visible variants with the same leaf both refuse.

The index can also retain an outer key with an empty contribution map: unit_variant_index_remove_binding reinserts contributions_without_key(...) even when the last contribution was removed during shadowing. Thus map_get may be Present for total 0 as well as total >1.

This head can therefore suppress the measured UnresolvedType for a leaf the ordinary route does not bind—the same two-route fork the PR is intended to eliminate.

Use one shared cardinality judgment over the index, approximately:

index observed
AND sum(all contribution.count for this leaf) == 1

Prefer moving/exposing that judgment so both lookup_unit_variant_phantom_type and the measurement arm consume the same authority, rather than reimplementing it twice.

Required REDs/controls:

  • two visible parent coproducts each declare Alpha -> Alpha remains unbound/charged;
  • one coproduct contains duplicate zero-payload Alpha arms -> remains unbound/charged;
  • shadowing removes the last visible contributor while the outer index key remains -> remains unbound/charged;
  • exactly one visible parent contribution -> accepted;
  • the existing transitive-only parent control -> still charged.

2. The maintenance receipt incorrectly says there is no public-surface growth

The PR adds a new top-level v1 declaration, type_ref_unit_variant_parent_is_bound, and the generated stage0 mirror exports it as:

pub fn type_ref_unit_variant_parent_is_bound(...)

The repository's own v1 maintenance standing records that stage0-added pub fn visibility counts as PublicSurfaceGrowth; that refusal class dominates ordinary purpose-test admission unless an exact exception is authorized. The new admission note nevertheless says “no public surface growth.” That classification is false.

The fastest repair is also the lower-concept shape: this helper has one consumer, so inline the exact-one-contribution fold into the existing type_ref_measure_binding_authority and regenerate the mirror. That removes the new exported declaration entirely. Otherwise the PR needs an explicit exact-subject authorization for the public growth, not a note saying the class does not fire.

After repair, rerun the three unit-variant claims plus the new ambiguity/shadow controls, the 8/8 phantom-marker population, the full 74-file census regression comparison, and exact-head required CI. I found no separate blocker in the measurement-only scope, the main-vs-mirror hunk discipline, or the corrected v1 purpose justification once the public-surface statement is made honest.

gunbai-bot Bot pushed a commit that referenced this pull request Sep 23, 2026
gunbc-ci-auto-heal @claude
Known-red admission: file_hold_plan harness census is red on this sta…
b55acde 
…ck for a main-wide measured-compile defect, dissolving on gunbc#12132

The census of the hold-store harness reaches std.measure through cas_slot_keys' std.decimal import
(bisected to 9945e92), and the measured compile charges std.measure's own unit variants, passed
as phantom type arguments, as missing imports. Enrolled through explicit_witness_admission
known_red_probe rather than deleted or loosened; the trigger names the capability #12132 delivers.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
gunbc-ci-auto-heal @claude
Measure consumes the resolver's exact-one unit-variant judgment; admi…
757bb62 
…ssion recorded as an annotation, not a String row

- lookup_unit_variant_phantom_type / UnitVariantPhantomLookup move from v1.compiler.infer_resolve to
  v1.compiler.infer_env (owner of unit_variant_index); infer_resolve and emit_rust import it from there.
- type_ref_measure_binding_authority consumes it inline (no new declaration), so the measure binds exactly
  what resolve resolves instead of any key present in the index (review: side chat 5289240229).
- The v1 purpose-test admission moves from a String data row to a // annotation on the admitted
  declaration, and states the relocation of the pub set instead of denying growth (review 70469).
- Answer controls for two parents, a duplicate arm and a shadow-emptied key, labelled as answer controls:
  a key-presence mutation leaves them green because the resolver/emitter route refuses first.
- Mirrors regenerated with --required-regen and installed by three-way merge against a clean-main
  candidate (merge-file committed clean mine), so pre-existing main drift is not taken.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
briansrls
briansrls requested changes Sep 23, 2026
View reviewed changes

@briansrls briansrls 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.

SOURCE HOLD at exact head 757bb621402212be81da216c0648eb46ca0fe15b.

The source construction itself now answers both findings from review 5289240229 correctly:

  • lookup_unit_variant_phantom_type has moved to v1.compiler.infer_env, beside the unit_variant_index it judges, and the resolver, Rust emitter, and measured-compile authority all consume that ONE exact-one judgment. Zero, many, duplicate-arm and unobserved states therefore cannot diverge through a second implementation.
  • The one-consumer type_ref_unit_variant_parent_is_bound declaration is gone. The maintenance annotation now states the actual public-set relocation instead of denying public-surface movement, and the generated mirrors are limited to the intended three-way-merged hunks.
  • The six census claims are honest answer controls. I accept that the two-parent, duplicate-arm, and shadow fixtures refuse before the measured binding predicate becomes outcome-determining; labeling them as answer controls rather than pretending the key-presence mutation discriminates them is correct.

One evidence wall remains, because there IS a direct route to the measured predicate with a non-exact-one key.

Exercise type_ref_measure_binding_authority directly through the existing generated-Rust test authority

src/v1/compiler_tests_rust.dag already uses exactly the needed fixture shape: it clones crate::v1_compiler_infer_env::empty_type_env(), authors unit_variant_index, sets unit_variant_index_observed = true, and drives an infer/emit function. Use that existing test authority to call the measured predicate itself:

let mut env_value = (*crate::v1_compiler_infer_env::empty_type_env()).clone();
env_value.str_bindings = empty;
env_value.ancestry_str_bindings = empty;
env_value.unit_variant_index_observed = true;
env_value.unit_variant_index = ...;
let env = Rc::new(env_value);

crate::v1_compiler_infer_env::type_ref_measure_binding_authority(
    env,
    "Alpha".to_string(),
)

With the ordinary binding maps empty, arm 1 cannot intercept the query, so the test reaches the new unit-variant arm directly. Author at least these cases:

  1. one parent contribution with count = 1 -> true;
  2. two parent contributions, each count = 1 -> false;
  3. one parent contribution with count = 2 -> false;
  4. an outer Alpha key whose inner contribution map is empty -> false;
  5. optionally, the same one-contribution map with unit_variant_index_observed = false -> false.

The variant node may be the same harmless node reused by the existing unit-variant emitter fixture; only contribution cardinality is under judgment.

The discriminating mutation is now available: replace the helper call/answer with bare outer-key presence. The many- and empty-key tests must fail while the exact-one positive remains green. This is the specific wall the end-to-end census controls cannot provide, because their resolver/emitter diagnostics dominate first.

Keep the six census claims. Together the two layers establish different facts:

  • census claims: the compiler's externally observed answers remain correct;
  • direct host test: the measured binding authority itself really consumes exact-one rather than key presence.

This is not a request to invent a production route that reaches an otherwise dominated branch. It uses the repository's existing white-box seed-test boundary for the exact helper this PR relocates and makes load-bearing.

I found no further source blocker in the shared helper placement, resolver/emitter repointing, measurement-only scope, public-surface account, or mirror merge discipline.

Landing also remains contingent on the completed 74-file fixed-vs-base census comparison and exact-head required CI. The PR body is still the pre-repair account (type_ref_unit_variant_parent_is_bound, key-presence semantics, three claims, String admission row); refresh it to the shared exact-one helper, six answer controls, direct discriminating test, and final 74-file receipt.

@gunbai-bot gunbai-bot Bot mentioned this pull request Sep 23, 2026
Merged
gunbc-ci-auto-heal @claude
White-box tests: the measure binds a unit variant exactly when one co…
038e3ac 
…ntribution is visible

Five generated compiler_tests (src/v1/compiler_tests_rust.dag ct_measure_unit_variant_exact_one_test)
call type_ref_measure_binding_authority over an env with EMPTY str_bindings, so only the contribution
map decides: one parent once binds; two parents, one parent twice, an emptied outer key, and an
unobserved index do not. Mutating the arm to bare key presence fails exactly the two-parent, twice and
emptied-key tests. Mirrors: v1_compiler_compiler_tests_rust.rs from the first regen, compiler_tests.rs
from the second generation, each installed by three-way merge against a clean-main candidate.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
briansrls
briansrls approved these changes Sep 23, 2026
View reviewed changes

@briansrls briansrls 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.

SOURCE APPROVE at exact head 038e3ac11ddee7f7d29e03b96bfc98e0444f661c, superseding REQUEST_CHANGES review 5290446232.

The requested discriminating boundary is now present and correctly generated.

Direct measured-authority route

ct_measure_unit_variant_exact_one_test generates five Rust tests that call v1_compiler_infer_env::type_ref_measure_binding_authority directly. The helper starts from empty_type_env(), changes only unit_variant_index and unit_variant_index_observed, and therefore leaves str_bindings, ancestry_str_bindings, parents and the symbol index empty. The ordinary-binding arm cannot answer the query; the tests reach the unit-variant branch itself.

The supplied states cover the load-bearing partition:

  • one parent with count 1 -> bound;
  • two parents with count 1 each -> unbound;
  • one parent with count 2 -> unbound;
  • an outer key with an empty contribution map -> unbound;
  • an otherwise valid contribution under an unobserved index -> unbound.

The reported mutation from the shared exact-one judgment to outer-key presence is genuinely discriminating: the two-parent, count-2 and empty-map tests fail while the exact-one positive and unobserved-index control remain green. This is the white-box complement the prior review required; the six census claims can remain honest answer controls for the externally observed compiler behavior.

Generated-source chain

The delta from 757bb621402 is one commit and exactly three files:

  • src/v1/compiler_tests_rust.dag, the generator authority;
  • src/v1/stage0/src/v1_compiler_compiler_tests_rust.rs, its first-generation mirror;
  • src/v1/stage0/src/compiler_tests.rs, the second-generation generated test output.

The generated Rust tests match the .dag source in helper shape, five cases and expectations. No production compiler source or previously approved exact-one construction moved in this delta, and no unrelated stage0 drift is present.

I therefore accept the shared-helper placement, resolver/emitter/measure convergence, inlined measure arm, public-surface relocation account, six census controls, and the new direct mutation-tested evidence.

Conditions before landing

This is source approval, contingent on the evidence the author already names:

  1. Complete the full 74-file fixed-versus-unfixed census sweep on this exact source and report every file. Any pass/fail delta beyond the intended unit-variant false positives needs adjudication rather than being summarized as a count.
  2. Refresh the PR body: it currently describes the superseded key-presence helper, three-claim evidence set and old mirror story. Record the shared exact-one judgment, six answer controls, five white-box tests, mutation result, full sweep and exact head.
  3. Require exact-head CI green. At review time clippy is green; compiler, floor and emit-build are running.

No additional source change is required by this review. If the sweep or CI moves the SHA, rebind the approval to the resulting exact head.

briansrls
briansrls approved these changes Sep 23, 2026
View reviewed changes

@briansrls briansrls 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.

SOURCE APPROVE at exact head 038e3ac11ddee7f7d29e03b96bfc98e0444f661c, superseding REQUEST_CHANGES review 5290446232.

The requested discriminating white-box boundary is now present and correctly generated.

ct_measure_unit_variant_exact_one_test constructs an otherwise-empty TypeEnv, writes only unit_variant_index, and calls v1_compiler_infer_env::type_ref_measure_binding_authority directly. Because str_bindings and ancestry_str_bindings remain empty, the ordinary binding arm cannot intercept. The five cases cover exactly the judgment at issue:

  • one contribution from one parent -> bound;
  • one contribution from each of two parents -> unbound;
  • count two from one parent -> unbound;
  • present outer key with an empty contribution map -> unbound;
  • unobserved index -> unbound.

The reported mutation from the shared exact-one judgment to bare outer-key presence is discriminating: it fails the two-parent, count-two, and empty-map tests while preserving the unique positive and unobserved-index control. That closes the evidence gap without inventing a fake census route. The six end-to-end census claims remain useful answer controls beside it.

The source-of-truth/generated relationship also holds: the test is authored in src/v1/compiler_tests_rust.dag; the first-generation v1_compiler_compiler_tests_rust.rs emits it; the second-generation compiler_tests.rs contains the formatted Rust tests. I found no hand-only insertion or new public helper.

No further source finding remains in the shared exact-one helper placement, measured-compile arm, resolver/emitter repointing, v1 admission annotation, or mirror-install method.

This is source approval, not a landing receipt. Before merge, require:

  1. the complete 74-file fixed-vs-unfixed census sweep on this exact source family, with every divergence accounted for;
  2. the PR body refreshed from its superseded key-presence/three-control description to the shared exact-one implementation, six answer controls, five direct compiler tests, and full sweep table;
  3. exact-head required CI on 038e3ac11dd (none was registered when reviewed).

Any source change made while refreshing evidence requires an exact-head rebind.

@gunbai-bot

gunbai-bot Bot commented Sep 23, 2026

Copy link
Copy Markdown
Contributor Author

Census regression sweep at exact head 038e3ac11dd (partial: 3 files rerunning)

Every witness file that calls compile_dag_diagnostic_census or compile_census_probe (78 on this head: the 74 comparable ones, this PR's new witness file, and 3 hits with no test fn). Each was run with claim_batch built from this head and with an unfixed claim_batch built from 034aef0, this branch's actual merge-base. Main was merged into the branch at 206d90a, so a 4716b4c base would have mixed unrelated main changes into the comparison. Each binary was built with its own CARGO_TARGET_DIR. Verdicts compare the per-claim PASS/FAIL sets.

Result so far: 71/71 fully-run comparable files match base claim for claim, and no file differs. The only red, algebra_receiver_alias_witness_test (3 FAIL), fails identically on the unfixed binary, so it is pre-existing. Rerunning now:

  • fleet_converge_launch_scope_non_interference_test: the base run was OOM-killed (rc=137) while both sweeps shared the host; fixed was 5/5.
  • the two src/v1/tests/claim/*occurrence_control_test files: they need --source-root src/v1, and the sweep did not pass it on either side.
file roster fixed P/F base P/F verdict
dag/gunbc/explicit_witness_admission.dag 0 0/0 0/0 no test fns
dag/test/claim/admit_callers_roster_integrity_witness_test.dag 6 6/0 6/0 match
dag/test/claim/algebra_expansion_evidence_witness_test.dag 2 2/0 2/0 match
dag/test/claim/algebra_receiver_alias_witness_test.dag 4 1/3 1/3 match
dag/test/claim/algebra_receiver_callable_witness_test.dag 11 5/6 5/6 match
dag/test/claim/argument_shape_at_a_declared_position_probe_test.dag 4 2/2 2/2 match
dag/test/claim/arity_zero_import_refusal_witness_test.dag 2 2/0 2/0 match
dag/test/claim/brand_nominal_identity_witness_test.dag 7 7/0 7/0 match
dag/test/claim/callable_candidate_ambiguity_witness_test.dag 5 5/0 5/0 match
dag/test/claim/census_app_acquisition_refusal_probe_witness_test.dag 4 4/0 4/0 match
dag/test/claim/citation_cause_subject_disjointness_witness_test.dag 2 2/0 2/0 match
dag/test/claim/compile_accepted_unevaluable_program_control_test.dag 6 3/3 3/3 match
dag/test/claim/compile_diagnostic_census_witness_test.dag 8 8/0 8/0 match
dag/test/claim/data_class_refusal_probe_witness_test.dag 8 8/0 8/0 match
dag/test/claim/declared_type_expected_type_path_witness_test.dag 19 13/6 13/6 match
dag/test/claim/declared_type_inhabitance_direct_call_witness_test.dag 33 32/1 32/1 match
dag/test/claim/declared_type_inhabitance_list_element_witness_test.dag 6 6/0 6/0 match
dag/test/claim/direct_call_argument_type_witness_test.dag 8 8/0 8/0 match
dag/test/claim/duplicate_definition_binding_probe_test.dag 2 2/0 2/0 match
dag/test/claim/duplicate_record_field_refusal_witness_test.dag 3 3/0 3/0 match
dag/test/claim/effectful_item_kind_collapse_witness_test.dag 5 5/0 5/0 match
dag/test/claim/emitter_unprojectable_construct_refuse_witness_test.dag 5 5/0 5/0 match
dag/test/claim/empty_list_element_fabrication_witness_test.dag 2 2/0 2/0 match
dag/test/claim/fabric/fabric_output_contract_resolution_witness_test.dag 11 11/0 11/0 match
dag/test/claim/fabric_m0_job_ending_seal_test.dag 4 4/0 4/0 match
dag/test/claim/fabric_m0_origin_readback_seal_test.dag 7 7/0 7/0 match
dag/test/claim/file_hold_plan_refusal_probe_witness_test.dag 6 6/0 6/0 match
dag/test/claim/fleet/fleet_converge_launch_scope_non_interference_test.dag 5 5/0 0/0 TRUNCATED
dag/test/claim/fleet_observation_seal_test.dag 9 9/0 9/0 match
dag/test/claim/generic_effectful_declaration_wall_witness_test.dag 3 3/0 3/0 match
dag/test/claim/generic_lambda_parameter_binding_witness_test.dag 9 9/0 9/0 match
dag/test/claim/guarantee_floor_class_probe_witness_test.dag 17 14/3 14/3 match
dag/test/claim/guarantee_probe_corpus_witness_test.dag 53 50/3 50/3 match
dag/test/claim/import_admission_closure_membership_witness_test.dag 4 4/0 4/0 match
dag/test/claim/infer_nullary_local_callable_witness_test.dag 6 6/0 6/0 match
dag/test/claim/infer_record_field_completeness_zero_field_witness_test.dag 5 5/0 5/0 match
dag/test/claim/infer_record_lit_variant_field_witness_test.dag 11 11/0 11/0 match
dag/test/claim/inventory_admitted_ledger_seal_livetree_witness_test.dag 3 3/0 3/0 match
dag/test/claim/lambda_receiver_type_witness_test.dag 2 2/0 2/0 match
dag/test/claim/long/where_refinement_enforcement_witness_test.dag 48 42/6 42/6 match
dag/test/claim/machine_intake/no_fallback_plan_ineligibility_wall_test.dag 4 4/0 4/0 match
dag/test/claim/machine_intake/predictive_claim_construction_witness_test.dag 4 4/0 4/0 match
dag/test/claim/marker_argument_not_consulted_witness_test.dag 4 4/0 4/0 match
dag/test/claim/match_exhaustiveness_coproduct_witness_test.dag 26 26/0 26/0 match
dag/test/claim/namespace_step0_subject_collector_witness_test.dag 15 15/0 15/0 match
dag/test/claim/operation_admit_callers_inert_witness_test.dag 2 1/1 1/1 match
dag/test/claim/optional_at_required_position_witness_test.dag 7 7/0 7/0 match
dag/test/claim/optional_cast_wall_witness_test.dag 4 4/0 4/0 match
dag/test/claim/payload_arm_excess_field_admission_test.dag 11 4/7 4/7 match
dag/test/claim/pipeline_transport_emit_rest_shell_witness_test.dag 25 25/0 25/0 match
dag/test/claim/printed_chassis_admitted_realization_seal_witness_test.dag 12 12/0 12/0 match
dag/test/claim/printed_chassis_manufacturing_manifest_livetree_witness_test.dag 3 3/0 3/0 match
dag/test/claim/process_lease_plan_is_not_a_refusal_witness_test.dag 2 2/0 2/0 match
dag/test/claim/qualified_pattern_head_witness_test.dag 2 2/0 2/0 match
dag/test/claim/record_decl_spelling_identity_witness_test.dag 7 7/0 7/0 match
dag/test/claim/record_literal_call_arg_handoff_witness_test.dag 3 3/0 3/0 match
dag/test/claim/refinement_seam_enforcement_witness_test.dag 7 7/0 7/0 match
dag/test/claim/review_sheet_batch_actuator_witness_test.dag 28 28/0 28/0 match
dag/test/claim/review_sheet_header_init_refusal_probe_witness_test.dag 6 6/0 6/0 match
dag/test/claim/review_sheet_header_schema_witness_test.dag 12 12/0 12/0 match
dag/test/claim/review_sheet_tab_refusal_probe_witness_test.dag 10 10/0 10/0 match
dag/test/claim/self_host_equality_admission_witness_test.dag 9 9/0 9/0 match
dag/test/claim/sole_constructor_completeness_audit_probe_test.dag 26 20/6 20/6 match
dag/test/claim/sole_constructor_type_head_exposure_witness_test.dag 6 5/1 5/1 match
dag/test/claim/spark/spark_declared_unit_digest_seal_probe_test.dag 3 3/0 3/0 match
dag/test/claim/test_reference_wall_witness_test.dag 11 11/0 11/0 match
dag/test/claim/transport_emission_not_modeled_witness_test.dag 8 8/0 8/0 match
dag/test/claim/transport_field_refusal_witness_test.dag 13 13/0 13/0 match
dag/test/claim/type_argument_kind_inhabitance_witness_test.dag 6 5/1 5/1 match
dag/test/claim/type_ref_hit_ne_bind_measure_witness_test.dag 8 8/0 8/0 match
dag/test/claim/type_reference_resolve_changeover_equivalence_witness_test.dag 4 4/0 4/0 match
dag/test/claim/unit_variant_binding_authority_witness_test.dag 6 6/0 -/- new on branch
dag/test/claim/where_refinement_predicate_vocabulary_witness_test.dag 9 9/0 9/0 match
dag/test/manual/command_runner_local_argv_receipt_test.dag 6 0/6 0/6 match
dag/test/retirement/model.dag 0 0/0 0/0 no test fns
src/v1/compiler_tests_rust.dag 0 0/0 0/0 no test fns
src/v1/tests/claim/bare_variant_reference_occurrence_control_test.dag 1 0/0 0/0 TRUNCATED
src/v1/tests/claim/type_declaration_occurrence_control_test.dag 1 0/0 0/0 TRUNCATED

@gunbai-bot

gunbai-bot Bot commented Sep 23, 2026

Copy link
Copy Markdown
Contributor Author

Census regression sweep at exact head 038e3ac11dd: FINAL

claim_batch built from this head vs an unfixed claim_batch built from the branch's merge-base 034aef0. Each binary was built with its own CARGO_TARGET_DIR, and each file compares its per-claim PASS/FAIL sets.

74/74 comparable files match base claim for claim. No file differs. The one red, algebra_receiver_alias_witness_test (3 FAIL), fails identically on the unfixed binary, so it is pre-existing and not this PR's. Since no answer moved in any existing witness, there were no false-charged leaves to show. This PR's own witness file passes 6/6 on this head; it has no base counterpart. Three census hits declare no test fn.

Reruns since the partial table: fleet_converge_launch_scope_non_interference_test on base, which the concurrent sweep had OOM-killed (now 5/5 on both), and the two src/v1 occurrence controls, run with --source-root src/v1 on both sides (1/1 each).

file roster fixed P/F base P/F verdict
dag/gunbc/explicit_witness_admission.dag 0 0/0 0/0 no test fns
dag/test/claim/admit_callers_roster_integrity_witness_test.dag 6 6/0 6/0 match
dag/test/claim/algebra_expansion_evidence_witness_test.dag 2 2/0 2/0 match
dag/test/claim/algebra_receiver_alias_witness_test.dag 4 1/3 1/3 match
dag/test/claim/algebra_receiver_callable_witness_test.dag 11 5/6 5/6 match
dag/test/claim/argument_shape_at_a_declared_position_probe_test.dag 4 2/2 2/2 match
dag/test/claim/arity_zero_import_refusal_witness_test.dag 2 2/0 2/0 match
dag/test/claim/brand_nominal_identity_witness_test.dag 7 7/0 7/0 match
dag/test/claim/callable_candidate_ambiguity_witness_test.dag 5 5/0 5/0 match
dag/test/claim/census_app_acquisition_refusal_probe_witness_test.dag 4 4/0 4/0 match
dag/test/claim/citation_cause_subject_disjointness_witness_test.dag 2 2/0 2/0 match
dag/test/claim/compile_accepted_unevaluable_program_control_test.dag 6 3/3 3/3 match
dag/test/claim/compile_diagnostic_census_witness_test.dag 8 8/0 8/0 match
dag/test/claim/data_class_refusal_probe_witness_test.dag 8 8/0 8/0 match
dag/test/claim/declared_type_expected_type_path_witness_test.dag 19 13/6 13/6 match
dag/test/claim/declared_type_inhabitance_direct_call_witness_test.dag 33 32/1 32/1 match
dag/test/claim/declared_type_inhabitance_list_element_witness_test.dag 6 6/0 6/0 match
dag/test/claim/direct_call_argument_type_witness_test.dag 8 8/0 8/0 match
dag/test/claim/duplicate_definition_binding_probe_test.dag 2 2/0 2/0 match
dag/test/claim/duplicate_record_field_refusal_witness_test.dag 3 3/0 3/0 match
dag/test/claim/effectful_item_kind_collapse_witness_test.dag 5 5/0 5/0 match
dag/test/claim/emitter_unprojectable_construct_refuse_witness_test.dag 5 5/0 5/0 match
dag/test/claim/empty_list_element_fabrication_witness_test.dag 2 2/0 2/0 match
dag/test/claim/fabric/fabric_output_contract_resolution_witness_test.dag 11 11/0 11/0 match
dag/test/claim/fabric_m0_job_ending_seal_test.dag 4 4/0 4/0 match
dag/test/claim/fabric_m0_origin_readback_seal_test.dag 7 7/0 7/0 match
dag/test/claim/file_hold_plan_refusal_probe_witness_test.dag 6 6/0 6/0 match
dag/test/claim/fleet/fleet_converge_launch_scope_non_interference_test.dag 5 5/0 5/0 match
dag/test/claim/fleet_observation_seal_test.dag 9 9/0 9/0 match
dag/test/claim/generic_effectful_declaration_wall_witness_test.dag 3 3/0 3/0 match
dag/test/claim/generic_lambda_parameter_binding_witness_test.dag 9 9/0 9/0 match
dag/test/claim/guarantee_floor_class_probe_witness_test.dag 17 14/3 14/3 match
dag/test/claim/guarantee_probe_corpus_witness_test.dag 53 50/3 50/3 match
dag/test/claim/import_admission_closure_membership_witness_test.dag 4 4/0 4/0 match
dag/test/claim/infer_nullary_local_callable_witness_test.dag 6 6/0 6/0 match
dag/test/claim/infer_record_field_completeness_zero_field_witness_test.dag 5 5/0 5/0 match
dag/test/claim/infer_record_lit_variant_field_witness_test.dag 11 11/0 11/0 match
dag/test/claim/inventory_admitted_ledger_seal_livetree_witness_test.dag 3 3/0 3/0 match
dag/test/claim/lambda_receiver_type_witness_test.dag 2 2/0 2/0 match
dag/test/claim/long/where_refinement_enforcement_witness_test.dag 48 42/6 42/6 match
dag/test/claim/machine_intake/no_fallback_plan_ineligibility_wall_test.dag 4 4/0 4/0 match
dag/test/claim/machine_intake/predictive_claim_construction_witness_test.dag 4 4/0 4/0 match
dag/test/claim/marker_argument_not_consulted_witness_test.dag 4 4/0 4/0 match
dag/test/claim/match_exhaustiveness_coproduct_witness_test.dag 26 26/0 26/0 match
dag/test/claim/namespace_step0_subject_collector_witness_test.dag 15 15/0 15/0 match
dag/test/claim/operation_admit_callers_inert_witness_test.dag 2 1/1 1/1 match
dag/test/claim/optional_at_required_position_witness_test.dag 7 7/0 7/0 match
dag/test/claim/optional_cast_wall_witness_test.dag 4 4/0 4/0 match
dag/test/claim/payload_arm_excess_field_admission_test.dag 11 4/7 4/7 match
dag/test/claim/pipeline_transport_emit_rest_shell_witness_test.dag 25 25/0 25/0 match
dag/test/claim/printed_chassis_admitted_realization_seal_witness_test.dag 12 12/0 12/0 match
dag/test/claim/printed_chassis_manufacturing_manifest_livetree_witness_test.dag 3 3/0 3/0 match
dag/test/claim/process_lease_plan_is_not_a_refusal_witness_test.dag 2 2/0 2/0 match
dag/test/claim/qualified_pattern_head_witness_test.dag 2 2/0 2/0 match
dag/test/claim/record_decl_spelling_identity_witness_test.dag 7 7/0 7/0 match
dag/test/claim/record_literal_call_arg_handoff_witness_test.dag 3 3/0 3/0 match
dag/test/claim/refinement_seam_enforcement_witness_test.dag 7 7/0 7/0 match
dag/test/claim/review_sheet_batch_actuator_witness_test.dag 28 28/0 28/0 match
dag/test/claim/review_sheet_header_init_refusal_probe_witness_test.dag 6 6/0 6/0 match
dag/test/claim/review_sheet_header_schema_witness_test.dag 12 12/0 12/0 match
dag/test/claim/review_sheet_tab_refusal_probe_witness_test.dag 10 10/0 10/0 match
dag/test/claim/self_host_equality_admission_witness_test.dag 9 9/0 9/0 match
dag/test/claim/sole_constructor_completeness_audit_probe_test.dag 26 20/6 20/6 match
dag/test/claim/sole_constructor_type_head_exposure_witness_test.dag 6 5/1 5/1 match
dag/test/claim/spark/spark_declared_unit_digest_seal_probe_test.dag 3 3/0 3/0 match
dag/test/claim/test_reference_wall_witness_test.dag 11 11/0 11/0 match
dag/test/claim/transport_emission_not_modeled_witness_test.dag 8 8/0 8/0 match
dag/test/claim/transport_field_refusal_witness_test.dag 13 13/0 13/0 match
dag/test/claim/type_argument_kind_inhabitance_witness_test.dag 6 5/1 5/1 match
dag/test/claim/type_ref_hit_ne_bind_measure_witness_test.dag 8 8/0 8/0 match
dag/test/claim/type_reference_resolve_changeover_equivalence_witness_test.dag 4 4/0 4/0 match
dag/test/claim/unit_variant_binding_authority_witness_test.dag 6 6/0 -/- new on branch
dag/test/claim/where_refinement_predicate_vocabulary_witness_test.dag 9 9/0 9/0 match
dag/test/manual/command_runner_local_argv_receipt_test.dag 6 0/6 0/6 match
dag/test/retirement/model.dag 0 0/0 0/0 no test fns
src/v1/compiler_tests_rust.dag 0 0/0 0/0 no test fns
src/v1/tests/claim/bare_variant_reference_occurrence_control_test.dag 1 1/0 1/0 match
src/v1/tests/claim/type_declaration_occurrence_control_test.dag 1 1/0 1/0 match

@gunbai-bot
gunbai-bot Bot added this pull request to the merge queue Sep 23, 2026
Merged via the queue into main with commit 951e536 Sep 23, 2026
5 checks passed
@gunbai-bot
gunbai-bot Bot deleted the session/bright-badger-206-unit-variant-binding branch September 23, 2026 21:35
gunbai-bot Bot pushed a commit that referenced this pull request Sep 23, 2026
gunbc-ci-auto-heal @claude
Regenerate the emit mirror against main at 951e536
57d31e1 
The merge of main (which carries #12132 and #11985, both touching the emitter) took main's
pristine v1_compiler_emit_rust.rs; --required-regen then produced the candidate from the joined
05_emit_rust.dag and only that file was installed. A second regeneration round produced a
byte-identical file (per-file fixed point, sha256 6297b8140753fc63). Main's regen round still
refuses on #12011's release_locus_seed_constants_generated.rs; that belongs to the
mirror-reconciliation lane and nothing is installed for it here.

All nine Pkg15 witness arms green on the regenerated mirror.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
gunbai-bot Bot pushed a commit that referenced this pull request Sep 29, 2026
gunbc-ci-auto-heal @claude
Quarantine: file_hold probe leaves the quarantine; it passes on main …
1479f86 
…(its dissolution, #12132, landed)

The #filter bare-provider refusal that hid its verdict was already retired by #12609. On main
785934a the probe PASSES, which its own dissolution names as the row's deletion condition.
The nine algebra_receiver probes remain red (re-measured on the same main).

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
@gunbai-bot gunbai-bot Bot mentioned this pull request Sep 29, 2026
Merged
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@briansrls briansrls briansrls approved these changes

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

1 participant

@briansrls