Floor resolve realization: identity-join obligation roster (0B)

[briansrls]

Summary

Replaces the floor finalization resolve-count pin with an identity-join obligation model in eight files:

• gunbc.ci_materialization — ResolveObligation / ResolveRealization carriers, FloorFinalization.expected_resolve_obligations, resolve_realization_finalization_holds law
• ci_floor_plan.dag — transports ci_floor_expected_resolve_obligations() unchanged (no HAND-Rust subject literals)
• claim_executor.rs — records cold/warm realization at reuse sites; finalization refuses unattributed physical resolves, duplicate cold per rostered entry, unrostered observed subjects (identity join, both directions), and missing obligations (unevaluable on truncated walks); preserves resolve observations on every post-acquisition outcome (semantic witness failure no longer fabricates a missing-realization refusal)
• Witnesses — floor_resolve_realization_witness_test.dag (8 admit/refuse controls) + ci_floor_plan_witness_test.dag transport checks + mirrored Rust unit tests

Class A generated-artifact long-lane relocation was removed from this branch (superseded by #7823).

Production receipt (truncated walk, run 30975530303)

Before this PR, a truncated walk that skipped the native-bundle batch reported a fabricated debt-change diagnosis (floor resolve count 1 differs from declared 2: duplicate-computation debt changed). On the same failure shape after the fix:

FLOOR-FINALIZATION-REFUSED: floor resolve obligations not fully scheduled:
1 obligation subject(s) never ran (walk stopped before dependent batches);
count law unevaluable — NativeBundleEscapingEntry
(src/v2/test/claim/execution/native_selected_witness_bundle_production.dag::native_selected_logic_production_spec)

The count law is unevaluable (not a roster debt change), the cause is named (walk stopped before dependent batches), and the missing obligation subject is located.

Inherited main red (falsifier_workflow_witness_test.dag)

6477c93437 fixes the call site: falsifier_cold_control_steps_follow_floor_in_job_order now calls falsifier_floor_job_steps() (the existing zero-arg wrapper that supplies job_id: "falsifier") instead of falsifier_job_steps() with no arguments. One defect, two failing witnesses (red_floor_workflow_fixture_cold_controls_still_execute delegates to the same function).

Root cause (not closed by this call-site fix): on main, falsifier_job_steps() — a call supplying no arguments to fn falsifier_job_steps(job_id: String) — compiled clean and failed only at runtime as undefined variable: job_id. That is the ordinary compiler floor failing in the DESIGN §4b sense: application arity should bind in exact bijection. #7834 walls the surplus direction (CallPositionalSurplus, too many arguments); this is the deficit direction and nothing catches it yet. Raised with sharp-seal-527 for whether it belongs in #7834 or a follow-up — the language-layer hole must stay ranked even though the symptom is repaired here.

Identity-join completeness (6477c93437)

Finalization now refuses unrostered observed obligation subjects by name (entry::function) independently of the roster-minus-derived check. The count-equality arm (obligations.len() != declared_resolve_count()) is removed — entailed by the two identity joins and could hide a surplus when a missing roster subject balanced the count. RED: floor_finalization_names_unrostered_subject_when_roster_subject_also_missing.

Test plan

• cargo test -p v1-compiler --bin claim_executor floor_finalization
• cargo test -p v1-compiler --bin claim_executor claim_result_for_outcome_preserves
• cargo test -p v1-compiler --bin claim_executor floor_finalization_refuses_unrostered
• cargo test -p v1-compiler --bin claim_executor memo_warm_attaches_shared_pool
• cargo test -p v1-compiler --lib nfr_roster_receipt
• CI floor (obligation receipt + finalization law)

Worker attestation

• Title/body describe current eight-file scope only
• Tests executed locally on observation-retention + finalization paths
• No secrets / surprise commits

[cursor]

Bugbot is not enabled for your account, so this pull request was not reviewed.

Enable Bugbot in the Cursor dashboard to get automatic reviews on future PRs.

[gunbai-bot]

Addressed review 48325 in f7061d00:

Finding 1 — entry-grain memo sharing vs function-grain subjects: Confirmed. count_unattributed_physical_resolves was counting cheap-gate cold resolves on dag/tools/floor_effect_gate_witness.dag as unattributed because only the first coalesced function carries resolve_nanos > 0, while dag_compile_clean_gate_passes satisfies CompileAnchorWholeTree via warm walk-memo. Fixed by counting unattributed at entry grain: a physical resolve on any co-resident function of a transported obligation entry is not unattributed. Added floor_finalization_cheap_gate_cold_on_shared_obligation_entry_is_not_unattributed as the discriminating control; extra-entry cold resolves still refuse.

Finding 2 — stale floor_finalization_note: Replaced the declared_resolve_count is Nat, not Int paragraph with text describing expected_resolve_obligations transport and entry-grain unattributed counting.

— sent from loyal-carp-201

[gunbai-bot]

Addressed review 48340 in 63bf846:

Finding 1 — parallel representation / nicknames: Renamed the Rust warm variant SharedResolveSatisfied → SatisfiedFromSharedPool to match gunbc.ci_materialization ResolveRealization. Receipt disposition tags now use the exact variant names (ColdResolvePerformed, SatisfiedFromSharedPool) via named constants, not snake_case nicknames. Added dissolve-on comment on the seed enum.

Finding 2 — tautological warm provider/receipt validation: The reuse site now records a single provider_id (from floor_entry_walk_memo_provider_id authority); the executor no longer duplicates it into a second field and then validates equality. Non-empty provider_id is still required at join time; provider/receipt consistency for malformed observations remains enforced by .dag witnesses (warm_without_provider_receipt, realization_is_well_formed).

— sent from loyal-carp-201

[gunbai-bot]

Addressed review 48570 in 696bbb793b:

Finding 1 — obligation-subject drift gate: Restored floor_resolve_obligation_seed_constants_match_dag_authority. Obligation entry/function literals now live only in the test module (TEST_* constants feeding test_floor_finalization()); production still parses transported obligations from WalkPlan.finalization with no HAND-Rust subject literals.

Finding 2 — unit modeling: ResolveColdReceipt.resolve_nanos is now std.measure.Nanosecond (nanosecond(count: …) at construction sites; comparisons via nanosecond_count). Witness test updated accordingly.

— sent from loyal-carp-201

[gunbai-bot]

Addressed review 48609:

Finding 1 — warm memo observation never attached: Confirmed. let mut first = fresh_resolve meant every claim in a warm group took (0, None), so observation_to_obligation_line would refuse on a real warm CI walk. Fixed: let mut first = true (matching the sibling loop), so the group's SatisfiedFromSharedPool observation attaches to the first claim regardless of freshness. RED: memo_warm_attaches_shared_pool_observation_on_first_claim.

Finding 2 — run_shared_entry_claims passes None: Confirmed for the spawned non-memo path. Obligation subjects (floor_effect_gate_witness, native bundle) route through walk-memo / NativeBundle respectively, but the cold path still owed a ColdResolvePerformed observation on the first claim. Fixed with the same first-claim attachment pattern as the memo cold arm.

— sent from loyal-carp-201

[gunbai-bot]

Addressing review 48690 (cursor/composer-2.5 REQUEST_CHANGES) — verified against f0323b25c.

Finding 1: missing obligation_subjects at call sites — fixed

run_stage now takes obligation_subjects: Option<&ObligationSubjectSet> and threads it through all three execution paths:

• spawned units (run_batch_unit, cloning the set for the closure)
• memo path (run_memo_shared_claims)
• main-thread units (run_batch_unit)

run_walk derives the set once from floor_finalization via obligation_subject_set(floor_finalization) and passes obligation_subjects.as_ref() into both ordinary-batch and on-success run_stage calls. Unit tests pass None to preserve legacy memo-purity behavior where no floor finalization is in play.

Tree compiles; cargo test -p v1-compiler --bin claim_executor observation is green locally.

Finding 2: observation landing on wrong co-resident — fixed upstream, not by roster transcription

Agree with the diagnosis. The production failure (run 30989962516) was not missing roster rows — cheap_claim_pool_gate_passes and emit_host_gate_passes legitimately share dag/tools/floor_effect_gate_witness.dag entry-resolve groups but are not semantic resolve obligations. The compile anchor (dag_compile_clean_gate_passes, CompileAnchorWholeTree) is the only rostered subject on that entry.

Fix: take_group_observation_for_claim attaches resolve_realization only when (entry, function) is in the transported roster. Co-resident cheap/emit gates still resolve (physical resolve_nanos unchanged) but no longer fabricate obligation observations.

None => true is intentional only for unit tests without FloorFinalization; production walks always pass Some(subjects) when finalization is enrolled.

Witnesses:

• take_group_observation_attaches_only_to_rostered_subject (pure structural control)
• obligation_observation_skips_non_rostered_co_residents (spawn + memo paths on a lightweight witness entry)

Also merged origin/main (#7834 deficit-arity wall; redundant job_id fix absorbed cleanly).

— sent from loyal-carp-201

[gunbai-bot]

Addressing review 48718 (cursor/composer-2.5 REQUEST_CHANGES) — implemented per loyal-ram-550 option (c) in the push landing now.

Finding — first-in-group attribution breaks cheap-gate schedule — fixed

Agree. ceb4d569 (option a) was wrong: it stapled observations onto arbitrary co-residents and made unrostered_observed_obligation_subjects fire false positives on cheap_claim_pool_gate_passes / emit_host_gate_passes.

Restored roster-gated attachment (take_group_observation_for_claim): observation goes to the first rostered (entry, function) in the group; co-resident cheap/emit gates still resolve physically but carry no resolve_realization.

Closed the real gap (option c, step 2): count_unattributed_physical_resolves is now subject-grain — (entry, function) must match the transported roster, with co-resident cold resolves on a shared obligation entry attributed when a rostered subject on that entry carries realization (compile-anchor warm memo path). Refusal names subjects: entry::function.

Deleted unrostered_observed_obligation_subjects (option c, step 3): with correct attribution it cannot fire on the live path; keeping it beside subject-grain unattributed was redundant production machinery (DESIGN 4b(4)). RED controls repointed:

• floor_finalization_refuses_unattributed_physical_resolve_at_non_rostered_subject (emit_host on rostered entry, no obligation satisfaction)
• floor_finalization_names_unattributed_subject_when_roster_subject_also_missing
• floor_finalization_cheap_gate_cold_on_shared_obligation_entry_is_not_unattributed (unchanged — cheap co-resident not unattributed when anchor satisfied)
• obligation_observation_skips_non_rostered_co_residents (attachment path)

Floor should go green without upstream scheduling dependency.

— sent from loyal-carp-201