Repository navigation
Step 0 census: reference-occurrence binding provenance (calibration matrix before the corpus run) - #10411
Step 0 census: reference-occurrence binding provenance (calibration matrix before the corpus run)#10411briansrls wants to merge 1 commit into
Conversation
…the import-authority scoping record TWO SUBJECTS, BOTH FALLOUT OF #10256. ONE -- THE NESTED-COMPILE BUILTIN I ADDED WAS NOT ON THE SURFACE. #10256 landed compile_dag_reference_occurrence_binding_census, which runs the frontend over supplied sources. That is unbounded host work of exactly the class compile_dag_rust_emit_check is on the surface for, and compile_dag_multi_module_fixture was ALREADY two rows above it in the same roster. It appeared in ZERO rows. So a claim reaching it would report CooperativelyPollable while doing unbounded work -- the same defect main repaired one commit earlier, where the surface was spelled with free_call. identities while the recorder compared bare spellings and 3602 of 3602 claims read benign. VERIFIED BY THE CHECK THAT DISCRIMINATES, not by the ones that do not. opaque_host_call_surface_is_grounded returns true, which is what proves roster_arms_for_identity RESOLVES the new identity to an authored spelling rather than leaving it in the unresolved list. A row that fails to resolve is precisely the decoration that reads as coverage, and 7/7 green on the reach-control and preemption witnesses does not substitute for it. TWO -- THE SCOPING RECORD FOR THE IMPORT-AUTHORITY CUT, registered so the work that cites it can name a stable slug rather than a PR body. It states TWO layers with citations (selective_func_env_view folding only over the listed names, so specific_names IS the population; and the two TypeEnv maps plus author_named_visibility), records the WRONG three-layer claim as a recorded error rather than silently replacing it, and states the parent_index order-dependence as a PRECONDITION of the population repoint rather than a ticket beside it -- it is latent today only because every resolution-path access is map_get(parent_index, imp.module_path), and it arms the moment lookups stop being keyed by import path. THE DOC-GRAPH GATES ARE RED AND THAT IS NOT THIS CHANGE. Measured against a stashed baseline rather than assumed: doc_graph_has_no_orphan_docs is false on main WITHOUT these files, and the population is dispositioned in gunbc.guarantee_stall.doc_graph_orphan_population_stall with a declared trigger. Those identities sit on floor_expected_red_roster, which means RUN EVERY PR AND HOLD THE RED -- so the hazard is an UNEXPECTED GREEN, not the red. All three are still red here, so no dissolution signal fires and the stall row stands. AND THE PLAN-ROOTS PROBE WAS CHECKED IN BOTH DIRECTIONS, because "it was already red" stops being an explanation once the change touches its subject. Registering a plan root is that subject. Measured: baseline total=215 admitted=72; with this change total=216 admitted=73. Both +1, so the new root IS admitted and the gap of 143 is untouched. The probe stays red for its pre-existing reason and this change adds no new one. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01RvMEJFpYsy55NRdJ1Az4Vd
|
Closing this. It cannot land as it stands, and I measured that rather than inferring it from the stale red. The red on this PR is a cancellation, not a content failure. All four jobs "failed" at ~39s on run 33875856188, whose overall conclusion is The real problem is the opposite of a missing row. The census builtin So this change registers a reach for an operation that is not in the grounded ops population, and lands a green witness in the red. Completing it would mean deciding whether the census builtin belongs in that population at all and registering both halves together — which is a different piece of work from the one this branch started, and belongs to whoever owns the opaque-host-call surface rather than to a closing census-calibration lane. The calibration work this branch was opened under has since landed by other routes — #10545 (the 4×7 matrix and its axis-completeness repair), #10608, #10592, #10638. Nothing here is needed for it. The plans doc is not being discarded. — sent from jolly-badger-374 |
1 participant
Auto-opened by session-dashboard for session
jolly-badger-374.Pushing to
work/census-builtin-opaque-surfaceadvances this PR.Worker attestation
Before flipping this PR to ready for review, confirm each item:
npm test,cargo test) and the result.Closes #Ndirective.Summary
TODO: replace this paragraph with one or two sentences naming the change and its motivation. Reviewers read this first.
Test plan