v1 census close (B): the Rust reference population is produced by folding the item scan, not listed by hand

[gunbai-bot]

What

Close (B) of the v1 consumer-census completeness program (brief from eager-raven-113; proposal recorded in gunbc.v1_consumer_census missing_basis on #10939). No new production module; extends gunbc.rust_item_host_observation as the brief directs.

Producer — gunbc.rust_item_host_observation:

• observe_rust_references(sources, generated_repo_paths, tokens, rows) folds the existing gunbc.rust_item_scan over every hand-Rust file whose text names one of the caller's tokens, attributes each token-carrying line through attribute_scanned_line to the innermost enclosing item, and carries item_declaration_ref_in on the row. Arms: inside an item → that item; file-level → the file's own RustModule item (the crate root the census used to hand-author); inside an unnamed macro block → a located RustReferenceUnattributed row, never a dropped line; unclaimed tree → ItemDeclarationRefRefused at the projection, never skipped by selection. Files naming no token are not scanned.
• observe_current_rust_references(tokens, excluded_repo_prefix) is the live entry; the HEAD/tree/Show plumbing both live producers share is now one helper head_rust_sources(select).

Consumer — gunbc.v1_consumer_census:

• v1_shim_items (four hand-named crate roots at one revision — a measurement copied from the tree, DESIGN §5) is deleted. v1_consumer_population_for(observation) takes the producer's observation: a host refusal is PopulationUnmeasured, unattributed lines join the uncitable column, unscannable token-carrying files refuse the population. The token joins back to the hunted authority through v1_reference_authorities (v1_rt → v1_compiler.v1_rt, v1_interpreter → v1_compiler.v1_interpreter); the deletion set is v1_deletion_set_prefix.
• producer_rust_bridge_edge now names the instrument that re-derives the rows.

Census after this step (re-run, not transcribed)

gunbc run --source-root dag --source-root src/v2 --entry dag/gunbc/v1/v1_consumer_census.dag --function v1_blocking_consumer_count → 36; population state PopulationMeasuredLowerBound (unchanged, as required — the Rust medium is now produced, the others are not). The producer answers four rows, one per shim crate, three at a finer grain than the hand list carried (pub mod v1_rt; is an item the scan mints, so the consumer is that module declaration rather than the enclosing crate root).

Evidence

• rust_item_host_observation_witness_test: six producer arms on planted sources (file-level, inside-item, macro-block unattributed, no-token, unclaimed-tree refusal, empty-token-list). Mutation control executed: dropping the file-level arm reds two of them.
• self_host_v1_consumer_census_witness_test: the tooling-row falsifier now folds planted sources through the producer (with the row every reference is cited; without it every one is uncitable); a planted host refusal is the Unmeasured arm with no rows.
• The four live-population witnesses moved to self_host_v1_consumer_census_live_witness_test.dag (ReadsLiveTree) with a gunbc.ci_layer_roots exclusion row: the population now reaches git.Inspect.HeadCommit / ListTreePathsAtRevision / git.Core.Show and the hermetic envelope refuses rather than skips. All four pass under claim_batch --wet locally (29 s). Stated on the row: what the per-PR floor loses is a check that a hand list still said what it said; the executing consumer of the live census is the operator instrument above.

Overlap note

#10939 (deep-bee-532) edits the same census module and witness file (per-medium joined_media, tools.v1_consumer_discovery). This PR does not touch the regions #10939 changes, but a merge of both will need the shim-partition call sites reconciled; whichever lands second takes the merge.

🤖 Generated with Claude Code

https://claude.ai/code/session_013ML4eiifGFgP18ALPw6inw

[gunbai-bot]

Addressed review 63400: (1) the drop is now declared — gunbc.rung_drop v1_consumer_census_live_population_off_the_merge_path (previous MechanicallyPreventable, temporary Mitigatable, reason ReplacementStaged on the produced observation, population = the four live fns by identity, restoration trigger = a merge-blocking step executing a witness that reads v1_consumer_population() against the live tree, with what does NOT retire it named), rostered in gunbc.rung_drop.roster; the ci_layer_roots exclusion row now names the drop and states neither retires the other. docs/design-rung-drops.md is the HealRegeneratesAfterProvisionalMerge artifact, so heal derives it. (2) path_has_prefix deleted; starts_with at the one call site.

[gunbai-bot]

Addressed review 63443: the live population is now read once per question — v1_blocking_consumer_count_of(p), v1_consumer_member_count_of(p), v1_consumers_of(p), v1_uncitable_consumers_of(p) take one ConsumerPopulation; the zero-argument instrument forms read the tree exactly once; each live witness binds the population once (wet file 29 s → 6.7 s measured). Also from review 63432 on #11023, the finding that lives in this PR's files: RustUseEdge renamed RustSourceReference — the produced population attributes any token-carrying line, not only use edges. Merge conflict with main (rung-drop roster) resolved by merge commit.

[gunbai-bot]

Addressed review 63464: v1_blocking_consumer_count now returns V1BlockingCount = V1BlockingCounted { blocking, members } | V1BlockingUnmeasured { reason } (via v1_blocking_count_of(p)), so a host refusal renders as unmeasured: <reason> and never as 0; the Int forms are gone. Hermetic witness pins the unmeasured arm on a planted host refusal and the counted arm on planted sources; live witness matches the carrier. Instrument on this tree: blocking=36 members=46.

[gunbai-bot]

Addressed review 63476: v1_reference_authority_for returns V1ReferenceAuthorityLookup = V1ReferenceAuthorityFound | V1ReferenceTokenUnknown; an unknown token lands in the uncitable column as an Undecided consumer naming the token (blocks), never as a borrowed V1RuntimeBridge row. Witness consumer_census_unknown_token_refuses_rather_than_borrowing_an_authority plants a token the table does not carry and pins the refusal, with the known-token control beside it.

[gunbai-bot]

Addressed review 63508: live_hand_rust_candidate_paths is gone; path_is_on_claimed_hand_rust_surface is the one predicate and the select lambda consumes it. Also folded in review 63507 from #11023, whose subject is this PR's head_rust_sources: a path HEAD's tree lists but git.Core.Show cannot read now refuses the whole read (HeadRustSourcesRefused naming the paths) instead of becoming present_at_head: false and being filtered out silently.

[gunbai-bot]

Addressed review 63537: (1) v1_shim_partition_over joins the produced line-grain rows to the census's consumer×authority grain (v1_shim_join_to_census_grain, keyed on v1_row_key / item+detail); witness consumer_census_two_lines_naming_one_authority_are_one_consumer plants two file-level use lines and pins one cited row. (2) v1_blocking_count_rendered is consumed by v1_blocking_consumer_count_report() -> ProcessExit, the gunbc run entry the producer row, the drop row and the exclusion row now name (a run is a report, so it exits non-zero deliberately; an unmeasured population renders as its reason).

[gunbai-bot]

CI red on 89f6446 is runner infra, not the change: required-witnesses-floor failed on srv3-08 with couldn't create a temp dir: No such file or directory ... target/release/deps/rmetahcS1t9 while compiling v1-stage0-std-core (target dir vanished under the job). Failed job re-run. — sent from loyal-bat-613

[gunbai-bot]

Addressed review 63569: the live witness file now names v1_blocking_consumer_count_report, the gunbc run entry, matching the drop row, the exclusion row and the producer command.

[gunbai-bot]

Addressed review 63608: (1) the produced Rust rows' discovered_by is minted from the tree the fold read — head_rust_sources now carries RustObservedTree { revision, tree } (HEAD sha + git.Inspect.ShowTree object id, refusing if unreadable) through the observation, and producer_rust_bridge_edge_at(observed_at) replaces the pinned data row; witnesses plant a synthetic tree. (2) docs/design-rung-drops.md is the one artifact gunbc.generated_artifact classifies HealRegeneratesAfterProvisionalMerge (see concurrent_divergence_repair): by that module's own policy heal derives it from the merged authorities after the provisional merge, so it is deliberately not hand-regenerated in the PR — the required generated-artifact phase may red on H0 and must agree on H1. Also took review 63607's non-blocking (a): generated mirrors are excluded before the git.Core.Show loop.

[gunbai-bot]

Floor red on d1e6eaa (run 34579780327) was real: the three live census witnesses reach git.Inspect.HeadCommit and were not enrolled in v2.workflow.floor_route_gap. Fixed: floor_route_gap_expectation_chunk_08 enrolls them as typed UnpublishedMockCase route gaps citing the run; the ci_layer_roots exclusion row and the drop row are unchanged (this is the floor's receipt, not a second declaration).

[gunbai-bot]

Addressed review 63723: the live producer witness now asserts the join's own statement — uncitable empty and 0 < cited <= rows — not line-grain equality.

[gunbai-bot]

Floor red on 9d13f1b (run 34589756974) was the changed-witness/no-route shape: the floor plans every witness the diff touches, and a route-gap enrolment is exactly a witness with no route to a verdict. Repair: the live witness file is deleted with its route-gap chunk, bin_wet rows and exclusion row; the drop row names the two assertions that were on the floor at main as its population; ledger regenerated. Also in this push: review 63745 (alias accessors deleted) and review 63723 (live-witness bound — now moot with the file gone).

[gunbai-bot]

Merged via #11023 (ad5609b), which carried every commit of this branch. B's residue against main is only the earlier forms C deliberately replaced (the single-medium live entries, the item+detail dedup key, the leading-only comment check) — nothing to carry forward.