DataItem decl_facts carry authored string values for lexeme-keyed discovery

[gunbai-bot]

Summary

• DataItem decl_facts nodes now attach every authored LitStr from the parse-tree initializer (bare literal, record field, list element, concat argument, string-interp fragment), not only the typechecked projection.
• Lexeme-keyed discovery (tools.v1_consumer_discovery) can therefore see values such as devboot_program_package = "v1-compiler".
• The enrolled census blindness row is flipped to a positive control; fixture witnesses cover the other initializer shapes.

Test plan

• cargo check -p v1-compiler --lib (remote)
• cargo clippy -p v1-compiler --lib -- -D warnings (remote)
• CI required lanes on this PR
• decl_facts_initializer_projection_witness_test string-lexeme rows
• discovery_fold_sees_a_bare_string_initializer

Made with Cursor

[gunbai-bot]

review 63976 (dashboard artifact /api/reviews/63976/artifacts/stdout.log) on 4e288de5 — both findings are already repaired on e5e08c; no further commit for those two items.

Silent widen/drop of children. with_authored_string_literals now refuses with InterpError::TypeError when the projection is not a Node record, when children is not a Value::List, or when the children field is missing. There is no Vec::new() fallback. Authored lexemes attach as named authored_string_literal edges beside the existing list, so parent/arm identity readers still see the typechecked projection.

Meaning fork on SkeletonMarshalAtomLexeme. The initializer witnesses query skeleton_contains_authored_string_literal (named-edge walk), not skeleton_contains_marshal_atom_lexeme. Marshal-atom walks skip authored_string_literal edges. The discriminator decl_facts_authored_string_does_not_fork_constructor_spelling_channel requires "SingleAuthority" visible on the authored-string channel and absent on the marshal-atom / constructor-spelling channel.

Pushed head e5e08c is still structurally red on a later typecheck of those skeleton walks (fold_list_right over node.children); that fix is local and held for the #11086 remarge push, not a reopening of these two findings.

— sent from stern-heron-554

[gunbai-bot]

review 64027 (dashboard artifact /api/reviews/64027/artifacts/stdout.log) on a011da69 — addressed on 1cb4ec85 except the fingerprint merge, which is the wrong home.

Do not widen data_init_literal_fingerprint / DataInitDecl. That row is one SOURCE fingerprint per declaration (str:"…", top-level literal only) consumed by v2.lens.no_dual_representation_test and v2.lens.effect_reach via lexeme_matches_literal_fp (equality of one encoding). Discovery keys atom_identities_in_node over the decl_facts Node, which needs N interned atoms (bare, list elements, concat fragments). One literal_fp cannot be that population; stuffing nested walks into it would fork the dual-rep contract. Coincidence that both surfaces contain "hello" for the planted specimen is not one concept. Dissolve-on is now collect_authored_string_literals_dissolve_on: delete the parse-tree attach when marshal_data_initializer_projection itself interns LitStr atoms.

Edge spelling. Dag readers compare authored_string_literal_edge_name (one intern). Host AUTHORED_STRING_LITERAL_EDGE remains the seed intern of that row; authored_string_literal_edge_host_spelling_dissolve_on is the emission trigger. The three skip sites already went through skeleton_edge_is_authored_string_literal.

Third walker. node_skeleton_marshal_atom_lexemes and node_skeleton_authored_string_literals now read AtomLexemeCensus from the existing fold_node. Authored edges contribute authored_string_lexemes and stay out of all_atom_lexemes.

No new closure-member import.

[gunbai-bot]

review 64040 (dashboard artifact /api/reviews/64040/artifacts/stdout.log) — census honesty landed on be60905cc32.

missing_basis clause (iii) and the dag-argv roster note no longer claim the producer is blind to a bare-string initializer. They name discovery_fold_sees_a_bare_string_initializer as the closed channel's positive control. Residual lower bound is the unenrolled whole-corpus join and the media without a DeclarationRef projection. Population stays PopulationMeasuredLowerBound.

[gunbai-bot]

review 64137 (dashboard artifact /api/reviews/64137/artifacts/stdout.log) — landed on ef61cbc49e4.

RFM. projection_drops_the_value_its_consumer_keys_on now reports mechanically preventable (enrolled value-projection pairs + discovery_fold_sees_a_bare_string_initializer). Next trigger is collect_authored_string_literals_dissolve_on: typechecked projection interns LitStr so the host attach deletes.

EdgeLabel Bool. skeleton_edge_is_authored_string_literal_predicate_disposition is 🟡 until named edges are a closed roster (same capability as that dissolve-on).

[gunbai-bot]

review 64153 (dashboard artifact /api/reviews/64153/artifacts/stdout.log) — landed on ccd5dc89ff1.

1. RFM uses found-at / rung-now: found at mitigation-by-review; rung now mechanically preventable.
2. Evidence cites test.claim.self_host_v1_consumer_census_witness_test discovery_fold_sees_a_bare_string_initializer.
3. Deleted the commentary String disposition; 🟡 EdgeLabel skip is on typed collect_authored_string_literals_dissolve_on.

[gunbai-bot]

review 64215 (dashboard artifact /api/reviews/64215/artifacts/stdout.log) — the three EXPECTED RED annotations in decl_fact_value_projection_witness_test.dag are retitled to REGRESSION CONTROL (and the positive-control comment no longer calls them reds). Local commit, not pushed: required build+floor on 3951f9b56df is still in progress.

— sent from stern-heron-554

[gunbai-bot]

review 64234 (artifact stdout.log): the new skeleton_edge_is_authored_string_literal matcher was a parallel EdgeLabel match beside v2.std.node labeled_named_is / edge_label_of. That is fixed locally (labeled_named_is(x: e, label_of: edge_label_of, wanted: authored_string_literal_edge_name)). Not on the PR head yet: e7abcbd0 still has in-flight required CI, and we will not cancel it with a mid-run push. Sibling skeleton_edge_is_construction_spelling stays as pre-existing debt, not expanded.

— sent from stern-heron-554

[gunbai-bot]

srv2 closure receipt for #11082 @ 50fd200 — re-taken after the sixth baseline (overlay on main 5b04cd1, the current tip; instrument closure_pr.sh 11082, producer built from the overlay, route and fold with GITHUB_SHA=<overlay tree>): clean. Fold rc=0; after-set 14 rows, identical to the sixth baseline at 62de3e7 (13 parse_g0_tokens_remain + 1 normalize_reason_post_normalize_not_well_formed on src/v2/std/runtime.dag, all pre-existing); regressed 0, fixed 0; manifest 167 modules, identical to main's.

Closure-manifest delta: no manifest member and no src/v1/** file changed on main between the overlay sha (5b04cd1) and the current tip (5b04cd1) — the receipt describes the landing composition.

— sent from eager-raven-113

[gunbai-bot]

srv2 closure re-take (B-vs-T), post-#11216 base — result: identical to B.

Binding: T = overlay 39d3028f2b6c63669a981c15732823b1165dfd51 of this PR's head c31790ffe4313c28cc90628a297669ea07aaef3d (three files applied cleanly: src/v2/std/data_initializer_identity.dag, src/v2/std/decl_facts_skeleton.dag, src/v2/workflow/floor_expected_red.dag) on B = refs/prs/11216 head 6511f69cda40fdac563174d0055a7d4fba0ffcc1 (= main 6c7b0819 + exact #11216, the composition capture n executed on); instrument closure_pr_base.sh sha256 f489038cf0c1677e…; producer claim_executor_pinned sha256 9573ed56397a0c45… (rebuilt from the B tree for this chain; capture n's producer was a9a541e39a7ccf18… from the overlay tree — different bytes, same source lineage); host srv2; start 00:31:20Z, end 01:47:04Z (chain log); emitted closure compiler BUILT on this composition (emitted crate built … exit_status=0 warning_count=0, 181 files, closure e6c29996…, binary sha256 0dbe7928d1b22403… — differs from main's 542a5c4e… because T changes v2 sources); fold census dag src/v2, _terminal: complete, EXIT=0, wall 387.32 s, rss 1.25 GB.

Executed-vs-classified: census mode folds files, so 0 identities executed with verdicts; 15 file refusals classified (14 parse_g0_tokens_remain, 1 normalize_reason_post_normalize_not_well_formed), manifest 176 files.

Joins (by path and fatal_reason, both directions):

• T vs B (capture n on Restore --v2-native-route: the emitted driver does not build on main #11216, comment 5649697682): refusal set identical (15 = 15, 0 added, 0 removed, 0 reason changes); manifest identical (176 = 176).
• T vs tenth baseline (main dfac90c3, 14 refusals): +1 src/v2/workflow/floor_route_gap.dag — the same single delta B carries against that baseline (v2-native route: closure-scoped ingest and native adjudication over a seed-prepared artifact (lands after #10990) #10940's), not this PR's.

Currency vs realized main 2a5baa914f (#11200, merged 01:23Z) — paths main carries that B lacks: dag/gunbc/compiler_frontend_program_interlock.dag, dag/gunbc/compiler_frontend_program_status.dag, dag/test/claim/compiler_frontend_program_status_witness_test.dag (#11200) — none in the 176-file closure manifest, none under src/v2, no declaration name shared with a manifest module (new module names); plus the #11110 delta (three recurring_failure_mode rows, each dispositioned on #11216: not in the manifest, not a test declaration, no bare-name collision). Applicability to realized main rests on those bounded per-path determinations, not on whole-tree identity.

Not covered by this receipt: the required floor on this head is red for two independent reasons — the fleet-wide CONSUMED admission rows (#10940's 181 relocation rows, cleared by #11156 c62d1e6378f landing and a remerge), and this PR's own ENROLMENT-MARGIN-REFUSED test.claim.decl_facts_initializer_projection_witness_test.decl_facts_concat_string_initializer_carries_authored_lexemes cause=enrolment_measured_over_margin (verdict=FloorRefused, claims_failed=0). The second is lane work.

— sent from eager-raven-113

Binding correction 2026-09-13 ~03:45Z: "N files applied cleanly" above lists only the hunks that took git apply --3way's fallback path (instrument log artefact); the overlay commit carries the PR's WHOLE diff (overlay vs base file count = PR changed-file count; verified for this capture in the overlay worktree). Subsequent captures print the full APPLIED file list. — sent from eager-raven-113

[gunbai-bot]

srv2 closure re-take (B-vs-T) on the remerged head — result: identical to B and to the 00:31Z re-take on c31790ff.

Binding: T = overlay 515ae3946f664cd54ac814859587e82b9880d696 of this PR's head a7179593f4801588ae1a0eabcdf269e64c1f9ed6 on B = refs/prs/11216 head 6511f69cda40fdac563174d0055a7d4fba0ffcc1 (= main 6c7b0819 + exact #11216); instrument closure_pr_base.sh sha256 cc7034c2edd6096b… (v5, prints the applied set); producer claim_executor_pinned sha256 b44b571526f82328… built from the overlay tree; host srv2; start 04:44:19Z, end 06:00:07Z (chain log), rc=0. APPLIED_FILES: pr_diff=84 overlay_vs_base=81 — of the 81 paths applied over B, 15 are this PR's own files (the PR's full GitHub file set, 3 under src/v2, 1 in the 176-file closure manifest: src/v2/workflow/floor_expected_red.dag) and 66 are main's d7b7ab96 remerge carried by the head (none in the manifest). Emitted closure compiler BUILT on this composition: exit_status=0 warning_count=0, binary sha256 0dbe7928d1b224039e9a4c83ccd459e8eca6fb570311f75f1c38fc381e7ffc97 — byte-identical to the 00:31Z re-take's binary (the remerge changed no closure input). Fold census dag src/v2, _terminal: complete, file_refusals: 15, wall 379.70 s, rss 1.25 GB.

Executed-vs-classified: census mode folds files, so 0 identities executed with verdicts; 15 file refusals classified (14 parse_g0_tokens_remain, 1 normalize_reason_post_normalize_not_well_formed at src/v2/std/runtime.dag); manifest 176 files.

Joins (path + fatal_reason, both directions): T(a717959) vs T(c31790f, comment 5650084644): refusal set identical (15 = 15, 0 added, 0 removed, 0 reason changes); manifest identical (176 = 176); emitted binary identical. T vs B (capture n, comment 5649697682): identical on the same joins.

Currency boundary (observation-specific, not whole-tree): this receipt is bound to composition B + T. Realized main is now a6ae1af6 (#11143, merged after this run ended); its delta from B carries one closure-manifest file, dag/gunbc/v1/v1_interpreter_primitive_surface.dag, plus src/v1/04_method.dag and five stage0 mirrors. Under the per-path currency rule a manifest-file change breaks applicability, so this receipt does not carry to a merge onto a6ae1af6; a re-take on refs/remotes/origin/main is owed before landing and is queued on the srv2 chain (after #11175's, which is first in the order). The required floor on this head is green (4/4, CLEAN at a7179593); that is CI evidence, separate from this closure receipt.

— sent from eager-raven-113