@weftlet/toolkit 0.6.1 → 0.7.1

package/docs/VALIDATION.md

@@ -1,5 +1,128 @@
 # Validation evidence
 
+## 0.7.0 large-graph field blocker follow-up, 2026-11-02
+
+Status: public capacity qualification and hosted Linux/Windows checks pass; patch release pending.
+Private consumer acceptance remains open.
+[ADR 040](ADR-040-bounded-graph-io.md) records the encoding, storage and strict SHACL boundaries.
+The consumer's reported 561,777-fact failure occurs before SHACL at whole-graph canonical
+string construction. Its prior snapshots remain readable; no private inputs were imported here.
+
+Local Linux checks currently establish:
+
+- Production build and 38 core tests pass, including an independently checked canonical hash
+  larger than V8's actual string ceiling. Existing encoding and persisted-JSON digest semantics
+  are retained, including optional fields and Unicode boundaries.
+- 65 focused lifecycle, RDF parity, streaming-storage and release-documentation tests pass.
+  The frozen 0.7.0 reference verifies named/union quad sets, strict positive/negative verdicts,
+  cross-subject proof checks, SPARQL aggregates and lossless traversal projection. A separate
+  100,001-fact test removes the old projection ceiling without truncation.
+- Fresh packed-consumer installation passes setup, preview/cache adoption, traversal hydration
+  and reuse, real SCIP, RDF/SPARQL/SHACL, HTML/DOT/SVG assets and MCP.
+- Inventory and license checks pass; all 79 original source import path/SHA pairs are unchanged.
+  Architecture passes with 15 advisory findings: two additional complexity advisories in the
+  canonical digest visitor and RDF matching function. No baseline, exception or gate changed.
+
+The full suite exposed a corrupt-JSON error-class compatibility regression. The streaming
+reader now preserves SyntaxError classification for malformed/truncated input while retaining
+I/O errors as infrastructure failures; cache-adoption CLI and strict parser regressions pass.
+The full local pass before those fixes completed with 977 passes / 8 failures / 1 optional skip.
+All corrupt-artifact cases pass on rerun (44 cache/preview/storage tests plus 3 CLI cases).
+An overlapping local rebuild interrupted a compiled benchmark subprocess; its 6 tests pass after
+the build. Seven sequential CLI launches exceeded the old aggregate 30-second setup-test budget;
+the test now bounds each process to 30 seconds and the sequence to 180 seconds, retaining all
+assertions (3 setup tests pass). A subsequent 29-test RDF/setup run passes. These corrections
+do not change production operation deadlines. At `85361dbb` both hosted runs
+[41532316889](https://github.com/jchewitt/weftlet/actions/runs/41532316889) and
+[41532326444](https://github.com/jchewitt/weftlet/actions/runs/41532326444) pass Linux and Windows.
+Linux passes 38 core + 985 toolkit tests (1 optional ontology integration skipped), architecture,
+inventory, licenses and fresh packed-consumer smoke. Windows passes 38 core + 878 selected tests,
+inventory and licenses. The final local packed-consumer rerun also passes.
+
+At `d1498a4`, hosted push run
+[41542970652](https://github.com/jchewitt/weftlet/actions/runs/41542970652) passes both platforms.
+The duplicate PR run
+[41542976891](https://github.com/jchewitt/weftlet/actions/runs/41542976891) passes Linux; Windows
+passes 877 selected tests but times out in the four-command failure-reporting test at its
+30-second aggregate deadline, then encounters cleanup EBUSY with the child still using its cwd.
+The harness now retains each 20-second command limit, allows 100 seconds for the combined
+sequence and fixture setup, and aborts/waits for child closure before cleanup. All four local
+failure-reporting tests pass; this changes no production timeout or assertion.
+At `4bc41fd1`, both subsequent hosted runs
+[41545800490](https://github.com/jchewitt/weftlet/actions/runs/41545800490) and
+[41545807452](https://github.com/jchewitt/weftlet/actions/runs/41545807452) pass Linux and Windows,
+including the corrected test and fresh packed-consumer installation.
+
+The optional broader test-source TypeScript check still reports existing diagnostics outside
+the production build. An isolated archive of the base commit has the same 42 diagnostics after
+normalizing checkout paths and ignoring its two absent-build-output errors; this patch adds none.
+
+The [public capacity result](benchmarks/large-graph-2026-11-02/result.json) passes the complete
+hydrate → strict validation → persist/read → unchanged repeat → clean comparison lifecycle.
+It uses 1,100 TypeScript modules with 40 functions each, actual SCIP, spec files and 18
+deterministic prose extractions, with zero live model calls. The run produced **554,516 facts /
+143,044 nodes**, a **1,021,806,445-byte snapshot**, and a **5,331,139,680-byte N-Quads export**
+containing 21,522,256 quads. Every exported record was parsed and matched by ID and canonical
+digest. Bounded real SPARQL and the complete record projection pass without dropping facts.
+
+All three full strict SHACL passes succeed: cold **2,099,768 ms**, repeat **2,225,093 ms**, and
+clean **2,096,881 ms** (about 35–37 minutes each). Both repeat and clean reuse all 18 artifacts
+with zero further extraction. Snapshot identity on repeat, logical/content digests, exact
+provenance, extractor versions, cache keys, original snapshot bytes and all semantic artifact
+bytes are retained. The complete harness took 8,841,003 ms, including three indexing/validation
+runs, large-file round trips, independent assertions and fixture cleanup. This establishes
+capacity/correctness, **not a 30-minute hydration budget** or private-workload performance.
+Other local checks overlapped part of the run; it is not an isolated comparative benchmark.
+
+An independent read of the saved 554,516 facts confirms that 938,267,270 characters of logical
+canonical encoding hash to
+`e6965407b8a8d76ed6d2d7ef8d2daa0297059b5fb327723cd49e73233865c001`
+under both the legacy per-record encoding with exact array separators and `stableJsonDigest`.
+That assertion is now also in the lifecycle harness; its scaled CI fixture passes.
+The long-running capacity process predates the final negative-path shape/error guards and
+harness corrections. Those changes pass separately in the hosted checks above; this is not
+claimed as an exact-final-head large-corpus rerun. Native Windows field acceptance, live provider
+quality and optional openCAESAR runtime remain separate checks.
+
+## 0.7.0 acceptance release disposition, 2026-11-02
+
+The maintainer authorized publishing the accumulated B1–B5 implementation for live TradePilot
+acceptance before running the new optional proposal integration. Its live model accuracy and
+usefulness validation are **tentative and deferred**. Earlier HDesk and reported consumer Codex
+hydration remain evidence for the established extraction flow at their recorded versions; they
+do not establish a live pass for the new proposal operation or this candidate.
+
+Implementation commit `ff053c23585809bbee1931c65cd411b70bb1af18` passed hosted Linux/Windows checks
+in run [41502699634](https://github.com/jchewitt/weftlet/actions/runs/41502699634). The initial
+0.7.0 version PR head `7b20066635547e2bd73dd1d017d3818ba1fdc5ee` also passed both hosted jobs in
+run [41502753047](https://github.com/jchewitt/weftlet/actions/runs/41502753047). Final release
+documentation, archive preparation, npm publication and exact source tags are verified by the
+release PR and publishing workflow; this source checkpoint does not claim their completion.
+
+The public pipeline extraction/SCIP/RDF test was rerun successfully to generate the saved HTML
+viewer. No live model request or browser interaction was performed. Candidate consumer
+hydration/reuse, agent usefulness and browser inspection remain acceptance work; official 1.0
+is not approved by this disposition.
+
+Release PR follow-up: at `8880cbfc3ccc4dce59bbedb4bf7192c1f72c2b0a`, push run
+[41508871410](https://github.com/jchewitt/weftlet/actions/runs/41508871410) passed Linux
+(988 deterministic tests and packed-consumer smoke) and Windows (35 core + 846 selected tests).
+The duplicate PR run [41508881495](https://github.com/jchewitt/weftlet/actions/runs/41508881495)
+passed Linux but hit a 30-second aggregate CLI test timeout and cleanup EBUSY on Windows.
+Its single failed-job retry passed that test, then hit synchronous frozen-study test deadlines
+and a Vitest worker RPC timeout. The harness correction runs the actual Git-backed study checker
+and compiler in bounded asynchronous child processes, and gives multi-command cases an aggregate
+deadline longer than their individual command budgets. No assertion, frozen input, production
+validator, provider timeout or graph behavior is changed. Final hosted results belong to the
+corrected release PR head; the earlier failures are retained rather than recorded as passes.
+At `39ca29f766563781281a57ff6a2194fa81d01267`, both Linux jobs and the push Windows job passed.
+The duplicate PR Windows job passed the corrected suites but exceeded the separate 60-second
+aggregate deadline for one hydration plus three indexed/RDF overlays in the traversal lifecycle
+case. That case now has a 180-second harness budget with every assertion retained. This is not
+a changed runtime deadline or a performance acceptance result; production code remains unchanged.
+
+## Environment and historical evidence
+
 Environment: Linux cloud, Node 24.19.0, pnpm 10.18.2. Source imported from
 `fda97e0146099543aeaa6b338ef2b0e9fab0d1d6`. No source application was built or changed.
 
@@ -8,21 +131,392 @@ The milestone entries below record the original cloud checks. Subsequent
 establishes live Codex success on Windows and the full deterministic checks on WSL,
 while identifying remaining native Windows issues.
 
-## Physical artifact filename fix, 2026-10-28
-
-The consumer requested bounded physical artifact/temp names while retaining logical snapshot
-identity and provenance. [ADR 028](ADR-028-bounded-artifact-paths.md) records the fix and migration
-boundary. The public traversal → unchanged repeat → clean-comparison regression reproduced an
-overlong SCIP staging basename before the change and passes locally after it. Physical names are
-bounded; full IDs, cited facts, cache keys and content digests remain unchanged. Legacy saved files
-remain readable and immutable. The same traversal test and added storage cases run in native
-Windows CI. Local release checks passed: build, 698 deterministic tests (17 core + 681 toolkit;
-one optional OML integration skipped), architecture with 14 advisory findings, inventory,
-licenses and fresh packed-consumer smoke. Hosted outcomes and patch publication are pending
-at this source revision.
+## B5 monorepo code coverage, 2026-10-30
+
+[ADR 038](ADR-038-monorepo-code-coverage.md) records saved source selection separately from
+actual SCIP documents and graph connections. The [public protocol](../evaluations/monorepo-coverage/README.md)
+uses real committed multi-package TypeScript/TSX, aliases/barrels, an independent selected service,
+an omitted service, an empty source, declaration-only targets and an unresolved external call.
+Six integration tests pass, comparing every indexed flag to the normalized binary SCIP document
+list and the derived overview to the lossless RDF projection. Source selection differs from code
+roots: the real probe did not emit an imported file outside the selected entries, and the report
+retains that absence plus references without an indexed definition.
+
+Bounded package/file pagination, cursor binding, unknown legacy coverage, corrupted inventory,
+shared query/viewer overview, API/CLI/MCP equivalence, dirty unavailable-provider queries, unchanged
+semantic reuse and clean comparison pass. Draft rename/delete and expanded committed scope update
+coverage while preserving historical snapshots and current-diff ineligibility. Three inventory
+unit tests and 17 strict interface tests complement the six integration tests. The existing
+commit-keyed hydration regression also passes after the deliberate assembly 12 bump. Strict test
+typechecking passes, and architecture retains the existing 13 advisories without new baselines
+or exceptions. Full publication and hosted Linux/Windows results are recorded in the change PR
+after execution, including installed coverage API/CLI/MCP checks.
+
+No private application, paid/model request, dependency, inference approval or npm publication is
+part of this increment. Scope is regular committed TypeScript-family files grouped by physical
+package manifest boundaries, not arbitrary compiler projects, all languages or runtime/business
+completeness. B3b's inference decision and B6 qualification/live acceptance remain open.
+
+## B4b selected contract comparison, 2026-10-30
+
+[ADR 037](ADR-037-selected-contract-compatibility.md) adds a read-only exact assertion-pair
+assessment through API/CLI/MCP. [The public protocol](../evaluations/contract-compatibility/README.md)
+predeclares eleven source cases and retrieval bounds. Fifteen integration tests exercise actual
+committed Markdown extraction, real SCIP/RDF/SHACL, numeric compatibility/insufficiency/disjointness,
+required revalidation, unknowns, unrelated domains and competing assertions. They retain exact
+facts, source authority and endpoint revisions; retrieve consumer symbol/test references; and
+execute the public consumer freshness check separately from the graph verdict. Assessment never
+claims code fulfillment or CI eligibility. The pure arithmetic suite checks all 81 operator/boundary
+combinations against a finite-set oracle plus context negatives; interface tests reject malformed
+requests. There are 42 new deterministic tests across these three files.
+
+Unchanged hydration reuses all 23 semantic artifacts with unchanged snapshot identity/content.
+API/CLI/MCP JSON agrees while dirty and the configured provider/indexer are unavailable. Draft
+edits, document renames and deletions replace/invalidate declarations without changing saved
+committed evidence. Strict test typechecking and architecture checks pass (the existing 13
+advisories, with no new baseline or exceptions). Full publication and hosted checks are recorded
+in the change PR after execution. No private source, model request, new dependency, extraction
+identity change or npm publication is part of this increment. B5, the B3b inference decision and
+B6 candidate/live acceptance remain open; this is not a product-wide compatibility proof.
+
+## B4a cited contract obligations, 2026-10-30
+
+[ADR 036](ADR-036-contract-obligation-declarations.md) adds optional explicit Markdown declarations
+for producer guarantees and consumer requirements. Eight new public extraction/query tests pass
+with real committed sources, TypeScript indexing, RDF/SHACL and an empty offline semantic provider.
+Fourteen public core-contract tests cover exact scoped identities, unknown fields and invalid input.
+The tests retain document fence spans and endpoint revisions, source approval separate from
+assertion authority, missing endpoints/counterparts, unrelated scopes and unknown versions. Agent
+pagination/paths/impact and SPARQL projection expose the same facts; the viewer retains endpoint
+navigation. Strict corruption checks and named/union SHACL parity pass. Repeat hydration reuses
+all nine semantic artifacts with unchanged snapshot identity/content; clean comparison, dirty saved
+queries and draft document/endpoint edits, deletes and renames pass. Strict test typechecking passes.
+
+At `a070ca68a3354f425cf8b506bdd3d0a76df32f5b`, the full local deterministic suite passes
+**886 tests** (35 core + 851 toolkit/example/script),
+with one optional OML integration test skipped. Build, release-documentation consistency,
+architecture (the existing 13 advisories), inventory, licenses and fresh packed npm/npx consumer
+smoke pass. The installed consumer exercises the public obligation extractor and both contract
+sides alongside real SCIP/RDF/SHACL/MCP. Hosted Linux/Windows results are recorded in the change PR.
+No new architecture baseline/exception, dependency, private consumer input or model request is used.
+All 79 original imported path/hash pairs remain unchanged. B4b compatibility acceptance, inference
+proposal review, monorepo coverage and candidate/live qualification remain separate queue items.
+Structured declaration parsing does not prove fulfillment, delivery, compatibility or model quality.
+
+Final review additionally checks source and endpoint text against the committed Git blob ID,
+rejecting lossy non-UTF-8 decoding before recording a digest. A real committed binary negative
+raises the focused suite to eight tests; final build, all eight tests, strict typechecking and
+architecture checks pass. Final installed-consumer and hosted checks are recorded in PR #61.
+
+## B3c static call evidence, 2026-10-29
+
+[ADR 035](ADR-035-static-call-evidence.md) adds actual TypeScript call/new syntax joined to exact
+SCIP identities. The [public observation](../evaluations/static-calls/README.md) records clean
+commit `98eafa99375fdd1d37333409e4e666bf7f5b0924`: 237 facts, 33 calls and nine explicit gaps;
+one cold semantic extraction, one repeat reuse and unchanged snapshot identity/content. The
+caller/callee query returns three callsites in two pages (19,181 bytes); the bounded helper-to-rule
+search exhausts its selected space with two paths and 52 candidate-edge examinations.
+
+Eight real-extraction acceptance tests cover private/local functions, recursive/repeated calls,
+shadowing, aliases, overloads, TSX, BOM/Unicode, callbacks/value uses, external/bodyless targets,
+decorators/initializers/defaults, source/proof validation, shared queries, line shifts, rename/delete
+and draft replacement. Negative SHACL/provenance/authority tests reject forged calls. Named and
+union shape verdicts agree; full RDF projection preserves assertions. The semantic schema still
+rejects call assertions, and its provider/prompt/cache identity remains unchanged. No paid or
+private consumer request was used. Strict typechecking passes for the evaluator and tests.
+
+Architecture remains at the existing 13 advisory findings; no new baseline or exception was
+accepted. A bounded 10,000-fact synthetic shared-node SHACL smoke completed in about 2.25 seconds
+of validation with unchanged serialized evidence. That case contains ordinary facts, not a
+large call graph, and is not a comparative benchmark or private workload result. The new call
+shapes target call records directly. Complete local publication and hosted Windows/Linux checks
+are required before merge; the packed consumer now exercises the shipped parser and call proofs.
+All 79 original source path/hash pairs and frozen B2/B3b inputs/results remain unchanged.
+
+B3c subsequently passed the complete local publication gate (865 deterministic tests), final-source
+focused/packed checks and [hosted PR Linux/Windows checks](https://github.com/jchewitt/weftlet/actions/runs/40991862219)
+at `00643bba3abb5899ed3d0c1ce92d991152bce7aa`. PR #60 merged as `2aaaf6749955162fa4b82a3f6ecfe8f98601e04e`;
+[main checks](https://github.com/jchewitt/weftlet/actions/runs/40992894407) also passed. No npm release was made.
+
+## B3b mapping gaps and Windows study reliability, 2026-10-29
+
+[ADR 034](ADR-034-mapping-gaps.md) records ten predeclared public cases, real SCIP/RDF/SHACL
+and exact agent queries. The clean evaluator run matches all ten states, publishes 111 facts,
+extracts eight documents cold and reuses all eight unchanged with identical snapshot identity
+and content digest. No misleading-name mapping or missed direct binding appears. Structural
+paths retain explicit truncation; they are not promoted into implementation claims. Public
+requests, responses, proofs, limits and measured timings are retained in the linked report.
+The provider is a fixed offline grammar; no paid request or model-quality result is claimed.
+
+GitHub PR #58's [retry](https://github.com/jchewitt/weftlet/actions/runs/40966207307) and
+[main checks](https://github.com/jchewitt/weftlet/actions/runs/40968675537) passed. Historical
+failed attempts remain visible. Their recurring native Windows frozen-study timeout motivated
+batched immutable Git reads, immutable fixture-input reuse and yielding between synchronous
+cases. All eight frozen-study cases pass locally without increasing the 30-second limit or
+changing frozen B2 inputs/results. Three new assessment tests and strict evaluator typechecking
+pass. Full local publication and hosted Windows/Linux checks are required for this change.
+
+The release version PR remains open. Browser interaction, live inference evaluation and private
+consumer acceptance remain separate; nothing in this assessment publishes an npm release.
+
+## B3a explicit document-to-symbol bindings, 2026-10-29
+
+[ADR 033](ADR-033-document-symbol-bindings.md) adds cited explicit mappings after semantic and
+code extraction. Focused public tests pass with real SCIP and an offline deterministic prose
+provider: functions/private methods, duplicate names, native qualification, overloads, re-export
+aliases, declaration-only targets, UTF-8 BOM offsets and reference-style Markdown. Missing,
+ambiguous and stale-byte targets produce gaps rather than guessed links. The public extractor
+checks committed source and definition bytes. Named-dataset and union SHACL agree on the new
+shape cases; consistency checks reject missing/mismatched proof facts and authority promotion.
+
+Agent neighborhoods/paths, impact scans, SPARQL and the viewer expose the same binding facts.
+Bindings retain selected occurrences, all indexed definitions, source citations and original
+subject assertion authority/review. Tests verify unchanged semantic reuse and clean comparison,
+dirty saved querying, overlay source edits/deletions/renames and invalidation of withdrawn rules.
+Traversal-selected validation documents can supply cited test mappings without model extraction
+or product-intent promotion. No private consumer or paid/live provider request was used.
+
+The initial full Linux publication gate passed **853 deterministic tests** (21 core + 832 toolkit;
+optional OML integration skipped), build, inventory, licenses and fresh packed npm/npx consumer
+smoke including real SCIP/RDF/SHACL and MCP. Follow-up review added code-derived contract-subject
+coverage, corrected omission of unset optional proof fields for exact JSON projection, and separated
+citation/parsing/proof checks into focused helpers. The 23-test selector/integration suite passes;
+architecture is back to the existing **13 advisory findings**, with no new baseline or exception.
+All 79 original source path/hash pairs remain intact. Final aggregate and hosted Windows/Linux
+checks are required on the change PR. Browser interaction, live model quality and private consumer
+acceptance remain separate.
+
+## B3a precise SCIP evidence, 2026-10-29
+
+[ADR 032](ADR-032-scip-code-evidence.md) adds typed code entities and exact occurrences to the
+existing projection. It does not yet implement explicit intent-to-symbol binding or actual calls.
+The new public integration fixture uses real committed TypeScript, SCIP, hydration, strict SHACL
+and RDF/SPARQL equivalence. Its 3,000-line file is explicitly padded with comments; the functional
+cases cover repeated same-line references, Unicode columns, recursive references, private methods,
+local arrow range limits, overload declaration spans, UTF-8 BOM first-line offsets, independent
+mocks and cross-file use.
+It is a location/isolation regression with a predeclared 300-fact/60-second hydration ceiling,
+not a representative monorepo benchmark.
+
+Exact neighbor pagination and paths preserve cited evidence. Shared unresolved external symbols
+remain queryable without automatically joining unrelated co-users in impact scans; diagnostic ID
+examples are bounded to 100 with total/omitted state. Repeat output is identical, and edited,
+renamed/deleted code is replaced in a draft overlay while prior evidence remains unchanged.
+New negative checks retain strict kind cardinality and reject malformed source/identity metadata,
+role/enclosure inconsistencies and invalid ranges. Native zero-width anchors and arrow initializer
+extents are preserved instead of imposing a false name-containment requirement.
+
+Existing coverage and traversal tests now include the additional code facts and follow continuation
+when they no longer fit on one page. Static test source remains implemented evidence, not an
+execution verdict. Semantic artifact identities are unchanged; extraction/assembly identity requires
+a code refresh. The frozen B2 comparison is unchanged and all 79 original source path/hash pairs
+remain intact. No private source or live model/provider request was used for this increment.
+
+Before the additional BOM regression, the local aggregate publication gate passes build, **828 deterministic tests** (21 core + 807
+toolkit; optional OML integration skipped), architecture with the existing 13 advisory findings,
+release documentation, inventory, license checks and fresh packed-consumer smoke. The installed
+consumer exercises the public core factory, new runtime modules, real SCIP/RDF/SHACL and MCP.
+Hosted Linux passes the same gate. The initial native Windows branch run passes 700 selected
+toolkit tests and 21 core tests. The concurrent PR run passes every new SCIP case but hits an
+existing timeout-fixture startup race: its 250 ms module deadline expires on the first source,
+before the deliberately blocked second source. The test-only deadline is now 2 seconds; timeout,
+completed-artifact reuse, privacy and old-snapshot assertions remain unchanged. Hosted results
+for that correction are recorded with the change PR. A final real-indexer probe also showed that
+TypeScript strips a leading UTF-8 BOM: the new projection now translates first-line columns back
+to committed text. Its 18 focused range/hydration tests pass, including first-line names, zero-width
+anchors, multiline extents, subsequent lines and unchanged repeat. Final aggregate/hosted results
+are recorded with the change PR. Browser interaction, live
+provider quality, private large-graph volume/usefulness and the 1.0 acceptance decision remain
+separate. Release PR #53 stays open; this increment does not publish npm packages.
+
+## B2b measured extractor decision, 2026-10-29
+
+[ADR 031](ADR-031-code-extractor-selection.md) selects retaining SCIP and exposing richer
+code entities/occurrences in B3a. This increment changes research tooling and records only;
+no production schema, indexer or dependency change and no npm publication.
+
+The [public evidence record](../evaluations/code-graph-results/2026-10-29/README.md) contains
+sealed setup/harness receipts, native bytes, per-case results, corrected projections, query
+witnesses and an explicit export inventory. Inputs remain the 72 frozen files at `f1844f3`,
+digest `04fa4584bd80a9b1a1d964420dcaf35b5e8c61a67515cc57e2476fd44fa798b3`.
+Sixteen fully recorded scenario runs are replayed after two disclosed normalization corrections;
+failed/interrupted attempts remain retained, and no scenario exceeds three native extractions.
+The 342.50-second total includes those attempts and replay, below the frozen 1,800-second budget.
+Extraction runs offline in Docker with 2 GiB cgroup memory enforcement. No paid/model or private
+consumer inputs. Graphify is an isolated pinned source-API experiment, not a verified published
+Python distribution or a new Weftlet runtime dependency.
+
+Cold labeled results: raw SCIP 15/15 declaration families and 16/16 references; Graphify 14/15
+families, 8/16 references and 7/10 native calls. Current Weftlet has coarse candidates for 7/15
+families and 12/16 references but lacks exact occurrences. No named false-join negative was
+asserted by the exact-track arms; unjudged edges and explicit/silent gaps remain reported.
+Calls remain unsupported by SCIP; a reference is not reclassified as a call.
+
+All 96 corrected graph/mode assemblies pass strict SHACL and exact RDF/SPARQL projection parity.
+Documented asynchronous business chains, the negative missing-handoff control, shared contracts,
+dispatch targets and assertion authority survive. Repeat/dirty normalized outputs are identical;
+rename/delete remove old current spans, excluded projects remain explicit, unrelated identities
+and historical snapshot hashes stay stable. These are synthetic deterministic retrieval results,
+not live semantic quality, deployed queue delivery, compatible rule checks or private acceptance.
+All 1,035 allowlisted artifact hashes and all 79 original extraction source path/hash pairs were
+verified. The six-megabyte compressed evidence archive is excluded from npm package contents.
+
+Local Linux build and **786 deterministic tests** pass (17 core + 769 toolkit; one optional OML
+integration skipped), with the existing 13 advisory architecture findings, inventory and license
+checks and fresh packed-consumer smoke. Twelve new scoring regressions join the native Windows
+CI selection; the frozen-input verifier retains seven checks. Hosted results are recorded with
+the exact change PR. This study does not validate Graphify on Windows/macOS or claim general speed ratios.
+Release PR #53 remains open for the later candidate; B3a is next.
+
+## B2a frozen code-extractor comparison, 2026-10-29
+
+The [comparison note](GRAPHIFY-EVALUATION.md) now points to a committed public corpus and
+predeclared protocol: current Weftlet SCIP projection, richer raw SCIP and pinned Graphify
+static extraction. There are 51 labeled declaration/reference/call cases, three explicit-gap
+requirements, three business controls and eight edit/coverage scenarios. The 72 input files
+have inventory digest `04fa4584bd80a9b1a1d964420dcaf35b5e8c61a67515cc57e2476fd44fa798b3`.
+The measurement receipt must identify the full freeze commit as well as this digest.
+
+Seven focused checks validate both configuration templates and typecheck both public corpora,
+reject changed/extra inputs, ambiguous anchors, contradictory labels, unsafe paths and broken
+edits/renames, and verify immutable Git blobs while the checkout is dirty. Real Git CRLF
+conversion remains supported. No indexer or provider is invoked by these input checks.
+Generated output is excluded; all 79 original extraction source path/hash pairs are unchanged.
+
+Local build and 774 deterministic tests pass (17 core + 757 toolkit, one optional OML integration
+skipped), along with architecture (the existing 13 advisory findings), inventory and licenses.
+Hosted checks and fresh packed-consumer installation are recorded with the exact change PR.
+The existing regression suite still exercises real SCIP and hydration on its established
+fixtures; that is not a comparative run on the new frozen corpus.
+
+Graphify's pinned manifest/lock project versions differ (0.9.97/0.9.96). The protocol requires an
+explicit setup receipt before measurement; no Graphify dependency, source or service was
+installed, and no comparative quality, timing, backend-selection or private acceptance result
+is claimed. B2b is next. The candidate release PR remains open, with no npm upload for B2a.
+
+## B1b bounded cited paths, 2026-10-29
+
+[ADR 030](ADR-030-bounded-evidence-paths.md) records `graph.paths` through the public API,
+CLI and MCP. Cited simple paths use deterministic assertion-ID depth-first order, exact saved
+snapshot endpoints, direction/relation filters and explicit page/depth/work bounds. Continuations
+reconstruct bounded state; replay and new steps are reported separately. Pending DFS prefixes,
+depth-cut examples and omitted counts remain explicit. Search completion only covers selected
+saved facts and depth, not unextracted dependencies, causality or business compatibility.
+
+The focused integration run passed 46 tests, including an independent oracle for all 4,096
+directed four-node graphs, competing branches/parallel citations, cycles/self-links, twelve-hop
+paths, empty work pages, frontier bounds, stale/malformed cursors and a 100,000-step hard ceiling.
+RDF projection comparisons retain exact facts. Actual public extraction/SCIP/RDF follows source
+files across the collection/execution boundary through real CLI/MCP continuation while dirty
+and with unavailable providers/indexers. Real overlays retain draft authority and capture/stale
+state, and old snapshots remain readable after deletion. All three original MCP startup tests
+remain enabled and pass. Missing targets in an incomplete document scope retain the exact
+selection/exclusion record and an explicit INCOMPLETE warning.
+
+Final local checks pass build, 767 deterministic tests (17 core + 750 toolkit; one optional OML
+integration skipped), architecture with the existing 13 advisory findings, release documentation,
+inventory, licenses and fresh packed-consumer smoke. The installed CLI/MCP exhaust work pages
+and preserve both a direct Markdown route and the semantic capability route with every citation.
+Shared snapshot metadata and CLI parsing keep neighbors/paths consistent; no architecture rule,
+baseline or exception was changed. All 79 original source path/hash pairs remain unchanged.
+
+The offline viewer embeds the same search function and provides next-path, direction, relation
+and depth controls. Executing rendered controls in a small DOM harness matches the RDF service's
+highlighted assertion IDs and continuation states, including duplicate labels and filtered
+evidence. This proves script/projection parity, not interactive browser rendering. The former
+viewer-only shortest-path behavior is replaced and labeled as cited deterministic paths.
+
+Hosted results belong to the exact change PR head. No npm publication, live model quality or
+private TradePilot acceptance is claimed. B2's frozen comparison protocol is the next task;
+the candidate version PR remains open while B2–B5 work accumulates.
+
+## B1a exact-node exploration, 2026-10-28
+
+The maintainer requested the remaining 1.0 items be tasked and started, with TradePilot live
+acceptance deferred until the next candidate. [The work queue](1.0-WORK-QUEUE.md) records B1–B6
+increments and dependencies. The maintainer believes the 0.6.1 filename issue was the main
+consumer blocker; this is not an additional private acceptance result.
+
+[ADR 029](ADR-029-exact-graph-neighborhoods.md) records the additive `graph.neighbors` public
+contract, CLI and MCP tool. Exact snapshot/node selection, direction/relation filters, assertion
+pagination and frontier counts retain citations, source/assertion authority and node variants.
+The shared read-side helper preserves dirty checkout, captured-overlay and subsequent-edit state.
+No new extraction, storage identity, backend or model/provider integration is introduced.
+
+Local release checks pass build, 736 deterministic tests (17 core + 719 toolkit; one optional
+OML integration skipped), architecture with 13 advisory findings, inventory, licenses and fresh
+packed-consumer installation. The existing 79 source path/hash pairs remain unchanged. The
+public predicate list is frozen without changing its ordered values. Relevant extraction gaps,
+including diagnostics for absent nodes, retain bounded items and explicit omitted counts.
+Hosted results belong to the exact change PR head; this source entry does not claim hosted
+completion or npm publication.
+
+New focused cases retain multiple citations, cycles and self-links; test exact IDs against
+same-label and malicious-looking inputs; exhaust pages against the RDF query; reject changed
+snapshots/filters/page sizes and malformed/stale cursors; and preserve old snapshots when newer
+ones no longer contain a node. Working-overlay tests cover draft authority, renames/deletions
+and stale captures through the new service. The actual public monorepo extraction test follows
+collection to execution's obligation and implementing file through real CLI/MCP transport while
+the checkout is dirty and configured provider/indexer are unavailable. It uses real SCIP and
+RDF/SHACL with a small deterministic prose provider, not a live model. Packed installation tests
+exercise the same new CLI/MCP operation and complete its two-page assertion set without loss.
+
+B1b multi-hop paths/viewer parity and B2's Graphify comparison remain open, followed by B3–B6.
+No new candidate, live semantic-quality result, browser interaction or private consumer rerun
+is claimed. A pending minor changeset accumulates toward the next acceptance candidate.
+
+## Published 0.6.1 physical artifact filename fix, 2026-10-28
+
+Both `@weftlet/core@0.6.1` and `@weftlet/toolkit@0.6.1` are published and verified as
+registry `latest`. [Fix PR 49](https://github.com/jchewitt/weftlet/pull/49) merged at
+`00e09ace22ea2cbd840d118f177b0508a60da9a7`; [release PR 50](https://github.com/jchewitt/weftlet/pull/50)
+committed both package versions to main at `a56e7dca734b40829325cac9f377037bcfdc765d` before
+[trusted publication](https://github.com/jchewitt/weftlet/actions/runs/40768151495). Preparation,
+both uploads and release records passed on the first attempt. Public provenance and both
+package-specific tags identify that release commit; both GitHub Releases exist.
+
+[ADR 028](ADR-028-bounded-artifact-paths.md) records the bounded physical naming policy and
+legacy read compatibility. The public traversal → unchanged repeat → clean-comparison regression
+reproduced `ENAMETOOLONG` in real SCIP before the fix, after the first hydration/repeat succeeded.
+It passes with full logical IDs, cited facts/provenance, cache keys and content digests preserved.
+Long physical keys are hashed and staging names are independent; existing immutable files remain
+readable without rewriting them. Cross-process publication guarantees and strict validation remain.
+
+Release head `56d2db53ab2a3871d1db72cb2b0d04c99e86c9bc` passed
+[hosted Ubuntu and Windows checks](https://github.com/jchewitt/weftlet/actions/runs/40767043720):
+698 deterministic tests (17 core + 681 toolkit; one optional OML integration skipped), architecture
+with 14 advisory findings, inventory, licenses and fresh packed-consumer smoke on Ubuntu;
+571 tests plus inventory/licenses on Windows. The requested Windows traversal/repeat/clean test
+passed, alongside 256-character keys, legacy bytes, immutable winners and concurrent heads.
+The fix PR also passed both platforms. Local build/release checks and package dry runs passed.
+Reviewed exports contain 13 core files and 258 toolkit files, including the runtime naming helper
+and license notices. All 79 original source path/hash pairs are unchanged.
+
+Downloaded public npm archives match the reviewed local and workflow-prepared archives byte for
+byte. Provenance source and subject digests were checked; independent certificate-chain validation
+is not claimed.
 
-This is a focused consumer blocker ahead of [B1–B6](PRE-1.0-BASELINE.md). No private consumer
-rerun, paid/model work or live optional OML compilation is claimed.
+| Package | Verified SHA-512 integrity |
+| --- | --- |
+| `@weftlet/core@0.6.1` | `sha512-xSEjqLulXW052KvPY/Dlu5DS5mTEJQYlHrv0idC3aZIPiuiq+GWA15nJ70T/tf9XE5S4ZbvF4dKJLsom3sIdDA==` |
+| `@weftlet/toolkit@0.6.1` | `sha512-S7iTCLWhGziTVbxBbeJIKu24j1FeuJKEnF+EbHFbH9oKRPveDzt0GnsO7s2AZn8PeDBJjUnl5+zB7X8IVL3AxA==` |
+
+An initial install during registry propagation returned `ETARGET`; a fresh-cache retry with
+`--prefer-online` passed. The fresh public-registry consumer on Linux / Node 24.19.0 installed
+both exact versions and ran configuration validation, traversal hydration, unchanged repeat,
+clean comparison and a saved
+snapshot query through the packaged CLI. The deterministic public grammar provider extracted
+six documents once; the repeat reused all six, and clean comparison made no further provider
+calls. Real SCIP and RDF/SHACL produced 230 facts and 200 query facts. The clean snapshot's
+226-character logical identity is retained, content digests and all facts/provenance apart from
+the containing snapshot ID agree, physical components are bounded, no staging files remain,
+and Git stays clean. This uses a public module provider, not a live model or Codex authentication.
+The Windows regression separately exercises the unchanged Codex adapter with an offline runner.
+
+This focused consumer blocker ships ahead of [B1–B6](PRE-1.0-BASELINE.md). TradePilot should
+upgrade to the patch and rerun its original sequence without clearing compatible semantic artifacts.
+No private consumer rerun, paid/model work or live optional OML compilation is claimed. B1 exact
+exploration and B2 Graphify/SCIP evaluation remain the next development items; no 1.0 is declared.
 
 ## Business-context baseline assessment, 2026-10-28
 
@@ -1861,3 +2355,16 @@ selected tests, including global/local npm launcher installation, discovery, log
 plumbing, process-tree cancellation, CRLF inventories, storage/Git and all original MCP startup
 cases. The repository inventory and complete license checks also passed natively. This is focused
 Windows coverage, not a full native Windows publication gate or a live Codex account/model test.
+
+
+## November 2, 2026 — bounded mapping proposals (candidate work)
+
+The opt-in implementation in ADR 039 adds actual committed-source/SCIP planning, exact cited
+proposal validation, isolated immutable artifacts, Codex/module contracts, and CLI/MCP inspection.
+Deterministic fixture providers are canned transport controls, not a semantic comprehension test.
+Focused checks cover scope/authority negatives, missing/declaration-only/unindexed code, budgets,
+cache round trips, unavailable-provider reuse while dirty, historical rename/delete, cancellation,
+timeouts and concurrent publication. Codex transport/authentication is tested without model calls.
+The public three-request live protocol was predeclared before any live proposal measurement.
+Final local publication gates and hosted platform results belong in the implementation PR once
+executed. No live quality, token-cost, private consumer, browser or 1.0 acceptance is claimed here.