Add provider-neutral semantic probes and independent validation calls

.github/workflows/links.yml

@@ -20,10 +20,10 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - name: Check out the repository
-        uses: actions/checkout@v4
+        uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1, Node 24
 
       - name: Set up Python
-        uses: actions/setup-python@v5
+        uses: actions/setup-python@5fda3b95a4ea91299a34e894583c3862153e4b97 # v7.0.0, Node 24
         with:
           python-version: '3.13'
 

.github/workflows/validate.yml

@@ -23,10 +23,14 @@ jobs:
         python-version: ['3.11', '3.13']
     steps:
       - name: Check out the repository
-        uses: actions/checkout@v4
+        uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1, Node 24
+        with:
+          # Conformance binds historical immutable normative commits. A shallow
+          # checkout cannot verify those Git objects and must not run the suite.
+          fetch-depth: 0
 
       - name: Set up Python
-        uses: actions/setup-python@v5
+        uses: actions/setup-python@5fda3b95a4ea91299a34e894583c3862153e4b97 # v7.0.0, Node 24
         with:
           python-version: ${{ matrix.python-version }}
 
@@ -41,5 +45,8 @@ jobs:
       - name: Validate production-readiness evidence
         run: make readiness
 
+      - name: Validate active publication surfaces
+        run: make publication
+
       - name: Run self-tests, including the negative cases
         run: make test

.learnings/LEARNINGS.md

@@ -0,0 +1,45 @@
+# Project learnings
+
+## [LRN-20260813-001] best_practice
+
+**Logged**: 2026-08-13T02:28:16+02:00
+**Priority**: high
+**Status**: resolved
+**Area**: docs
+
+### Summary
+
+Verify every pinned GitHub blob link against the exact commit before publishing outreach.
+
+### Details
+
+The targeted implementation draft initially linked to `IMPLEMENTATION_CALL.md`
+at the corrected frozen commit, but that path does not exist in the tree.
+Pre-publication `git cat-file -e <commit>:<path>` validation caught the defect
+before any external comment was written. Full link validation then found the
+same dead path in an older ignored draft, showing that publication artifacts
+must be checked even when they are not tracked.
+
+### Suggested Action
+
+For every external comment containing a pinned blob URL, extract its commit and
+path, verify the object with `git cat-file -e`, then confirm public HTTP
+resolution before posting.
+
+### Metadata
+
+- Source: error
+- Related Files: `.scratch/g2-publication/issue-5-targeted-implementation.md`
+- Tags: github, outreach, immutable-target, link-validation
+- Pattern-Key: publication.verify_pinned_blob_before_post
+- Recurrence-Count: 1
+- First-Seen: 2026-08-13
+- Last-Seen: 2026-08-13
+
+### Resolution
+
+- **Resolved**: 2026-08-13T02:28:16+02:00
+- **Commit/PR**: agent/g2-publication publication follow-up
+- **Notes**: Replaced nonexistent blob URLs with Issue #5 links and verified all remaining pinned paths before publication.
+
+---

.superpowers/sdd/2026-08-12-semantic-probes-and-publication/task-2-report.md

@@ -0,0 +1,48 @@
+# Task 2 report
+
+## Loop state — revision 1
+
+- Goal: publish offline semantic-probe conformance fixtures and `errata semantic-test`.
+- Success measure: verified, failed, and unknown fixtures return `0`, `1`, and
+  `2`; malformed input returns `1` without traceback; focused tests, full suite,
+  and `make check` pass.
+- Invariants: standard library only; synthetic data only; erasure records remain
+  content-free; canonical JSON report; no Phase 1 receipt behavior changes.
+- Authority: modify only Task 2-owned files in this worktree; preserve unrelated
+  concurrent work.
+- Primary loop: TDD. Rejected: debugging (no unexplained failure), multi-agent
+  checker (task scope is cohesive and deterministic).
+- Budget: three vertical test/implementation slices, one full validation run,
+  one diff review. Stop and report if Task 1 interface conflicts or a named gate
+  remains red.
+- Gates: PASS — CLI exit behavior (`verified=0`, `failed=1`, `unknown=2`);
+  PASS — malformed JSON returns `1` without traceback; PASS — canonical JSON,
+  operation binding, and content-free erasure evidence; PASS — focused tests,
+  full suite, and `make check`.
+- Evidence: `python3 -m unittest tests.test_cli tests.test_semantic -v` (43
+  passed); `python3 -m unittest discover -s tests -t .` (210 passed); `make
+  check` (repository, claim, readiness, 210 tests, and intentional demo exit-2
+  gate passed).
+- Review: inspected owned diff; no scope creep or Phase 1 receipt changes.
+  `PHASE3_SYSTEMS.md` was unrelated untracked work and was not touched.
+- Learning: no promotion; implementation-specific fixture layout is documented
+  locally and no reusable policy change was discovered.
+- Verdict: PASS.
+
+## Review remediation — revision 2
+
+- Scope: Task 2 review findings only.
+- Change: `--case` defaults to `verified-correction`; exact approved invocation
+  now succeeds without it. Added named provider-error, missing-response,
+  duplicate-response, configuration-drift, and nonconforming-output fixtures.
+- Evidence gate: `tests.test_cli.SemanticProbeConformance` enumerates all eight
+  cases, their required exit codes, canonical output, and expected limitation
+  evidence. Provider error uses structured `error`; nonconforming output uses
+  a structurally valid operation mismatch. Malformed JSON remains invalid input.
+- Gates: PASS — focused CLI/semantic suite (44 passed); PASS — full suite (211
+  passed); PASS — `make check`; PASS — remediation diff has no whitespace or
+  ownership violations.
+- Evidence: exact invocation, every named semantic case, required coverage
+  status, canonical JSON, and expected limitation are asserted in
+  `SemanticProbeConformance`.
+- Verdict: PASS.

AGENTS.md

@@ -98,6 +98,24 @@ Never convert missing lineage, inaccessible stores, skipped tests, timeouts, or
 - Respect quotation, copyright, license, confidentiality, and personal-data boundaries.
 - Preserve contrary and partial-collision evidence. Do not select sources only because they support the proposal.
 
+## Publication discipline
+
+- `publication/active-surfaces.json` is the tracked authority for current public
+  review, implementation, validator, and operated-system calls. Historical
+  GitHub comments stay visible and are superseded append-only; never edit or
+  delete them to make an old target appear current.
+- Before publishing a pinned GitHub blob URL, run
+  `git cat-file -e <commit>:<path>` and verify the public URL. After publishing,
+  read the rendered artifact back and record its author, URL, body, and mentions.
+- Recruitment evidence is not independent readiness evidence. Invitations,
+  mentions, publication, maintainer work, and CI cannot upgrade G2 through G6.
+- Run `make publication` whenever an active surface, frozen target, evidence
+  role, mention, supersession URL, or implementation-licence posture changes.
+- Targeted outreach must begin with a plain-language description of the user
+  problem and exact bounded question. Do not send a specialist only an internal
+  gate label, digest, or repository jargon and expect them to infer why the
+  request matters.
+
 ## Implementation and test expectations
 
 - Use synthetic fixtures; never add real personal, customer, employer, credential, or confidential data.
@@ -108,6 +126,11 @@ Never convert missing lineage, inaccessible stores, skipped tests, timeouts, or
 - Verify receipt binding to the erratum and relevant pre-repair and post-repair state.
 - Keep deterministic structural checks separate from model-graded behavioral probes.
 - Never mark an opaque required store green merely because accessible stores passed.
+- Never hardcode checkpoint success from enumeration. Preserve adapter-supplied
+  quarantine-phase coverage and carry a worse checkpoint result into the receipt.
+- Keep the declared adapter protocol synchronized with every controller and
+  strategy call; independent adapters must own their repair inputs rather than
+  depending silently on reference-ledger registration.
 
 ## Required completion check
 
@@ -122,11 +145,12 @@ Before presenting work as complete:
    make check
    ```
 
-   This runs four validation components, and each answers a different question:
+   This runs five validation components, and each answers a different question:
 
    - `scripts/validate_repo.py` — are repository structure and metadata well formed?
    - `scripts/claim_guard.py` — does the documentation still state the bounded claim? It anchors to exact sentences in named files, because keyword presence over the whole corpus cannot tell a hedge from its inversion.
    - `scripts/check_readiness.py` — is readiness evidence structurally honest and synchronized with `PRODUCTION_READINESS.md`?
+   - `scripts/check_publication.py` — do active public calls bind the corrected immutable target, current licence posture, distinct evidence roles, and honest evidence boundaries?
    - `tests/` — do the checkers actually reject the faults they claim to catch? A checker that has only ever passed has not been shown to work.
 
    Readiness-check exit `0` validates structural honesty of the recorded evidence. It does **not** mean `PROD_READY`; the ledger and human matrix state the current verdict.

CHANGELOG.md

@@ -6,10 +6,110 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
 
 Changes to the bounded novelty statement or the source comparison are recorded here even when they narrow or retire part of the claim. That is the intended direction of travel, not an exception.
 
-## [Unreleased](https://github.com/thomaswillner/llm-errata/compare/v0.3.0...HEAD)
+## [Unreleased](https://github.com/thomaswillner/llm-errata/compare/v0.4.0...HEAD)
+
+## [0.4.0](https://github.com/thomaswillner/llm-errata/releases/tag/v0.4.0) - 2026-08-13
+
+### Added — adapter conformance and validator hardening
+
+- Added five provider-neutral adapter cases with complete checkpoint,
+  aggregate, triad, store, and receipt outcomes. Calls are traced through the
+  exact target adapter instance rather than matched globally by function name.
+- Added bounded proposition-multiplicity preservation. Synthetic conformance
+  bindings expose stable provider-local proposition identity and active count;
+  an unobservable count is `unknown`, not a pass.
+- Added exact semantic mutation controls. A mutation must complete and produce
+  its declared counter-result; an exception or different failure cannot earn
+  credit.
+- Added three executable validator attacks covering empty receipts, incomplete
+  no-op feed acceptance, and constant-`unknown` semantic aggregation.
+- Added `errata adapter-conformance`, canonical JSON reporting, and distinct
+  pass, failed-control, and invalid-evidence exit codes.
+
+Rastislav Drahos/DanceNitra materially advanced this release by reporting the
+adapter-coverage gap, a real duplicate-preservation defect, candidate adapter
+behaviors, and the three anti-vacuity attacks. The immutable MIT-licensed source
+artifact is
+`DanceNitra/agora@2ba1e299b3483b9038d03387345702427608b90b`; Inspeximus is a G4
+candidate and the source commit discloses Claude Opus 5 co-authorship, so the
+input remains interested-party evidence. LLM Errata independently authored its
+corpus and validator without copying or vendoring that runner or fixture.
+
+### Status and ownership
+
+- Repository files authored here remain copyright Thomas Rainer Willner.
+  Contributor credit and MIT permission do not transfer copyright.
+- Existing licence terms are unchanged: independent commercial and
+  non-commercial specification implementations are permitted under the
+  attributed implementation grant; Reference Code remains restricted.
+- Verdict remains `NOT_PROD_READY`. G2 through G6 remain blocked by their
+  existing external evidence requirements.
 
 ### Added — production-readiness evidence
 
+- Durable quarantine checkpoints now preserve adapter-supplied phase coverage
+  instead of hardcoding every enumerable adapter to `verified`; a later final
+  success cannot erase an earlier `partial`, `unknown`, or `failed` result.
+- `StoreAdapter` now declares the complete controller and repair call surface,
+  including quarantine coverage, source mapping, store-owned repair inputs,
+  retirement, rebuild, and recall. Independent adapters no longer have to copy
+  their lineage into the reference `LineageLedger`.
+- Recorded Inspeximus `v2.7.0` as a tagged external adapter candidate with its
+  preserved v2.6.1 contamination disclosure and claimed clean-room rewrite. It
+  targets historical LLM Errata commit `a477fe4` and remains unvalidated G4
+  evidence until provenance review and current-target rebinding are complete.
+
+- Corrected two conflict-disclosed external findings. Importers now remember
+  same-view sequence conflicts across separate observe calls, while receipts
+  explicitly disclaim global non-equivocation across split views. Required
+  adapters must establish root-specific lineage completeness before an empty
+  enumeration can receive `verified`; unsupported empty walks become `unknown`
+  with a signed limitation. The earlier categorical comparison claiming LLM
+  Errata was stricter than Inspeximus is withdrawn and replaced with the
+  verified limits and remediations in both projects.
+- Upgraded both GitHub Actions workflows to immutable `actions/checkout`
+  v7.0.1 and `actions/setup-python` v7.0.0 commit pins. Both action releases use
+  Node 24, removing GitHub's Node 20 deprecation path without changing the
+  Python 3.11/3.13 validation matrix or local Node runtime. Repository lint and
+  negative tests now reject missing, mutable, or downgraded action references.
+- Granted irrevocable, worldwide, royalty-free rights for independently
+  authored commercial and non-commercial implementations of the specification,
+  conditioned on accessible product attribution to LLM Errata and Thomas
+  Willner. Reference code remains personal-use, no patent or certification
+  right is granted, and the historical Apache-2.0 grant remains unchanged.
+- Split the callback requirement into independently evaluated authenticity and
+  coverage-truthfulness properties. A valid signature authenticates the
+  importer and receipt bytes; it cannot turn overstated, missing, or opaque
+  coverage into a truthful claim.
+- Reordered the production-cryptography qualification to foreground observed
+  refusal of malformed lengths, non-canonical scalars, tampered messages and
+  signatures, and wrong keys. Successful RFC vectors remain compatibility
+  evidence and do not establish constant-time or production assurance.
+
+- Corrected the Inspeximus comparison after conflict-disclosed maintainer
+  feedback and source verification. `retract_lineage` plus `rederive` is a
+  stronger local quarantine/rebuild collision than the earlier erasure-focused
+  row; its coverage audit remains weaker than required-store aggregation.
+- Completed internal Phase 2 conformance surface with owner-key rotation,
+  rotated-key refusal, concurrent sequence-conflict, invalid-target,
+  content-free confidentiality, and every-field receipt-binding vectors. G2
+  remains `BLOCKED` until dated independent review of exact committed surface.
+- Added a ten-scope G6 operational evidence contract and fail-closed checker.
+  One independent report must bind exact commit and deployment, declare
+  workload, platform, failure domain, observation window, numeric thresholds,
+  units, comparators, and raw artifacts, then pass every measurement. Internal
+  tests validate the contract but leave G6 `BLOCKED`.
+- Added durable `errata quarantine` checkpoints. CLI repair now requires an
+  authenticated, atomic checkpoint bound to erratum, sequence, target,
+  inspectable pre-state, adapters, opaque limitations, and gated artifacts;
+  consumption occurs only after durable receipt and applied-state writeback.
+  G2 remains `BLOCKED` for remaining Phase 2 vectors and independent review.
+- Added a fail-closed production-cryptography qualification record. PyCA
+  reproduced the repository's RFC 8032 vectors but documents no external
+  project audit; libsodium has a published assessment for older releases, not
+  the current 1.0.22 build. G3 remains `BLOCKED` pending exact-build
+  qualification, lifecycle implementation, and independent security review.
+- Added provider-neutral semantic probes, deterministic recorded fixtures, and `errata semantic-test` documentation as Phase 2 item 6 internal implementation. Opened independent review, independent implementation, and Phase 3 system-nomination programs. G2 remains `BLOCKED`: repository artifacts and local tests do not substitute for dated independent external review.
 - Added a fail-closed production-readiness ledger and human verification matrix. Local green tests cannot produce `PROD_READY`; every required gate must pass, and external gates require dated evidence naming an independent producer.
 - Added drift checks binding README maturity and SECURITY support policy to `VERSION`. The current verdict remains `NOT_PROD_READY`: only G1, document consistency, is complete.
 - Normalize complete-cell Markdown presentation decoration before readiness-matrix duplicate and contradiction checks.
@@ -89,7 +189,7 @@ All six were found by adversarial review after the suite was green, and each has
 
 - `CHANGELOG.md` used `[version]` heading syntax with no link definitions, so the headings rendered as literal brackets, and it listed a placeholder as though it were a change. Version headings are now plain until the repository has a public URL to compare against.
 
-## [0.1.0](https://github.com/thomaswillner/llm-errata/releases/tag/v0.1.0) - 2026-08-01
+## 0.1.0 - 2026-08-01
 
 ### Added
 

CITATION.cff

@@ -5,8 +5,8 @@ type: software
 authors:
   - family-names: "Willner"
     given-names: "Thomas Rainer"
-version: 0.3.0
-date-released: 2026-08-07
+version: 0.4.0
+date-released: 2026-08-13
 license: LicenseRef-LLM-Errata-Personal-Use
 license-url: "https://github.com/thomaswillner/llm-errata/blob/main/LICENSE"
 abstract: >-
@@ -21,4 +21,3 @@ keywords:
   - provenance
   - conformance
   - verifiable repair
-

CONTEXT.md

@@ -0,0 +1,36 @@
+# LLM Errata Publication Context
+
+This context defines the language used to distinguish current public instructions
+from retained publication history without weakening evidence provenance.
+
+## Language
+
+**Active surface**:
+A tracked public artifact that currently instructs a reviewer, implementer,
+validator, or system operator which immutable target and evidence role to use.
+_Avoid_: Latest comment, current main, final link
+
+**Historical surface**:
+A retained public artifact that accurately records an earlier project state but
+must not be used as current instructions.
+_Avoid_: Bad comment, deleted record
+
+**Frozen review target**:
+An immutable source commit paired with the canonical G2 surface digest that a
+review or independent implementation must bind.
+_Avoid_: Branch head, latest build
+
+**Supersession**:
+An append-only correction that explicitly marks an earlier public instruction
+historical and publishes its current replacement.
+_Avoid_: Rewrite, deletion
+
+**Evidence role**:
+One independently attributable responsibility in the readiness program, such as
+reviewer, adapter author, validator author, or operated-system owner.
+_Avoid_: Contributor, interested party
+
+**Recruitment evidence**:
+Proof that a bounded request was published to a relevant party. It is not proof
+that the party accepted, performed, or passed the requested work.
+_Avoid_: External evidence, review evidence