Repository navigation
CBA-KB v2.1.1 Engine Release
CBA-KB Engine v2.1.1 — Engine-only Release
Scope: Engine code only. This Release publishes the v2.1.1 local Engine code. It does NOT authorize formal data release, Identity release, E4a, E-1, Production, Consumer acceptance, full version closure, formal current changes, or any real Drive write. Issue #226 stays OPEN.
Release identity
- Tag:
v2.1.1(annotated), target commita12ae3ddb4f66b63bdf318f0068b1e065f1ec792(PR #229 merge commit; parents: mainb89b7f48…+ PR headf8ae6946…). - Git tree:
ed9d4ae644b762d358a55a53a2cc14a2806f5fff(merge tree == PR head tree, no drift). - Q2-reviewed HEAD:
f8ae6946cd7c2abc4889d5e414f5074cdcf3fbf4(PR #229 head; independent Q2 PASS, limited static scope — see below). - Dependency lock:
requirements.lock(39 pinned lines), SHA-2564f4feb47f72405e2f1bbe95193083e4d956a8ae5a201b66f95c916296a525882. - LACD: v1.0.2 (
.lacd/lock.json: lacd_cli_version 1.0.2, SkyCjq/LACD ref v1.0.2). Build environment: GitHub Actionsubuntu-latest+ Python 3.11. - Package version vs Tag (independent naming, v2.1.0 precedent): the source package version is actually
2.0.2(pyproject.tomlandsrc/cba_kb/__init__.pyagree). Tagv2.1.1is the Engine Release version identifier. The numbers differ by design; no source code was changed to match them.
Delivered Engine functions (local code behavior)
- Fact sidecar guarantee; missing-evidence facts report UNKNOWN only.
- Claim trusted event association (version-bound canonical event reference or formal association assertion); OK / CONFLICT / UNKNOWN per evidence; event_key alone is a candidate index only; bounded bidirectional event-key judgment.
- Identity controlled entry (six-item proof, target-first, I/V scope separation, approved-alias filtering with approved aliases currently none, pagination completeness independently bound, same-name clarification, F-02 safe-failure semantics).
- Local Producer: canonical → trusted mapping → snapshot/manifest → Markdown navigation + machine data, local output only; all consumer outputs marked
simulated: True; formal and synthetic sources never merged; identity HOLD without a formal identity object. - Trust-boundary fixes: V-out-of-index reason internalized (generic external reason, specific cause internal only); session anchor recursively frozen at construction; direct resolver clarified as an internal projection primitive with call-graph / divergence / signature guard tests; formal assertion artifact requirement; session identity authorization with shared validation;
ref in anchor_textpartial-match fix; identity candidate row-total reconciliation. - Positive capabilities preserved: canonical anchor binding, formal assertion CONFLICT, different-event OK, Claim OK/CONFLICT, Identity scoped RESOLVED, same-name clarification, post-revocation render refusal, original Fact/Stats paths.
Quality evidence
- Merges: PR #227 (
b89b7f48…, 2026-10-31) and PR #229 (a12ae3dd…, 2026-11-01), both Human-approved and Human-merged; Bot-authored PRs via Bot Broker. - CI: PR head branch checks 7/7 success (policy x2, lacd-verify x2, cleanroom-smoke, evidence, tests); main push checks 3/3 success (lacd-verify, cleanroom-smoke, tests). Note: the main push tests job is the push subset; PR-only evidence steps do not run on main push. The full evidence suite comes from the branch checks on the byte-identical tree.
- Independent Q2: PASS, limited to this round's code obligations and classification honesty (static review; reviewer did not run tests/CI, did not verify the GitHub HEAD, did not read baseline source outside the diff). Session (must be preserved): https://chatgpt.com/c/6ae69d46-ebe8-83ea-ab98-c8a7b4474869 . PASS is not full product acceptance and not a data or Production PASS.
- Test migration (1 place, legitimate behavior fix, not a weakening): the V-out-of-index assertion now requires the generic external reason, absence of the specific reason in output JSON, and retention of the internal diagnostic.
- Local full suite (author-reported, Q2 did not independently verify): 2295 passed, 6 skipped, 24 failed + 24 errors, name-by-name identical to baseline, zero new failures; attack drill 14/14; secrets scan passed.
- Local 24 failed + 24 errors are known environment differences (test_release googleapiclient local env, 1 PDF case, test_archive_read_retry local env). They are explicitly NOT formally exempted and NOT claimed resolved. CI tests success and local failures coexist because the run environments differ.
LOCAL residual limitations (Human risk decision 2026-11-01; trusted single-process use only)
- The direct resolver is an internal computation primitive: direct calls with a self-built anchor can yield a bare result, but that never confers formal trusted Consumer qualification. Accepted with the existing product call boundary and guard tests; no new trust platform is built.
- The local single-owner session model keeps authorization sets mutable (revocation support). V-subset-of-I does NOT block the owner from widening V within I. This must NOT be presented as cross-account ACL or arbitrary-caller isolation.
- Anchor immutability covers the verified objects and call boundary only: in-place dict/registry mutation is blocked, but same-process owner rebinding of the
session._trust_anchorattribute and unlisted custom mutable objects are outside the guarantee. - Local
simulatedproof does not authenticate a real publisher; the section 5.4 local verifier covers only auxiliary self-consistency of local synthetic projections. - Call-graph guard tests are text-search based and cannot exclude dynamic calls; the full call graph is author-provided, not independently traversed. This is a proof-power boundary, not a confirmed vulnerability.
If a real Consumer deployment allows untrusted parties to directly access or rewrite sessions, these conclusions must be re-evaluated at the real deployment Gate. No further code repair or Q2 rounds target these same disclosed limitations unless a new really reachable counterexample overturns the product call-boundary judgment.
Uncompleted gates (all NOT PASS; unchanged by this Release)
- B-1 formal event evidence release: candidate complete, RELEASE blocked (key-only bidirectional identification unproven).
- C-1 formal identity index or bounded HOLD: candidate complete, formal index pending.
- E4a formal trusted identity entry: HOLD (NOT_VERIFIED).
- E-1 corrected receipt: OPEN; E4b results must not be claimed as passing the prerequisite evidence gate before the receipt satisfies its agreed conditions; the existing receipt is Muse-produced, not independent adjudication.
- Formal data release, Identity release, real Drive, Phase-2, P0, cross-account ACL, Production: all NOT_VERIFIED, each under its original acceptance requirements.
- Five states after this Release: ENGINE_RELEASED = YES; PRODUCTION_COMPLETE = NO; CONSUMER_ACCEPTANCE_PASS = NO; PROJECT_CLOSED = NO.
What this Release does not do
- Does not modify the v2.1.0 Tag or Release. Does not move any Production/current pointer. Does not close Issue #226. Does not publish any candidate data as formal. Notes factual errors, if found, will be corrected by a formal correction record, never by rewriting the Tag history.