fix(reporting): integrate reviewed reporting and signing corrections

[bokelley]

Publication validator correction

Installed production conformance requested an rc4 revision validator that the rc6 package does not ship. This child changes the publication-clock helper's request and validator pins to the shipped rc6 contract. All 30 assertions, historical offline controls, SDK code, packaged schemas, version policy, workflow selectors and budgets remain unchanged.

Current head 8d6eba964b86c209a26df5decfef4f47d1c0a576, tree ae7e33d462bde2eb8862000e94479e6d3445b0d0, sole parent a124afd3b72f3fb81e30e40b298222a683e71d5f; base remains 940c95e0c2d93758ed334b3edff59cbe933362d4. Child: one file, +2/-2, binary-diff SHA256 da32f920940cc03b018761ef0b8dd58084d44b5c7b1a1d18dffe41aeab6c3fd0. Full 34-path composition diff SHA256 ef3b9c4831750624c1930c0e8fee809d23b0a9a0e899e2dfbe1844ead2937084.

Fresh exact-source Python3.12.14 validation: 11 passed, 11 existing PostgreSQL-URL-absent skips. Ordinary commit hooks passed on the first attempt without changing bytes. Separately, the same candidate test bytes against the retained original CI wheel d484dfd9fe20f0bad212b2948d0feee41be07a464f783c0f0ef54b22e5df14b9 yielded 24/24 real PostgreSQL/HTTP cases on Python3.10.21, and 11 passed/11 driver-route skips in the base probe; the original helper reproduced 4 failures/7 passes/11 skips. That wheel was built at original merge e85285428b859a9abd5a43a3a080dae6d03d9e6f, not this new child. These diagnostics do not qualify a new child distribution. The named revision-validator closure is five documents: four identical, and the revision schema differs only in versioned references; this is no whole-protocol equivalence claim.

Parent a124's six original attempt-1 runs are sealed. CI36124859608 remains FAILURE: 39 jobs, 34 success/5 failure. All four installed-production routes failed (base:1053 passed/4 failed/386 skipped; PG:1435 passed/8 failed/0 skipped), and the strict aggregate correctly failed. Their failures, original artifacts, full logs and independently passing controls are retained under seal 1265eeae72e3fe3e78c6790a911b9474e43b0b28833c6656c3afded2c147259d; no original was retried, cancelled or relabelled. New independent review and complete fresh CI/installed controls are pending for this head.

Fresh head-specific Gitleaks disposition SHA256 eb84c226c810fcb272e9b4d6caf01a7bfa42893d6871eb433799ac8d7de57bfc: all 34 files and 17 commits/merge-parent additions covered, positive/negative controls passed, 113 public schema/example matches freshly adjudicated with zero unadjudicated matches or confirmed private credentials. This is owner evidence, not a vendor success or policy substitution. No merge, release or complete rollout acceptance is claimed.

Retained a124 history and acceptance map

The following evidence belongs to its explicitly named earlier heads; current-head wording inside this retained history refers to a124 or its stated parent. All new-head gates above remain required.

Prior a124 correction and validation

Current head: a124afd3b72f3fb81e30e40b298222a683e71d5f, tree 6d9aa7b80597e806d9cf7fef6cb13eea6f3e702b, sole parent c28907bbadadd83dadfa1e1c1dc2e80694ffabbe. Base remains 940c95e0c2d93758ed334b3edff59cbe933362d4.

Installed conformance failed to collect because four signing fixtures selected the historical rc4 compliance directory, which is not shipped. The current rc6 directory contains the same 31 webhook vector paths and byte-identical blobs. The child selects those shipped files while retaining historical version/model tests. It also removes the unused generated Mapping import, makes the generator emit it only for the scope model that uses it, and strengthens the generator idempotence/import regression. No schema, signed vector, packaging, workflow, runtime guard or timeout is relaxed.

The seven-file child is +19/-8 (Git binary/abbrev=8 SHA256 6f92f8cd2fd6f982fe2bd2cb9a0cf91d711546fdaf50d962abba9506f11e44f5). The full range remains the same 34 paths, +4644/-72 (SHA256 c0e176fdf1aebbb090df81a0c8a54d3a02af22465d199a27f978be4ff872df73).

Focused source validation on Python 3.12.14: 94 passed, 0 skipped, including generator/idempotence, scope/exact selector models, signing vectors/emission and a separate real HTTP receiver. Ordinary commit hooks passed without adjustments or bypass. Command: python -m pytest -q tests/test_reporting_selector_generation.py tests/test_reporting_scope_models.py tests/test_reporting_exact_request_models.py tests/conformance/signing/test_webhook_rc4_vectors.py tests/conformance/signing/test_webhook_signature_emission.py tests/conformance/signing/test_webhook_signature_http.py. This is source validation, not new installed qualification.

Parent original CI36118568262 remains 34 success / 5 failure, 39 jobs. The four installed-production routes had two collection errors each and zero executed cases; their strict aggregate correctly failed. Its terminal guard seal e926b7e56a4a2681595efb57599d3b810df4ae29f44bc4d62d0cd81e5b67b53a and all originals are preserved. The parent source approval and successful controls do not transfer to this child. Fresh independent child review and all 39 CI jobs, four installed routes, complete lifecycle controls, current scanner disposition and final checks are required.

Fresh exact-child scan disposition: 1b3bce5150a4ec5d96378ec1078c5702aedf33d232ba43b371419353b1181744; current 34 files and all 16 commit/merge-parent additions independently reconciled, positive/negative controls executed, 113 public schema/example matches source-adjudicated with no unadjudicated match. This is owner evidence, not GitGuardian success or policy acceptance. One finalizer invocation used a system Python without tomllib; the completed scanner outputs were unchanged and finalization then used the existing Python3.12 environment.

Typed reporting requests could add incompatible selectors or defaults, numeric cursor binding could change received values, and webhook emission used the wrong Signature encoding. A continuously due account could also exclude newly enrolled accounts, while acquisition-time finality could be published with an earlier revision creation time. This integrates the reviewed corrections with actual rc6 main, preserving main's restatement checkpoints and version policy.

Reviewed parent composition and retained behavior

• Base: 940c95e0c2d93758ed334b3edff59cbe933362d4 (PR1193 merge, tree 1185c7483786a4f4583f83b4c09d44ba07ff5248).
• Original feature: 25e0c7278a19493345975881d1578c76fc64dc1b; original feature base 1f953c40d761be71d11fde84c78359ff2074fe7c.
• Composition: 1acd02dc2885242976519b483bb363cd350e958b, ordered parents [25e0c727, 940c95e0].
• Candidate: c28907bbadadd83dadfa1e1c1dc2e80694ffabbe, tree ed191eee6c08a919803fc96dc1bc73894db0178f, sole parent composition.
• The original 34 feature paths are retained with no added or lost paths. Full base-to-head diff: 34 files, +4633/-72; Git binary/abbrev=8 SHA256 021f4fd3590699250a2eecf7745b420f8f05ea6ca9103bcfeb1dd25891b9f429. Corrective composition-to-head diff: 8 files, +51/-29; SHA256 a0a326d1d4ba1e886e4e676aaf07cf9179697744b345117f9dcd807e4154df39.

All five component branches and ordinary merge ancestry remain reachable: wire/numeric/webhook b203bbcf (#1202), runtime progress dba15b6b (#1206), publication/replay 26151fc5 (#1207), revocation classification ccb7713f (#1205), and fallback hygiene fe1a1cbd (#1204). Those verdicts and executions remain attributed to their original objects. This composition requires fresh review and CI.

Integration details

The sole textual conflict was in ReportingProducer.acquire_obligation. The resolution keeps main's unchanged-snapshot early return and both restatement-checkpoint paths. It samples the publication clock after staged reads and the unchanged-snapshot branch, rejects clock regression, commits using that publication instant, and then records the checkpoint using the original dispatch instant. Reversing only that publication delta recovers main's acquisition-method AST exactly; both checkpoint blocks equal main. Immutable replay, finality bounds and changed-content rejection from the feature are retained.

The materializer clears its sampling hint only after a successful committed account turn. Main's native timestamp ordering, account/row locking, fences and rollback remain; lock misses retain continuation. The hint remains unsynchronized instance state: correctness is protected by database locking, while concurrent hint races can affect sampling fairness.

Other retained corrections preserve typed scope and exact-read selectors, exact received integral values, webhook Base64URL emission and mixed-alphabet rejection, the documented revocation fetch/parse/freshness classification, and sanitized fallback behavior. Generated-model changes have a matching generator repair; no schema-cache regeneration occurs. The Content-Digest wording discrepancy remains separate. Credential-key screening remains best effort, not a proof that metadata contains no credentials.

The corrective child changes tests only. Five current live-route pins in three harness/test files move from rc4 to rc6, matching the existing packaged default and resolver. Historical offline webhook vectors and the direct legacy-contract tests retain their original versions. Main already rejects the old explicit live pin; no SDK version support or guard is relaxed. Nineteen awaited predicates across five existing files move to ordinary assignments (five outer-await and fourteen whole-predicate forms); inverse AST reconstruction preserves expressions, short circuit, assertion messages and all other logic after accounting for the declared live-pin edit.

All eight schema manifests remain byte-identical to main (1693 entries). SQL, schema cache, version tables, project/dependency metadata and signed compliance inputs are unchanged. Every CI job object equals main: 24 keys expand to 39 jobs, with the strict eleven-result required gate, receipt shard verification, permissions, budgets and coverage floor intact. Only the two inherited pull-request branch-filter additions differ from main's workflows. Current installed-production module selection derives to 63 unique module keys from 46 literal occurrences and 20 glob matches; duplicates are deduplicated. The corrective installed-contract append adds decisioning.dispatch and signing.verifier to the origin/hash map so every changed runtime module is covered, and adds the later revocation boundary file to the installed test selection. The 16 assets remain unchanged. Fresh distribution bytes and actual test counts must be derived again, not inherited.

Local validation and limits

• Focused source composition: 259 passed / 20 database-absent skips / 14 warnings. This predates the corrective test edits and is not installed or full-suite qualification.
• The unchanged current-live fixture rejected the old explicit pin in both memory notification modes (two expected constructor failures). After the input correction, the focused source controls passed 84 with 20 database-absent skips and 14 warnings; counts overlap the earlier selection and are not summed.
• Focused memory/libc PostgreSQL and real public HTTP integration: 62 passed / 0 skipped in 588.75s. This tested the frozen SDK and test candidate before the final three-line installed-map/selection addition. Selected hashes and worktree identity were stable throughout, all source test helpers remained source/editable Python 3.12, and the owned database cluster was cleaned. Subsequent formatting was AST-neutral. This is not an installed wheel, Python 3.10, frozen-artifact, full-suite or coverage result.
• Actual workflow command collection: 4735 unique nodes, pairwise disjoint lane sets and exact full union. Receipt rolling uses the complete eight-node collection; this is not eight-shard execution. The final packaging-only append leaves all these selectors and declarations unchanged.
• Final installed-selector source collection: 1444 collected minus the one existing deselection = 1443 selected across 47 files. This is source collection, not a prediction forced onto installed CI results.
• Both commits passed normal hooks. The corrective hook initially removed redundant generator-expression parentheses; its original stop is retained, full-file AST equality was verified, and the ordinary retry passed without suppression. The exact conventional-commit workflow body, rendered only for the actual main base, passed all 15 commits with no merge-title exemptions.

The original normal composition hook attempt stopped during editable build provisioning with ENOSPC. After cleaning only disposable adcp build-cache files, the same normal hooks passed and the composition committed. Both streams are retained. One read-only structural recorder initially selected the wrong checkpoint method name; its zero-match comparison supplies no proof. The corrected exact-name read independently matched both real checkpoint blocks. Neither incident is SDK or CI evidence.

Fresh CI must complete all 39 jobs, the eleven-result gate, complete Python coverage, all installed-production VCS/sdist and base/PG routes, receipt shards, materializer/feed/hardening/production frozen controls, real installed origins and whole job lifecycles. No old four-cell artifact, source-only probe, collection or partial log qualifies this head. The inherited Python 60-minute job / 45-minute coverage step and 80% floor are unchanged. Translator PARTIAL 8 pass / 47 skip remains a bound, not full interoperability.

Historical and standing boundaries

PR1203 remains superseded evidence, not rewritten: its invalid conventional-merge subjects and original failed CI remain unchanged. PR1208's original 25e0c727 ancestry replaced those metadata merges without changing their corresponding source trees; that content/history attribution is retained. Earlier component supervisors, unknown original failures, separate complete runs and any partial observations stay on their original objects. No historical failure is waived or relabelled here.

Nine rolling release-note boundaries remain open: pre-17ee A, pre-0f34 B, pre-967 C, pre-5487 B1, pre-3fd B2.1, pre-09fd B2.2, pre-2d777 B2.3, pre-e16 hardening and pre-34c production. Historical schema availability is offline support, not live-version advertisement. The 9rLB source half is attributable to merged PR1191 only, with no under-load or closure claim; schema materialization caching is a distinct object. F-SX remains inherited fail-closed with no snapshot-retention promise. Epoch-zero, quarantine and advertisement/readiness non-promotion remain.

Secret scanning is separate: fresh exact-head Gitleaks disposition 6deceb8fd3d1219416f6f9e851585cb6937f0017ffbaf30e7450c1982a4c9d4c is prepared but not yet accepted as a substitute. Root alone may apply standing train policy after authenticating fresh GitGuardian output. Neutral/size-skipped is never clean-scan credit. Genuine CodeQL full output and annotations must be inspected, including any positive alert despite a success conclusion. The documented wildcard/optional-auth/five-B104 exposure remains recorded; no new exposure is waived. The missing literal check/App binding, visible 14/17 inventories and full-protection 403 remain separate, without exclusive-blocker claims.

No artifact, release, activation, #1172, #1199 or TypeScript acceptance. Guard #1201 remains draft d6e7c228 and LAST; legacy workflow 204238826 remains disabled. Final main artifacts must independently retain Pydantic 2.13.0 / MCP 2.0.0 floor support; candidate version metadata is not registry artifact identity.

[aao-secretariat]

Ladon verdict: Approve

Approve — clean integration of six previously-reviewed reporting/signing corrections onto rc6 main.

What I checked:

• No Critical, High, or Medium findings from the reviewer.
• high_risk: true flags only (modified) files (decisioning/dispatch, a2a_server, signing/*). Per the reading of the high_risk flag, modified sensitive files with no medium-or-higher findings are presumed safe; row 5 requires a medium finding, and none exist.
• gated_paths: false, so row 2 does not fire despite review_decision: REVIEW_REQUIRED.
• No no-auto-approve team match (row 7 N/A).
• Source changes fail closed and are covered by new tests: publication-timing clock-regression guards, fail-closed revocation classification, unpadded Base64URL webhook signatures with mixed-alphabet precheck, exact feed-number handling, materializer cursor reset, and generated-model selector validators paired with a matching generator repair.
• The dispatch/CLAUDE.md credential-gate change is documentation-only; _CREDENTIAL_SHAPED_KEY_SUFFIXES and gate logic are byte-identical — no weakening of the no-credentials-in-ctx_metadata guard.

None of rows 1–8 fire; falls through to row 9 = approve.

[aao-secretariat]

Ladon verdict: Approve

Approve — clean review, no blocking or medium findings.

Checked the decision table top-down:

• Rows 1, 4, 5, 6, 8: no findings at any severity (reviewer reports zero Critical/High/Medium).
• Row 2 (gated paths): gated_paths: false — does not fire despite review_decision: REVIEW_REQUIRED.
• Row 3 (high-risk deletions): high_risk is true, but all five high-risk reasons are (modified), none (deleted).
• Row 7 (no-auto-approve team): no team match.
• Falls through to row 9 → approve.

The high_risk flag is true only from modifications to decisioning/, server/, and signing/ paths; per the change-kind guidance, modifications with no medium-or-higher findings are presumed safe. The reviewer confirms the changes are spec-faithful and internally consistent: typed reporting selector validators paired with matching generator repair (generated code stays regeneration output), the Scope.all_media_buys wire-fidelity fix, fail-closed revocation-checker classification, consistent webhook 3.1 profile on emit/verify, publication-clock-regression rejection, and a materializer cursor reset scoped to committed turns. _CREDENTIAL_SHAPED_KEY_SUFFIXES is unchanged (doc-only reframing), so the no-credentials gate is intact.

[aao-secretariat]

Ladon verdict: Approve

Approve

Clean delta pass with no critical, high, or medium findings.

What I checked:

• The Mapping import in fix_reporting_request_selectors is now gated behind a needs_mapping flag, added only when the scope class matches (exactly where Mapping is used) — sound conditional-import behavior with added test coverage.
• The now-unused import was removed from the regenerated delivery-request model — consistent with generated-code posture.
• Webhook signing tests repointed from rc.4 to rc.6 vectors, verified byte-identical.
• No public-surface/semver break, no import-layering violation, no credential-in-metadata concern, no CI-gate disabling.

High-risk flag: true, but all reasons are (modified) files with no accompanying medium-or-higher finding — presumed safe per the change-kind heuristic. No deletions.

Gate checks: gated_paths is false, so row 2 does not fire despite review_decision: REVIEW_REQUIRED. No no-auto-approve team match. Prior decision was approve, not a sticky escalation. No findings of any tier.

Rows 1–8 do not fire; falls through to row 9 → approve.

[aao-secretariat]

Ladon verdict: Approve

Approve — clean subsequent pass, no blocking or medium findings.

This delta-only review covers a single test file change (tests/conformance/reporting/test_reporting_publication_time.py, +2/-2) bumping the request adcp_version pin 3.2-rc.4→3.2-rc.6 and the validator pin 3.2.0-rc.4→3.2.0-rc.6. Both pins are correct: main now gates on is_adcp_version_at_least "3.2-rc.6" (ADCP_VERSION=3.2.0-rc.6) and the 3.2.0-rc.6/core/reporting-revision.json schema is shipped so the validator resolves. Test-only change — no public surface, schema, generated-code, or credential path touched.

Decision path: no critical/high/medium findings (rows 1, 4, 5, 8 don't fire). gated_paths is false (row 2 n/a). high_risk is true but all high-risk reasons are (modified) with no medium finding on them, and none are (deleted) — rows 3 and 5 don't fire. Prior decision was approve, so row 6 is n/a. No no-auto-approve team match (row 7 n/a). Falls through to row 9 → approve.

Note: high_risk flag is a heuristic on the modified signing/decisioning/server files; the reviewer surfaced no concerns there in this delta, so the flag alone does not warrant escalation.