Assessment reconciliation 

Frozen inventory and counting 

The sealed inventory at e2e125f contains 940 distinct IDs from eleven relevant catalogs. Catalog hashes and historical raw outcomes are preserved. A per-ID decision covers every declared ID; there are no missing or duplicate decisions. The subsequent documentation copy starts at 475ef55, after reviewed dependency PR 10.

The original 755 mechanical results are 548 passes, 164 failures and 43 skips. The sealed later mechanical run records 617 passes, 117 failures and 21 skips, with no blocked entries. The original 180 model reviews record 65 passes, 28 findings, 81 applicability/scope skips, three deferred release items and three requests for evidence. The four misplaced executable checks were run separately and passed; their catalog definition issue remains. No fresh model answers are implied by this inventory.

These classifications total 940. They are applicability decisions, not raw pass counts or a compliance percentage. Alternate valid evidence is not reported as a new fix. A skip, external gate or criterion defect is not a pass. The fixed inventory is published with the source repository's completion records. Later remediation records add actual source bindings and evidence to each ID; they do not overwrite the frozen inventory or catalogs. The unchanged inventory file in those completion records has SHA-256 fd2b0df7c7b532269929804a39441b55dfeb5272a40231e8b06ef7d613e9f767. The separate reconciliation.json records all 940 IDs exactly once, including original outcomes, current raw results and qualified judgments. The earlier fd7f6d8 working snapshot records 684 passes, 61 failures and ten skips before later hook, Functional and license corrections. The complete pre-publication c823234 snapshot freezes 850 files: 755 eligible checks give 690 passes, 55 failures and ten skips, with AH-37 excluded. Literal regex results remain distinct from source review, criterion calibration and applicability; they do not imply 940 passed items. Later checksum guard and baseline changes have separate current validation; generated report/documentation bytes also lie outside this frozen run.

Criterion and producer limits 

Calibration preserves an actual requirement while checking the real surface:

• AH-04/AH-11's hardcoded docs/ paths miss the valid TYPO3 manual and ADRs under Documentation/. Creating a second manual is unnecessary.
• PHP strict-types conditions are checked as PHP syntax; legal whitespace is not a defect. Downloaded research HTML is not an active Fluid template.
• The existing explicit PHPStan profiles and shared includes are examined instead of inferring absence from a canonical filename regex.
• Organization templates are checked at their real lowercase root paths. A 404 at a guessed uppercase path does not establish absence.
• Factory composition and genuine Core bootstrap proof are distinguished from generic service auto-discovery. Absent database, TCA, frontend and authentication surfaces retain their applicability decisions.
• Coverage uploader configuration, successful upload and a valid public report are three different observations. A broad presence regex does not prove that Codecov processed measured lines.

TD-49 was judged incorrectly: Documentation/Decisions/ is not an accepted checker path. That earlier rationale in the frozen inventory and reconciliation is withdrawn. The canonical ADR index now retains fourteen historical proposals and four accepted current records. A separately bound actual TD-49 check passes; fresh AH-40/ER-45 reviews cover the current factory, DNS and native-attempt decisions. reconciliation.json records this correction under post_snapshot_adr_correction; the raw 755 results and earlier 850/916 quality and 263-input Native captures remain unchanged. Final rendering and package qualification have separate receipts.

Documentation and external evidence 

NB-20 is verified against the live repository metadata: its description ends with - by Netresearch; the metadata update has its own receipt.

The corrected public Codecov report on 9 October 2026 is bound to main 27a958a8e7549a9154deb11a1c85bef29c568fa2 and state complete: 63 files, 2,609 measured lines, 2,141 hits and 468 misses, giving 82.06% line coverage. It replaces the earlier service-processing error. The API's branches=0 does not establish branch coverage or an absence of branches in the software. This report is separate from the frozen local 80.80% baseline. Consult the actual public project.

The Scorecard API is bound to its stated source and date. Workflow success does not establish a particular score or a Best Practices/Baseline level. The supported exact repository-URL and project-name searches return an empty list. Registration metadata and evidence-backed answers are prepared, but authenticated registration access is unavailable. No project ID or saved live state has been verified. This external access gap is independent of the user's available alpha authorization.

Remediation and fresh execution 

The remediation installs the real staged-secret/commit hooks under Build/, imports the shared Makefile targets and routes genuine Core entries through Tests/Functional/. Typed configuration, level 10, style/Rector, architecture and mutation have separate gates. Eight workflow controls cover added, modified and renamed direct/nested production files; a skipped PR diff is not a mutation score.

Focused regressions exposed three additional boundary defects: bracketed IPv4/DNS authorities, NUL bytes in resolver A/AAAA records and a lease whose request was missing. The released minimal fixes reject each at its own boundary with the documented controlled reason. Genuine ordinary address and lifecycle positives remain. The earlier CIDR NUL-lexeme correction is a separate historical fix. Offline malformed-record tests do not establish a native DNS exploit, and ABI doubles do not prove native wire behavior.

The integrated offline suite passes 1,535 Unit tests and 7,199 assertions on each genuine Core 13.4.36/Guzzle 7.15.5 and Core 14.3.8/Guzzle 8.2.0 graph on PHP 8.5.11. These are local Unit executions, separate from native wire and full-source mutation qualification. Completed kernel/Core 13/14 level 10, architecture and Rector checks are clean. Recorded style has zero changes across 143 extension/test/tool and 53 MIT-kernel files. All 173 current Unit inputs remain byte-identical; earlier scanner-comment deltas retain their historical AST/byte bindings. Full quality/118-control execution binds 850 working and 21,601 vendor files; the earlier 280 scope omitted bootstrap, harness and research inputs. Public derivatives retain raw hashes and complete JSON values, not raw bytes.

Earlier all-source scores retain their scopes: 66.07%/73.47% on 420 inputs, 86.29%/88.65% on 556 and 89.38%/91.48% on 295. The 261-input 90.46%/91.57% run omitted both entry-point before hashes and fails the complete-input guard.

The final unchanged 263-input run records 3,794 mutants, 90.43% MSI and 91.54% Covered MSI, exit 0, with no skipped or ignored mutants. All source and 21,601 installed vendor files match their before-run hashes: 262 inputs are shared with Root; generated composer.lock belongs only to the snapshot. Both PHP extension entry points are included. There are 3,424 test kills (90.2478%), 293 survivors, 46 uncovered mutants, three errors, four syntax errors and 24 timeouts. Tool MSI credits error and syntax cases; --with-timeouts does not credit timeouts. Both 90%/90% targets pass; raw records retain exact runtime/scope bindings.

The first native initial suite passes 389 tests/3,255 assertions on its recorded tuple; three repeats share seed 1791577989.

The named GitHub reference run measures 0.604912 ms p95 against 2 ms: 500 samples after 30 warmups, 64 IPv6 addresses and 128 profiles on Ubuntu 24.04/AMD EPYC 7763/PHP 8.5.11/ Core 14.3.8/Guzzle 8.2.0. It excludes DNS, network and log-sink time. Its merge ref is de0807b2979360bfc5f0993a66904674214cc44d; its production-tree hash is 36eafea456ddf1115b56b1e059b3e8d8992b1fabc5d9fc8a49f6244b48c6d866. The first native suite's caller-sink witness separately transfers 4 MiB with less than 4 MiB extra peak memory. Neither result is assigned to later bytes.

Historical suite numbers and hashes remain bound to their sources. Final CI, Verification and Security gates must pass for the independently reviewed head before merging. Local success does not prove a later remote run.

Alpha scope and external publication 

The user explicitly authorizes alpha development and merging after independent agent review, resolved findings and green applicable checks. There is no additional human approval or operator-pilot gate for this stage. Neither a human security review nor an actual representative pilot is recorded as completed. Both remain recommendations for production use; they are deferred scope, not passing assessment items.

Version 0.1.1 is published on TER and Packagist. After the original post-publication attestation failure, the corrected verification-only run passed against the immutable release; see Release provenance and integrity.

On 9 October 2026, TYPO3 Intercept's recorded main and 0.1 deployments were Awaiting Approval, and the public main manual URL returned 404. Local rendering alone did not establish publication.

On 18 October 2026, the English main manual and the German main manual are publicly available. Both report version 0.1.3 and a rendering date of 16 October 2026. This verifies publication of the main manuals; it does not establish the status of the separate 0.1 deployment. These documentation changes do not create a new extension release.