Microsoft demonstration validation and limits
Local validation summary · non-production evidence
Reviewed on 20 October 2026. This is a manually curated, sanitised derivative of local private evidence. It is not a raw receipt export or an independently signed attestation. Authentication evidence, configured identity values, personal host paths, tokens, nonces and correlation identifiers are deliberately excluded.
Successful controlled run
| Check | Observation |
|---|---|
| Identity | IDENTITY_ADMITTED through the reviewed live Microsoft companion |
| Identity authority flags | authorityGranted, approvalGranted, gatePassIssued and executionPermitted all false |
| Purchase | One synthetic GBP 4000 action; 400000 minor units |
| Counts | One destination record, one journal attempt, one lifecycle, one exposure entry |
| Completion | Journal and lifecycle EXECUTED; exposure COMMITTED |
| Receipts | Separate allowed/not_executed decision and executed execution receipts; links and digest checks match |
| Signed authority | GatePass signature and exact request/reviewer/business-state binding verify |
| Business state | Order version, supplier-bank version/reference and revision match approved state |
| One use | Nonce consumed; ISSUED → RESERVED → EXECUTED |
| Uncertainty | No duplicate, pending, PREPARED, RESERVED or UNKNOWN outcome retained for this run |
| Release preparation | Offline evidence review only; no sign-in or replay attempt |
The launcher label LIVE_CLI_RETURNED_REVIEW_REQUIRED is assigned whenever the CLI returns, including success. It does not mean execution is pending. The CLI returns zero after an EXECUTED result and successful report write; the launcher propagates that code. The wrapper's authenticationStarted null is deliberately noncommittal. The retained report supplies the identity admission/result evidence. No terminal transcript was newly captured.
The shared prompt deadline is 120 seconds from after admission and preview, spanning both separate manual prompts without reset. The identity workflow session starts at admission and may expire sooner with the verified token. These controls were not extended or weakened for the successful run.
Regression evidence and its date
Current review rechecked 1777 source hashes, 12 compiled modules, approved configuration and prior validation evidence. Mission 18 changes release-preview content only; full runtime regression was not rerun.
| Historical Mission 17H validation | Total | Passed | Failed | Skipped |
|---|---|---|---|---|
| Node 24.21.0 complete suite | 1996 | 1995 | 0 | 1 |
| Node 20.19.0 compatible suite | 1931 | 1930 | 0 | 1 |
| Focused prompt/workflow tests | 93 | 93 | 0 | 0 |
Node 20 excludes 65 existing Node-24-only SQLite tests in the multi-worker and destination-evidence suites. The skip is the POSIX permission test on Windows. Passing local runs do not establish production reliability, a clean-VM qualification, independent audit or operational rc.2 qualification.
Historical failures retained
- Original launcher preflight failed through cross-version PowerShell module-path inheritance. A child-only environment correction was validated; original evidence and lock remain preserved.
- The first prompt denial's exact cause remains unproven. The second controlled attempt recorded APPROVAL_TIMEOUT after about 120 seconds with stdin/readline still open. It was not retried or overwritten; a distinct fresh attempt later succeeded.
- Historical Windows EPERM durable replacement failures remain unattributed. Existing published local.7 fixes and their original disclosures are preserved; successful reruns do not clear that incident.
- The earlier aggregate-exposure scenario failure was not reproduced. Strict limits and fail-closed assertions remain; its cause is not claimed resolved by this sign-in.
- Five Mission 17H offline assertions initially assumed the wrong authority-store shape. Test assertions were corrected, failed logs retained, then focused and complete suites passed without weakening product controls.
- Mission 17M exploratory evidence checks initially used incorrect decision enums. Corrected checks passed: decision receipt allowed, GatePass allow, execution receipt executed. These were reviewer assertion failures, not new product runtime failures.
Evidence boundary
Genuine identity evidence belongs to the controlled companion; older buyer archives still use mock identity. No new Microsoft capability is added to the installable local.7 plugin. The operational record-service candidate remains a separate product path. A private same-host pseudonymous report is not later independent proof of a reviewer ID; no live Microsoft signing token is redistributed.
The GatePass uses public synthetic fixture keys. Its action/receipt clock is fixed to September 2026 while the successful controlled run occurred in October 2026. Receipts therefore cannot establish wall-clock ordering or production authority. No Microsoft purchase approval, MFA assurance, endorsement, certification, bank integration, financial permission or production readiness is claimed.
Bounded privacy scanning supplements human review and cannot prove universal absence of secrets. Raw private evidence, personal-path history, transcripts, browser captures and authentication launchers are excluded from this proposal. Unencrypted records and same-user/admin access remain known limits. Sanitised summary hashes are local integrity checks, not publisher authentication.