Source identity
| Field | Recorded value |
|---|---|
| Source type | GitHub source |
| Original record | Janady13/missionmatch / docs/evidence/phase2/trial/results.txt |
| Date | Not specified in source |
| Repository | Janady13/missionmatch |
| Revision | 720d47fae476b0d38477ae4401697d372b1ea2bc |
| Branch | feature/opportunity-to-outcome |
| Path or archive member | docs/evidence/phase2/trial/results.txt |
| Original SHA-256 | 3eab9cceca913054baa678912d1b8dae488aefd42a9804f4f5c84b25c68643bb |
| Git blob SHA | 2ddca09d773e3e29047489bd4660b4c2e05c5d54 |
| Reading-copy SHA-256 | 3eab9cceca913054baa678912d1b8dae488aefd42a9804f4f5c84b25c68643bb |
Recovered source transcript
Download complete source text ↓54 lines · 2.9 KB
PACKAGE A AND B TRIAL, RESULTS
29 July 2026. Real federal solicitation packages, through the genuine HTTP upload
intake, on staging. Findings are in findings.txt and were recorded BEFORE fixes.
THE PACKAGES
pkgA_eos_fieldcraft 9 files, 5.2 MB, notice c91da3fd4ff7414ba2bdbcc21a66d1f5
pkgB_elevator_pm 7 files, 13 MB, notice 3ebc5efaf311429e92ee2381fad752eb
All 16 checksums matched the expected values exactly. No re uploads, no 404s.
INTAKE, through POST /api/opportunities/{id}/documents with a real session
16 of 16 accepted, HTTP 201
ALL 16 BORN tenant_upload. Not one declared it; migration 0082 stamped it.
DUPLICATE DETECTION: re uploading an identical file returned 409.
EXTRACTION
done 10 documents, 719 chunks, all provenance tenant_upload
encrypted_or_locked 4 documents, named reason, batch continued
unsupported_format 1 document, the xlsx, no office reader in this build
Largest single document: 05_PWS_amended_23JUL26.pdf, 300 chunks.
THE LOCKED PWS, which was the point of shipping one
04_PWS_FCTx_FY27_EOS.pdf is genuinely encrypted. It failed with a NAMED reason,
the batch continued, and the unlocked variant shipped alongside it yielded 33
chunks. Graceful survival of a locked file, proven on a real locked file rather
than a synthetic one.
THE WAGE DETERMINATION, which was the open question
Its text layer measures as TABULAR, not prose: 231 candidate spans, of which the
prose gate accepted 1 after the gate was widened to admit rate tables. The
document is present with correct provenance. Extracting its rate grid properly
is TABLE extraction, a different capability from text extraction, and is
recorded as a known limit rather than forced through a threshold.
THE AMENDMENT EXERCISE, acceptance item 11, THE SAFETY PROMISE
requirement md5 BEFORE: 74e08dcc97928dc6b5ab3e7e5a42a3f5
5 material changes detected, 1 approved requirement flagged for a named human
requirement md5 AFTER: 74e08dcc97928dc6b5ab3e7e5a42a3f5
alerts: 1, opportunity_amendment, unread
BYTE FOR BYTE IDENTICAL. The guarantee is not that the rewrite is careful, it is
that there is no rewrite, and that now holds on real amendment text.
WHAT THE TRIAL COST, and why it was worth it
SIX findings, five of them defects that every prior test had passed over:
F1 the store path was read only under systemd sandboxing
F1b and root owned when that was granted, three layers for one upload
F2 nginx refused every file over 1 MB before the app ever saw it
F3 a REAL CODE BUG, created_at scanned as timestamptz into a string, so the
upload path had NEVER written a row despite being committed and reported
as built
F4 permission restricted and genuinely encrypted PDFs were conflated
F5 the prose gate refused an entire wage determination
F6 the diff pairs documents by recency rather than by type
F3 alone justifies the trial. The intake was tested, reviewed, and wrong.