Janady13/missionmatch / docs/evidence/phase2/functional/document-trial-rehearsal.txt

1PHASE 2 DOCUMENT TRIAL, REHEARSAL AGAINST A SYNTHETIC PACKAGE 228 July 2026. Target: missionmatch_staging. Everything used was invented: no real 3notice, no real person, no real award. All of it was removed afterwards and the 4removal was verified, 0 rehearsal rows remain. 5 6WHY REHEARSE A CHECKLIST. docs/PHASE2_DOCUMENT_TRIAL.md is the thing that runs 7the day the real package arrives, and until today nobody had ever executed it. A 8checklist that has never been run is a plan, not a procedure. This rehearsal is 9that checklist's own positive control: it exists to make the checklist fail while 10failing is cheap. 11 12IT FAILED IN FOUR PLACES. That is the result. The refusals it does prove are 13listed after them, but the four defects are the point of the exercise. 14 15================================================================================ 16CHECKLIST DEFECT 1, STEPS 1 AND 2 NAME MACHINERY THAT DOES NOT EXIST 17================================================================================ 18 Step 1 says "upload through the real intake path". 19 Step 2 says "run cmd/opportunity-extract, deterministic only". 20 21 There is no cmd/opportunity-extract. There is no upload handler. The tables 22 opportunity_documents and opportunity_document_chunks EXIST, are correct, and 23 carry provenance bound at origin, but the only Go file in the repository that 24 names either of them is internal/platform/scrub/binding_sites_test.go. 25 26 So the document path is SCHEMA WITHOUT CODE. Nothing writes an uploaded 27 document, and nothing chunks one. 28 29 This is not a surprise in the sense that the two deferred Phase 1 items are 30 exactly the upload screen and the attachment sync. It IS a surprise in the 31 checklist, which reads as though steps 1 and 2 are ready to execute and only 32 the file is missing. On the day, that reading would have cost an hour before 33 anyone realised the blocker is engineering and not paperwork. 34 35================================================================================ 36CHECKLIST DEFECT 2, STEP 3 NORMALIZES METADATA, NOT A PERFORMANCE WORK STATEMENT 37================================================================================ 38 Step 3 reads as though the model pass consumes the uploaded PWS. It does not. 39 cmd/opportunity-scope-normalize joins tracked_opportunities to 40 sam_opportunities and composes its payload from NINE METADATA FIELDS: title, 41 department, sub agency, NAICS, PSC, set aside, notice type, and the place of 42 performance city and state. Verified by reading the query and by the byte 43 count of the composed payload at run time, 250 bytes. 44 45 A real PWS is tens of thousands of bytes of clauses. The normalizer has never 46 seen one and cannot, because nothing puts one in front of it. 47 48 CONSEQUENCE FOR THE TRIAL: the scope items it produces on the day will restate 49 the notice header, not the statement of work. In this rehearsal, from a notice 50 titled PACAF Cyber Systems Operations Support, it produced exactly one cited 51 item titled "PACAF Cyber Systems Operations Support". That is the honest 52 ceiling of the current pipeline and it should be said before the package 53 arrives, not discovered while Joseph is watching. 54 55================================================================================ 56CHECKLIST DEFECT 3, THE COLUMN NAMES IN THE CHECKLIST ARE NOT THE SCHEMA 57================================================================================ 58 Four inserts written from the checklist's own vocabulary were rejected by the 59 database before any assertion could run: 60 sam_opportunities has no "description" column, and requires raw and poc 61 opportunity_documents has file_name, not filename 62 requirement_drafts has no "title", it has labor_category 63 requirement_drafts review_status is draft, in_review, approved, rejected, 64 and NOT "pending" 65 Small, and exactly the kind of thing that turns a twenty minute trial into an 66 afternoon. Corrected in the checklist. 67 68================================================================================ 69CHECKLIST DEFECT 4, STEP 5 NAMES ONE CONSTRAINT AND THERE ARE TWO 70================================================================================ 71 Step 5 says promotion records the named human and the schema refuses an 72 approved draft with no reviewer. True, but incomplete. There are TWO 73 constraints and they fail in a fixed order: 74 ck_drafts_approved_has_requirement fires FIRST, because an approved draft 75 must carry approved_requirement_id 76 ck_drafts_decided_has_reviewer fires only once the first is satisfied 77 An operator asserting "no reviewer is refused" would see the FIRST error and 78 conclude the reviewer rule works without ever exercising it. Both are proven 79 separately below. 80 81================================================================================ 82WHAT THE REHEARSAL DID PROVE, each with the refusal quoted 83================================================================================ 84 85 PROVENANCE IS BORN, NOT DECLARED. The synthetic notice was inserted with NO 86 provenance column supplied: 87 notice provenance BORN as: public_cache 88 89 PROVENANCE IS IMMUTABLE. An update straight at the column: 90 ERROR: provenance is immutable: public_cache cannot become tenant_upload 91 on table sam_opportunities 92 CONTEXT: PL/pgSQL function provenance_is_born_not_chosen() 93 94 A CALLER CANNOT LAUNDER A TENANT DOCUMENT. The insert DECLARED public_cache: 95 caller declared public_cache, row was born: tenant_upload 96 This is the single most important line in the rehearsal. It is the sharpening 97 Joseph asked for, proven against the real table rather than a fixture: the 98 egress decision derives from what the database says about where a row came 99 from, and a caller who lies about it is simply overwritten. 100 101 THE EGRESS GATE READ PROVENANCE FROM THE DATABASE AND ALLOWED A PUBLIC NOTICE: 102 egress allowed {"backend":"megamind-gateway","byte_count":250, 103 "content_sha256":"bf1161ce...","payload_class":"REQUIREMENT_NORMALIZE", 104 "provenance":["public_cache"]} 105 Then the LOCAL model produced one cited item, and the run wrote nothing 106 because it was a dry run. 107 108 AN UNCITED SCOPE ITEM IS REFUSED: 109 ERROR: new row for relation "opportunity_scope_items" violates check 110 constraint "ck_scope_items_cited" 111 112 AN APPROVED DRAFT WITHOUT ITS REQUIREMENT IS REFUSED: 113 ERROR: violates check constraint "ck_drafts_approved_has_requirement" 114 115 AN APPROVED DRAFT WITH A REQUIREMENT BUT NO NAMED HUMAN IS REFUSED: 116 ERROR: violates check constraint "ck_drafts_decided_has_reviewer" 117 118 POSITIVE CONTROL ON THOSE TWO REFUSALS, because a table that refused 119 everything would produce the same evidence: 120 draft accepted, review_status=draft, reviewer=none 121 122 THE AMENDMENT DOES NOT REACH THE REVIEWED WORK. The notice was rewritten 123 underneath a reviewed scope item, title changed and deadline moved: 124 notice amended to: PACAF Cyber Systems Operations Support, AMENDMENT 0001 125 the reviewed scope item is UNCHANGED: PACAF Cyber Systems Operations 126 Support (hash a1b2c3d4) 127 and its provenance is STILL: tenant_record 128 129================================================================================ 130WHAT THIS CHANGES ABOUT THE DAY THE PACKAGE ARRIVES 131================================================================================ 132 The package is STILL the one blocker for the parts of Gate A that only a real 133 document can close. It is no longer the only thing standing between here and 134 steps 1 and 2, and pretending otherwise was the checklist's biggest defect. 135 136 On the day, steps 3 through 6 run as written and prove what is listed above on 137 real notice metadata. Steps 1, 2, and 7 need the deferred Phase 1 engineering 138 first: the upload screen, the extraction binary, and the attachment sync. Those 139 were deferred on documents existing. The rehearsal shows the sequencing is the 140 other way around for two of them: the code has to exist before the document can 141 be useful to it. 142