87
attributed utterances — redacted entirely
14
mentions inside other people’s material — minimised to “[name removed]”
9/ 17
claims that would lose all support — 8 keep independent support
Claims that disappear (9)
- Case pack was the blocking gap: you cannot order in units the supplier does not ship. Lead times affect forecast accuracy rather than function.
- Supplier lead times were populated but distrusted: many were exactly seven days, the signature of a defaulted value rather than a measured one.
- The remediation scope for the pilot cohort was 4,214 articles, of which 1,317 were missing case-pack quantity and 3,105 had missing or defaulted supplier lead times.
- The same assessment found the store master and the product hierarchy clean.
- Kwame Boateng’s 24 November 2025 answer on historic files: OP_ID is present in the raw landing zone for every daily file since 14 February; it is not carried into the aggregated waste tables; it is present in the June and September reconciliation extracts (row level), two of which were emailed as attachments — once to Priya Nair, once to a distribution list of nine people including three at Meridian. He could purge the landing zone; the attachments he could not.
- The first measured figure: shelf life was populated on 48% of the pilot-cohort articles — reported by Kwame Boateng in the 24 September 2024 master-data workshop and repeated in his 30 September remediation-scope email.
- By the December 2025 fresh cutover, shelf life was populated on 61% of articles in the three in-scope fresh categories (produce, dairy, chilled ready meals), up from 48% after a remediation pass by Priya’s team in October 2025.
- Core-replenishment UAT in January 2025 ran eleven business scenarios at least once: nine passed, two failed with the same root cause — rounding on articles where case pack and minimum order quantity disagree.
- The two open UAT defects were fixed and re-tested, and all nine passing scenarios were re-confirmed on 24 January 2025 — as reported by Kwame Boateng, the consultant who ran the retest.
Claims that survive re-cited (8)
- The DC-2 cover gap: Kwame Boateng was the only person watching the feed and was on holiday 10–24 March 2025; the monitoring fired at him in Portugal. Ana Duarte acknowledged she raised the cover question in January and did not close it. · 4 surviving receipts
- A file-size check was added the week after the 14 March failure: any file under 1KB now fails the job loudly. Since then the archive records no silent failures — four loud ones by December 2025. · 1 surviving receipt
- The DC-2 permanent fix slipped: failures continued — twice to three times a week in May 2025, still failing at reduced pace in September 2025, roughly fortnightly by January 2026 with fourteen failures in January (eleven silent). The archive contains no record that the permanent fix ever completed. · 3 surviving receipts
- The September 2024 master-data assessment found case-pack quantity missing on 31% of active articles. · 1 surviving receipt
- For lead times, the agreed workaround was to derive actuals from receipt history — less accurate for new articles, fine for everything else. Proposed by Kwame Boateng, agreed by Priya Nair. · 1 surviving receipt
- OP_ID was excluded at the extract from 21 November 2025, confirmed in writing to Priya Nair and recorded in the weekly update of 24 November. · 1 surviving receipt
- The operator identifier was first raised in the September 2024 master-data workshop: Kwame Boateng reported an unexpected operator column on write-off movement records and said he did not need to receive it; Priya Nair agreed to take it out (“a five minute change”), and Ana Duarte noted it as an action. · 2 surviving receipts
- Shelf life did not block the first go-live because fresh was not in scope; Priya Nair stopped chasing the warehouse question for that reason. · 2 surviving receipts
Disclosed boundaries
- Original source files outside this application are untouched; this erasure covers everything the agent derived and stores: segments, citations, claims, the commitment trail, document summaries, cached answers, signals and both full-text indexes.
- The person is kept as a tombstone in the deletion register so the erasure itself remains auditable; their name survives nowhere else in the working memory.
- Name variants recorded for this person are erased with them.
Effect on practice question P1 (live preview)
The current P1 answer, before erasure. After erasure the same composition runs again — figures only this person reported will vanish from it.
Erase Kwame Boateng
Transactional and irreversible within this application: working corpus, citation store, claim memory, commitment trails, document summaries, cached answers, signals and both full-text indexes are rebuilt in one transaction. Type the full name to confirm.