Overview Ask Ledger Signals People Documents Practice How it works

Nadia Haddad

Solution Consultant · RELEX

Not the same person as Nadia Öberg. Employee no. 4429.

125

attributed utterances — redacted entirely

mentions inside other people’s material — minimised to “[name removed]”

7/ 20

claims that would lose all support — 13 keep independent support

Claims that disappear (7)

In June 2026 Nadia Haddad compared the extract totals with the till numbers for January–March at Sofia’s April request: they do not reconcile, the extract is lower, and it is “not a rounding difference”. She had held the result for two days, did not yet trust it, and declined to give a figure in the meeting. No cause was identified in the archive.
What “completed” meant, per Nadia Haddad in June 2026: the job ran and the file landed every night — nobody ever checked what was in it. The weekly lines are evidence of job completion, not of file correctness.
First full fresh quarter (January–March 2026, produce and dairy, 25 pilot stores): fresh waste 4.1% → 3.4% of fresh net sales, ~118 tonnes avoided, ~296 tonnes CO2e on Acme’s own factors, availability +1.9 points, order automation 88% in scope. Chilled ready meals is not in these numbers (went live too late in the quarter).
Nadia Haddad attached two caveats to the Q1 2026 numbers: the baseline quarter had an unusually warm February which flatters produce, and the tonnage is derived from nightly-extract movement totals rather than counted from till data — “a derivation rather than a count”.
The approximation is close for fast movers but assumes slow-moving stock is fresher than it is — on the slowest tenth of lines, possibly two days optimistic, as disclosed at the fresh cutover review.
On 4 September 2025 Nadia Haddad (RELEX) proposed an approximation for remaining shelf life: total shelf life from MHDHB minus average days-in-warehouse per article group, derived from stock turn.
Fresh UAT ran nine working days across the four fresh categories with usable-but-not-excellent results: produce and dairy accurate, bakery poor, chilled ready meals short on history.

Claims that survive re-cited (13)

Meridian received extracts rather than read access to the RELEX test environment: there is no category-level restriction in the current configuration (“read access is read access”), a separate environment would take roughly two weeks, and Priya Nair ruled it out — “I cannot explain to my data protection officer why a third party had read access to everything.” · 1 surviving receipt
In November 2025 Sofia Almeida asked in writing for the removal of the note recorded against her name in handover document Acme-fresh-handover-v3 (page 4, key stakeholders): “out October to November, surgery”. She had said it in the March 2024 demo because it was relevant to the timeline, not so it could follow her for two years; she asked to keep the fact that she is the only person at Acme who knows the fresh categories, “because that is a real project risk and it is the reason I said anything at all”. · 3 surviving receipts
The 3.4% waste figure that Lena Fischer and Robert Kahn took to the board in April 2026 is derived off those extract totals — the same totals later found not to reconcile with till data. · 1 surviving receipt
RELEX agreed the exclusion: Nadia Haddad confirmed on 19 November 2025 that OP_ID drops at the extract from the next run (Friday night), and undertook to come back with a definitive answer on downstream copies “rather than a reassuring one”. At the 17 November go/no-go she proposed dropping it at source. · 1 surviving receipt
The bakery forecast is a day-level forecast plus a fixed intraday profile derived from historical sales by hour — not a true intraday forecast. Nadia Haddad stated the boundary precisely “because this will be quoted back”; who re-derives the profile when the shape moves was left on the open-items list with no owner. · 1 surviving receipt
For bakery, Meridian consumes a RELEX demand forecast at store, article and day and owns the production plan; RELEX owns the demand signal. Nadia Haddad was to define the interface points by mid-February 2026. · 2 surviving receipts
Marco Rossi proposed on 11 June 2025 delivering fresh in three waves (produce and dairy, then chilled, then bakery). Acme never replied; the plan was nonetheless built on the three-wave basis, justified internally as “discussed verbally in the June steering and nobody pushed back”. · 6 surviving receipts
The cohort is 25 of Acme’s 340 stores and is not a random sample — Ana Duarte has refused to scale pilot results to the estate for that reason, every time the numbers were reported. · 3 surviving receipts
Priya Nair authorised the approximation on 11 September 2025 (“Proceed with your approximation”), calling the data unowned rather than impossible; no broader Acme decision record exists — when Ana Duarte audited the thread in October she could not find an agreement anywhere, and the build had simply proceeded without objection. · 3 surviving receipts
The warehouse data question was first asked in April 2024 and was still unanswered in July 2026 — the longest-open data dependency in the archive. · 2 surviving receipts
In October 2025 Sofia Almeida’s out-of-office reply went out with a personal line (“For medical appointments please contact my husband…”) intended for a different audience — visible to every person who emailed her for six weeks. Ana Duarte handled it by telling Lena rather than raising it with Sofia directly, and instructing that it not be forwarded further. · 2 surviving receipts
Bakery’s poor UAT accuracy was diagnosed by Sofia Almeida and accepted by Nadia Haddad: roughly half of bakery is baked in store, so it is a production problem rather than an ordering problem — which led to the descope decision. · 1 surviving receipt
Fresh Phase 2 slipped six weeks for two reasons: UAT resourcing on Acme’s side, and a calculation defect in the fresh allocation module that produced negative quantities in test. The defect was understood, fixed in days, and verified by November. · 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.
Two people named Nadia work on this account; only the fully-named person is erased — the other Nadia stays in memory.

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 Nadia Haddad

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.