# Football recommendation containment — September 13, 2026

Prospective amendment authorized by the September 13 remediation request:
automated recommendations remain paused until a new strategy is validated.
`delivery_policy.PAUSED` blocks both Discord modes, including --force.
`board.py` also refuses new weekly recommendation boards/writeups while paused.
Raw capture, game freeze evidence, existing-board settlement and reveal continue.
No committed selection, result, cohort, historical arithmetic or frozen study
is changed. A future strategy must preserve the official status and settlement
of all existing fp-v0.4 positions; merely changing record_policy.VERSION is unsafe.

The new `market-input-v1` module is for prospective research observations only.
It does not supply official selections or alter legacy market.py. Its strict
market timestamp requirement deliberately rejects older bookmaker-timestamp-only
captures; endpoint/scope/budget policy must be settled before adopting it.
The existing normalizer retains provider market timestamps when supplied,
additively, without guessing missing values or making extra API requests.

Reopening requires a separately preregistered strategy, untouched evaluation,
independent review and explicit approval. Debugging the incident or requiring
a positive price discount does not demonstrate an edge. Failed studies and the
historical holdout are not reopened by this amendment. Original picks continue
to grade and reveal, including losses. There is no further record reset.

## Named result identity adjudication

The September 13 grading run left one committed position pending because the
results provider reports 03:59 UTC and the odds commitment reports 04:00 UTC.
The same fixture and final result were verified directly on ESPN event
401864578. `data/football/result_identity/2026-09-13-ncaaf-401864578.json`
adds an exact, dated cross-reference scoped to both the original board and
game hashes. It grants no generic time tolerance. Original kickoff, side,
price and hashes remain unchanged; the appended result records the resolution
identifier and provider kickoff separately. Changed provider identity/timing
fails closed. CLV admits no capture after the earlier of the two kickoffs.
Future adjudications must be separate files, never edits to this record.

## Record classification amendment

At the user's direction, the exact incident board hash
`da33da68ad7e9f444c18dc1337c81af4fc172a22a361bc83d6b65f6f92e55a34` is now
excluded from the official football record. Its two settled rows remain in the
append-only ledger, with original prices, outcomes, CLV, event IDs and hashes.
They render as **INVALIDATED INCIDENT** in a separate disclosed incident
history and contribute no official W-L, CLV or hypothetical return. This is a
hash-scoped cohort classification, not deletion, result editing, or a general
loss-exclusion rule. Future classifications require their own dated amendment.

This September 13 amendment supersedes the earlier no-further-reset statement
only for this exact board. It is a post-result classification: both featured
positions had already settled as losses. The full slate contains 72 cards:
two featured picks and 70 additional coverage games. All are invalidated as
recommendations; coverage cards were never part of the official aggregate.
The original ledger, board, encrypted board and commitments are preserved.
Settlement deduplication uses board hash and tier regardless of classification.
The public hub and original weekly page disclose this exclusion. A 0-0
official aggregate is not an undefeated lifetime record. Automated football
delivery remains paused pending validation of a separately registered strategy.