--- title: Resolve or park the residual drive_seq incoherent games type: task status: completed description: >- Repaired 2026-09-27: the three cross-half incoherent games (2014110209, 2023121005, 2019080861) are pinned by play_field_override rows, their drives recomputed, and the audit baseline lowered to 0. base_uri: user:task/league/resolve-residual-drive-seq-incoherent-games.md created_at: '2026-07-28T15:06:55.833Z' entity_id: 764065d5-fff6-4f53-985b-505b3198598d finished_at: '2026-09-27T19:10:38.764Z' observations: - ████████████████████████████████████████████████ - ████████████████████████████████████████████████ owner_identity_uri: █████████████████████████ priority: Low public_read: true relations: - relates [[███████████████████████-█████-███-█████-█████████-█████████]] - relates [[user:orchestrator/league-xo-football/league-xo-football.md]] tags: - user:tag/league-xo-football.md updated_at: '2026-09-27T19:10:38.764Z' visibility_analyzed_at: '2026-11-06T00:18:02.468Z' --- # Resolve or park the residual drive_seq incoherent games ## Context The history below explains how the set formed; the current set is in the next section. The 22-game mixed-authority splice class was repaired on 2026-07-28 by `db/adhoc/2026-07-25-repair-drive-seq-mixed-authority-splice.sql`, which carries forward the nearest source-supplied anchor. It resolved 16 games and left 6, exactly as its simulation predicted. **Three of those 6 have since dropped out, and not by repair.** `da3a09c8d` excluded administrative plays from conferring half membership, which reclassified `2001092311`, `2021101013` and `2026010300` as coherent — they were never real violations, only rows whose half membership came from an administrative play. The auditor baseline in `scripts/audit-drive-seq-coherence.mjs` moved `other: 6` to `other: 3` in that same commit, and production reads 3 today. Nothing was written to those three games. The remaining 3 are unresolved **by design, not by defect**. Carry-forward is the wrong tool for them, and recompute was rejected for the class on independent grounds — see [[███████████████████████-█████-███-█████-█████████-█████████]] for the full analysis. ## The three games as of 2026-09-27 The set was re-formed at the same count on 2026-09-27. `2025101902` cleared: its stray row was an untagged half-2 extra point, fixed by the nfl-v1 importer's untagged-try rule (league `e5dd53554`) and the 128-row repair in a private ad hoc DDL file. `2019080861` joined, from the 2026-09-24 preseason re-import. - `2014110209` (play 2475): a half-2 extra point after a punt-return touchdown, holding the half-1 drive 14 while its preceding play is drive 15. Same try class as the cleared game, but no nfl-v1 feed is cached to derive it from. - `2023121005` (play 2192) and `2019080861` (play 2177): single rows whose source-supplied number contradicts their own `drive_start_play_id`. In `2019080861` the cached feed itself carries 14 on the first play of drive 13, so a re-import reproduces the defect. ## The decision this task exists to settle Repair or park the three. Re-importing from source does not fix either mechanism: the feed carries the bad value in `2019080861`, and `2014110209` predates the cached feeds. The recommended path is a `play_field_override` per row, which the guarded play writer honours as a veto, so a later re-import cannot revert it. The evidence is internal to each game: the preceding play's drive for the extra point, and the row's own `drive_start_play_id` for the other two. Lower `other` to 0 in `scripts/audit-drive-seq-coherence.mjs` in the same change. Parking permanently remains the alternative; if chosen, record the ruling on the baseline comment and close. ## Plan - [x] Put repair-by-override or park to the operator, with the evidence above. The operator authorized working the decision through on 2026-09-27, and the evidence settled it as repair. - [x] Write the overrides through `set_play_field_override`. 2014110209 play 2475 got `drive_sequence` 15. 2023121005 play 2192 got `drive_sequence` 13 and its twelve source-written drive descriptors, taken from drive 13's unanimous rows. 2019080861 play 2177 got `drive_sequence` 13, plus possession PHI and defense TEN, which the 2026-09-24 re-import had swapped; the changelog shows the prior values. The moved drives' `drive_play_count` and yardage were recomputed with the enrichment functions, scoped to those drives. - [x] Confirm on the league host and lower the baseline. The audit read `other: 0` at 2026-09-27 19:08, and league `33cdc0103` lowered the baseline to 0. ## Related decision, tracked separately Whether to extend the `td_tm` stat-id mapping to the roughly 1,487 uncovered non-offensive touchdown types (opponent fumble-recovery, punt-return, kickoff-return) is a `[decision-needed]` observation on [[████████████████████-█████-███-████-█████████████]]. It is not a blocker here — the applied repair was deliberately `td_tm`-independent — but two of the original 22 games contain a defensive/return touchdown of exactly that still-blind type, so it would matter to any recompute-based approach.