# Semifinal Fact Mining Status

Updated: 2026-07-13T05:20:00+00:00

This is the operating board for the 1000-fact factory. A fact is not model-valid merely because a slot exists.

## Global Counts

- `pending_mining`: 495
- `seed_verified_needs_crosscheck`: 40
- `source_verified_pending_review`: 465
- `team_strength_rating_rows`: 11
- `accepted_strength_ratings`: 11
- `official_crosscheck_resolution_rows`: 12
- `regulation_substitute_source_candidate_rows`: 3
- `parse_review_queue_rows`: 3
- `parse_review_ready_candidate_rows`: 1
- `model_valid_rows`: 0

## By Match

### ENG_ARG_SF (500 slots)
- `pending_mining`: 247
- `seed_verified_needs_crosscheck`: 20
- `source_verified_pending_review`: 233

### FRA_ESP_SF (500 slots)
- `pending_mining`: 248
- `seed_verified_needs_crosscheck`: 20
- `source_verified_pending_review`: 232

## By Owner

### Aster (200 slots)
- `pending_mining`: 200

### Cicero (390 slots)
- `pending_mining`: 132
- `seed_verified_needs_crosscheck`: 21
- `source_verified_pending_review`: 237

### Vela (350 slots)
- `pending_mining`: 163
- `seed_verified_needs_crosscheck`: 19
- `source_verified_pending_review`: 168

### Villanova (60 slots)
- `source_verified_pending_review`: 60

## Continuation Loop

1. Agent opens its work order.
2. Agent fills only source-backed facts in its output CSV/markdown.
3. Aster reviews source, cutoff, confidence, and no-leak status.
4. Aster promotes accepted rows into `fact_cards.csv`.
5. Re-run `python tools\manage_semifinal_fact_factory.py`.
6. Rebuild AI-readable layer and deploy only after public state remains truthful.

No prediction lock exists while pending mining dominates the queue. Accepted strength ratings may now feed opponent-realm filtering, but they do not create score predictions or market guidance.

## Round 15 Crosscheck Layer

`official_crosscheck_resolution_round15.csv` now separates public-result confirmation from model-valid promotion. It confirms that several standard-score rows can move toward parse review, but it also blocks all after-extra-time finals from regulation exact-score use and keeps duplicate rows from being counted twice.

Promotion remains closed: no row becomes model-valid until parse status, primary/reconciled source proof, duplicate handling, cutoff fields, and Aster review are complete.

## Round 16 Regulation Substitute Source Layer

`regulation_substitute_source_candidates_round16.csv` now separates regulation-score search directions from parse-ready substitute candidates.

Current state:

- England-Norway 2026 regulation candidate `1-1`: source chronology supports it, but explicit regulation wording is still required before parse.
- Argentina-Switzerland 2026 regulation candidate `1-1`: source chronology supports it, but explicit regulation wording and duplicate collapse are still required before parse.
- France-Argentina 2022 regulation candidate `2-2`: rendered reference evidence is direct enough for mechanical parse review, but primary/reconciled source acceptance is still required before model use.

`parse_review_queue_round16.csv` therefore has one parse-review candidate and zero model-valid promotions.