# A missing historical result and the September 12 briefing

Prepared 8 September 2026, before the event. [All thirteen picks and reasons](September-12-UFC-Predictions.md) include a separately labelled, limited calculation for Garcia–Rongzhu.

The [independent adjudication addendum](ADJUDICATION-ADDENDUM.md) retains the prototype's **8.55/10** quality rating. It awards no additional predictive credit and records that independent future confirmation is still required.

At the user's request, the briefing also includes [likely means of victory and supporting evidence](victory-methods.json). These are qualitative judgements conditional on the selected fighter winning, not a fitted or validated method classifier. The displayed winner percentages are unchanged and are not joint winner-and-method probabilities. This addition has no method-accuracy score and was made after the above quality adjudication.

UFC's [Rongzhu profile](https://www.ufc.com/athlete/zhu-rong?page=1) records a 28 May 2026 Road To UFC win over Victor Martinez, by strikes at 1:10 of round one. This bout is absent from the original acquired archive. The current card lists Garcia–Rongzhu; the original September 7 forecast for Garcia was 63.9710%.

This bounded calculation adds only the independently verified result, paired identities, event date and observed duration. Both rows of strike, takedown, control and round statistics remain unavailable; none is invented or represented as observed zero. The identical fitted V3 weights and unchanged feature rules then produce 60.5118% for Garcia. The other twelve current-card estimates remain unchanged within numerical precision. No future outcome is used.

A missing-stat bout still advances the frozen history window and recency weights. Consequently some derived recent-rate inputs change even though no new strike or control counts were added. The difference is the model's response to this limited historical supplement, not a causal effect or a measured improvement in accuracy.

The original data, all model files, original 897-bout test and sealed September 12 pilot remain unchanged. This supplement does not silently replace the prospective primary issue. The main hypothetical-matchup service continues to use the original acquired history; the separate briefing explicitly shows this supplementary estimate.

## The finite coverage check

All 26 current-card official athlete pages were retrieved successfully. Among 65 dated 2025–26 narratives, Rongzhu–Victor Martinez was the only confirmed omitted recent UFC/Road result. The other unmatched entries were three explicitly excluded DWCS results, a Pitbull/Freire name alias and a stale Rosas announcement without a result. Sean King had no dated history. This is a finite narrative-source check, not exhaustive professional-career coverage.

The whole May 28 event is absent from the frozen corpus and primary collector's saved event list. Its absence from the earlier observed UFCStats index is inferred from the retained count/hash reconciliation; a fresh live UFCStats-index verification was unavailable. Index coverage must not be presented as complete UFC/Road history.

Rongzhu's profile also records three older Road tournament wins that are absent from the core. The [four-result ledger](road_result_ledger.json) preserves all four omitted results and their source/identity limits. The one-result calculation above inserts only the May 2026 result. It does not reconstruct all Road history or establish how results from different competitive settings should be weighted.

[Coverage ledger](coverage_audit.json), [detailed May result receipt](may28_result_receipt.json), [independent calculation review](independent_review.md).

## Reproduce the bounded calculation

Use the workbench's prepared Python environment from the workbench directory:

```powershell
python results/history_supplement_2026_09_08/REPRODUCE.py
```

The script validates the original thirteen probabilities against the preserved fitted model, inserts the one supplementary result chronologically, retains all original same-date history updates, compares all thirteen results with the saved summary and writes `reproduction-run.json`. It refuses to overwrite that record; pass `--output` with a fresh path for another run. The script never fits a model or writes to original data or sealed files.

[Saved summary](summary.json) records all thirteen before/after probabilities and the exact model hash. [Detailed evidence](September-12-Forecast-Evidence.json) retains the earlier profiles, fight histories and explanation boundaries. Official live-card/source coverage and independent verification are recorded with this supplement when available.