# Postgame Audit: Argentina vs Egypt

Settlement source: Guardian live report, `Argentina 3-2 Egypt: World Cup 2026 last 16`.

Settlement URL: `https://www.theguardian.com/football/live/2026/jul/07/argentina-v-egypt-world-cup-2026-last-16-live`.

Final regulation score: `Argentina 3-2 Egypt`.

Locked pregame C4: `2-0, 3-0, 2-1, 1-1`.

Official C4 result: `MISS`.

Support settlement:

- `Over 1.5 YES`: `HIT`. This support was coherent with every locked C4 score because each node had at least two total goals.
- `Under 3.5`: `MISS` if treated as a support candidate. It was too confident because the model suppressed the high-total stress branch too aggressively.

## What Went Right

Turtle did not erase Egypt entirely. The locked card preserved `2-1` and `1-1`, so the model admitted that Egypt scoring was live against Argentina. The post-lock `Over 1.5` support read was also correct: the cloud required goals rather than a dead 0-0 or 1-0 type match.

The separate human macro read that Argentina would survive and advance was also correct, but it is not the Turtle exact-score record. It belongs in a future advance-survival module, not in the public score-cloud accuracy count.

## What Failed

The exact-score engine failed. The final score `3-2` was outside C4.

The main failure was not variance. Turtle compressed Egypt's scoring into a one-goal branch. It allowed `2-1` and `1-1`, but it did not properly test the second Egypt goal. Once Egypt scoring survived the contradiction audit, the model should have compared `2-1` against `2-2` and `3-2` with a dedicated second-underdog-goal test.

The rank-spread prior also overpowered the live tactical reality. The target sample said stronger clean sheet `69.2%` and over 3.5 only `7.7%`, so Turtle placed two clean-sheet cushion scores into C4. That made the card too brittle. Egypt had weak elite-realm sample size, but its shrunk realm scoring rate was still recorded as `0.792`; that was too strong to be represented by only one Egypt-goal score.

## Failure Class

`SECOND_UNDERDOG_GOAL_UNDERMODELED`

Secondary classes:

- `RANK_SPREAD_CLEAN_SHEET_OVERWEIGHT`
- `HIGH_DRAW_AND_FIVE_GOAL_STRESS_SUPPRESSED`
- `SUPPORT_MARKET_LOCK_CLARITY_NEEDED`

## Rule Correction

Add `Mutual-Scoring Stress Override`:

If the underdog scoring branch survives and `P(BTTS) >= 0.40`, C4 may not allocate two clean-sheet scores to the favorite unless the proof explicitly defeats both `2-2` and the nearest favorite-win high-total branch by numeric comparison.

Add `Second Underdog Goal Test`:

When `2-1` or `1-1` is included because the underdog scoring branch survives, Turtle must compute and publish the comparisons:

- `2-1` vs `2-2`
- `2-1` vs `3-2`
- `1-1` vs `2-2`

At least one of `2-2` or `3-2` must either enter C4 or receive a named exclusion with stronger evidence than broad over-3.5 rarity.

Add `Advance-Survival Module`:

When a giant favorite falls behind live but still owns the structural qualification pressure, Turtle may write a separate analysis note. That module must not rewrite the exact-score C4.

## Minimal Honest Summary

Argentina 3-2 Egypt was a C4 miss. Turtle saw Egypt scoring, but did not respect the second Egypt goal or the high-total stress route. The next protocol patch is not cosmetic: when BTTS survives, Turtle must attack the one-goal cap itself before locking clean-sheet-heavy cards.

Do not rewrite `PREDICTION_LOCK.md`.