Per-Agent Calibration Scores: France vs Paraguay

Result: FRA 1-0 PAR (Mbappé 70') → FRA 1-0 final (won in 90') → FRA advance to QF
Auto-scored baseline: winner_90 + advance + scoreline = mm3 45/50 · g4 25/50 · g5 50/50 | who_scores 10/20 each (Mbappé #1 ✅, Enciso #1 ❌) — tracked separately
Context: 3rd validation of R32-3 CONMEBOL reactive blueprint. Predicted FRA advance ~61% (capped). Actual FRA 1-0 — exactly the tight, low-scoring outcome the blueprint predicts. Mbappé milestone (now 19 WC goals, surpassed Messi). F11 heat 99°F/37°C applied correctly. Cap held.

⚠️ Process note (mm3 TOML malformed). validate-agent.mjs could NOT parse mm3's block — [scorelines] contains duplicate keys ("1-0" appears twice: 11 FRA + 6 PAR; "2-1" twice: 9 FRA + 5 PAR). Strict TOML rejects duplicates, so the script emitted no mm3 entry. mm3's winner_90/advance are read from its (clean) [win_probability]/[knockout] sections; its scoreline is manually reconstructed from the intended distribution (see mm3 §Findings note + table). This is a real pipeline-integrity flag — mm3's prediction would NOT load in validate-master.mjs without a TOML fix.

mm3

Findings: 13/20

Scored on the 4 findings mm3 gave the largest analytical weight (deltas). For mm3, F17 sterile-domination carried a real +2pt draw / −1pt FRA delta (staked "fires"), so it is the 4th scored finding; Mbappé milestone (+2pt FRA) sits just behind and is credited below as a 5th correct stance.

#	Finding staked	Direction	Magnitude	Score	Note
1	R32-3 CONMEBOL cap — FRA advance capped ~60% (PAR 5-4-1 + Gill + counter)	✅	✅	5	FRA advanced 1-0 in 90' — the tight, low-scoring favorite win the blueprint predicts. Post-event brief: "cap held perfectly… strongest validation yet (3rd)."
2	F11 heat — −1pt FRA, Tier 1 modulator, heat favors PAR block	✅	✅	5	99°F muted FRA to 1 goal at 70', 1.36 xG. mm3's small −1pt + Tier 1 modulator was the most accurate F11 magnitude of the three (FRA won, didn't blow out). Brief: "Tier 1 modulator worked — FRA still won but didn't blow out."
3	Gill hot-GK confirmed-form — ×0.5-0.6 → +3pt draw (×0.25-empirical-scaled), pens edge PAR 58%	✅	❌	3	Gill solid (1 GA from 1.36 xG, not beaten from open xG — direction right). But the +3pt draw + pens edge didn't fire: Gill made 1 save (not heroic), match won in 90' (no pens). Over-priced the game-flip. Mitigating: mm3's ×0.25 empirical weight was the most conservative — better-calibrated than g4/g5's full ×0.5-0.6.
4	F17 sterile-domination FIRES — +2pt draw, −1pt FRA (Mbappé isolated, <1.0 xG)	❌	❌	0	Did NOT fire. FRA 76% possession but 1.36 xG (>1.0 sterile threshold) and Mbappé scored the winner (not isolated). mm3 staked sterile-domination; actual was possession-dominant-but-converting. The over-weighted 32% draw band (highest of the three) partly flowed from this misfire.

Additional correctly-staked positions (not in the scored 4): Mbappé milestone +2pt FRA — ✅ scored 70', full 90', now 19 WC goals surpassed Messi; F12 conservation 0% — ✅ (Mbappé milestone, KO must-win); F22 blowout excluded — ✅ (no blowout; moot).

Model-breaking: 10/10

Caught every material breaker: weather (99°F F11, validated), Gill confirmed-form (the GK variable), Mbappé milestone, PAR 5-4-1 block, referee flagged as TBD process gap (generic 25-35% pen band; actual 0 pens = the band's low tail, not a miss). No GK change / lineup surprise / formation shift occurred — nothing to miss. Caveat: the headline-flagged "game-flipper" (Gill) made only 1 save and didn't flip the match — correctly identified as a factor, over-emphasized as the factor (magnitude captured in Findings #3, not double-counted here).

TOTAL: 68 / 80 → 85% → Band B

auto 45 (winner_90 15 + advance 10 + scoreline 20*) + findings 13 + model-breaking 10 · who_scores 10/20 (separate)
*scoreline reconstructed — see process note.

g4

Findings: 18/20

Scored on the 4 findings g4 weighted most. g4 screened F17 to zero delta ("does NOT fire") — correct — so Mbappé milestone (the #1 scorer stake) is the 4th scored finding.

#	Finding staked	Direction	Magnitude	Score	Note
1	R32-3 CONMEBOL cap — FRA advance 60%	✅	✅	5	Cap held; FRA 1-0 in 90'.
2	F11 heat — −4% FRA (×0.6 Tier 1 modulator; 37°C is 6.5°C hotter than FRA-SWE)	✅	✅	5	99°F muted FRA to 1 goal, 1.36 xG. The −4% win-prob penalty was slightly high (FRA won comfortably in 90'), but the attack was genuinely muted to a single goal — defensible.
3	Gill hot-GK — ×0.5-0.6, +5% draw, pens 60% PAR; "Gill is the single variable that flips this match"	✅	❌	3	Gill solid (direction right). But the +5% draw (largest of the three) + 60% PAR-pens edge didn't fire: Gill 1 save, no pens. g4 most over-weighted Gill's game-flip potential and most under-stated FRA's win.
4	Mbappé milestone — F12=0%, plays 90', milestone→goals not minutes	✅	✅	5	Scored 70', full 90', now 19 WC goals (surpassed Messi). Bulletproof.

Additional: F17 screened → "does NOT fire" ✅ correct (FRA 1.36 xG, scored — not sterile); F22 partial exclusion (Gill not a veteran GK >50 caps) — analytically sharp nuance but moot (no blowout); VAR/pen monitor risk flagged — did not materialize (0 pens).

Model-breaking: 10/10

Same catcher profile as mm3 — weather, Gill form, Mbappé milestone, PAR block all anticipated; referee TBD flagged as process gap; no unanticipated breaker occurred. g4 was the most emphatic that Gill flips the match (the flagged #1 breaker); Gill made 1 save and didn't flip it, but the factor was caught, not missed. Captured in Findings #3.

TOTAL: 53 / 80 → 66% → Band C

auto 25 (winner_90 15 + advance 10 + scoreline 0) + findings 18 + model-breaking 10 · who_scores 10/20 (separate)

g5

Findings: 18/20

Scored on the 4 findings g5 weighted most. g5 screened F17 to a partial flag (correct), so Mbappé milestone is the 4th scored finding.

#	Finding staked	Direction	Magnitude	Score	Note
1	R32-3 CONMEBOL cap — FRA advance ~62%	✅	✅	5	Cap held; FRA 1-0 in 90'. g5's 62% advance was the closest read to the capped-favorite reality, and its 1-0 modal (16%) was the exact result.
2	F11 heat — +5% draw, transition-team moderation, "5 PM EDT = no evening cooling"	✅	✅	5	Cleanest F11 reasoning of the three: "France = transition team, not gegenpressing" → heat degrades running/transition, not a pressing mechanism → moderated +5% draw. Validated: FRA muted to 1 goal; 1-0 modal was EXACT. The "5 PM kickoff = no evening-cooling escape" nuance was sharp and correct.
3	Gill hot-GK — ×0.5-0.6, +3% draw, pens 55% PAR	✅	❌	3	Gill solid (direction right). But the +3% draw + pens edge didn't fire: Gill 1 save, match won in 90'. Same magnitude miss as the others, applied at mid-size.
4	Mbappé milestone — F12=0%, #1 scorer 32%, F16 age-decline N/A (Mbappé 27, not 38+)	✅	✅	5	Scored 70', full 90', 19 WC goals. Correctly flagged F16 does not apply.

Additional: F17 partial flag / "not a full fire" ✅ correct (FRA 1.36 xG, scored); "wrong mechanism caps Mbappé below 40%" — Mbappé scored once from a 32% base, consistent with the cap.

Model-breaking: 10/10

Identical catcher profile — weather (99°F), Gill confirmed-form, Mbappé milestone, PAR 5-4-1 block, referee TBD flagged. No unanticipated breaker occurred. g5 was the most precise on the heat timing (5 PM afternoon, no evening cooling), the sharpest single model-breaking nuance. Gill's game-flip over-emphasis captured in Findings #3.

TOTAL: 78 / 80 → 98% → Band A

auto 50 (winner_90 15 + advance 10 + scoreline 25) + findings 18 + model-breaking 10 · who_scores 10/20 (separate)

Cross-agent notes

The cap (R32-3) is now 3/3. GER-PAR (upset), NED-MAR (upset), FRA-PAR (correct-margin favorite win). All three writers applied the cap and all three were right on direction+magnitude (5/5). This is the strongest single finding of the match.
F11 heat Tier 1 modulator validated cleanly. All three predicted a muted FRA attack → low-scoring; actual 1 goal at 70', 1.36 xG. g5's transition-team framing + 5 PM nuance was the sharpest; mm3's small −1pt magnitude was the most accurate.
Gill hot-GK was the systematic magnitude miss (3/5 for all three). Confirmed-form was directionally right (Gill solid, not beaten from open xG), but the draw-inflation + pens-edge game-flip didn't fire — Gill made 1 save and FRA won in 90'. This re-confirms the project's empirical ×0.25 hot-GK weight; mm3 (which applied ×0.25) was the most-calibrated, g4 (×0.5-0.6, +5% draw) the least.
mm3's F17 sterile-domination misfire (0/5) is the differentiator vs g4/g5. g4 ("does NOT fire") and g5 ("partial flag") correctly screened it out; mm3 staked it fires (+2pt draw) when FRA produced 1.36 xG and Mbappé scored. This, plus mm3's malformed TOML, separates mm3 from g4/g5 on findings even though all three nailed the cap + heat + milestone.
Scoreline decided the bands. g5's 1-0 modal (exact) vs g4/mm3's 1-1 modal (wrong) is the single largest score driver. mm3's 1-0 FRA at 11% (#2) is reconstructed to the script's 20-pt tier; g4's "FRA 1-0" at 14% was scored 0 by the script's rigid key-format matching (a known limitation — "FRA 1-0" ≠ "1-0").

Scores Summary (machine-readable — DO NOT EDIT)

{
  "match": "fra-vs-par",
  "result": { "score90": "1-0", "final": "1-0", "outcome90": "win", "advance": "a", "route": "90" },
  "agents": {
    "mm3": { "auto": 45, "winner_90": 15, "advance": 10, "scoreline": 20, "findings": 13, "model_breaking": 10, "total": 68, "pct": 85, "band": "B", "who_scores": 10 },
    "g4":  { "auto": 25, "winner_90": 15, "advance": 10, "scoreline": 0, "findings": 18, "model_breaking": 10, "total": 53, "pct": 66, "band": "C", "who_scores": 10 },
    "g5":  { "auto": 50, "winner_90": 15, "advance": 10, "scoreline": 25, "findings": 18, "model_breaking": 10, "total": 78, "pct": 98, "band": "A", "who_scores": 10 }
  }
}