Project Turtle / Portugal-Spain / July 6, 2026

Portugal vs Spain exposed a Turtle protocol failure.

The locked card missed the basic Spain clean-sheet branch. Spain's verified non-concession streak and Neohm's named 0-1 Spain signal should have forced a clean-sheet score into C4 unless the proof defeated it numerically. It did not.

LOCKED: 2026-07-06 2:00 PM ETNo-leak cutoff: 2026-07-06Rows mined: Portugal 50, Spain 50Model-usable: Spain 49 after one missing-score row is flagged out

Locked four-score card / 2:00 PM ET / postgame miss

C4: 1-1, 1-2, 2-1, 2-2. Missing branch: 0-1 Spain.

Card CSV Prediction lock Postgame audit

C4-1 / draw_low 1-1

four-score prediction. Index 0.132955; count probability 0.124624.

C4-2 / spain_by_1 1-2

four-score prediction. Index 0.098638; count probability 0.082953.

C4-3 / portugal_by_1 2-1

four-score prediction. Index 0.094139; count probability 0.080920.

score 4 / draw_high 2-2

fourth score prediction. Index 0.042145; count probability 0.053863.

This locked card is preserved as the failed pregame record. The final was Portugal 0-1 Spain. The method now requires Clean-Sheet Override, Human Signal Preservation, and Recursive Attack before any future C4 can lock.

Postgame audit / protocol failure

The proof should have preserved 0-1 Spain.

Audit MD

Locked record

The locked C4 was 1-1, 1-2, 2-1, 2-2. The final score was Portugal 0-1 Spain, so the public card missed. The miss is not classified as variance: the model ignored the simplest decisive branch.

Correct missing branch

s^\* = 0\text{-}1\ \mathrm{Spain}.

Spain's elite non-concession streak and Neohm's pre-lock signal made this branch mandatory to test. Turtle excluded it without a valid defeat inequality.

Failure class

\mathrm{Failure}=BTTS_{\mathrm{overstack}}+\mathrm{CleanSheet}_{\mathrm{ignored}}+\mathrm{HumanSignal}_{\mathrm{dropped}}.

The card put every score in a both-teams-score state. That was not justified against Spain's verified defensive state.

Corrected C4

C4_{\mathrm{corrected}}=\{0\text{-}1,\ 1\text{-}1,\ 2\text{-}1,\ 2\text{-}2\}.

This preserves the named clean-sheet branch while keeping draw, Portugal-by-one, and high-draw failure modes alive.

Stricter C4

C4_{\mathrm{strict}}=\{0\text{-}1,\ 0\text{-}2,\ 1\text{-}1,\ 2\text{-}1\}.

If Spain's non-concession streak is treated as the dominant opposing fact, the second Spain clean-sheet score also survives.

item	postgame finding	method correction
Locked C4	1-1, 1-2, 2-1, 2-2	Preserve as failed pregame lock.
Actual	Portugal 0-1 Spain	Grade C4 as miss.
Missing branch	0-1 Spain	Clean-Sheet Override must force a clean-sheet score unless defeated numerically.
Human signal	Neohm named 0-1 Spain before lock	Human Signal Preservation requires inclusion or numeric rejection.
Proof defect	2-2 was promoted over 0-1 without defeating Spain clean-sheet evidence	Recursive Attack must attack every C4 score with the strongest opposing fact.
Support guidance	Overconfident and not backed by EV/risk/correlation tables	Public Turtle is analysis-only until those tables exist.

Data mining

Evidence first: last-50 plus opponent-threshold realms.

Portugal CSV Spain CSV

Portugal and Spain are both elite. The model therefore does not ask only how often they score in general; it asks how Portugal performs against Spain-realm opponents and how Spain performs against Portugal-realm opponents.

team	GF50	GA50	GF10	GA10	GFWC	GAWC	realm_rows	realm_GF_shrunk	realm_GA_shrunk	realm_BTTS_rate	realm_over_2_5	realm_over_3_5	readiness
Portugal	2.520	0.820	2.300	0.700	2.000	0.500	10	2.031	0.878	0.534	0.605	0.324	strong
Spain	2.551	0.755	2.100	0.400	2.000	0.000	13	2.257	1.217	0.653	0.702	0.433	strong

Model threads

The score cloud comes after result, total, and BTTS threads.

Model CSV

metric	value	formula
lambda_POR	1.2986	((POR attack profile + ESP defensive weakness profile)/2)0.920.96
lambda_ESP	1.3313	((ESP attack profile + POR defensive weakness profile)/2)0.920.96
P(POR regulation win)	0.3613	Poisson grid result mass
P(draw regulation)	0.2622	Poisson grid result mass
P(ESP regulation win)	0.3765	Poisson grid result mass
P(under 3.5)	0.7296	sum_{a+b<=3} P(a,b)

The key read was incomplete: Spain's regulation edge was visible, but the clean-sheet state was not preserved. A future proof must attack every BTTS score with the Spain non-concession fact before admitting it.

Candidate grid

Every S10 score receives a status.

Candidate CSV

score	poisson_probability	turtle_index	branch	total_bucket	btts_clean_bucket	status
1-1	0.124624	0.132955	draw_low	exactly2	btts	SCORE
1-2	0.082953	0.098638	spain_by_1	exactly3	btts	SCORE
2-1	0.080920	0.094139	portugal_by_1	exactly3	btts	SCORE
0-1	0.095967	0.093435	spain_by_1	u1_5	spain_clean	EXCLUDED
1-0	0.093614	0.088184	portugal_by_1	u1_5	portugal_clean	EXCLUDED
0-2	0.063878	0.057943	spain_by_2plus	exactly2	spain_clean	EXCLUDED
2-0	0.060785	0.052762	portugal_by_2plus	exactly2	portugal_clean	EXCLUDED
0-0	0.072087	0.047164	draw_low	u1_5	no_score	EXCLUDED
1-3	0.036811	0.043533	spain_by_2plus	o3_5	btts	EXCLUDED
2-2	0.053863	0.042145	draw_high	o3_5	btts	SCORE_4

Locked proof board / failed audit

The reasoning was not strong enough because it did not recursively attack BTTS.

D1Locked universe

The pregame proof considered Portugal-Spain exact scores using opponent-threshold ledgers and a Poisson count kernel. The locked public C4 was {1-1, 1-2, 2-1, 2-2}.

D2Missed evidence

Spain entered with an elite verified non-concession streak. Neohm also named the 0-1 Spain branch before lock. These are not narrative facts; they are mandatory candidate attacks.

Failure 1Overstacked BTTS

Every locked score required both teams to score. The proof did not show that BTTS evidence defeated Spain's clean-sheet state.

Failure 2Human signal dropped

The 0-1 Spain signal was not included and was not rejected by a numeric rule. Under the patched method, this alone invalidates the proof.

Failure 3Recursive attack absent

2-2 was promoted over 0-1 by a tail-role argument, but the argument never attacked 2-2 with the strongest opposing fact: Spain's non-concession streak.

Patch R1Clean-Sheet Override

An extreme verified non-concession streak forces at least one clean-sheet score into C4 unless every clean-sheet candidate is defeated by a stronger inequality.

Patch R2Human Signal Preservation

If Neohm names a score branch before lock, the proof must include it or reject it with a numeric rule-based argument.

Patch R3Recursive Attack

Every score admitted to C4 must survive the strongest opposing fact. No score survives by narrative or by branch language alone.

Corrected theorem0-1 must be represented

Under the patched rules, a Portugal-Spain C4 that excludes every Spain clean-sheet score is invalid unless it explicitly defeats the Spain clean-sheet streak. The locked proof did not do that.

FAILLocked C4 = {1-1, 1-2, 2-1, 2-2}; actual = 0-1 Spain.The theorem is preserved as the failed pregame proof, not defended postgame.

Support analysis after lock

Support sizing guidance was overconfident and formally unjustified.

Support CSV

market	model probability	status	settlement	protocol note
Over 1.5 goals	0.7383	overconfident	missed	Depended on the same overstacked BTTS premise that failed the card.
BTTS Yes	0.5350	overstacked	missed	Was not allowed to defeat Spain's clean-sheet evidence without a numeric rule.
Under 3.5 goals	0.7296	aligned	hit	Directionally coherent, but sizing guidance was not justified without EV/risk/correlation tables.
Over 2.5 goals	0.4890	partial	missed	Contradicted the clean-sheet branch Turtle failed to preserve.

Public Turtle is analysis-only until a formal EV table, bankroll cap, risk table, and correlation table exist. Support markets may be classified as aligned, partial, or contradictory; they do not publish allocation-size guidance.

Sources and artifacts