Project Turtle / prediction lock / July 4, 2026

Paraguay vs France

France pressure is active, but Paraguay's low-block/draw evidence cannot be deleted by favorite strength alone. The proof must compare France cushion against low-event containment.

Lock: 2026-07-04 06:00 PM EDTKickoff: 2026-07-04 5:00 PM ETWorld Cup knockoutLincoln Financial Field, Philadelphia

Late analysis lock: generated after scheduled kickoff from pre-cutoff ledgers only; do not grade as a clean pregame lock.

Locked four-score card

Card: 0-2, 0-1, 1-1. Score 4: 1-3.

Card CSV

Prediction lock2026-07-04 06:00 PM EDT

Visible timestamp; the card cannot move after state changes.

Four-score card0-2, 0-1, 1-1

$10 standard per score score.

Score 41-3

Score 4 in the same C4 card.

SCORE 1 0-2

B_away_cushion | P=0.140093 | I=0.740957

SCORE 2 0-1

B_away_win | P=0.141735 | I=0.664791

SCORE 3 1-1

B_draw_low | P=0.093326 | I=0.225854

SCORE 4 1-3

B_away_cushion | P=0.060785 | I=0.507267

Pregame / postgame analysis

Pregame proof and postgame analysis live together.

Pregame analysisProblem. Using only pre-cutoff data for Paraguay and France, let G={0,...,7} x {0,...,7} and P_count(h,a)=Pois(h;0.6585)Pois(a;1.9768). Filter each ledger by the opponent's strength realm, build result/spread/total/BTTS-clean threads, then rank S10 by LL(s). Determine the unique four-score card C4.

Lineup conflict unresolved in public scout sources: Mbappe status, Diego Gomez return and Paraguay defensive availability are treated as bounded modifiers rather than hard score facts.

Market prices are excluded from this proof. Prices may size or reject a public card only after the score cloud is locked.

Postgame analysisFinal: Paraguay 0-1 France. Turtle's four-score card hit, but the single-branch execution missed.

Final score0-1

Paraguay goals first; France goals second.

OfficialFour-score card hit

Four-score accuracy is actual score in C4.

Score-card resultContainment hit

Four-score accuracy is actual score in C4.

SupportOver 3.5 goals NO: hit

Support markets do not rewrite the score prediction.

The locked late-analysis C4 was 0-2, 0-1, 1-1, 1-3. The match settled on 0-1, so the four-score card contained the exact score and the default Under 3.5 support thesis also landed. The public correction is not that Turtle should chase only the top-ranked cushion branch; it is that all three score scores are equal protocol slots.

Lock honesty. Late analysis lock: generated after scheduled kickoff from pre-cutoff ledgers only; do not grade as a clean pregame lock.

What went right

The opponent-threshold realm did its job: Paraguay's France-realm attack was sparse and low-scoring, while France's Paraguay-realm profile preserved a strong clean-sheet/control branch.
R3 kept 0-1 as the exact-cell anchor even though R2 gave first display position to the 0-2 cushion branch.
The total-goals thread was directionally correct: every four-score card score was under 3.5 and the match finished with one total goal.

What changes

Ranking is not staking permission. When 0-2 and 0-1 live in the same favorite-clean-sheet family, Turtle must keep equal-stake protocol explicit instead of letting the first displayed score look like a solo bet.
If the exact-cell leader is the lower-total sibling and its Poisson probability is within 0.5 percentage points of the spread-thread leader, label both as co-leaders in the public card.
Because the lock was generated after scheduled kickoff, grade this as a late-analysis hit, not a clean pregame hit.

Support settlement

Market	Side	Model probability	Settlement
Over 1.5 goals	YES	0.687703	miss
Over 2.5 goals	NO	0.490415	hit
Over 3.5 goals	NO	0.695473	hit
Both teams to score	YES	0.412643	miss

Score cloud

S10 before branch proof.

Full grid

Rank	Score	Probability	Cumulative Probability	Total Goals	Btts
1	0-1	0.141735	0.141735	1	0
2	0-2	0.140093	0.281828	2	0
3	1-1	0.093326	0.375154	2	1
4	0-3	0.092314	0.467468	3	0
5	1-2	0.092246	0.559714	3	1
6	0-0	0.071698	0.631411	0	0
7	1-3	0.060785	0.692196	4	1
8	1-0	0.047210	0.739406	1	0
9	0-4	0.045622	0.785029	4	0
10	2-1	0.030726	0.815755	3	1

Mathematical model

Data ledgers become lambdas, then a theorem.

Model CSV

A_PAR0.791

Blended attack

D_PAR1.490

Blended defensive weakness

A_FRA2.363

Blended attack

D_FRA0.702

Blended defensive weakness

Formula. A_i=w50 GF50_i+w10 GF10_i+wWC GFWC_i and D_i=w50 GA50_i+w10 GA10_i+wWC GAWC_i. Then lambda_home_raw=(A_home+D_away)/2 and lambda_away_raw=(A_away+D_home)/2. Context gates are bounded and recorded in the CSV.

P (h, a) = Pois (h; λ_h) Pois (a; λ_a)

Opponent-threshold model

The last 50 is filtered by today's opponent realm.

Threshold ledger Realm summaries

A flat last-50 table asks how a team performs in general. Turtle now asks how Paraguay performs against France-strength opponents, and how France performs against Paraguay-strength opponents. Sparse samples are marked and shrunk toward the full ledger.

Team Code	Team	Team Rating	Target Opponent Code	Target Opponent	Target Rating	Epsilon Used	Raw Rows In Realm	Effective Rows	Readiness	Readiness Score	Shrinkage Alpha	Goals For Per Match	Goals Against Per Match	Scored Rate	Failed To Score Rate	Conceded Rate	Clean Sheet Rate	Btts Rate	Over 1 5 Rate	Over 2 5 Rate	Over 3 5 Rate	Win Rate	Draw Rate	Loss Rate	Win By 1 Rate	Win By 2 Plus Rate	Loss By 1 Rate	Loss By 2 Plus Rate	Draw Low Rate	Draw High Rate
PAR	Paraguay	6.60	FRA	France	9.40	0.75	6	4.875	usable	0.780	0.0	0.6896	1.7396	0.5158	0.4842	0.8227	0.1773	0.3385	0.4794	0.4794	0.3057	0.3511	0.0000	0.6489	0.3511	0.0000	0.3433	0.3057	0.0000	0.0000
FRA	France	9.40	PAR	Paraguay	6.60	0.75	22	15.226	strong	1.000	0.0	2.1417	0.6883	0.8717	0.1283	0.5311	0.4689	0.4809	0.7441	0.6004	0.4049	0.6970	0.2171	0.0859	0.2022	0.4948	0.0859	0.0000	0.1243	0.0928

Thread model

Result, spread, total, and clean-sheet evidence before scores.

Thread CSV

Thread	State	Probability	Model Component	Evidence	Rule Note
result	home	0.161315	grid=0.114362; evidence=0.218500	Poisson count grid blended with opponent-threshold W/D/L realm evidence.	Exact scores inherit this result mass before branch selection.
result	draw	0.159026	grid=0.200163; evidence=0.108550	Poisson count grid blended with opponent-threshold W/D/L realm evidence.	Exact scores inherit this result mass before branch selection.
result	away	0.679659	grid=0.684455; evidence=0.672950	Poisson count grid blended with opponent-threshold W/D/L realm evidence.	Exact scores inherit this result mass before branch selection.
spread	home_by_1	0.152009	grid=0.085374; evidence=0.218489	Spread family is built before exact-score disclosure.	The strongest spread family receives the first representative.
spread	home_by_2_plus	0.014502	grid=0.028988; evidence=0.000000	Spread family is built before exact-score disclosure.	The strongest spread family receives the first representative.
spread	draw	0.154433	grid=0.200163; evidence=0.108545	Spread family is built before exact-score disclosure.	The strongest spread family receives the first representative.
spread	away_by_1	0.264659	grid=0.256311; evidence=0.272736	Spread family is built before exact-score disclosure.	The strongest spread family receives the first representative.
spread	away_by_2_plus	0.414398	grid=0.428144; evidence=0.400230	Spread family is built before exact-score disclosure.	The strongest spread family receives the first representative.
total_btts_clean	over_1_5	0.687703	count grid plus opponent-threshold event rates	Support thread; it can protect or reject a branch but cannot create a score by itself.	Used inside LL(s), not as a standalone prediction.
total_btts_clean	over_2_5	0.509585	count grid plus opponent-threshold event rates	Support thread; it can protect or reject a branch but cannot create a score by itself.	Used inside LL(s), not as a standalone prediction.
total_btts_clean	over_3_5	0.304527	count grid plus opponent-threshold event rates	Support thread; it can protect or reject a branch but cannot create a score by itself.	Used inside LL(s), not as a standalone prediction.
total_btts_clean	btts_yes	0.412643	count grid plus opponent-threshold event rates	Support thread; it can protect or reject a branch but cannot create a score by itself.	Used inside LL(s), not as a standalone prediction.
total_btts_clean	home_clean_sheet	0.144938	count grid plus opponent-threshold event rates	Support thread; it can protect or reject a branch but cannot create a score by itself.	Used inside LL(s), not as a standalone prediction.
total_btts_clean	away_clean_sheet	0.498864	count grid plus opponent-threshold event rates	Support thread; it can protect or reject a branch but cannot create a score by itself.	Used inside LL(s), not as a standalone prediction.

Candidate table

Every S10 score receives one status.

Candidate CSV

Rank	Score	Poisson Probability	Result Thread Probability	Spread Thread Probability	Total Thread Probability	Btts Clean Thread Probability	Turtle Index	Status	Rule
1	0-1	0.141735	0.679659	0.264659	0.312297	0.498864	0.664791	SCORE	R3 exact-cell anchor
2	0-2	0.140093	0.679659	0.414398	0.178118	0.498864	0.740957	SCORE	R2 spread-thread leader
3	1-1	0.093326	0.159026	0.154433	0.178118	0.412643	0.225854	SCORE	R4 alternate-thread representative
4	0-3	0.092314	0.679659	0.414398	0.205058	0.498864	0.588489	EXCLUDED_BY_SPREAD_THREAD	R2/R5 spread comparison
5	1-2	0.092246	0.679659	0.264659	0.205058	0.412643	0.457659	EXCLUDED_BY_SPREAD_THREAD	R2/R5 spread comparison
6	0-0	0.071698	0.159026	0.154433	0.312297	0.321901	0.262779	EXCLUDED_BY_TOTAL_THREAD	R1 total gate
7	1-3	0.060785	0.679659	0.414398	0.304527	0.412643	0.507267	SCORE_4	R5 fourth-score
8	1-0	0.047210	0.161315	0.152009	0.312297	0.144938	0.211614	EXCLUDED_BY_SPREAD_THREAD	R4 spread-order comparison
9	0-4	0.045622	0.679659	0.414398	0.304527	0.498864	0.437231	EXCLUDED_BY_SPREAD_THREAD	R2/R5 spread comparison
10	2-1	0.030726	0.161315	0.152009	0.205058	0.412643	0.231824	EXCLUDED_BY_SPREAD_THREAD	R4 spread-order comparison

Data analysis

Only computed evidence, not clutter.

PAR ledger FRA ledger

Team Code	Team	Sample	Matches	Wins	Draws	Losses	Goals For Per Match	Goals Against Per Match	Scored Rate	Conceded Rate	Clean Sheet Rate	Btts Rate	Over 1 5 Rate	Over 2 5 Rate	Over 3 5 Rate	Two Plus Scored Rate
PAR	Paraguay	last_50	50	16	14	20	0.880	1.120	0.5800	0.5800	0.4200	0.3400	0.5000	0.3800	0.2200	0.2400
PAR	Paraguay	last_10	10	4	2	4	1.200	1.200	0.8000	0.6000	0.4000	0.5000	0.7000	0.5000	0.2000	0.2000
PAR	Paraguay	world_cup_2026	4	1	2	1	0.750	1.250	0.7500	0.5000	0.5000	0.5000	0.5000	0.2500	0.2500	0.0000
FRA	France	last_50	50	34	8	8	2.280	0.880	0.8600	0.5600	0.4400	0.5000	0.8200	0.6400	0.3600	0.6800
FRA	France	last_10	10	9	0	1	2.900	0.800	1.0000	0.7000	0.3000	0.7000	1.0000	1.0000	0.6000	0.9000
FRA	France	world_cup_2026	4	4	0	0	3.250	0.500	1.0000	0.5000	0.5000	0.5000	1.0000	1.0000	0.5000	1.0000

Win/draw/loss score modes

Team Code	Team	Bucket	Matches	Avg Goals For	Avg Goals Against	Avg Total Goals	Scoreline Modes
PAR	Paraguay	Wins	16	1.625	0.250	1.875	1-0 (9/16); 2-1 (3/16); 2-0 (2/16); 3-1 (1/16); 4-0 (1/16)
PAR	Paraguay	Draws	14	0.643	0.643	1.286	0-0 (9/14); 2-2 (4/14); 1-1 (1/14)
PAR	Paraguay	Losses	20	0.450	2.150	2.600	0-1 (7/20); 1-2 (4/20); 0-2 (3/20); 1-4 (3/20); 0-4 (1/20)
FRA	France	Wins	34	2.882	0.500	3.382	2-1 (6/34); 3-1 (6/34); 2-0 (6/34); 3-0 (5/34); 1-0 (4/34)
FRA	France	Draws	8	1.000	1.000	2.000	0-0 (4/8); 2-2 (2/8); 3-3 (1/8); 1-1 (1/8)
FRA	France	Losses	8	1.000	2.375	3.375	1-2 (3/8); 0-2 (2/8); 0-1 (1/8); 1-3 (1/8); 4-5 (1/8)

Support market note

Totals/BTTS are support, not the prediction.

Support CSV

Market	Side	Model Probability	Fair Price	Max Entry Price	Suggested Stake Usd	Portfolio Role	Decision
Over 1.5 goals	YES	0.687703	0.69	0.63	0-5	Supports score clouds requiring at least two total goals.	coherent only when at least two score scores need 2+ goals
Over 2.5 goals	NO	0.490415	0.49	0.43	5-10	Protects low-total cards but can kill 2-1/1-2 branches.	use cautiously if a score score has three total goals
Over 3.5 goals	NO	0.695473	0.70	0.64	10-15	Protects every cloud that stays below four total goals.	default support when all C4 scores are below four total goals
Both teams to score	YES	0.412643	0.41	0.35	0-5	Only supports if the four-score card actually requires both teams scoring.	not a shortcut; compare to exact-score cloud first

Aster proof board

Definitions, data, rules, lemmas, theorem.

Proof MD

01Definition 1
Let G={0,...,7} x {0,...,7}. S10 is the ten scores with largest P_count(s), tie-broken lexicographically by (home goals, away goals): {0-1, 0-2, 1-1, 0-3, 1-2, 0-0, 1-3, 1-0, 0-4, 2-1}. No exact score outside S10 can enter C4; it may enter only a named post-lock/live-state audit.
02Definition 2
For s=(h,a), P(s)=Pois(h;lambda_PAR) Pois(a;lambda_FRA), where lambda_PAR=0.6585 and lambda_FRA=1.9768.
03Definition 3
Let r_PAR=6.60 and r_FRA=9.40. PAR evidence is filtered to opponents within epsilon of r_FRA; FRA evidence is filtered to opponents within epsilon of r_PAR.
04Definition 4
Each historical row receives weight recency * competition * squad_overlap * exp(-distance^2/(2 sigma^2)), sigma=0.75. If the realm has fewer than five usable rows, epsilon widens to 1.25. If effective n still stays below five, the realm statistic is shrunk toward the last-50 statistic and readiness is marked weak.
05Definition 5
LL(s)=log P_count(s)+0.70 log P_result(s)+0.85 log P_spread(s)+0.45 log P_total(s)+0.45 log P_BTTS_clean(s). TurtleIndex normalizes exp(LL) and adds bucket, branch, form, readiness, and pressure support.
06Definition 6
C4 is the ordered set of four scores selected by R2-R6; score 4 is selected by R5 and graded inside C4.
07Data 1
Paraguay last-50 W-D-L is 16-14-20; GF/M=0.880, GA/M=1.120.
08Data 2
France last-50 W-D-L is 34-8-8; GF/M=2.280, GA/M=0.880.
09Data 3
Opponent-threshold realm: Paraguay vs France-realm has 6 raw rows, effective n=4.875, readiness=usable, GF/M=0.6896, GA/M=1.7396, scored=0.5158, clean-sheet=0.1773.
10Data 4
Opponent-threshold realm: France vs Paraguay-realm has 22 raw rows, effective n=15.226, readiness=strong, GF/M=2.1417, GA/M=0.6883, scored=0.8717, clean-sheet=0.4689.
11Data 5
Result thread probabilities are PAR win=0.161, draw=0.159, FRA win=0.680.
12Data 6
Spread thread order is away_by_2_plus=0.414; away_by_1=0.265; draw=0.154; home_by_1=0.152; home_by_2_plus=0.015. Therefore the strongest spread before exact-score selection is away_by_2_plus.
13Data 7
Total and clean-sheet threads are over1.5=0.688, over2.5=0.510, over3.5=0.305, BTTS=0.413, PAR clean sheet=0.145, FRA clean sheet=0.499.
14Data 8
Recent scoring remains visible but cannot delete clean sheets by itself: Paraguay scored in 80.0% of last ten; France scored in 100.0% of last ten. Current World Cup samples: 4 and 4.
15Data 9
Lineup conflict unresolved in public scout sources: Mbappe status, Diego Gomez return and Paraguay defensive availability are treated as bounded modifiers rather than hard score facts.
16Data 10
Scout memo: France hold the strength edge through ranking, squad quality and tournament scoring; Paraguay carry low-block evidence.
17Rule R1
The only hard pre-selection removal is 0-0 when P(over 1.5) >= 0.640000. Raw scoring rate alone cannot delete a clean-sheet score.
18Rule R2
Let B* be the spread state with maximum spread-thread probability. The first MAIN score is argmax TurtleIndex among eligible S10 scores with spread_key=B*.
19Rule R3
After R2, the exact-cell anchor is the remaining eligible S10 score with maximum P_count(s).
20Rule R4
After R2 and R3, scan spread states by descending spread-thread mass and skip already represented states. The first unrepresented spread whose representative has spread mass >=0.140000 or TurtleIndex >=0.280000 enters C4.
21Rule R5
The fourth score is the remaining eligible score maximizing 0.35*P_spread + 0.25*form + 0.25*I + 0.15*P_BTTS_clean. It is part of C4, not a separate containment grade.
22Rule R6
If fewer than three score scores survive, fill by highest remaining eligible I(s).
23Inclusion Lemma 1
0-2 is in C4 by R2 spread-thread leader: P_count=0.140093, result=0.679659, spread=0.414398, total=0.178118, BTTS/clean=0.498864, I=0.740957. 0-2 maximizes I(s) inside the strongest spread thread away_by_2_plus=0.4144 and beats 0-3 by I=0.740957 > 0.588489.
24Inclusion Lemma 2
0-1 is in C4 by R3 exact-cell anchor: P_count=0.141735, result=0.679659, spread=0.264659, total=0.312297, BTTS/clean=0.498864, I=0.664791. 0-1 is the exact-cell anchor after the spread leader and beats the next remaining count cell 1-1 by P_count=0.141735 > 0.093326.
25Inclusion Lemma 3
1-1 is in C4 by R4 alternate-thread representative: P_count=0.093326, result=0.159026, spread=0.154433, total=0.178118, BTTS/clean=0.412643, I=0.225854. 1-1 represents the strongest unrepresented spread thread draw with mass 0.1544 >= 0.1400 with I=0.225854.
26Fourth Score Lemma
1-3 is score 4 by R5: spread=0.414398, form=0.727500, I=0.507267, BTTS/clean=0.412643. 1-3 is the best excluded thesis-compatible branch by spread/form/index/BTTS-clean comparator and beats 0-3 by score 4 0.515627 > 0.503866.
27Exclusion Lemma
0-3 has status EXCLUDED_BY_SPREAD_THREAD by R2/R5 spread comparison: 0-3 shares away_by_2_plus with 0-2 and loses I=0.588489 < 0.740957.
28Exclusion Lemma
1-2 has status EXCLUDED_BY_SPREAD_THREAD by R2/R5 spread comparison: 1-2 shares away_by_1 with 0-1 and loses I=0.457659 < 0.664791.
29Exclusion Lemma
0-0 has status EXCLUDED_BY_TOTAL_THREAD by R1 total gate: 0-0 fails R1 because P(over 1.5)=0.687703 >= 0.640000, so the no-goal cell is removed before branch selection.
30Exclusion Lemma
1-0 has status EXCLUDED_BY_SPREAD_THREAD by R4 spread-order comparison: 1-0 belongs to home_by_1, but R4 selected 1-1 from draw because P_spread(draw)=0.154433 > P_spread(home_by_1)=0.152009.
31Exclusion Lemma
0-4 has status EXCLUDED_BY_SPREAD_THREAD by R2/R5 spread comparison: 0-4 shares away_by_2_plus with 0-2 and loses I=0.437231 < 0.740957.
32Exclusion Lemma
2-1 has status EXCLUDED_BY_SPREAD_THREAD by R4 spread-order comparison: 2-1 belongs to home_by_1, but R4 selected 1-1 from draw because P_spread(draw)=0.154433 > P_spread(home_by_1)=0.152009.
33Theorem
Under Definitions 1-6, Data 1-10, and Rules R1-R6, the unique disclosed four-score card is C4={0-2, 0-1, 1-1, 1-3}.

Data room

Artifacts for audit.

Audit note

Strength ratings Threshold ledger Realm summaries Thread table Candidate table Proof ledger Prediction lock

Public sources used.