Transparent model tracking

Navurex model performance.

These results use frozen pre-match prediction snapshots. Once kickoff passes, the model probabilities are not rewritten using the final score or future matches.

Integrity-verified history

129 historical settled rows are excluded by the integrity gate and do not affect these metrics.

Settled predictions

1793

403 pending snapshots

Pick accuracy

43.7%

783 correct top-outcome picks

High confidence

49.1%

1151 settled at ≥70 confidence

Brier score

0.6571

Lower is better · evaluates all 1X2 probabilities

Log loss

1.1017

Lower is better · strongly penalizes confident wrong probabilities

Recent window

43.6%

Last 30 days pick accuracy

1624 matches

How to read this

Pick accuracy measures the top 1X2 choice. Brier score evaluates all three probabilities, while log loss penalizes confident wrong probabilities most strongly. Calibration checks whether events predicted at a given probability happen at roughly that rate.

V1 vs V2 benchmark

Is Contextual V2 beating the baseline?

Only settled, integrity-verified matches with both V1 shadow and V2 pre-news probabilities are included. News adjustments are excluded so this comparison isolates the contextual engine.

498 compared matches

Compared matches

498

220 improved · 176 worsened · 102 neutral

Brier · V1 → V2

0.6674 → 0.6622

+0.0052 · +0.8%

Log loss · V1 → V2

1.1248 → 1.1155

+0.0093

Pick accuracy · V1 → V2

43.8% → 45.8%

Top 1X2 outcome only; probability scores remain the primary benchmark.

Average probability movement

1.80 pp

Raw V2 stored

498 / 498

Diagnostics stored

498 / 498

V2 snapshot write health

Are new Raw V2 diagnostics being frozen correctly?

Healthy

Recent contextual snapshots

latest 20 window

Raw V2 coverage

100.0%

20 / 20

Diagnostics coverage

100.0%

20 / 20

Complete V2 snapshots

100.0%

20 / 20

Latest diagnostics capture: 20 Oct 2026

Keep the current blend unchanged until the sample grows. Navurex reports the benchmark without automatically tuning the model from a small sample.

League comparison

CompetitionSampleV1 BrierV2 BrierV2 Brier delta

Championship

0.6287

0.6423

-0.0136

3. Liga

0.8497

0.8411

+0.0086

UEFA Champions League

0.7496

0.7367

+0.0129

UEFA Europa League

0.6426

0.6350

+0.0076

UEFA Conference League

0.6775

0.6791

-0.0016

2. Lig

0.4432

0.4415

+0.0018

1. Division

0.5429

0.5450

-0.0021

League Two

0.7475

0.7181

+0.0294

League One

0.7172

0.7104

+0.0067

La Liga 2

0.5229

0.5353

-0.0124

V2 vs V3 shadow benchmark

Is V3 improving the model?

Only settled, integrity-verified matches with both V2 pre-news and V3 shadow probabilities are included.

498 compared matches

Compared matches

498

258 improved · 227 worsened · 13 neutral

Brier · V2 → V3

0.6622 → 0.6289

+0.0333 · +5.0%

Log loss · V2 → V3

1.1155 → 1.0516

+0.0639

Pick accuracy · V2 → V3

45.8% → 47.2%

Top 1X2 outcome only; probability scores remain the primary benchmark.

Market coverage

391 / 498

Average probability movement

6.45 pp

V3 promotion gate

Keep V2 production until the V3 sample is large enough and probability scores are consistently better.

League comparison

CompetitionSampleV2 BrierV3 BrierV3 Brier delta

Championship

0.6423

0.6072

+0.0351

3. Liga

0.8411

0.7716

+0.0695

UEFA Conference League

0.6791

0.5880

+0.0911

UEFA Champions League

0.7367

0.6615

+0.0752

UEFA Europa League

0.6350

0.5669

+0.0681

2. Lig

0.4415

0.4378

+0.0037

1. Division

0.5450

0.5564

-0.0114

League One

0.7104

0.6180

+0.0925

League Two

0.7181

0.6340

+0.0841

La Liga 2

0.5353

0.5648

-0.0295

V3 vs market benchmark

Does V3 add information beyond the bookmaker consensus?

The market baseline uses frozen pre-kickoff no-vig consensus from eligible bookmaker snapshots. V2, V3 and market metrics are calculated on the exact same settled matches.

386 eligible market snapshots

Market benchmark sample

386

386 strict V3.1 quality verified · 0 legacy eligible · 5 reference-only excluded

Brier · V3 vs market

0.6146 → 0.5695

V3 vs market Brier edge -0.0451 · -7.9%

Positive means V3 better; negative means market better.

Log loss · V3 vs market

1.0269 → 0.9616

-0.0653

Pick accuracy · V3 vs market

47.7% → 56.7%

V3-market pick agreement 68.9%

1X2 calibration error

4.32 pp → 3.79 pp

same verified market sample

Average bookmakers

8.1

Average market blend

34.4%

V3-market pick agreement

68.9%

Average disagreement

12.36 pp

Strict market quality split

Does the market advantage persist after V3.1 strict quality validation?

Strict V3.1 snapshots and older legacy-eligible snapshots are reported separately so market-quality changes cannot hide inside one combined benchmark.

Strict V3.1 cohort

Only snapshots with an explicit V3.1 market quality gate are included.

n=386

Brier · V3 → market

0.6146 → 0.5695

-0.0451

Log loss · V3 → market

1.0269 → 0.9616

-0.0653

Pick accuracy · V3 → market

47.7% → 56.7%

Agreement 68.9%

Calibration error · V3 → market

4.32 pp → 3.79 pp

Bookmakers

8.1

Market blend

34.4%

Disagreement

12.36 pp

Structural → final → market Brier: 0.6454 → 0.6146 → 0.5695 · n=386

Positive means V3 better; negative means market better.

Strict disagreement buckets

DisagreementSampleV3 BrierMarket BrierBrier edge

0–5 pp

n=51

0.4305

0.4279

-0.0026

5–10 pp

n=131

0.6013

0.5696

-0.0318

10–20 pp

n=157

0.6547

0.6340

-0.0208

20+ pp

n=47

0.7170

0.5072

-0.2097

V3.4 dynamic market anchor shadow

Does stronger market anchoring improve V3 when disagreement is large?

This is a frozen pre-match shadow A/B test. Current V3 probabilities remain unchanged. No historical backfill is used.

Collecting V3.4 shadow settlements

No settled V3.4 shadow sample yet. Future eligible snapshots will populate this panel.

Pending V3.4 snapshots

159

Settled V3.4 payloads

337

Strict A/B payloads

241

V3.4 payloads are already being frozen before kickoff. Metrics appear after eligible matches settle.

Independent V3 signal check

Structural V3 is measured before bookmaker blending. Comparing structural V3, final V3 and the market helps show whether the model adds information independently or mostly follows the market.

Structural sample

386

Brier · structural → final → market

0.6454 → 0.6146 → 0.5695

Market-blend effect

+0.0308

Structural V3 vs market edge

-0.0760

Where does V3 disagree with the market?

DisagreementSampleV3 BrierMarket BrierBrier edgeV3 pickMarket pick

0–5 pp

0.4305

0.4279

-0.0026

74.5%

82.4%

5–10 pp

131

0.6013

0.5696

-0.0318

53.4%

55.7%

10–20 pp

157

0.6547

0.6340

-0.0208

40.8%

48.4%

20+ pp

0.7170

0.5072

-0.2097

25.5%

59.6%

Market calibration audit

V3 and market calibration on the same settled benchmark sample.

BandV3 predicted / observedMarket predicted / observed

40–50%

43.3% / 50.3% n=147

44.6% / 51.3% n=152

50–60%

53.6% / 53.7% n=67

54.2% / 55.0% n=80

60–70%

63.8% / 57.8% n=45

65.1% / 69.8% n=43

70–80%

74.1% / 83.3% n=30

74.7% / 92.9% n=28

80–90%

80.3% / 33.3% n=3

85.2% / 62.5% n=8

90–100%

—

91.5% / 100.0% n=1

League comparison

CompetitionSampleV3 BrierMarket BrierBrier edge

Championship

0.6072

0.6320

+0.0249

3. Liga

0.7716

0.7067

-0.0649

UEFA Champions League

0.6615

0.5674

-0.0941

UEFA Europa League

0.5669

0.4563

-0.1106

UEFA Conference League

0.5880

0.3948

-0.1932

1. Division

0.5564

0.5426

-0.0138

League Two

0.6340

0.5938

-0.0402

League One

0.6180

0.5523

-0.0657

La Liga 2

0.5691

0.6047

+0.0356

2. Lig

0.4495

0.4412

-0.0083

Market baseline is a benchmark, not ground truth. Reference-only or explicitly unusable markets are excluded.

Probability calibration

Do Navurex probabilities behave like probabilities?

Each settled 1X2 probability is grouped into a 10-point band. A well-calibrated model should have predicted probability and observed frequency close together.

Calibration error

7.60 pp

5379 probability observations

V5.5 calibration monitor

Monitor overconfidence before changing probabilities.

This monitor is diagnostic only. It does not recalibrate published probabilities. Navurex waits for enough V2 comparisons and enough observations inside the watched probability bands before considering a calibration change.

Calibration candidate

V2 comparison sample

498

Review at 50 · calibration candidate at 100

50–60% watch band

+2.8 pp

n=332

53.6% predicted · 50.9% observed

60–70% watch band

+5.9 pp

n=229

64.0% predicted · 58.1% observed

The monitoring thresholds are met. This only means a calibration experiment can be evaluated; it does not mean Navurex should automatically change production probabilities.

BandPredictedObservedGapSample

0–10%

8.0%

0.0%

+8.0 pp

1 observations

10–20%

13.8%

25.3%

-11.4 pp

1251 observations

20–30%

24.4%

30.1%

-5.6 pp

1091 observations

30–40%

34.6%

31.0%

+3.6 pp

1022 observations

40–50%

42.1%

32.3%

+9.8 pp

1262 observations

50–60%

53.6%

50.9%

+2.8 pp

332 observations

60–70%

64.0%

58.1%

+5.9 pp

229 observations

70–80%

75.2%

63.9%

+11.4 pp

191 observations

News Intelligence impact

Did verified match news improve the prediction?

For settled matches where verified model-eligible news changed the probabilities, Navurex compares the base model with the final news-adjusted prediction against the actual result.

Settled with news

Improved

Worsened

Brier improvement

-1.3%

Average Brier · before news

0.6118

Average Brier · after news

0.6197

By competition

Where the model has been tested.

Sample size is shown beside every league so a small run of results is not mistaken for long-term evidence.

CompetitionSampleAccuracyBrierLog loss

UEFA Nations League

UNL

Sample96

Accuracy42.7%

Brier0.6648

Log loss1.0961

2. Lig

TR2

Sample66

Accuracy56.1%

Brier0.5944

Log loss1.0019

La Liga 2

Sample55

Accuracy41.8%

Brier0.6566

Log loss1.1066

League Two

EL2

Sample54

Accuracy33.3%

Brier0.6875

Log loss1.1426

Championship

ELC

Sample48

Accuracy37.5%

Brier0.6825

Log loss1.1336

National League

ENL

Sample47

Accuracy61.7%

Brier0.5628

Log loss0.9696

Serie B

BRB

Sample46

Accuracy39.1%

Brier0.6598

Log loss1.1032

League One

EL1

Sample43

Accuracy41.9%

Brier0.6692

Log loss1.1248

Campeonato Brasileiro Série A

BSA

Sample41

Accuracy43.9%

Brier0.6176

Log loss1.0334

J2 League

Sample40

Accuracy30.0%

Brier0.7235

Log loss1.2182

J3 League

Sample40

Accuracy45.0%

Brier0.6266

Log loss1.0503

Serie C - Girone A

Sample39

Accuracy43.6%

Brier0.6854

Log loss1.1356

La Liga

Sample37

Accuracy51.4%

Brier0.5499

Log loss0.9454

Serie C - Girone B

SCB

Sample37

Accuracy51.4%

Brier0.6207

Log loss1.0573

Serie C - Girone C

SCC

Sample37

Accuracy40.5%

Brier0.6858

Log loss1.1500

Serie D - Girone E

SDE

Sample36

Accuracy41.7%

Brier0.6841

Log loss1.1355

Serie D - Girone F

SDF

Sample36

Accuracy33.3%

Brier0.6803

Log loss1.1411

Serie D - Girone G

SDG

Sample36

Accuracy30.6%

Brier0.7477

Log loss1.2380

Serie D - Girone C

SDC

Sample36

Accuracy33.3%

Brier0.7466

Log loss1.2510

Serie D - Girone A

SDA

Sample36

Accuracy52.8%

Brier0.6522

Log loss1.1043

Serie D - Girone D

SDD

Sample35

Accuracy51.4%

Brier0.6308

Log loss1.0515

Serie D - Girone H

SDH

Sample35

Accuracy40.0%

Brier0.6868

Log loss1.1575

National League - South

ENLS

Sample35

Accuracy57.1%

Brier0.5596

Log loss0.9396

Serie D - Girone I

SDI

Sample35

Accuracy40.0%

Brier0.6858

Log loss1.1501

National

Sample35

Accuracy42.9%

Brier0.6647

Log loss1.0978

Serie D - Girone B

SDB

Sample34

Accuracy38.2%

Brier0.6984

Log loss1.1437

Serie A

Sample32

Accuracy56.3%

Brier0.6130

Log loss1.0756

Paulista - U20

BRU20

Sample31

Accuracy64.5%

Brier0.5220

Log loss0.9052

3. Liga

BL3

Sample31

Accuracy38.7%

Brier0.7482

Log loss1.2685

National League - North

ENLN

Sample31

Accuracy22.6%

Brier0.7720

Log loss1.2762

Serie B

Sample30

Accuracy43.3%

Brier0.6430

Log loss1.0673

UEFA Europa League

UEL

Sample30

Accuracy53.3%

Brier0.6255

Log loss1.0646

Premier League

Sample30

Accuracy30.0%

Brier0.6961

Log loss1.1775

1. Division

NOR2

Sample29

Accuracy55.2%

Brier0.5481

Log loss0.9341

Bundesliga

BL1

Sample27

Accuracy37.0%

Brier0.7663

Log loss1.2903

Ligue 1

FL1

Sample27

Accuracy48.1%

Brier0.7112

Log loss1.1865

Ligue 2

FL2

Sample27

Accuracy48.1%

Brier0.6362

Log loss1.0439

2. Bundesliga

BL2

Sample27

Accuracy37.0%

Brier0.7354

Log loss1.2366

Primera División Femenina

ESPW

Sample24

Accuracy50.0%

Brier0.5744

Log loss0.9595

2. Division

DEN2

Sample24

Accuracy33.3%

Brier0.7311

Log loss1.1898

Women's Super League

WSL

Sample21

Accuracy47.6%

Brier0.5110

Log loss0.8747

Frauen-Bundesliga

FBL

Sample20

Accuracy45.0%

Brier0.6264

Log loss1.0873

Liga Portugal Betclic

PPL

Sample19

Accuracy42.1%

Brier0.6029

Log loss1.0198

UEFA Conference League

UECL

Sample18

Accuracy50.0%

Brier0.6791

Log loss1.1345

Süper Lig

TRSL

Sample18

Accuracy61.1%

Brier0.5126

Log loss0.8607

UEFA Champions League

UCL

Sample18

Accuracy38.9%

Brier0.7367

Log loss1.2764

Serie C

Sample18

Accuracy55.6%

Brier0.5975

Log loss0.9688

Ykkönen

FIN3

Sample18

Accuracy61.1%

Brier0.5837

Log loss0.9781

Eliteserien

NOR1

Sample16

Accuracy62.5%

Brier0.4949

Log loss0.8537

Ykkösliiga

FIN2

Sample15

Accuracy13.3%

Brier0.8784

Log loss1.4317

Veikkausliiga

FIN1

Sample15

Accuracy40.0%

Brier0.6662

Log loss1.1148

Serie C

BRC

Sample13

Accuracy23.1%

Brier0.8007

Log loss1.3052

Superliga

DENSL

Sample12

Accuracy33.3%

Brier0.8061

Log loss1.3257

Super League Women

BEW

Sample12

Accuracy41.7%

Brier0.7273

Log loss1.1712

1. Division

DEN1

Sample12

Accuracy58.3%

Brier0.5538

Log loss0.9289

Jupiler Pro League

BE1

Sample9

Accuracy66.7%

Brier0.5490

Log loss0.9607

Kakkonen - Lohko A

FIKA

Sample6

Accuracy16.7%

Brier0.6535

Log loss1.0544

Kakkonen - Lohko C

FIKC

Sample6

Accuracy33.3%

Brier0.6931

Log loss1.1423

Super League

CHSL

Sample6

Accuracy33.3%

Brier0.6488

Log loss1.1057

Kakkonen - Lohko B

FIKB

Sample6

Accuracy0.0%

Brier0.9129

Log loss1.6094

Audit trail

Latest settled predictions.

Each row links back to the match page and uses the frozen pre-kickoff snapshot.

IncorrectCampeonato Brasileiro Série A · 19 Oct 2026

Corinthians 1–1 Vitoria

Pick

Vitoria

Pick %

46%

Confidence

72%

Model

navurex-v1.0-contextual-form

CorrectSerie B · 19 Oct 2026

Goias 0–0 Nautico Recife

Pick

Draw

Pick %

40%

Confidence

72%

Model

navurex-v1.0-contextual-form

IncorrectLiga Portugal Betclic · 19 Oct 2026

SC Braga 1–2 GIL Vicente

Pick

Draw

Pick %

40%

Confidence

62%

Model

navurex-v1.0-contextual-form

CorrectLa Liga · 19 Oct 2026

Getafe 1–1 Rayo Vallecano

Pick

Draw

Pick %

40%

Confidence

66%

Model

navurex-v1.0-contextual-form

IncorrectPremier League · 19 Oct 2026

Tottenham 3–1 Coventry

Pick

Coventry

Pick %

45%

Confidence

65%

Model

navurex-v1.0-contextual-form

CorrectLigue 2 · 19 Oct 2026

Rodez 0–3 Montpellier

Pick

Montpellier

Pick %

56%

Confidence

85%

Model

navurex-v1.0-contextual-form

CorrectSerie A · 19 Oct 2026

Parma 0–3 Torino

Pick

Torino

Pick %

43%

Confidence

69%

Model

navurex-v1.0-contextual-form

IncorrectSerie C - Girone A · 19 Oct 2026

Juventus U23 2–1 Renate

Pick

Renate

Pick %

46%

Confidence

63%

Model

navurex-v1.0-contextual-form

CorrectSerie C - Girone A · 19 Oct 2026

Cittadella 1–0 Arzignano Valchiampo

Pick

Cittadella

Pick %

55%

Confidence

70%

Model

navurex-v1.0-contextual-form

IncorrectSerie C - Girone C · 19 Oct 2026

Giugliano 3–1 Foggia

Pick

Foggia

Pick %

45%

Confidence

73%

Model

navurex-v1.0-contextual-form

IncorrectSerie C - Girone A · 19 Oct 2026

Giana Erminio 2–1 Union Brescia

Pick

Union Brescia

Pick %

71%

Confidence

88%

Model

navurex-v1.0-contextual-form

CorrectSerie C - Girone A · 19 Oct 2026

Dolomiti Bellunesi 0–3 Trento

Pick

Trento

Pick %

61%

Confidence

86%

Model

navurex-v1.0-contextual-form

Methodology note

Performance is only meaningful when history cannot be rewritten.

Navurex counts only prediction rows that existed before kickoff and were later settled against the final result. Old matches with no saved pre-match snapshot are excluded instead of being reconstructed after the result is known.