⚽ Betting Punter

Calibration — predicted vs actual

Confidence bucket	n	Predicted	Actual
50%-60%	64	55.3%	62.5%
60%-70%	104	64.9%	64.4%
70%-80%	157	75.5%	72.6%
80%-90%	2351	87.1%	88.6%
90%-100%	1584	91.8%	91.6%

When "actual" tracks "predicted", the probabilities are honest — that's what the nightly recalibration optimizes for.

By market

Market	n	Hit rate	Brier
1X2	97	52.6%	0.2304
AH	273	61.5%	0.2000
BTTS	3	33.3%	0.3131
CS	3	0.0%	0.0092
DC	10	90.0%	0.0979
DNB	207	78.3%	0.1580
OU	3798	89.8%	0.0909

Ticket ROI (flat ₦100 stakes)

Tier	Tickets	Wins	ROI
BETBUILDER	8	3	+641.4%
JACKPOT · fewer games	2	0	-100.0%
SAME_GAME_1	9	3	+500.9%
VALUE	2	0	-100.0%
WEEKLY_KELLY	1	0	-100.0%

Day by day

Date	Picks settled	Hit rate	Expected
2026-08-11	7	100.0%	88.3%
2026-08-10	78	82.1%	82.9%
2026-08-09	645	82.5%	84.3%
2026-08-08	717	87.7%	86.1%
2026-08-07	168	90.5%	86.2%
2026-08-06	93	95.7%	89.2%
2026-08-05	62	96.8%	89.6%
2026-08-04	78	82.1%	88.7%
2026-08-03	76	84.2%	87.6%
2026-08-02	362	85.4%	84.8%
2026-08-01	339	86.1%	87.1%
2026-07-31	131	90.8%	83.1%
2026-07-30	97	88.7%	89.9%
2026-07-29	83	88.0%	86.9%

Hit rate above "expected" = the model is beating its own probabilities; consistently below = it's overconfident (the nightly recalibration corrects this).

Correct-score experiment — modal vs value (same games)

Arm	Legs	Settled	Hits	Expected hits	Avg odds	P&L (flat 1u)	4-folds won
MODAL — most likely score	28	25	3	2.9	7.0	-4.65	0/7
VALUE — best price vs model	28	25	0	1.9	12.7	-25.00	0/7

2026-08

Arm	Settled	Hits	P&L (flat 1u)
Modal	6	1	+2.25
Value	6	0	-6.00

2026-07

Arm	Settled	Hits	P&L (flat 1u)
Modal	15	2	-2.90
Value	15	0	-15.00

2026-06

Arm	Settled	Hits	P&L (flat 1u)
Modal	4	0	-4.00
Value	4	0	-4.00

Both arms bet the SAME four games each week: modal takes the most likely scoreline at the offered price (the tipster strategy), value takes the scoreline the model prices as most underpaid. Compare P&L, not hit count — modal will hit more by design.

What the model has learned (self-training state)

Market	a	b	Samples	Brier before → after	Fitted
1X2	1.069	0.006	1038	0.1863 → 0.1863	2026-07-02
AH	1.002	-0.022	2443	0.1825 → 0.1825	2026-07-17
DC	1.140	-0.101	976	0.1828 → 0.1826	2026-07-02
DNB	1.126	-0.114	248	0.1475 → 0.1473	2026-08-07
OU	1.214	-0.505	3690	0.0939 → 0.0938	2026-08-06

All markets — calibration progress

Market	Settled picks (90d)	Status
1X2	1158	✓ calibrated
OU	4988	✓ calibrated
DC	993	✓ calibrated
BTTS	30	calibrating — 30/80
DNB	411	✓ calibrated
AH	2690	✓ calibrated
OE	38	calibrating — 38/80
CS	69	calibrating — 69/80

A market only gets a calibrator once it has ≥80 settled picks and the fit improves the Brier score. Until then it runs on the raw model+market blend (identity calibration) — accurate, just not yet self-tuned.

League reliability weights

Division	Weight	Samples
ARG2	1.11	112
DNK	1.08	51
AUT	1.05	37
B1	1.05	40
NOR	1.04	156
MEX	1.04	26
ROU	1.03	116
P1	1.03	45
N1	1.02	36
POL	1.02	84
CHN	1.02	193
RUS	1.02	96

>1.00 = picks in this league outperform their probabilities (ranked up); <1.00 = underperform (ranked down). Updated nightly.

Market reliability

Market	Weight	Samples
OU	1.00	4971
DNB	0.97	409
AH	0.97	2686
OE	0.96	38
1X2	0.95	1157
DC	0.93	991
CS	0.92	21
BTTS	0.92	30

How much each market is trusted in selection from its realized results: >1.00 outperforming (more share), <1.00 trailing its predictions (less share until it earns it back). Shrinks the calibrated probability used for picks. Updated nightly.

League model training

League	Matches trained	Last updated
ARG2	1641	2026-08-10
AUT	767	2026-08-10
B1	939	2026-08-10
BRA	1526	2026-08-10
BRA2	924	2026-08-10
CHN	1097	2026-08-07
D2	920	2026-08-10
DNK	761	2026-08-10
E0	1140	2026-08-08
E1	1656	2026-08-10
E2	1656	2026-08-10
E3	1656	2026-08-10