TOUCHLINE TIPS

Season 2026/27 · tl-poisson-v1.1

How accurate is the model?

Every locked prediction stores the model's probability for all 15 priced markets, not just the Super Tip. Once a final score is verified, each of those markets is settled too. That judges the model itself, separately from the rule that picks one tip per match. Only predictions locked before kickoff are counted.

Live evaluation

10 of 10 locked matches settled

10/10Locked matches with a verified score
150Market outcomes settled (15 per match, pushes excluded)
+1.2%Brier skill, all markets, vs base rates; backtest +4.4%
1.045Match-result log loss (lower is better); backtest 1.009, base rates 1.084
3/6Super Tips won, 50% against 65% expected

Calibration, every market

When the model says 70%, does it happen about 70% of the time?

0%10%20%30%40%50%60%70%80%90%100%10–20%: landed 33% of 3 (plausible 6–79%); model expected 16%10–20%n = 333%20–30%: landed 26% of 27 (plausible 13–45%); model expected 25%20–30%n = 2726%30–40%: landed 40% of 25 (plausible 23–59%); model expected 33%30–40%n = 2540%40–50%: landed 63% of 32 (plausible 45–77%); model expected 46%40–50%n = 3263%50–60%: landed 38% of 29 (plausible 23–56%); model expected 54%50–60%n = 2938%60–70%: landed 50% of 12 (plausible 25–75%); model expected 65%60–70%n = 1250%70–80%: landed 75% of 20 (plausible 53–89%); model expected 74%70–80%n = 2075%80–90%: landed 50% of 2 (plausible 9–91%); model expected 83%80–90%n = 250%
150 settled market outcomes, grouped by stated probability. Dot: what actually happened. Band: plausible range for that sample size. Dark tick: what the model expected.

Calibration, Super Tips

6 settled tips; 3 of 4 passes would have won.

40%50%60%70%80%90%100%50–60%: landed 0% of 2 (plausible 0–66%); model expected 57%50–60%n = 20%60–70%: landed 50% of 2 (plausible 9–91%); model expected 66%60–70%n = 250%70–80%: landed 100% of 2 (plausible 34–100%); model expected 73%70–80%n = 2100%
Live Super Tips grouped by stated probability. Dot: what actually happened. Band: plausible range for that sample size. Dark tick: what the model expected.

Small sample. 10 matches is not enough to judge the model. Markets on the same match move together (a 3–0 home win settles most of them at once), so the 150 outcomes carry much less independent evidence than the count suggests. Read the plausible ranges, not the dots, and expect Brier skill to swing by several points from week to week.

By market

Skill compares the model's Brier score with always predicting the market's long-run base rate; positive is better. The last column is the same measure on 840 out-of-sample backtest matches.

MarketOutcomesModel saidHappenedPlausible rangeBrier (model)Brier (base rate)SkillBacktest skill
Home to win1046%50%24%–76%0.2690.255−5.7%+11.5%
Draw1025%20%6%–51%0.1650.163−1.6%+0.9%
Away to win1029%30%11%–60%0.2000.210+5.1%+10.9%
Home win or draw1071%70%40%–89%0.2000.210+5.1%+10.9%
Away win or draw1054%50%24%–76%0.2690.255−5.7%+11.5%
Over 2.5 total goals1052%20%6%–51%0.2800.304+8.1%+0.3%
Under 2.5 total goals1048%80%49%–94%0.2800.304+8.1%+0.3%
Under 3.5 total goals1070%80%49%–94%0.1870.182−2.7%−0.7%
Over 1.5 total goals1076%70%40%–89%0.2210.224+1.6%+0.3%
Both teams to score1054%30%11%–60%0.2910.286−1.5%0.0%
At least one team fails to score1046%70%40%–89%0.2910.286−1.5%0.0%
Home over 1.5 goals1046%60%31%–83%0.2610.252−3.7%+5.9%
Away over 1.5 goals1033%10%2%–40%0.1590.177+10.0%+4.1%
Home not to score1022%30%11%–60%0.2150.216+0.7%+6.2%
Away not to score1031%40%17%–69%0.2570.261+1.7%+4.2%

By market family

FamilyOutcomesModel saidHappenedPlausible rangeBrier (model)Brier (base rate)Skill
Match result3033%33%19%–51%0.2110.209−1.0%
Double chance2062%60%39%–78%0.2350.233−0.8%
Total goals4061%63%47%–76%0.2420.254+4.7%
Both teams to score2050%50%30%–70%0.2910.286−1.5%
Team goals4033%35%22%–50%0.2230.227+1.6%

Match result as one forecast

Matches10
Log loss, model (backtest 1.009)1.045
Log loss, base rates (backtest 1.084)1.041
Ranked probability score, model0.235
Ranked probability score, base rates0.233
Model's most likely result happened6/10 (60%)

Super Tips by confidence

30%40%50%60%70%80%90%100%Strong: landed 100% of 3 (plausible 44–100%); model expected 72%Strongn = 3100%Lean: landed 0% of 3 (plausible 0–56%); model expected 59%Leann = 30%
Live Super Tips by confidence tier. Dot: what actually happened. Band: plausible range for that sample size. Dark tick: what the model expected.

Round by round

RoundMatches settledMarket outcomesBrier (model)SkillSuper Tips won
Matchweek 9101500.236+1.2%3/6Scorecard

Backtest reference

840 out-of-sample matches, tl-poisson-v1.1

Calibration, every market

What good calibration looked like before the live record began.

0%10%20%30%40%50%60%70%80%90%100%0–10%: landed 8% of 88 (plausible 4–16%); model expected 8%0–10%n = 888%10–20%: landed 15% of 784 (plausible 12–17%); model expected 16%10–20%n = 78415%20–30%: landed 26% of 2046 (plausible 25–28%); model expected 25%20–30%n = 204626%30–40%: landed 37% of 1882 (plausible 34–39%); model expected 35%30–40%n = 188237%40–50%: landed 43% of 2320 (plausible 41–45%); model expected 45%40–50%n = 232043%50–60%: landed 58% of 2205 (plausible 56–60%); model expected 55%50–60%n = 220558%60–70%: landed 64% of 1638 (plausible 61–66%); model expected 65%60–70%n = 163864%70–80%: landed 76% of 1055 (plausible 73–78%); model expected 75%70–80%n = 105576%80–90%: landed 83% of 557 (plausible 80–86%); model expected 83%80–90%n = 55783%90–100%: landed 92% of 25 (plausible 75–98%); model expected 92%90–100%n = 2592%
Backtest market outcomes, grouped by stated probability. Dot: what actually happened. Band: plausible range for that sample size. Dark tick: what the model expected.

Expected skill by market

MarketModel saidHappenedSkill
Home to win43%41%+11.5%
Draw23%26%+0.9%
Away to win34%32%+10.9%
Home win or draw66%68%+10.9%
Away win or draw57%59%+11.5%
Over 2.5 total goals56%55%+0.3%
Under 2.5 total goals44%45%+0.3%
Under 3.5 total goals66%69%−0.7%
Over 1.5 total goals79%80%+0.3%
Both teams to score56%56%0.0%
At least one team fails to score44%44%0.0%
Home over 1.5 goals46%47%+5.9%
Away over 1.5 goals39%38%+4.1%
Home not to score22%24%+6.2%
Away not to score27%26%+4.2%

Brier score: mean squared error of a probability (0 is perfect). Skill: 1 − model Brier ÷ base-rate Brier. Log loss and ranked probability score treat home/draw/away as one forecast. Plausible ranges are 95% Wilson intervals. The model is frozen; these numbers are for reading, never for tuning it (see Method).