Alternative goal lines
Sevilla–Osasuna ended 2–0: selected Over 1.5 won, rejected Over 2.5 lost. Test adjacent totals prospectively rather than selecting the winner in hindsight.
01 · 26 October CORE Research · Reviewed 27 October
Five matches, five favourable outcomes
5/5Distinct research markets
0.0376Worksheet Brier Score
0.2092Worksheet Log Loss
| Match | Market | Frozen p* | VS | Ref. odds | Final | Outcome |
|---|---|---|---|---|---|---|
| Santa Clara–Benfica | Under 4.5 | 89.6% | 34 | 1.16 | 1–2 | WIN |
| Copenhagen–Odense | Under 4.5 | 82.8% | 41 | 1.26 | 1–1 | WIN |
| Beşiktaş–Başakşehir | Under 4.5 | 82.0% | 56 | 1.36 | 2–0 | WIN |
| Sevilla–Osasuna | Over 1.5 | 76.0% | 34 | 1.43 | 2–0 | WIN |
| PSG–Lyon | Under 4.5 | 76.0% | 34 | 1.42 | 2–0 | WIN |
* Archived diagnostic research probabilities, not independently certified AIONORA win probabilities. Market prices are references, not verified Greek executable offers. Zero canonical official BET settlements claimed.
Research Doubler and illustrative return (not realised ROI)
Santa Clara Under 3.5 (1.42) × Sevilla–Osasuna Over 1.5 (1.43), combined reference 2.0306, hit 2/2. The Sevilla event repeats the Top 5 result, so these are not independent new fixtures. The archived stress-EV test was negative. Five 1-unit reference selections produce a hypothetical +1.63 units / +32.6%, not actual profit. No trustworthy CLV available.
02 · Errors & False Negatives
What the results can teach us
Rejected favourites
Benfica, Beşiktaş and PSG won. Copenhagen did not. Outcome alone cannot determine whether the original NO-BET choice had negative value at the price then offered.
Forecast calibration
Five favourable outcomes do not establish a reliable probability model. Preserve each original forecast version, compare to simple baselines and test out of sample.
Doubler robustness
Both archived legs hit, but that does not invalidate a negative conservative EV decision. Track HOLD as well as selected combinations.
03 · Learning & Validation
Next measurable improvement
Freeze event identity, exact market/line, model version, prematch probability, reference/executable price, time, and alternative NO-BET passports before each event starts. Reconcile independent final scores and evaluate calibration after an adequate chronological sample (research target: 200 independent events).
No retrospective rewriting of forecasts, no unsupported change to champion model weights, no verified CLV or official ROI. The Full Audit worksheet has not been authenticated as canonical Blob history sync.