How we price a match — and why we publish everything.

1 · The fair price

For each match the headline probabilities come from the sharpest source available: the sharp market price with the bookmaker margin stripped out. Decades of research — and our own testing on tens of thousands of matches — show nothing public beats it. Alongside it we run our own calibrated model (Elo team strength plus recent form, home and away) as a second opinion, and as the only source for leagues without sharp coverage. From the fair probabilities comes a fair price — odds with no margin and no bias. When no single outcome reaches 55% probability, our pick covers the two most likely outcomes instead (a double chance, exactly as bookmakers offer it) — calling a single winner in a balanced match is guessing, and we do not guess.

2 · The benchmark: the sharp market

We benchmark against the sharpest price in the market (Pinnacle-style closing odds). Sharp books are, collectively, extremely hard to beat — their closing line is the market’s best estimate of the truth. We treat it with respect, not contempt. If our model disagrees with it wildly, the honest assumption is that the market is right and we are wrong — so we discard those outliers rather than dress them up as “huge value”.

3 · Closing line value — the strongest leading indicator of an edge

A pick is not validated by whether it won. One match is a coin toss. It is validated by closing line value (CLV): did the odds we took beat the closing price? Consistently beating the close is the single behaviour that research links to a genuine long-term edge. That is why CLV — not a flashy win rate — is the headline of our track record.

4 · Value, and the discipline to pass

We flag value only when a soft bookmaker prices a selection above the sharp fair price, within a plausible band (the exact thresholds are in section 8). A claimed edge of 30%, 50%, 80% is not a goldmine: it is bad data, and we filter it out. We also never flag long shots (fair probability under 30%): we measured them across ~32,000 matches and they are systematically the worst value on the board. The result: on most matches we tell you there is no value and to pass. Discipline is the product. Value exists only when the market is wrong, and that is rare.

5 · What we do not do

No guaranteed profit. No “sure things”. No VIP slips. No deleted losses. We do not promise you will win — we promise you will see the truth: the fair price, where value exists, and a full record of how our calls landed, with the value-flag history before 26 September 2026 labelled as a reconstruction.

6 · Play responsibly

Betting carries real financial risk and should never be chased. Stake only what you can afford to lose, treat any edge as small and slow, and walk away when it stops being disciplined. 18+. If gambling stops being under your control, seek help.

7 · The formulas

• Bookmaker margin (overround) = 1/odds(1) + 1/odds(X) + 1/odds(2) − 1, from each bookmaker's latest pre-kick-off prices.
• Fair probability (power devig): p = (1/odds)^k for each outcome, with k chosen so that the three probabilities sum to exactly 1. Fair odds = 1 / p.
• Edge (value gap) = fair probability × soft-bookmaker odds − 1.
• Closing line value (CLV) = odds taken / closing odds − 1. Value flags, Value Alerts and AutoBet are graded against the devigged Pinnacle close; predictions compare Pinnacle's own price at logging with Pinnacle's close; tipster and site picks are compared with the devigged Pinnacle close on the same selection (since 9 October 2026; before that, the raw closing price).
• Realizable CLV = the same, using the first price the same bookmaker quoted after the flag (within 60 minutes, before kick-off).
• Break-even accuracy = 1 / average odds of the picks.

8 · Value rules (v2, since 30 September 2026)

• Only soft bookmakers count as the price to beat; exchanges (Betfair Exchange, Matchbook) charge commission and are excluded.
• Both the soft price and the Pinnacle price must have been seen within the last 90 minutes. A gap seen only on older prices is shown as “not re-checked”, not as value.
• The first gap of +0.5% to +12% on each match, on an outcome with a fair probability of at least 30%, is recorded as its value flag and counted in the track record. On the match tables gaps are labelled by size: under 1.5% = Micro-edge (monitor only), 1.5–3% = Value flag, 3% or more = Strong value.
• Value Alerts record every gap of at least 2% and publish an alert at a gap of at least 5% with odds of at most 3.5.
• Flags recorded under the earlier rule keep their label (flag-v1); nothing is reclassified retroactively.

9 · What counts in the record

• Every prediction is logged before kick-off; its log time is exported with it, and predictions are snapshotted daily to a public Git repository and timestamped with OpenTimestamps.
• Value flags shown on the site have been recorded live since 26 September 2026. The value-flag history before that date is a backtest reconstructed with the same rule, and is labelled as such everywhere.
• A return (ROI) is published only from 100 settled results, and a 95% confidence interval only from 30.
• World Cup and UEFA Nations League matches (neutral venues, national teams) are shown on the site but excluded from the audited figures.
• Display-only competitions show Pinnacle's price with the margin removed: they carry no model, no value flags and are not part of the track record.
• Match results (1X2) and goal markets are settled on the score after 90 minutes, because the prices are for 90 minutes: extra time and penalties do not count. Pre-registered experiments follow the rules published in their own specification.
• Most likely wins lists the wins that Pinnacle's current price, margin removed, gives at least 65%; its record is kept separately from the track record.

10 · How tipsters and prediction sites are audited

• We read only public picks, posted before kick-off, from the tipster's channel or the site's own page, once a day or as they are posted.
• Each pick is matched to a real fixture, settled on the official result and, where we hold Pinnacle's close on the same market, graded by CLV.
• Picks on competitions we do not price are counted separately as out of scope, never as losses.
• A tipster channel is scored from 15 settled picks; a verdict needs 30 picks with a measurable closing price. A prediction site gets a verdict from 30 settled predictions. Every figure is shown with its denominator: posts read, picks matched, picks settled, picks with a measurable close.
• We never edit or backfill a pick after the result is known.

11 · Changes to this method

• 18 July 2026: margin removal moved from proportional to power devig after a 35,454-match calibration study.
• 26 September 2026: the value-flag record was rebuilt from the published rule and relabelled as a backtest; flags have been recorded live since.
• 30 September 2026: rule v2 (no exchanges, prices of at most 90 minutes) and realizable CLV published next to CLV at signal.

The corrections we announced, dated →

Methodology v2 (18 July 2026): margin removal upgraded from proportional to power devig, after a 35,454-match calibration study showed it 2–3× more accurate at the extremes (bookmakers load more margin onto longshots — proportional devig overestimated them by ~1.1pp and underestimated strong favourites by ~2.1pp). Predictions logged before this date keep the method they were flagged under; nothing was reclassified retroactively.