On 17 July 2026, before the Italian season started, we froze the rules of a forward test and published them with a SHA-256 hash and an OpenTimestamps proof: back the draw, flat one unit, on every Serie A and Serie B match where Pinnacle's opening draw price is 3.50 or shorter. No other filter, no staking plan, no early stopping. Primary metric: closing line value against Pinnacle's own close. Duration: the whole 2026-27 season.
The reason it exists is a six-season backtest in which blind draws in Serie B returned +7.2% at opening prices (95% interval +0.5% to +14.3%, positive in six of six seasons) and the combined "Italy, draw at 3.50 or shorter" band returned +13.6%. Backtests like that are exactly what we do not trust, which is why the test is forward and the rules were fixed first.
Five weeks in, here is what the ledger says, and two things it does not say on its own.
| Value | |
|---|---|
| Selections logged (before kick-off) | 51 |
| Settled | 38 |
| Draws | 12 (31.6%) |
| Average opening price | 3.24 (implied 30.9%) |
| Break-even strike rate at that price | 30.9% |
| Profit, flat 1 unit | +1.03 units |
| Flat-stake ROI | +2.7%, 95% interval −41.6% to +53.7% |
| Average closing line value | −0.14%, 95% interval −1.84% to +1.47% |
| Closing line beaten | 18 of 38 |
By league: Serie A 15 selections, 5 draws, +1.69 units, CLV +0.92%; Serie B 23 selections, 7 draws, −0.66 units, CLV −0.83%.
The strike rate is within a single draw of the market's implied rate, the profit is one unit, and the interval on ROI spans ninety-five percentage points. Day to day it looks like this: 5 September, zero draws from four, −4 units; 6 September, four from six, +7.07; 12 September, five from six, +9.96. That is what 38 bets look like, and nothing about it should update anyone's belief in either direction. The number that will settle the question is CLV, whose interval is already narrow at ±1.7 points and will be around ±0.5 by June; ROI will still be noise then.
Rule 6 of the specification says technical failures are documented, not repaired. Between 12 August and 1 September our odds feed ran out of provider credits. Matchweek 2 of both leagues (28 to 31 August, twenty matches) therefore has no opening prices in our database and no selections in the ledger. We could reconstruct those prices from other sources and add the selections now; we know the results, so we will not. Roughly a third of the sample we should have by now is missing, and it stays missing. The ledger carries the note.
Until yesterday the ledger showed an average closing line value of +30%. That is not a good number; it is an impossible one. A draw that opens at 3.37 does not close at 1.20 before kick-off — but at 0–0 in the 80th minute it does, and that is where the price came from.
Our odds collector stores whatever the provider returns, and the provider returns in-play prices for matches that have started. The settlement code took "the latest Pinnacle snapshot" as the closing price without stopping at kick-off. Seventeen of the 38 settled selections had their "close" taken during the match. With a kick-off cutoff and a recomputation, the average CLV went from +30.4% to −0.14%. Results and profit were never affected; only the closing prices and the CLV column were wrong. The same contamination touched our bookmaker margin index, where about 12% of bookmaker-match pairs used an in-play price as "the last pre-match price"; it has been rebuilt on pre-kick-off snapshots only.
Two things are worth stating plainly. First, pre-registration prevents you from choosing what to count; it does nothing to stop you counting it wrong. The defence against that is publishing bet-level data so that other people can find what you missed, and treating any number that looks too good as a bug until proven otherwise — which is how this one was found. Second, a +30% CLV on a public ledger would have been the most quoted figure on this site within a month. We are publishing the correction with the same prominence we would have given the claim.
The success criterion was declared in advance: season-end average CLV positive with a 95% interval that excludes zero, measured against Pinnacle's close on the same market. At the current rate of twelve to fourteen selections per matchweek there will be somewhere around four hundred by June. Until then the ledger updates every six hours, every selection is logged before kick-off, and the daily snapshot of the whole record is timestamped externally. If the draw edge from the backtest is real, it will show up in CLV long before it shows up in profit. If it is not, we will say so, in the same place.
Method notes: universe and rules as in TIPS-EXP-001. Closing price = last Pinnacle draw price captured at or before the scheduled kick-off; CLV = opening price ÷ closing price − 1. Intervals are 20,000-sample bootstraps over the settled selections. All strategy pages: Strategy Board. Background: closing line value, variance and sample size.