For three months — 8 June to 3 September 2026 — an automated pipeline read every public post on 47 football tipster channels on Telegram, extracted the picks, matched them to real fixtures, settled them on the official result, and, where possible, graded each one against Pinnacle's closing price. No pick was edited or removed after kick-off; every profile shows its own denominators.
We expected the interesting number to be who beat the closing line. It turned out to be something upstream of that.
| Stage | Picks | Share of everything read |
|---|---|---|
| Public posts parsed into a single-match football pick | 1,106 | 100% |
| Carried a usable price in the message | 683 | 61.8% |
| Matched to a fixture and settled on the result | 475 | 42.9% |
| Had a price and a Pinnacle closing price on the same market — i.e. a measurable closing line value | 129 | 11.7% |
Of the 631 picks that could not be settled, 233 were on other sports or on leagues we do not cover (not failures, just out of scope), 369 could not be tied to a fixture at all — vague team names, no date, "tonight's banker" — and 29 are still pending.
That last row is the point of this article. A tipster's advertised record is whatever survived their filter. Ours publishes the whole funnel, because the size of the measurable part decides how much any number below it can mean.
Across the 129 picks with a measurable closing line value, the average CLV is +3.85% (median +4.45%). Read that with the denominator in view: 129 picks spread over five channels, dominated by whoever posts prices at all.
Per channel, for everyone with at least 20 settled picks:
| Channel | Settled | Win rate | Flat-stake ROI | Avg CLV (n measurable) |
|---|---|---|---|---|
| AnalysisTips | 124 | 70% | −0.3% | +2.5% (12) |
| شرطبندی · پیشبینی · جام جهانی | 114 | 62% | −7.2% | −8.0% (36) |
| Puskás Tippek | 75 | 64% | +10.4% | +16.4% (13) |
| انالیز تیپ | 59 | 80% | −1.4% | −12.7% (1) |
| VBH · Value Bets & Sports Betting | 25 | 40% | −30.0% | +25.7% (5) |
Three things stand out, and none of them is "this tipster is good".
The largest measurable sample is negative. The one channel with a meaningful CLV sample (36 picks) sits at −8.0%: the market moved against its picks after they were posted, which is the signature of picks with no information in them. The channels that look impressive — +16%, +26% — carry 13 and 5 measurable picks respectively, which is a good fortnight, not evidence.
Win rate keeps telling you nothing. A 70% win rate produced a −0.3% return; an 80% win rate produced −1.4%. At the short odds those picks carry, the break-even win rate is higher than the win rate achieved. The arithmetic is in why a 67% win rate can still lose you money.
Nobody clears the bar for a verdict. Our pre-registered standard for calling a tipster skilled is a positive average CLV whose 95% confidence interval excludes zero, which needs on the order of a hundred measurable picks. After three months and 1,106 picks, no channel has reached it — not because they are bad, but because they publish prices too rarely for anyone to check.
Most tipster "verification" services record the picks a tipster chooses to submit, at the price the tipster states, and report a win rate. Every step of that is under the tipster's control.
We do the opposite: we read what was posted publicly, we settle on official results, we price against the sharpest market's close on the same market, and we publish how much of the record could not be measured. When a profile on this site says "CLV +2.5% on 12 picks", the "on 12" is the important part.
If you follow a tipster, the two questions that matter are not "what is your win rate" but "what is your average odds" and "how many of your picks can be checked against the closing line". Anyone who cannot answer the second is asking to be trusted, not verified.
Every audited profile is public at /tipster-audit, with the full funnel, the closing-line test and a luck test on each. If your tipster is not there, Audit My Tipster runs the same arithmetic on any record you paste in.
Method notes: picks were extracted by a language model from public Telegram posts and deduplicated; only single-match football picks were kept. Matching to fixtures uses team-name normalisation and a ±7-day window around the post; picks whose fixture could not be identified are marked unverifiable, and picks on uncovered sports or leagues out of scope, so that neither is counted as a loss. CLV = advertised price ÷ Pinnacle closing price − 1, same market. Nothing is edited after kick-off; the daily ledger is timestamped on GitHub and OpenTimestamps.