A track record is a claim about the past with choices hidden inside it: who recorded it, when, at what price, which picks were counted, over what period. Here is the order in which to check them, for a service's own page, a screenshot, or a profile on this site. Each item is a question with a checkable answer.
1. Who timestamped it
A record kept by the service can be edited; a record kept by a third party that saved the pick before the event started cannot be edited after the result. On PickAuditor the timestamp is ours — the moment the public page was captured — and a pick captured after kickoff is void, not a win or a loss.
2. Sample size, period, p-value
n first, always. Below about 30 picks any hit rate is consistent with no edge; below 200, differences of a few points between services are mostly noise. The p-value on a profile is the formal version of that check: how often a record at least this far from break-even would appear by chance — around 0.40 for 60% over 30 picks, around 0.008 for 60% over 300. A 30-day and a 365-day record of the same service can differ in sign; prefer the longest period with a meaningful n.
3. Public or premium
Was every pick in the record visible to a reader who had not paid? A record of premium picks cannot be checked by anyone outside; a record that mixes the two may be carried by whichever set did better. PickAuditor records public-tier picks only and says so on every page that shows a record.
4. Which odds, from where, when
A record without odds cannot produce a return. Ask which price each pick was graded at — the service's own quote or a market price, and from which bookmaker — and when it was taken. On PickAuditor each pick shows whether its odds were the service's quote or the market at capture, with the closing price beside it.
5. Hit rate against break-even
Divide one by the average odds. That is the hit rate needed to break even, and the record's hit rate only means something relative to it: a 70% record at 1.30 loses, a 40% record at 2.80 profits. The arithmetic is in why win rate lies.
6. Flat stake or “units”
ROI at one unit per pick can be checked from the picks alone. A profit in units under a staking plan — 3 units on confident picks, 1 on the rest — cannot, unless every stake was published in advance and the plan never changed. PickAuditor grades everything at one flat unit.
7. Pushes and voids
A push — a total landing on the line, a tie in a market with no draw option — returns the stake. It belongs outside the hit rate and counts as zero in ROI. A void pick (event postponed or cancelled, pick published late) counts as nothing. A record that voids losses more readily than wins is not a record.
8. Drawdown and streaks
Two numbers describe what following a record would have felt like. Max drawdown is the largest fall from a running peak in cumulative units; the longest losing streak is the longest run of consecutive losses. Both are ordinary features of profitable records, and their size is mostly a function of the odds and the number of picks.
- Picks over the period
- 300
- Hit rate at average odds 1.91
- 55.0% (break-even 52.4%)
- Final result
- +15.1 units · +5.0% ROI
- Running total at its peak (pick 141)
- +21.3 units
- Lowest point after that peak (pick 188)
- +9.7 units
- Max drawdown
- 11.6 units
- Longest losing streak
- 7
A follower watched more than half of the peak profit disappear over 47 picks and lost seven in a row once. At a 55% hit rate over 300 picks a run of six or seven losses is expected somewhere; its presence is not evidence of anything. A drawdown far larger than the final profit, on the other hand, means the result depends on when you started.
9. Claims against the record
Put the service's headline figure next to the measured one, with the scope of each. “72% NBA accuracy” is a claim about a sport, a market and a period; the comparable figure is the NBA match-winner hit rate over the same period with its n, not the overall rate. A gap can be an overstatement, a different scope, or an unverifiable claim about premium picks. The claim check table on each profile makes this comparison with the scope written out.
10. Right of reply
A record about a service should be open to the service. Check whether disputes exist, whether they were answered, and whether accepted ones changed the record. On PickAuditor every dispute is public with its resolution, and accepted ones are dated in the changelog.
Reading one line
58.4% hit · +4.1% ROI · n=147 · 90d · CLV +1.2% (n=131) · Brier — · max drawdown 9.5u · longest losing streak 6 · p=0.15
147 graded public picks in 90 days: enough to display, not enough to be confident. Positive return and positive CLV agree in sign, which is the combination worth noticing. No Brier score, so the service does not state probabilities. A drawdown of 9.5 units against a profit of about six means the follower was down more at one point than they are now up. A p-value of 0.15 says a record like this appears by chance roughly one time in seven — so the next thing to do is open the 365-day row.
Definitions for every figure are in the methodology; the leaderboard shows them side by side with n on every row.