Football prediction methodology: picks, prices and results
Forecast Record methodology 2.2 has applied since . Its published rules explain who predicted what, when the pick was recorded, which price applies and how the result is settled.
Summary
This page answers five questions about every number Forecast Record publishes: what was predicted, who made the prediction, when we captured it, which price was used and what could change the result. Method version 2.2 has been in force since , and the rules below describe what the code actually does (source: our own methodology record).
Check one record
Pendikspor v Istanbulspor — Pendikspor (1X2, match result) by İddaaTahmin, from the live record.
- 1 · The prediction and original source
- Pendikspor (1X2, match result), in Pendikspor v Istanbulspor, Türkiye 1 Lig, kick-off . The author’s words: “Pendikspor - İstanbulspor
1
1.72” (Stated).
İddaaTahmin. The source does not name a separate tipster, so the selection is attributed to the byline or channel owner.
- 2 · Recorded before kick-off
- The page gives its publication time as (Stated). We captured it at (Measured), less than an hour before kick-off, so it belongs to the live record.
- 3 · Prices kept separately
- Our price feed gave the same selection a median bookmakers’ price of 1.83 (Measured), from a quote taken at , within a minute of our capture; the headline return of the live record uses it. The author stated 1.72 (Stated); that price is a separate line on the same pick.
A closing price is not supplied for this example.
- 4 · Final score and settlement
- Lost: the final score was Pendikspor 0-2 Istanbulspor (Measured, regulation time). At 100 points per pick the snapshot series records −100 points (Measured). This changes only if the result provider restates the score, the reading of the pick or its attribution is corrected, or the method version changes. Each such change is logged with the old and the new value.
- 5 · Where the pick enters the record
- The settled loss belongs to the live record at our recorded market price. The author’s stated-price result remains a separate series; it is never added to the market-price result.
Five questions about every record
Where predictions come from
Public statements stay with their original source and quoted words.
Where predictions come from — explanation
We collect what tipsters and editorial sources publish on websites, feeds, videos and podcasts. Each statement is kept with its source and the author’s exact words. A prediction is a statement we can read as a pick on one match, market and selection; opinions and lines that are not about a match are not predictions. Sources that returned nothing are shown separately, not as zero; anything we did not use is listed with the reason.
Who gets attributed and when
A person and a publisher have distinct records; the timing evidence decides the origin.
Who gets attributed and when — explanation
Which price is used
Live comparisons use our recorded market price; the author’s price and closing price stay separate.
Which price is used — explanation
The headline return of the live record uses our own market price of the same selection: the median of bookmakers’ prices in our price feed at the moment we first recorded the pick, always before kick-off (the delay of that price after our record is under a minute at the median: the price is taken when we record the pick). The author’s stated price and the closing price are separate lines on the same picks. Records archived before kick-off is priced at the last pre-match price in archive (~8–15 h before kick-off), not at a closing price, because the market price at the moment of publication cannot be reconstructed. A stated price more than twice or less than half the market price of the same selection is marked “price disputed” and left out of returns at the author’s price until it is checked.
How outcomes and returns are calculated
Each eligible pick has the same points stake and is checked against its match result.
How outcomes and returns are calculated — explanation
A pick is settled from the regulation-time result under the published rule for its market. The recorded settlements name 3 result sources. Where the two commercial providers both cover a match and disagree, the pick waits. A prediction on a national-team match outside the match calendar stays in the record, unsettled.
Every qualifying pick is a level stake of 100 points. The return on stakes is the net result divided by the total staked, separately for each price series, with a 95% interval from a bootstrap over matches.
What stays separate and how to reply
Live and archived records stay separate, and every correction can be checked and answered.
What stays separate and how to reply — explanation
Tier is earned on the whole record: Established means 100+ predictions checked against the final score with at least 90 days of history. In the tipster comparison, a position goes only to a live record with at least 30 settled picks, not counting voided ones, on at least 15 matches, from a tipster with an Established record; records without a position are listed by sample size, not by return. Levels within a market or competition (No.1, Top 3, Top 5, Top 10) and titles start at Tracked. The live record and Records archived before kick-off are never compared or added together, returns at different prices are never mixed, and a joint byline is not treated as one person’s record.
Every number carries the version of the method that produced it; a new version recalculates the result and never creates a second bet. A correction changes a number in the open: the corrections log keeps the old and the new value with the date, and an author can reply. Links to bookmakers, where they appear, never change a number or the order of any table, and nothing in a comparison is for sale.
Time of recording and status of a number
Recorded before kick-off
Captured in our live record before the match starts.
Archived before kick-off
Supported by an independent archive copy taken before the match starts.
Found after kick-off
Kept separate from pre-match returns.
- Measured
- Observed in a timestamped record or checked against the final score. A return is calculated from the recorded inputs under a published rule.
- Stated
- Reported by the source; not independently confirmed. A publication date written on the page and the author’s quoted price are stated.
- Unknown
- Awaiting sufficient evidence. A missing closing price leaves closing-line value unknown.
Statement → Prediction → Settled pick
Definitions dictionary
Each term keeps its original address. Definitions are open so a link from any record lands on the explanation.
- Statement
A statement is a fragment of an author’s public text, video or audio that we found and recorded together with its source. The same fragment found twice is one statement. Source advertising and site boilerplate are not statements and stay out of the record.
Example. One sentence in a match preview that names a score and a price is one statement; a line asking readers to sign up for a newsletter is not.
- Prediction
A prediction is a statement we could read as a pick on a specific match, market and selection: a priced pick, a pick without a price or a score forecast. Opinions and statements that are not about a match are not predictions. The same pick repeated by the same author on the same match is one prediction.
Example. A preview sentence backing a 2-1 home win at a stated price is one prediction on the correct-score market; if the same author repeats that pick in a video, it is still one.
- Coverage of a record
How much of what a source published we found and could use, step by step: statements read, predictions recognised in them, qualifying picks and settled picks. The sources covered are listed on the page “What we collect”.
Example. A source with 200 statements read, 120 predictions, 40 qualifying picks and 30 settled picks has the coverage line 200 → 120 → 40 → 30.
- Attribution
Who is speaking in the recorded source: the channel owner (the author of the column, feed or show), a guest, an unsigned editorial line of the channel or someone being quoted.
- Origin of a record
Every pick carries the label of its strongest timing evidence: Submitted to us before kick-off; Recorded before kick-off; Archived before kick-off; Source-reported history; Found after kick-off. Submitted and live-captured picks belong to the live record. Archived before kick-off, Source-reported history and Found after kick-off remain separate from it. The headline number of a profile is the live record from 10 settled picks; until then it is the part archived before kick-off, reconstructed at archived prices, with the live record so far shown beside it as picks won out of picks settled. Tier is earned on the whole record: Established means 100+ predictions checked against the final score with at least 90 days of history. In the tipster comparison, a position goes only to a live record with at least 30 settled picks, not counting voided ones, on at least 15 matches, from a tipster with an Established record; records without a position are listed by sample size, not by return. Levels within a market or competition (No.1, Top 3, Top 5, Top 10) and titles start at Tracked.
Example. A pick we archived on Friday for a Saturday match is recorded before kick-off. A pick we read on Monday from an old article is Found after kick-off.
- Settlement and net result
A pick is settled from the result of the match under the rule published for its market; 1X2 and totals use regulation time. Won, Lost, Void, Half won and Half lost are the only outcomes. The net result is the sum of what each settled pick won or lost, in points.
Example. A pick of “over 2.5 goals” at 1.90 on a match that ends 2-1 wins 90 points; on a match that ends 1-0 it loses 100 points.
- Settled pick
A settled pick is a qualifying pick whose level stake has been settled against the final score: won, lost, half won, half lost or void. A pick waiting for its result is not settled. Returns are calculated on settled picks only, and voids are shown separately because they return the stake.
Example. If that match finishes 2-2, the 2-1 pick becomes a settled pick with the outcome lost.
- Match
A match is one football fixture. Several predictions and several settled picks can belong to one match, so we always say how many matches a number covers: uncertainty is estimated per match, not per pick.
Example. A result is described as 10 settled picks across 6 matches.
The price the author published next to the pick. A return at this price shows what the author claimed; it assumes that price was available.
Example. The author writes “Arsenal to win at 2.10”: 2.10 is the stated price.
- Independent odds snapshot
The median of bookmakers’ prices for the same selection in our price feed when we first recorded the pick, from a quote no older than 5 minutes. Betting exchanges, prediction markets and prices of 1.01 or lower are excluded. If there is no such price at that moment, we use the first price within 2 hours, provided it is before kick-off. Otherwise, the pick has no return at the market price, and no other price is substituted. A pick recorded before kick-off but checked only after kick-off settles the same way, at the price when we recorded it, and is marked “checked after kick-off”. Earlier figures stay in the corrections log. The snapshot records the market at that moment, not a promise that a reader could have got that price.
Example. We recorded a pick at GMT on Monday; the median bookmakers’ price for that selection in our feed at GMT, two minutes earlier, was 1.80, so this series settles the pick at 1.80.
- Closing-line value (CLV)
How a price compares with the closing price of the same selection: CLV = 100 × (price ÷ closing price − 1). For the author’s stated price, it is calculated wherever a closing price of the same selection exists and shown as a median under “Stated price vs market”. Where no closing price of the same selection exists, the CLV is unknown, not zero.
Example. A pick quoted at 2.10 that closes at 1.95 has a CLV of 100 × (2.10 ÷ 1.95 − 1) = +7.7%.
- Overlap of the two price series
The share of settled picks in one price series that were also settled in the other (the author’s stated price and our independent odds snapshot). The two returns are only directly comparable on that shared part; a low overlap means they describe different picks.
Example. If 4 picks are settled at the author’s price and none of them could be priced by our snapshot, the overlap is 0.0%.
- Qualifying pick
A qualifying pick is a prediction that met the rules of a Forecast Record series: the market is one we can settle, the pick was recorded before kick-off and a price was available, so a level stake was placed. It is counted per series: the same prediction can qualify at the author’s stated price and not at an independent odds snapshot.
Example. A correct-score pick with a stated price qualifies for the series at the author’s stated price even when no independent price for that score was available.
- Return on stakes
The net result of the settled picks divided by the total staked, at 100 points per pick. Void picks are left out of the total staked. Each series has its own return: at the author's stated price and at our independent odds snapshot.
Example. Ten settled picks at 100 points each are 1,000 points staked. If they return 1,080 points in total, the net result is +80 points and the return on stakes is +8.0%.
- Forecast Record reconstruction from the predicted score
A Forecast Record series created by our rules from the author’s score: a predicted score also settles as a match-result pick for the side the score favours. It is shown separately and never added to the author’s own picks.
Example. The author predicts USA 3-1 Peru and the match ends 4-1: the score prediction lost; the derived “USA to win” pick won.
- Uncertainty
A 95% interval around a return, from a clustered bootstrap over matches. With too few matches it is reported as not estimated rather than guessed.
Example. A return of +12.0% with an interval from −20.0% to +40.0% means the sample is compatible with a loss as well as a gain.
- Top-up
When a points bankroll cannot cover the next 100-point stake, 10,000 points are added. Top-ups are counted separately and never mixed into the result.
- Drawdown
The largest fall of the cumulative result from a previous high, in points, within one series.
Example. A result that goes from +300 to −200 before recovering has a drawdown of 500 points.
- Probability accuracy
How close the stated chances are to what happened: high chances on picks that win and low chances on picks that lose score well, and confident chances on picks that lose score badly. It needs our own probability for each pick, so it is shown only once those probabilities are published.
Example. A pick given a 60.0% chance that wins scores better than the same pick given 30.0%.
- Comparison threshold
The minimum sample a live record needs before it is marked as meeting the comparison criteria in a window. Tier is earned on the whole record: Established means 100+ predictions checked against the final score with at least 90 days of history. In the tipster comparison, a position goes only to a live record with at least 30 settled picks, not counting voided ones, on at least 15 matches, from a tipster with an Established record; records without a position are listed by sample size, not by return. Levels within a market or competition (No.1, Top 3, Top 5, Top 10) and titles start at Tracked.
- Current method
The current method applies the published rules; earlier versions and effective dates are kept in the method history below.
- Tipster
A tipster is an individual whose football predictions are recorded with their source and timing evidence.
- Source
A source is the website, publication or channel where a prediction appears. A source record and an individual tipster record stay separate.
- Recorded before kick-off
A prediction saved by Forecast Record from its source before the match started.
Tracking levels. Every tipster has one tracking level, by fixed thresholds on predictions checked against the final score, recorded before kick-off or archived before kick-off, in every market and at either price; this count can differ from the settled picks of the headline number:
- New: registered with a confirmed channel and awaiting admission; no page on the site.
- Probation: admitted; the record is shown on its own page from 10 predictions checked against the final score.
- Tracked: at least 30 predictions checked against the final score over at least 30 days, in at least 2 different weeks, with no rule breaches.
- Established: at least 100 predictions checked against the final score over at least 90 days, and everything required for Tracked.
Days are counted from the first to the last settled match. Tier is earned on the whole record: Established means 100+ predictions checked against the final score with at least 90 days of history. In the tipster comparison, a position goes only to a live record with at least 30 settled picks, not counting voided ones, on at least 15 matches, from a tipster with an Established record; records without a position are listed by sample size, not by return. Levels within a market or competition (No.1, Top 3, Top 5, Top 10) and titles start at Tracked.
Rules for checking a result
Each rule explains which records qualify, how a market is settled and what can change a result.
What a measurement here means · What counts as a pick and what is executed · What is executed and what is not · ROI · Confidence interval · The points bankroll · Settlement · Which markets we settle · Where the results come from · Disputed prices · National-team matches · Records archived before kick-off · Late-found picks, priced at publication time · How to check the original evidence · Comparisons and positions · Checking a published number · Seven days to reply · How we measure our own visibility
What counts as a pick
A statement with an explicit price and stake is executed in two series: at the price the author stated and at our own market snapshot. A statement that names a pick without a price is executed at our market snapshot only, and the return at the author's own price is reported as not stated. A correct-score forecast is measured for score accuracy, and a separate Forecast Record reconstruction from the predicted score settles the 1X2 outcome implied by that score. That derivation is a Forecast Record rule and is reported apart from the author's own market return. An opinion, a conditional or a joke is kept as an observation and never executed. Only statements about an event are counted. Advertising and site boilerplate captured from a source page (bonus offers, betting buttons, affiliate disclosures, newsletter prompts, score widgets, menus) are not the author's words and are not listed as statements. On a match page, the split of what tipsters and sources picked counts each tipster with 10 or more predictions checked against the final score to date once per market, with the pick made before kick-off; predictions from tipsters and sources with fewer are listed separately, without shares.
What a measurement here means
We record what a tipster published, price it under a published rule (for the live record, the market price in our own price feed at the moment we first recorded the pick, always before kick-off) and settle it against the final score. Every number on this site therefore has four things attached to it: the sample size, the window, the status of the data and the version of the method.
- Measured: directly observed in a source or timestamped capture;
- Stated: reported by the source; not independently confirmed;
- Calculated: derived from the listed inputs under a published rule;
- Unknown: not established from the evidence available. Unknown is never written as zero.
A status belongs to one field, not to a whole record: the time of our own capture is measured, while a publication date written on the page is stated.
Archived before kick-off
Part of some tipsters’ history is read later and kept separate from the live record. For a set of authors we re-read what they published before matches played between and — the Premier League first, then the Championship, League Cup, Champions League and Europa League — and settled it against the final score. Every pick carries the label of its strongest evidence of time, from one list used on every page. Submitted to us before kick-off: the author sent the prediction to us directly and we stamped the time when we received it; it belongs to the live record. Recorded before kick-off: we archived it ourselves before the match; this is the live record; only it appears in the live comparison table and the points bankroll, and only a live record can earn a position there. Archived before kick-off: an independent public web archive holds a copy of the page taken before kick-off, and the pick is in that copy. Source-reported history: the time and the pick are known only from the source's own history; it is used to check the source's own record and is never part of the evidenced total. Found after kick-off; date as stated by the source: found later and dated by the source's own publication date — the lowest of the levels of evidence. Evidence obtained after kick-off carries a second line, Added to our record on (date); the pre-match record of the match and of each author stays as it was.
Historical picks are shown as a reconstruction at the last pre-match price in archive (~8–15 h before kick-off), for picks archived before kick-off and picks Found after kick-off separately. The author's own quoted price and, where it exists, the aggregator closing price are separate lines on the same picks. Outcome accuracy is shown separately from return. From the minimum of 10 checked predictions on 2 dates a profile needs before it is indexed counts only the evidenced part of the record: picks Recorded before kick-off and picks archived before kick-off. Every published profile has predictions in the evidenced part and states on its first screen how many of its predictions come from each part.
Since Records archived before kick-off covers the Championship, League Cup, Champions League and Europa League as well, with prices back to August 2024. The last pre-match archive price is the median of each bookmaker's last price before kick-off, from at least two bookmakers, in a series of its own. Score forecasts are settled the way the live record settles them: on the exact-score market only when the author explicitly recommends the bet, and otherwise as the match result the score implies, in a series of its own, separate from the author's picks.
Late-found picks, priced at publication time
A pick published before kick-off and Found after kick-off has started stays outside the record kept before kick-off. Late-found picks form a separate series, priced at publication time, shown on the author's profile as its own line and marked on the pick itself.
A late-found pick enters the series only if the time of publication is evidenced: by the platform's own timestamp (a post or a video), by the publisher's feed date checked against our own records (not later than our first capture, not later than kick-off and not more than 30 days before it), by our own copy of the page taken before kick-off, or by a copy in an independent public web archive taken before kick-off that contains the quoted words. For a web page the date the page shows about itself is not used as evidence; the moment used is the time of the archive copy, which can only be later than the real publication. Pages that list picks for several matches and pages whose date does not hold up against our own records are never included. Anything published after kick-off, such as in-play picks, is never included.
All or nothing: every late-found pick that meets these rules is included under the same rule, before the result is read. A pick, once included, stays.
The price is the last market price recorded within 30 minutes before the evidenced moment of publication; a pick without such a price stays in the series unsettled. The series starts on ; for earlier picks the closing price is shown for reference. The author's own stated price, where given, is a separate line, as everywhere on this site. Settlement uses the same final score and the same market rules as the live record. Late-found picks stay outside the live record, the rankings and the points bankroll, and the rule, not a person, decides which of them count.
How to check the original evidence
The original source material stays with the recorded prediction. Open a pick’s evidence to check its quoted words, source link and recording time, then compare that time with kick-off. Later changes to the source page do not replace the evidence already recorded.
Disputed prices
A price the author states is compared with the market price of the same selection: our snapshot for the live record; For records archived before kick-off, the aggregator closing price or, where there is none, the last pre-match price in archive (~8–15 h before kick-off). When the two differ more than twofold (the author's price is more than twice the market price, or less than half of it), the price is marked "price disputed". A gap that large is almost never market movement: it usually means the price was read from a different market, a promotional offer or a misprint. A disputed price is left out of every return at the author's price and out of the "Stated price vs market" comparison, and the pick shows "price disputed" in the author's list of picks. It returns to both only after a manual check of the source confirms the price, and that decision is recorded with its date. The return at our market snapshot and the reconstruction at the last pre-match archive price do not use the author's price, so they are not affected.
Settlement
An outcome is settled from the final score of the event, as reported by our match-data providers, under the rule published for that market: 1X2 and totals are settled on regulation time only. A match that was abandoned, awarded or never played is void, not a loss. A restatement of the result by the data provider produces a new version of the settlement, not a second bet: the previous version is reversed in the ledger and the change is visible in the corrections log.
Which markets we settle
We settle match result, double chance, draw no bet, Asian and European handicaps, total goals, team goals, both teams to score (also combined with the result), odd/even goals, correct score, total corners, corner handicap, total cards and both teams to be carded, each on regulation time, under the rule published for that market. Player markets are settled too: "player to score" and "player shots on target". The player is identified only from the two squads of that match in our result data, and only when the quoted name matches exactly one player; if it matches none or several, the pick is recorded but not settled. A player who did not play is void. Player statistics are only available for the whole match, so a match that went to extra time leaves player markets unsettled. Of the player-shot markets we settle only shots "on target"; a "total shots" pick is recorded and left unsettled for now, with that reason shown on the pick. A bet builder or multiple is a set of legs: its outcome is known only when every leg is settled (any lost leg means it lost), and it is never executed as a bet, because no independent price for the combination exists. The card market quoted by our price source counts a red card as two, while our rule counts it as one, so no independent price is used for total cards. Picks recorded before this version were never re-priced or re-staked: where a published card gave the wrong reason for a missing bet, the correction log records it.
Where the results come from
Results of matches are currently read from a commercial sports-data aggregator covering 19 football competitions; scores are read from regulation time. Competitions it does not cover are read from a second commercial football-data provider covering 34 competitions; scores are read from regulation time. If both cover a match, we compare the scores. A mismatch leaves the match unsettled until they agree. Our prices are independent market snapshots, taken from a multi-bookmaker aggregator before the event and stored per bookmaker. The closing line comes from the same market data where supplied, or from our last snapshot before kick-off. If the aggregator misses a match or market, we use a second independent provider, kept separate. For records archived before kick-off, that second provider's price is the last pre-match archive price (~8–15 h before kick-off), not a closing price. Our market snapshots, and the priced live record, start on . Settlements of the live record so far, one per price series, were made from: a commercial sports-data aggregator covering 19 football competitions (18,884), a second commercial football-data provider covering 34 competitions (5,767), another result source (2,582). Records archived before kick-off is settled in its own series, described below. 1,898 more executed picks are awaiting a confirmed result.
Every settlement records which provider supplied its final score, and changing provider does not rewrite the past: a settlement already made keeps the source it was made from. When a provider restates a score, the settlement gets a new version and the change appears in the corrections log. Scores are always read from regulation time: 90 minutes plus stoppage, never extra time or penalties, in any competition.
National-team matches
The record covers club football and the national-team matches in its match calendar, such as friendlies. Predictions on the Nations League, World Cup and European qualifiers, the Africa Cup of Nations and similar tournaments stay in the author's record, labelled "National-team match", outside the settled record.
When a pick qualifies
A statement is executed only on an exact match of event, market, line and pick, and only if we saw it before the event started. Our timestamp is the moment the material entered our archive: a publication time at the source can be rewritten, ours cannot. A statement without a price is executed at the market price at the moment we first recorded it, always before kick-off (the rule is under ROI below); without an eligible quote nothing is executed. The same pick repeated across several channels is one observation and one bet per series. Reprocessing the same signal never creates a second bet.
ROI
ROI is the net result divided by the total staked, at an equal stake of 100 points per settled pick. It is published separately for each series: at the price the author stated, and at the independent odds snapshot of the same selection. Void picks are excluded from the denominator and reported separately. The headline number of a profile follows one fixed rule (from ). For the live record it is the return at our recorded market price. That price is the median of bookmakers' prices for the same selection in our price feed at the moment we first recorded the pick, from a quote no older than 5 minutes; betting exchanges, prediction markets and prices of 1.01 or lower are left out. If there is no such price at that moment, the first price within 2 hours after it is used, still before kick-off. Otherwise the pick has no return at the market price, and no other price is substituted. The delay of that price after the record is under a minute at the median: the price is taken when we record the pick. Figures under the earlier rule stay in the corrections log. The author's quoted price and the closing price are shown as separate lines on exactly the same picks. Without a live return, a profile shows a reconstruction at the last pre-match price in archive (~8–15 h before kick-off) on picks archived before kick-off only; picks first Found after kick-off never make the headline number. Half-won and half-lost picks on quarter-goal Asian-handicap lines are settled picks and count in full. The two returns are compared on a common sample, and the share of each sample that falls into that common set is shown next to the number.
One set per number: money figures (return, average price, cumulative result, drawdown, bankroll) belong to one series and name it; outcome figures (share of settled picks that won, share voided, correct-score accuracy, runs) count each prediction once, even when it was settled in two series.
Confidence interval
Bets on the same match are not independent, so the 95% interval is produced by a clustered bootstrap over matches with a fixed seed. Where there are too few matches to resample, the interval is reported as not estimated rather than guessed.
The points bankroll
Every tracked author starts with 10,000 points in each series and stakes 100 per qualifying pick. When the free balance falls below 100 and no pick is waiting for its result, another 10,000 points are added and the remainder is kept. Top-ups are counted and drawn separately and are never mixed into the result. Each bankroll on a profile names its series, and the bankroll shown is the starting points plus top-ups plus the net result of settled picks; stakes on picks still waiting for a result are part of it and are shown as such. Its history is read from the settlement journal in order, one point per settlement. No money is involved anywhere on this site: the bankroll is a unit of measurement, not a wallet.
Comparisons and positions
A comparison table shows records over one window side by side. A live record meets the sample criteria when it has at least 30 settled picks (excluding voided ones) on at least 15 different matches within the window, priced at our recorded market price. Tier is earned on the whole record: Established means 100+ predictions checked against the final score with at least 90 days of history. In the tipster comparison, a position goes only to a live record with at least 30 settled picks, not counting voided ones, on at least 15 matches, from a tipster with an Established record; records without a position are listed by sample size, not by return. Levels within a market or competition (No.1, Top 3, Top 5, Top 10) and titles start at Tracked. Reconstructed history is shown in its own table and is never compared with live records. An author with no qualifying pick for 30 days keeps the profile with a note and leaves the current window. Nothing in a comparison is for sale.
Checking a published number
Use a record’s JSON or CSV download to check the picks, prices and outcomes behind its published total. Each metric identifies the settlement rules used; the corrections log shows changes to those rules and to published results.
Seven days to reply
Authors can answer a record before search engines see it, and after any major change. We publish the answer word for word next to the record. A profile gets an address from its first recorded statement and stays hidden from search engines until it reaches the publication threshold. This gives time to check and answer the whole record before anyone finds it through search.
A major change is a correction to a settled outcome, a profile's headline number, or the series a pick belongs to. We make it immediately. It appears in the corrections log with its date, the old value, and the new one. A reply sent within seven days is published as a reply to the change, next to the changed number; later replies appear next to the current numbers. The deadlines and form are on the right of reply page.
Public review method — how public search results and answers are checked.
Frequently asked questions
How is a tipster's return on stakes (ROI) calculated?
The net result of the settled picks divided by the total staked, at 100 points per pick. Void picks are left out of the total staked. Each series has its own return: at the author's stated price and at our independent odds snapshot.
How is a pick settled?
A pick is settled from the result of the match under the rule published for its market; 1X2 and totals use regulation time. Won, Lost, Void, Half won and Half lost are the only outcomes. The net result is the sum of what each settled pick won or lost, in points.
What is closing-line value (CLV)?
How a price compares with the closing price of the same selection: CLV = 100 × (price ÷ closing price − 1). For the author’s stated price, it is calculated wherever a closing price of the same selection exists and shown as a median under “Stated price vs market”. Where no closing price of the same selection exists, the CLV is unknown, not zero.
Where do the prices come from?
The median of bookmakers’ prices for the same selection in our price feed when we first recorded the pick, from a quote no older than 5 minutes. Betting exchanges, prediction markets and prices of 1.01 or lower are excluded. If there is no such price at that moment, we use the first price within 2 hours, provided it is before kick-off. Otherwise, the pick has no return at the market price, and no other price is substituted. A pick recorded before kick-off but checked only after kick-off settles the same way, at the price when we recorded it, and is marked “checked after kick-off”. Earlier figures stay in the corrections log. The snapshot records the market at that moment, not a promise that a reader could have got that price.
What is the difference between a live and an Archived before kick-off record?
Every pick carries the label of its strongest timing evidence: Submitted to us before kick-off; Recorded before kick-off; Archived before kick-off; Source-reported history; Found after kick-off. Submitted and live-captured picks belong to the live record. Archived before kick-off, Source-reported history and Found after kick-off remain separate from it. The headline number of a profile is the live record from 10 settled picks; until then it is the part archived before kick-off, reconstructed at archived prices, with the live record so far shown beside it as picks won out of picks settled. Tier is earned on the whole record: Established means 100+ predictions checked against the final score with at least 90 days of history. In the tipster comparison, a position goes only to a live record with at least 30 settled picks, not counting voided ones, on at least 15 matches, from a tipster with an Established record; records without a position are listed by sample size, not by return. Levels within a market or competition (No.1, Top 3, Top 5, Top 10) and titles start at Tracked.
Can a tipster delete losing picks to improve the record?
Not in the live record: a pick enters it only if we recorded it before kick-off, with the page archived at that moment. A post deleted later stays in the record. History first found after kick-off is different: it is read later from pages still online, so an author can delete losing picks before we see them. That is why history found after kick-off is labelled, kept in its own series and never mixed with the live record.
Which records get a position?
The minimum sample a live record needs before it is marked as meeting the comparison criteria in a window. Tier is earned on the whole record: Established means 100+ predictions checked against the final score with at least 90 days of history. In the tipster comparison, a position goes only to a live record with at least 30 settled picks, not counting voided ones, on at least 15 matches, from a tipster with an Established record; records without a position are listed by sample size, not by return. Levels within a market or competition (No.1, Top 3, Top 5, Top 10) and titles start at Tracked.
How is the uncertainty of a return estimated?
A 95% interval around a return, from a clustered bootstrap over matches. With too few matches it is reported as not estimated rather than guessed.
What does "unknown" mean next to a number?
Unknown marks a value awaiting sufficient evidence. Zero is a measured value. Measured describes an observed fact or a result calculated from recorded inputs under a published rule; Stated describes the source’s own claim.
Research using this method
- Which collected football predictions can be checked against a result only some collected predictions can be checked against a match result; this page counts only those
Method history
Published rule versions retained from the methodology record, with their effective dates.
Earlier published rule versions and effective dates
What a measurement here means version 1.0
We record what a tipster published, price it under a published rule (for the live record, the market price in our own price feed at the moment we first recorded the pick, always before kick-off) and settle it against the final score. Every number on this site therefore has four things attached to it: the sample size, the window, the status of the data and the version of the method.
- Measured: directly observed in a source or timestamped capture;
- Stated: reported by the source; not independently confirmed;
- Calculated: derived from the listed inputs under a published rule;
- Unknown: not established from the evidence available. Unknown is never written as zero.
A status belongs to one field, not to a whole record: the time of our own capture is measured, while a publication date written on the page is stated.
What counts as a pick and what is executed version 1.1
A statement with an explicit price and stake is executed in two series: at the price the author stated and at our own market snapshot. A statement that names a pick without a price is executed at our market snapshot only, and the return at the author's own price is reported as not stated. A correct-score forecast is measured for score accuracy, and a separate derived series settles the 1X2 outcome implied by that score. That derivation is a Forecast Record rule and is reported apart from the author's own market return. An opinion, a conditional or a joke is kept as an observation and never executed. Only statements about an event are counted. Advertising and site boilerplate captured from a source page (bonus offers, betting buttons, affiliate disclosures, newsletter prompts, score widgets, menus) are not the author's words and are not listed as statements. On a match page, the split of what tipsters and sources picked counts each tipster with 10 or more predictions checked against the final score to date once per market, with the pick made before kick-off; predictions from tipsters and sources with fewer are listed separately, without shares.
What is executed and what is not version 2.2 · in force since
A statement is executed only on an exact match of event, market, line and pick, and only if we saw it before the event started. Our timestamp is the moment the material entered our archive: a publication time at the source can be rewritten, ours cannot. A statement without a price is executed at the market price at the moment we first recorded it, always before kick-off (the rule is under ROI below); without an eligible quote nothing is executed. The same pick repeated across several channels is one observation and one bet per series. Reprocessing the same signal never creates a second bet.
ROI version 2.2 · in force since
ROI is the net result divided by the total staked, at an equal stake of 100 points per settled pick. It is published separately for each series: at the price the author stated, and at the independent odds snapshot of the same selection. Void picks are excluded from the denominator and reported separately. The headline number of a profile follows one fixed rule (from ). For the live record it is the return at our recorded market price. From method version 2.2 () that price is the median of bookmakers' prices for the same selection in our price feed at the moment we first recorded the pick, from a quote no older than 5 minutes; betting exchanges, prediction markets and prices of 1.01 or lower are left out. If there is no such price at that moment, the first price within 2 hours after it is used, still before kick-off. Otherwise the pick has no return at the market price, and no other price is substituted. The delay of that price after the record is under a minute at the median: the price is taken when we record the pick. Figures under the earlier rule stay in the corrections log. The author's quoted price and the closing price are shown as separate lines on exactly the same picks. Without a live return, a profile shows a reconstruction at the last pre-match price in archive (~8–15 h before kick-off) on picks archived before kick-off only; picks first found after the match never make the headline number. Half-won and half-lost picks on quarter-goal Asian-handicap lines are settled picks and count in full (from version 2.1). The two returns are compared on a common sample, and the share of each sample that falls into that common set is shown next to the number.
One set per number: money figures (return, average price, cumulative result, drawdown, bankroll) belong to one series and name it; outcome figures (share of settled picks that won, share voided, correct-score accuracy, runs) count each prediction once, even when it was settled in two series.
Confidence interval version 1.0
Bets on the same match are not independent, so the 95% interval is produced by a clustered bootstrap over matches with a fixed seed. Where there are too few matches to resample, the interval is reported as not estimated rather than guessed.
The points bankroll version 2.1 · in force since
Every tracked author starts with 10,000 points in each series and stakes 100 per qualifying pick. When the free balance falls below 100 and no pick is waiting for its result, another 10,000 points are added and the remainder is kept. Top-ups are counted and drawn separately and are never mixed into the result. Each bankroll on a profile names its series, and the bankroll shown is the starting points plus top-ups plus the net result of settled picks; stakes on picks still waiting for a result are part of it and are shown as such. Its history is read from the settlement journal in order, one point per settlement (version 2.1; before it every past point repeated the current balance). No money is involved anywhere on this site: the bankroll is a unit of measurement, not a wallet.
Settlement version 1.0
An outcome is settled from the final score of the event, as reported by our match-data providers, under the rule published for that market: 1X2 and totals are settled on regulation time only. A match that was abandoned, awarded or never played is void, not a loss. A restatement of the result by the data provider produces a new version of the settlement, not a second bet: the previous version is reversed in the ledger and the change is visible in the corrections log.
Which markets we settle version 2.1 · in force since
We settle match result, double chance, draw no bet, Asian and European handicaps, total goals, team goals, both teams to score (also combined with the result), odd/even goals, correct score, total corners, corner handicap, total cards and both teams to be carded, each on regulation time, under the rule published for that market. Player markets are settled too: "player to score" and "player shots on target". The player is identified only from the two squads of that match in our result data, and only when the quoted name matches exactly one player; if it matches none or several, the pick is recorded but not settled. A player who did not play is void. Player statistics are only available for the whole match, so a match that went to extra time leaves player markets unsettled. Of the player-shot markets we settle only shots "on target"; a "total shots" pick is recorded and left unsettled for now, with that reason shown on the pick. A bet builder or multiple is a set of legs: its outcome is known only when every leg is settled (any lost leg means it lost), and it is never executed as a bet, because no independent price for the combination exists. The card market quoted by our price source counts a red card as two, while our rule counts it as one, so no independent price is used for total cards. Picks recorded before this version were never re-priced or re-staked: where a published card gave the wrong reason for a missing bet, the correction log records it.
Where the results come from version 1.0
Results of matches are currently read from a commercial sports-data aggregator covering 19 football competitions; scores are read from regulation time. Competitions it does not cover are read from a second commercial football-data provider covering 34 competitions; scores are read from regulation time. If both cover a match, we compare the scores. A mismatch leaves the match unsettled until they agree. Our prices are independent market snapshots, taken from a multi-bookmaker aggregator before the event and stored per bookmaker. The closing line comes from the same market data where supplied, or from our last snapshot before kick-off. If the aggregator misses a match or market, we use a second independent provider, kept separate. For restored history, that second provider's price is the last pre-match archive price (~8–15 h before kick-off), not a closing price. Our market snapshots, and the priced live record, start on . Settlements of the live record so far, one per price series, were made from: a commercial sports-data aggregator covering 19 football competitions (18,884), a second commercial football-data provider covering 34 competitions (5,767), another result source (2,582). Restored history is settled in its own series, described below. 1,898 more executed picks are awaiting a confirmed result.
Every settlement records which provider supplied its final score, and changing provider does not rewrite the past: a settlement already made keeps the source it was made from. When a provider restates a score, the settlement gets a new version and the change appears in the corrections log. Scores are always read from regulation time: 90 minutes plus stoppage, never extra time or penalties, in any competition.
Disputed prices version 2.2 · in force since
A price the author states is compared with the market price of the same selection: our snapshot for the live record; for restored history, the aggregator closing price or, where there is none, the last pre-match price in archive (~8–15 h before kick-off). When the two differ more than twofold (the author's price is more than twice the market price, or less than half of it), the price is marked "price disputed". A gap that large is almost never market movement: it usually means the price was read from a different market, a promotional offer or a misprint. A disputed price is left out of every return at the author's price and out of the "Stated price vs market" comparison, and the pick shows "price disputed" in the author's list of picks. It returns to both only after a manual check of the source confirms the price, and that decision is recorded with its date. The return at our market snapshot and the reconstruction at the last pre-match archive price do not use the author's price, so they are not affected.
National-team matches version 2.2 · in force since
The record covers club football and the national-team matches in its match calendar, such as friendlies. Predictions on the Nations League, World Cup and European qualifiers, the Africa Cup of Nations and similar tournaments stay in the author's record, labelled "National-team match", outside the settled record.
Restored history version 1.1
Part of some authors' history is restored rather than recorded. For a set of authors we re-read what they published before matches played between and — the Premier League first, then the Championship, League Cup, Champions League and Europa League — and settled it against the final score. Every pick carries the label of its strongest evidence of time, from one list used on every page. Submitted to us before kick-off: the author sent the prediction to us directly and we stamped the time when we received it; it belongs to the live record. Recorded by us before kick-off: we archived it ourselves before the match; this is the live record; only it appears in the live comparison table and the points bankroll, and only a live record can earn a position there. Archived before kick-off: an independent public web archive holds a copy of the page taken before kick-off, and the pick is in that copy. Source-reported history: the time and the pick are known only from the source's own history; it is used to check the source's own record and is never part of the evidenced total. Found after the match; date as stated by the source: found later and dated by the source's own publication date — the lowest of the levels of evidence. Evidence obtained after kick-off carries a second line, Added to our record on (date); the pre-match record of the match and of each author stays as it was.
Restored picks are shown as a reconstruction at the last pre-match price in archive (~8–15 h before kick-off), for picks archived before kick-off and picks found after the match separately. The author's own quoted price and, where it exists, the aggregator closing price are separate lines on the same picks. Outcome accuracy is shown separately from return. From the minimum of 10 checked predictions on 2 dates a profile needs before it is indexed counts only the evidenced part of the record: picks recorded by us before kick-off and picks archived before kick-off. Every published profile has predictions in the evidenced part and states on its first screen how many of its predictions come from each part.
Since restored history covers the Championship, League Cup, Champions League and Europa League as well, with prices back to August 2024. The last pre-match archive price is the median of each bookmaker's last price before kick-off, from at least two bookmakers, in a series of its own. Score forecasts are settled the way the live record settles them: on the exact-score market only when the author explicitly recommends the bet, and otherwise as the match result the score implies, in a series of its own, separate from the author's picks.
Late-found picks, priced at publication time version 1.0 · in force since
A pick published before kick-off and found after the match has started stays outside the record kept before kick-off. Late-found picks form a separate series, priced at publication time, shown on the author's profile as its own line and marked on the pick itself.
A late-found pick enters the series only if the time of publication is evidenced: by the platform's own timestamp (a post or a video), by the publisher's feed date checked against our own records (not later than our first capture, not later than kick-off and not more than 30 days before it), by our own copy of the page taken before kick-off, or by a copy in an independent public web archive taken before kick-off that contains the quoted words. For a web page the date the page shows about itself is not used as evidence; the moment used is the time of the archive copy, which can only be later than the real publication. Pages that list picks for several matches and pages whose date does not hold up against our own records are never included. Anything published after kick-off, such as in-play tips, is never included.
All or nothing: every late-found pick that meets these rules is included automatically, by one rule, before the result is read. A pick, once included, stays.
The price is the last market price recorded within 30 minutes before the evidenced moment of publication; a pick without such a price stays in the series unsettled. The series starts on ; for earlier picks the closing price is shown for reference. The author's own stated price, where given, is a separate line, as everywhere on this site. Settlement uses the same final score and the same market rules as the live record. Late-found picks stay outside the live record, the rankings and the points bankroll, and the rule, not a person, decides which of them count.
How captured files are kept version 2.2 · in force since
Every captured file is kept in two separate places and checked by its fingerprint before it counts as stored. Files captured before were copied unchanged and checked the same way.
Comparisons and positions version 1.1
A comparison table shows records over one window side by side. A live record meets the sample criteria when it has at least 30 settled picks (excluding voided ones) on at least 15 different matches within the window, priced at our recorded market price. In the tipster comparison, a position goes only to a live record with at least 30 settled picks, not counting voided ones, on at least 15 matches, from a tipster with an Established record; records without a position are listed by sample size, not by return. Levels within a market or competition (No.1, Top 3, Top 5, Top 10) and titles start at Tracked. Reconstructed history is shown in its own table and is never compared with live records. An author with no qualifying pick for 30 days keeps the profile with a note and leaves the current window. Nothing in a comparison is for sale.
Reproducibility version 1.0
Every published number can be recomputed from the journal of bets and outcomes, and the recomputation has to produce the same number. That is a test the build runs against itself. Each metric carries the version of the method that produced it, and a change of method changes the version rather than the past.
Seven days to reply version 1.0 · in force since
Authors can answer a record before search engines see it, and after any major change. We publish the answer word for word next to the record. A profile gets an address from its first recorded statement and stays hidden from search engines until it reaches the publication threshold. This gives time to check and answer the whole record before anyone finds it through search.
A major change is a correction to a settled outcome, a profile's headline number, or the series a pick belongs to. We make it immediately. It appears in the corrections log with its date, the old value, and the new one. A reply sent within seven days is published as a reply to the change, next to the changed number; later replies appear next to the current numbers. The deadlines and form are on the right of reply page.
How we measure our own visibility version 1.0
We also measure how search engines and AI assistants see us using a separate method: fixed questions, fixed surfaces, and failures kept apart from “not mentioned”. The full method is on its own page.
How we measure our own visibility — the questions, the surfaces, what is recorded and the limits.
Earlier method-version explanation
Every number carries the version of the method that produced it, for example “method v2.1”. A pick is placed under the version in force at that moment and keeps it, so a profile can say “picks placed under method v2.0” while its numbers are calculated under version 2.1: a new version recalculates the result and never creates a second bet. What changed in each version is listed under “Rules and versions”.
Example. A pick placed on under version 2.0 and settled after reads “placed under method v2.0” and is calculated under version 2.1.
Earlier profile headline rule
Every pick carries one of three labels: recorded by us before kick-off; restored from a pre-match web archive copy; found after the match with the date the source states. Only the first is the live record; the other two are shown in their own parts of every profile and never added to it. The headline number of a profile is the live record from 10 settled picks; until then it is the part archived before kick-off, reconstructed at archived prices, with the live record so far shown beside it as picks won out of picks settled (rule of ). In the tipster comparison, a position goes only to a live record with at least 30 settled picks, not counting voided ones, on at least 15 matches, from a tipster with an Established record; records without a position are listed by sample size, not by return. Levels within a market or competition (No.1, Top 3, Top 5, Top 10) and titles start at Tracked.
The same data, machine-readable
The same numbers in machine-readable form.
Check us