Prediction methodology
How FairPlay's cricket predictions and rates work
FairPlay rates every team with its own model, built only from verified match results, and uses those ratings for the match estimates on the analysis page and the FairPlay rates. This page sets out how the model works, how often it has been right on past matches and how an estimate becomes a rate.
Last updated:
The model right now
- Model in use
- Version 1, awaiting sign-off
- Verified results learned from
- 6,003
- Past matches it was tested on
- 2,414
- Ratings last rebuilt
- Sat, 3 Oct, 20:37 IST
- Toss rates showing now
- 67
- Match-winner rates showing now
- 31
Read from FairPlay's own records when this page loaded. Accuracy figures below are recalculated from the same results each time.
What the model is
It is a team-rating model of the Elo kind, the approach used for chess and many sports rankings. Each team has one rating per format: T20, one-day and Test. A higher rating means a stronger team on recent verified results.
Before a match, the gap between the two ratings gives each side a chance of winning. A 100-point gap gives the higher-rated side about 64%. After the result is verified, the winner gains points and the loser loses the same number. A surprise result moves both ratings more than an expected one.
When a team has little history, its estimate is pulled towards 50/50. For example, a raw 70% chance becomes 60% if one of the teams has only 10 results in the format.
Settings in use
| Setting | Value | What it does |
|---|---|---|
| Starting rating | 1,500 | Every team starts here in each format, before any verified result. |
| Update size (K) | T20 24 · ODI 28 · Test 30 | How many points a single result can move a rating. Longer formats move a little more per match because they are played less often. |
| Host-side allowance | 30 points | Added to the team the data feed lists as the local side before the estimate is made. |
| Full-confidence history | 20 matches | Until both teams have this many results in the format, the estimate is pulled towards 50/50. |
| Minimum history | 8 matches each | Below this, FairPlay makes no estimate and offers no match-winner rate. |
What it learns from
Only finished matches with a verified winner, from FairPlay's own database. Ties, no-results, abandoned matches and draws do not change any rating. FairPlay does not use betting odds or rates from anyone else, as input or as a check.
Results come from the same data feed as the live scores, described on the data methodology page. The ratings are rebuilt from scratch at regular intervals and soon after new results arrive.
How formats are grouped
- T20: T20 internationals and leagues, plus T10 and 100-ball cricket.
- One-day: ODIs and other List A one-day matches.
- Test: Tests and other first-class matches of four or five days.
- Matches in any other format are not rated.
How often it has been right
FairPlay replays every verified result in date order. Each match is estimated using only the results before it, then its result is added. A match is scored only when both teams already had at least 8 results in the format, the same rule used for rates.
| Format | Results | Tested on | Favourite won | Host side won | Brier score | Log loss |
|---|---|---|---|---|---|---|
| T20 | 4,496 | 1,854 | 56.0% | 51.8% | 0.244 | 0.682 |
| One-day (ODI) | 961 | 440 | 60.7% | 59.3% | 0.233 | 0.658 |
| Test and first-class | 546 | 120 | 60.8% | 55.8% | 0.239 | 0.671 |
| All formats | 6,003 | 2,414 | 57.1% | 53.4% | 0.242 | 0.677 |
- Favourite won
- How often the side the model gave the higher chance went on to win.
- Host side won
- A simple comparison: how often the data feed's local side won the same matches. The model is only useful if it beats this.
- Brier score
- The average squared gap between the estimate and what happened. Lower is better. Calling every match 50/50 scores 0.250.
- Log loss
- Like the Brier score, but punishes confident wrong calls harder. Lower is better. Calling every match 50/50 scores 0.693.
In plain terms: the model has an edge, and it is modest. It picks the winner more often than the simple comparison in every format, and it beats calling every match 50/50, but cricket stays hard to call. Whether its percentages hold up in each range is shown in the calibration table on the analysis page.
From an estimate to a FairPlay rate
Step 1
Fair chance
The model's chance for each side. For the toss, both sides are 50/50.
Step 2
Confidence check
If either side's fair chance is above 97%, no rate is offered.
Step 3
House margin
The margin is added with the power method, which puts slightly more of it on the less likely side.
Step 4
Price limits
Each rate is capped at 26.00 and must be at least 1.01.
Step 5
Rounded towards FairPlay
Every rate is rounded down to the nearest step on FairPlay's price ladder.
Step 6
Band check
The total margin must stay inside the policy's band after rounding, or the rate is not published.
Worked example
A match-winner market where the model gives the two sides 60% and 40%, run through the same engine that prices live matches.
| Side | Fair chance | Fair price | FairPlay rate | Indian rate | ₹100 returns |
|---|---|---|---|---|---|
| Side A | 60% | 1.67 | 1.60 | 60 | ₹160 |
| Side B | 40% | 2.50 | 2.32 | 132 | ₹232 |
The two FairPlay rates add up to a 105.6% book: the 5.6% above 100% is FairPlay's margin. A toss is priced the same way from 50/50, giving 1.90 on each side. “₹100 returns” is what a winning ₹100 stake would pay back, stake included.
Margin policies
Each market type has a written margin policy. Only an approved policy is used for rates, and every change creates a new version.
| Market | Target book | Allowed band | Rate limits | Valid for | Status |
|---|---|---|---|---|---|
| Toss winnerVersion 1 | 105% | 103%–106% | 1.01–26.00 | 15 min | Approved by FairPlay, 2 October 2026 |
| Match winner (T20 and one-day)Version 1 | 105% | 104%–106% | 1.01–26.00 | 15 min | Approved by FairPlay, 2 October 2026 |
| Test match resultVersion 1 | 107% | 106%–108% | 1.01–26.00 | 15 min | Proposed — not used for any rate |
The price ladder
FairPlay rates move in fixed steps, so the same chance always gives the same rate. Steps are small near evens and wider for long shots. The Indian rate shown on FairPlay is the decimal rate minus 1, times 100: 1.85 is shown as 85.
| Decimal rate | Step |
|---|---|
| 1.01 to 2.00 | 0.01 |
| 2.00 to 3.00 | 0.02 |
| 3.00 to 4.00 | 0.05 |
| 4.00 to 6.00 | 0.10 |
| 6.00 to 10.00 | 0.20 |
| 10.00 to 20.00 | 0.50 |
| 20.00 to 50.00 | 1.00 |
When rates open, close or are suspended
- Which matches are priced
- Scheduled matches with both teams confirmed, starting within the next 7 days.
- Toss rates close early
- 45 minutes before the scheduled start, or as soon as a toss result is recorded, whichever comes first.
- Match-winner rates close at the start
- At the scheduled start time. FairPlay does not offer in-play rates.
- Suspended when data is late
- If FairPlay has not confirmed a match with the data feed for 15 minutes, its rates are suspended. They return once the match is confirmed again.
- Suspended when a match changes
- A change to the start time, teams or venue suspends that market's rates. They are not re-opened automatically: they stay suspended until the change has been reviewed.
- Closed if the match does not go ahead as planned
- A match that is postponed, abandoned, cancelled or no longer scheduled has its rates closed.
- Never shown past their validity
- Each price is valid for up to 15 minutes and never beyond the market's close. A price that is not re-confirmed in time is suspended, never left on show.
What the model does not know
The model sees team results and nothing else. Read every estimate and rate with these limits in mind.
- Players. The model does not know playing XIs, injuries, rested players or captaincy changes.
- Conditions. It does not use the pitch, the weather or the result of the toss.
- Neutral venues. The host-side allowance goes to the team the data feed lists as the local side, even when neither team is at home.
- In-play. It makes pre-match estimates only. There are no in-play rates.
- Test matches. Draws need a separate model, so Test results are not priced yet. Test ratings are kept and tested, but not used for any rate.
- New or rarely seen teams. Teams with fewer than 8 results in a format are not estimated.
- Separate sides. Men's, women's, age-group, A and franchise teams are rated separately, as the data feed lists them.
- The future. A model that was right more often than not on past matches can still be wrong about the next one.
Records and changes
- Every published rate is stored as its own version, with the model, the margin policy and the inputs it came from. A published version is never edited.
- Every time a market opens, is suspended or closes, the reason is recorded.
- A change to the model's settings or to a margin policy creates a new version. Older versions are kept, and the figures on this page always describe the version in use.
Estimates are not guarantees
A model estimate is a statistical view of past results, not a tip, and a FairPlay rate includes FairPlay's margin. Neither promises any outcome. Read the responsible use guide and the FairPlay rates page for how rates are shown.
Betting is not open yet. No deposits, withdrawals or bets are possible today. INR betting opens only after payment, identity (KYC), anti-money-laundering and location checks are connected and verified.