Metrics

19 metrics7 categories

Every statistic on FTC Secrets is defined here: what it measures, how it is computed, and where it can mislead. For data sources, the update schedule, and which matches count, see the methodology.

How ratings work

Ratings come from replaying the whole season from its first match, so a corrected result changes every rating after it. Results move through five steps.

Starting rating
1500
Every team, every season
Rating scale
400
A 200-point gap is a 76% expected result
K-factor
32 → 16
Floor reached after 30 eligible matches
  1. Official match results

    Each day we pull played matches: the teams on each alliance, any surrogate, no-show, or disqualification flags, and the reported final totals, which include penalty points. For seasons with verified score fields we also keep non-penalty scores and the score breakdown.

  2. Match eligibility

    A match counts only if both alliances have two distinct teams and a reported final total. A surrogate, no-show, or disqualified team is skipped for its own update while the other teams keep the result. The methodology has the full rules.

  3. Season Elo replay

    Every team starts the season at 1500. Matches are replayed from the first one, in the order their results were posted. Each alliance is the mean of its two ratings, and only the outcome moves ratings: win, loss, or tie. Margin of victory, playoff status, and recency play no part. See Elo rating.

  4. OPR per event

    Where the season’s score fields are verified, each event’s non-penalty alliance scores are solved by least squares for one value per team. Events whose schedule cannot separate every team get no OPR, and matches with a flagged team are dropped. See OPR.

  5. Derived metrics

    Percentiles, season gain, opponent strength, field strength, win probability, and the rest are computed from the ratings, OPR, and results above. Each is defined below with its formula, an example, and its limits.

Strength

4 metrics

Elo rating

Season rating built from every official match result. Starts at 1500; beating stronger alliances moves it more.

Scope
Team
Higher is better
Yes
Unit
rating
Range
Unbounded. In 2025–26 most teams finished between about 1400 and 1650 (median 1486); the top 1% finished above about 1750.
Data required
Official final scores for both alliances. Matches with surrogate, no-show, or disqualified teams freeze those teams’ ratings.

Formula

Alliance rating R = mean of its two teams’ ratingsExpected red result E = 1 / (1 + 10^((R_blue − R_red) / 400))Actual result S = 1 (red win), 0.5 (tie), 0 (blue win), from official final scoresK = max(16, 32 / (1 + n / 30)), n = the team’s prior eligible matchesEach red team: R += K̄ × (S − E); each blue team: R −= K̄ × (S − E)K̄ = mean of the current K of the match’s eligible teams; surrogate, no-show, and disqualified teams are not updated

Interpretation

Every team starts the season at 1500. A team that consistently beats alliances it was expected to lose to climbs quickly. Because the update depends only on win, loss, or tie, Elo measures winning against the field you played, not how many points you scored. Differences matter more than absolute values: a 200-point gap between two alliances corresponds to roughly a 76% expected win rate for the stronger one.

Example

A 1500 team partnered with a 1600 team (alliance 1550) beats an alliance averaging 1650. Expected result was 0.36, so with K = 32 each red team gains 32 × (1 − 0.36) ≈ +20.

Limitations

Ratings only compare teams that are connected through common opponents. Regions that rarely play each other can drift apart, so cross-region comparisons are less reliable than within-region ones. Elo cannot separate a team from its partners within a match, so early-season ratings are noisy. There is no carryover between seasons and no minimum match count for a rating to exist.

Related metrics

Elo percentile

Share of rated teams with a lower Elo this season (ties count half).

Scope
Team
Higher is better
Yes
Unit
percentile
Range
0–100
Data required
Elo ratings for the comparison pool.

Formula

Percentile = 100 × (teams below + tied teams / 2) / all rated teamsTied teams include the team itself; rounded to one decimal

Interpretation

The 90th percentile means the team is rated above 90% of teams in the comparison pool. Region percentiles use the same formula restricted to teams from the same FIRST region.

Example

Team ranked 1,068th of 8,366 rated teams sits at about the 87th percentile.

Limitations

The pool includes every team with a rating, including teams that have played only a handful of matches, which compresses the middle of the distribution early in the season.

Related metrics

Elo change at last event

How much the team’s Elo moved during its most recent event.

Scope
Team
Higher is better
Yes
Unit
rating points
Range
Usually within about ±60 per event (median size about 23); rarely beyond ±300.
Data required
Per-event rating snapshots.

Formula

Δ = Elo after last match at the event − Elo entering the event

Interpretation

A large positive change means the team beat stronger alliances than its rating predicted. Combine with the number of matches played at the event: ten matches can move a rating much more than five.

Example

1880 entering the championship, 1974 after it: Δ = +94.

Limitations

Single-event changes are noisy and partly reflect partner luck.

Related metrics

Season Elo gain

Rating gained since the season started at 1500.

Scope
Team
Higher is better
Yes
Unit
rating points
Range
Unbounded; in 2025–26 usually −100 to +250.
Data required
Elo rating.

Formula

Season gain = current Elo − 1500

Interpretation

Every team starts each season at 1500, so this is the total rating earned this season. It grows with both quality and the number of matches played.

Example

A team at 1650 has gained +150 this season.

Limitations

Teams that have played few matches have small gains regardless of quality.

Related metrics

Scoring

5 metrics

OPR

Estimated non-penalty points the team contributes to its alliance, solved from match scores at its most recent event.

Scope
Team
Higher is better
Yes
Unit
points
Range
Points; can be negative for teams that rarely score.
Data required
Non-penalty alliance scores and a schedule that identifies every team’s contribution (a full-rank, well-conditioned system); otherwise OPR is unavailable for that event.

Formula

For each alliance in each match: score_np = OPR_a + OPR_bSolve the over-determined system for every team at the event with least squaresMatches with surrogate, no-show, or disqualified teams are excludedUnavailable when the system does not identify every team (rank-deficient or condition number above 10¹⁰)

Interpretation

OPR (offensive power rating) splits each alliance’s non-penalty score between its two teams by finding the contributions that best explain every match at an event. Unlike Elo it rewards points scored, not wins, so it is the better measure of raw scoring ability.

Example

Three matches with alliance scores 120, 90, and 150 across varying partners might solve to OPR 68 for the team.

Limitations

With 5–8 matches per team the estimate has an error of roughly 8–12 points. It assumes contributions add up and ignores defense and partner synergy. OPR from different events is not directly comparable because fields differ. Rows computed before the September 2026 metrics rewrite hold a season-cumulative OPR instead of the latest event’s; they change at the next ratings rebuild. Unavailable until the season’s score schema is verified.

Related metrics

Best event OPR

The highest OPR the team produced at any single event this season.

Scope
Team
Higher is better
Yes
Unit
points
Range
Points.
Data required
At least one event with a solvable OPR.

Formula

Best event OPR = max over events of that event’s OPR

Interpretation

Shows a team’s scoring ceiling. Useful for spotting teams whose latest event underrepresents what they can do.

Example

Event OPRs of 25, 41, 48, 78, 95, 69 give a best event OPR of 95.

Limitations

Taking the maximum of several noisy estimates is biased upward, especially for teams with many events.

Related metrics

Highest alliance score

Best non-penalty score of any alliance the team played on this season.

Scope
Team
Higher is better
Yes
Unit
points
Range
Points.
Data required
Non-penalty alliance scores.

Formula

max over the team’s eligible matches of its alliance’s non-penalty score

Interpretation

An alliance-level ceiling. A high value shows the team has been part of a very high-scoring alliance.

Example

An alliance score of 326 with a partner is credited to both teams.

Limitations

Includes the partner’s contribution, so it overstates individual ability when paired with a strong partner.

Related metrics

Average alliance score

Mean non-penalty score of the team’s alliances across its played matches.

Scope
Team
Higher is better
Yes
Unit
points
Range
Points.
Data required
Non-penalty alliance scores.

Formula

mean of alliance non-penalty scores over the team’s played matches

Interpretation

A steadier view of scoring than the high score. Computed on team pages from the team’s match list; no season-wide average is stored, so it is not available in rankings.

Example

Alliance scores of 54, 39, 61, 45 average to 49.8.

Limitations

Includes partner contributions and depends on the season’s game, so it is not comparable across seasons.

Related metrics

Score consistency

How tightly the team’s alliance scores cluster around their average. 100 means every score was identical.

Scope
Team
Higher is better
Yes
Unit
0–100
Range
0–100
Data required
At least two non-penalty alliance scores.

Formula

Consistency = clamp(100 × (1 − σ / μ), 0, 100)σ, μ = population standard deviation and mean of alliance non-penalty scores

Interpretation

Higher values mean more predictable output. Teams with similar averages but different consistency carry different risk as alliance partners.

Example

Scores averaging 100 with a standard deviation of 20 give a consistency of 80.

Limitations

Based on alliance scores, so partner variation adds noise. High scorers tend to look more consistent because random variation is a smaller fraction of their mean. Needs at least two matches; fewer than ~8 matches is noisy. Older rows stored this as a 0–1 fraction; it is rescaled for display.

Related metrics

Results

3 metrics

Win rate

Share of matches won, counting ties as half a win.

Scope
Team
Higher is better
Yes
Unit
%
Range
0–100%
Data required
Official final scores.

Formula

Win rate = (wins + ties / 2) / matches with official results

Interpretation

A direct measure of results. Unlike Elo it does not account for opponent strength, so it favours teams in weaker fields.

Example

23 wins, 22 losses, 0 ties = 51%.

Limitations

Includes partner contribution and is noisy for small samples.

Related metrics

Record

Wins, losses, and ties in matches with official final scores.

Scope
Team
Higher is better
Yes
Data required
Official final scores.

Formula

Win, loss, or tie is decided by official final scores including penalties

Interpretation

Counts qualification and playoff matches. Non-penalty scores never decide a result.

Example

78–10–0

Limitations

Matches without official final totals, incomplete lineups, and the team’s own surrogate, no-show, or disqualified matches are excluded.

Related metrics

Matches played

Matches the team has appeared in this season.

Scope
Team
Higher is better
Neutral
Unit
matches

Formula

Count of qualification and playoff matches the team appears in, excluding its no-shows (surrogate and disqualified appearances count)

Interpretation

Sample size behind the team’s other statistics. Treat ratings from fewer than ~10 matches with caution.

Limitations

Counts appearances, including matches excluded from some metrics.

Related metrics

Schedule

1 metric

Opponent strength

Average season Elo of the opponents the team faced.

Scope
Team
Higher is better
Neutral
Unit
rating
Range
Rating scale; in 2025–26 most schedules fell between about 1490 and 1550.
Data required
Opponent Elo ratings.

Formula

mean over eligible matches and opponents of each opponent’s season-final Elo

Interpretation

Higher values mean a harder schedule. Compare a team’s record with its opponent strength: a winning record against strong opponents is more impressive than the same record against weak ones.

Example

Facing opponents averaging 1600 Elo is a much harder schedule than the 2025–26 median of about 1515.

Limitations

Uses end-of-season ratings, so it describes how strong opponents turned out to be, not how strong they were at the time of the match. It does not account for partner strength.

Related metrics

Ranking points

2 metrics

Ranking point rate

Share of the team’s matches in which its alliance earned a given bonus ranking point.

Scope
Team
Higher is better
Yes
Unit
%
Range
0–100%
Data required
Official per-alliance RP flags.

Formula

Rate = matches where the alliance earned the RP / eligible matches with an official RP flag

Interpretation

Each season defines bonus ranking points for specific achievements. A high rate means the team’s alliances reliably hit that threshold, which matters for qualification rank.

Example

41 of 45 matches with the goal RP = 91%.

Limitations

Alliance-level, so partners contribute. Thresholds differ between regular events and championships. Unavailable until the season’s score schema is verified.

Related metrics

Ranking point outlook

Smoothed historical estimate of how often the team’s alliances earn a given RP.

Scope
Team
Higher is better
Yes
Unit
%
Range
0–100%
Data required
RP flags for at least one match.

Formula

Prior: Beta(0.5, 1.5), mean 25%p = (1 − b) × posterior mean + b × recency-weighted rate, b = min(n / 5, 0.9)Recency weight of a match = e^(−0.15 × matches ago)

Interpretation

Pulls small samples toward a 25% prior and leans on recent matches once enough are available.

Limitations

A heuristic blend, not a calibrated probability. Partners and event thresholds affect it.

Related metrics

Matches & predictions

2 metrics

Win probability

Model estimate of the red alliance’s chance to win, from each team’s Elo before the match.

Scope
Match
Higher is better
Neutral
Unit
%
Range
0–100%
Data required
Elo ratings for all four teams.

Formula

P(red) = 1 / (1 + 10^((mean Elo blue − mean Elo red) / 400))

Interpretation

An estimate, not a certainty. A 70% favourite is expected to lose about three times in ten. Event-page predictions use only results posted before the match was scheduled.

Example

Red alliance 1700 vs blue 1550: P(red) = 70%.

Limitations

Not calibrated. On 2025–26 data the model was under-confident: when it gave red 50–60%, red won 61.7% of those matches; at 40–50%, red won 40.4%. Season-start ratings are all 1500, so early predictions are close to 50%.

Walk-forward backtest on 2025–26 DECODE matches. Brier score and log loss: lower is better.
EvaluationMatchesBrierLog lossWinner accuracy
Elo, full season35,9260.22690.649560.4%
50% benchmark, full season35,9260.24830.6931—
Elo, later 30% of matches10,7780.22120.636765.3%
50% benchmark, later 30%10,7780.24860.6931—

Evaluated on 2026-09-23 with an earlier revision of the model. That version dropped a whole match when any participant was flagged surrogate, no-show, or disqualified; the current one skips only the flagged team. Treat these figures as evidence that Elo beats a coin flip, not as a benchmark for the model as it runs today. The 50% benchmark names no winner, so it has no accuracy.

Calibration was loose. Red’s observed outcome rate was 40.4% in matches rated 40–50% and 61.7% in matches rated 50–60%, so predictions sat too close to 50%. Scrimmage and off-season events were left out of the backtest.

Related metrics

Upset

A match the pre-match favourite lost, when the favourite had at least a 60% win probability.

Scope
Match
Higher is better
Neutral
Data required
Pre-match predictions.

Formula

Upset if P(favourite) ≥ 60% and the favourite lost on official final scoresUpset size = favourite’s pre-match win probability

Interpretation

Bigger favourites losing are bigger upsets. Frequent upsets at an event suggest the field’s ratings were not yet settled.

Example

A 78% favourite loses: an upset of size 78%.

Limitations

Inherits every limitation of the win-probability model.

Related metrics

Events

2 metrics

Field strength

Average Elo of the eight highest-rated teams at an event.

Scope
Event
Higher is better
Yes
Unit
rating
Range
Rating scale.
Data required
Elo for the event’s teams.

Formula

Field strength = mean Elo of the top 8 teams at the event

Interpretation

Captures how hard it is to win the event. The top 8 are used because alliance captains and first picks decide most playoffs.

Example

A top-8 average near 1800 is an elite field: those teams are in roughly the top 1% by Elo.

Limitations

Uses current ratings, so it reflects how strong the teams are now, not necessarily on event day.

Related metrics

Event difficulty

Field strength placed on a 1–10 scale against season-wide rating percentiles.

Scope
Event
Higher is better
Neutral
Unit
1–10
Range
1–10
Data required
Field strength and season percentiles.

Formula

Map top-8 average Elo onto season team-Elo percentiles by linear interpolation: p10 → 1, p50 → 3, p70 → 5, p90 → 8, p99 and above → 10The completed-event (actual) score adds a small modifier from how the event’s non-penalty scores rank season-wideLabels: ≥ 9 Elite, ≥ 6.5 High, ≥ 4 Moderate, ≥ 2 Low, otherwise Beginner

Interpretation

A quick label for comparing events. The underlying field strength is the more precise number.

Limitations

Compares an average of the top 8 against individual-team percentiles, which pushes scores upward. Uses current ratings.

Related metrics

What we don’t compute

These are deliberately not shown, with the reason for each. A dash on the site means we have no defensible number, never zero.

Foul attribution
The source records penalty points per alliance and does not say which team caused them. A per-team foul rating would be a guess, so there is no foul OPR or foul rate.
Clutch
Clutch would need each team’s endgame contribution, which is neither defined nor available in the match data.
Schedule luck and grades
Earlier versions graded schedules with final ratings and placeholder values. A fair grade needs each opponent’s rating at match time and a validated outcome model. Cached grades from those versions are ignored.
Team upset factor
Needs pre-match ratings stored for every match and a way to credit an upset to one team on an alliance. Individual upsets are still flagged per match.
Recency-weighted Elo
Ratings move only when a match result is replayed, never with the passing of time, so the same match data always produces the same ratings. We do not publish a form-adjusted or time-decayed variant.
Momentum and form
Not shown. The stored values come from a retired formula and do not match any definition we publish.
Score forecasts and margins
There is no validated score model, and OPR is descriptive rather than a forecast. Win probability is the only match-result prediction we show.
Cross-season ratings
Every season restarts at 1500 and is not normalized against other seasons, so ratings from different seasons cannot be compared.