Accuracy

How Accurate Is FPL Tactics?

We replayed four full Premier League seasons, 2022/23 to 2025/26, with the planner as it runs today. Every headline test starts from a squad a manager could actually pick before the season, and each season’s model was trained only on the seasons before it. This page covers what to expect, the best case, what is not included, and the experiments that didn’t work.

2218 ptsSteady’s average season from a realistic pre-season squad (4 test seasons, no chips)
+50 ptsClimber’s lift per season over Steady from a deliberately weak squad (ahead on 14 of 20 squads)
2410 ptsBest case, not a forecast: Steady from a squad picked with hindsight no manager has at GW1
29–46%Bias reduction from the xGI + Elo ensemble vs an Elo-only baseline

What to expect: about 2218 points a season from a good pre-season squad, before any chip is played. Third-party estimates put a top-100k finish at roughly 2150–2250 points and a top-10k finish at roughly 2250–2400. These are estimates, and the cut-offs move every season, but on them the planner lands around top-100k level without a single chip. Chips are not in these totals: a manager who plays Wildcard, Free Hit, Bench Boost and Triple Captain well should score more, but we have not measured how much.

How we tested

Four test seasons, walk-forward. When we test a season, the model has never seen it: it is trained only on the seasons before.

  • Testing on 2022/23 → the projection model was trained on 2019/20, 2020/21, and 2021/22 only.
  • Testing on 2023/24 → training adds 2022/23.
  • Testing on 2024/25 → training adds 2023/24.
  • Testing on 2025/26 → training adds 2024/25.

Training on a season and then testing on it flatters a model, because it has seen the answers. Walk-forward closes that gap: each test season is held strictly out of training.

Clean data. In November 2026 we checked the stored data season by season and rebuilt it. Missing expected-goals rows were filled in, players stored under two identities were merged, double gameweeks were stored as two matches, and weeks where expected goals were never published now count as unknown rather than zero. Every number in the headline results was re-run on the rebuilt data.

Realistic starting squads. Each test starts from a pre-season squad picked only from what a manager knows at the GW1 deadline: each player’s Fantasy Rating going into the season and his expected goals and assists from the season before. Every squad fits the real rules: £100.0m, two goalkeepers, five defenders, five midfielders, three forwards, and no more than three players from one club. We built five good squads and five deliberately bad squads for each season, 40 in all.

The rules. The simulator enforces every FPL rule that matters: the budget cap, the three-per-club limit, free transfers, the −4 hit cost, the half-rise sell mechanic, and bench autosubs. Each strategy makes its own transfers, captain picks and bench calls gameweek by gameweek, against the real fixtures and the real prices at the time. No chips are played. Re-running any test gives exactly the same result.

Headline results

From a realistic pre-season squad (five good squads a season, 20 in all):

Strategy2022/232023/242024/252025/26AverageTransfersHits
Steady (live)21262321220922182218360
Steady, no flags21262321219322102212360
Climber, no flags21432235219821862190403

Average net points a season (hits deducted), with transfers and hits per season, no chips. “Live” is Steady as the live planner runs it, reading FPL’s injury and availability flags; we have those stored only from 2024/25, so 2022/23 and 2023/24 run without them. The two “no flags” rows compare the strategies like for like.

From a deliberately weak squad (five bad squads a season, 20 in all; each costs about £66–68m, so the planner has a rebuild to do):

Strategy2022/232023/242024/252025/26AverageTransfersHits
Climber21412105220320792132447
Steady21562096206720072082360
Climber − Steady−15+9+136+72+50

Transfers and hits are per season. Both without availability flags, no chips.

Three things to take from this:

  • From a good squad, expect about 2200 points a season before chips. Steady averaged 2218 as the live planner runs it. Single seasons vary: from 2126 in 2022/23 to 2321 in 2023/24.
  • From a good squad, the two strategies finish level. Like for like, Climber ended 22 points a season behind Steady and came out ahead on 8 of the 20 squads (season by season +18, −86, +5, −24). A gap that size is within the noise of a test this small, so we read it as level rather than as a win for Steady.
  • From a weak squad, Climber is worth about 50 points a season. It came out ahead on 14 of the 20 squads, and by 136 points a season on average in its best season (2024/25). A planner that may take a −4 can dig out of a hole faster; one that can’t has to wait for free transfers.

The best case: a squad picked with hindsight

Earlier versions of this page led with a stronger starting squad: the 15 players who scored the most points over gameweeks 1–3, fitted inside £100m. That page reported 2391 a season for Climber and 2372 for Steady from it, and called that comfortably inside top-10k territory. No manager can pick that squad: at the GW1 deadline nobody knows who will score in the first three weeks. So it is not what you should expect, and it is no longer our headline.

We still show it, re-run on the rebuilt data with today’s planner, as a ceiling: what the planner does from about the best start possible.

Strategy2022/232023/242024/252025/26Average
Steady23772375248923972410
Climber23892291238723302349

One hindsight squad a season, no chips.

From this squad the planner does at least as well as the old page said, so the planner itself has held up. The difference between 2410 and the 2218 above is the starting squad, not the planner: picking players with hindsight is worth about 200 points a season on its own. With only one squad a season these figures are also noisy, as the caveats below explain.

How the settings were chosen

Both strategies look six gameweeks ahead. Climber asks every transfer to gain 2.5 points and may take up to two −4 hits a gameweek. Steady never takes a hit and asks less of a transfer the more free transfers it has banked; the methodology explains how.

We first picked these settings from a grid of 48 combinations of look-ahead, transfer threshold and hit limit, run before the data rebuild from the hindsight squads above. On the rebuilt data and realistic squads we then re-tested the main choices on two earlier seasons, 2020/21 and 2021/22, and checked anything promising on the four test seasons:

  • Six weeks of look-ahead beat four and eight. For Steady, four weeks cost about 32 points a season and eight about 20. For Climber both did clearly worse from weak squads, by about 77 and 54 points a season. For Steady, five and seven weeks were within the noise of six; for Climber they did worse.
  • Lower transfer bars for Steady did not hold up. They looked promising on the two earlier seasons, then finished 17 points a season behind the shipped bars on the four test seasons, behind in every one, so they did not ship.
  • Steady’s wider planning is no worse than the old Steady. Banking free transfers and always considering every premium scored 11 points a season more than the older, simpler Steady on the rebuilt data. That is within the noise, so we count it as level rather than better.

xGI ensemble validation

We also tested the projection model on its own, separately from the strategy backtest. First we trained it on 2022/23 and 2023/24 and checked it against 2024/25, a season it had never seen. Then we trained it through 2024/25 and checked it against 2025/26. That is about 7,800 player performances in the first test and 5,900 in the second. Each range below covers both tests:

MetricResultWhat it means
Bias reduction29 – 46%Bias is how far off-centre the projections sit. Elo on its own guesses too low for almost every player, by about 1.3 to 1.8 points a match. Adding xGI closes most of that gap. Attackers gain the most: they go from around 1.7 points too low to roughly spot on.
RMSE reduction5 – 6%This is the typical size of a miss, counting big misses heavily. The ensemble lands closer to the points players actually score.
MAE1 – 3% worseThis is the average size of a miss on a single match, and it gets slightly worse. That is the trade we chose. Players tend to either haul or blank, so a projection sitting nearer the true average can end up further from any one score. Across a run of gameweeks, guessing low adds up while single-match misses cancel out, so the bias fix is worth more.

Component-Elo v2d (which added the defensive-contribution component for the 2025/26 CBIT rule) originally shipped at T = 3.0 under the previous archetype. The walk-forward redesign re-tuned the threshold down to T = 2.5 alongside the horizon move from h=5 to h=6.

Minutes and appearance points. How long a player is on the pitch drives most of his projection, and FPL pays 1 point for playing and 2 for playing an hour. The projection used to pay the full 2 to anyone expected to play an hour or more, which over-rates starters: even a player expected to play the full 90 misses the whole match about one week in nine. A learned model of minutes now gives each outfield player the chance he plays an hour or more, the chance he plays less, and how long he is likely to be on, and appearance points count at their average value. Trained only on the seasons before each one it was tested on, it cut the error of a player’s projected points over weeks 3–6 of a plan by 8–10% in each of the four seasons from 2022/23 to 2025/26. In the planner backtest over those seasons, from 11 starting squads each, Climber scored about 96 more points a season and came out ahead in 40 of the 44 runs; Steady gained about 130. Goalkeepers keep the old rule.

Experiments that failed

An accuracy page that only lists wins is a marketing page. The experiments below were run, didn’t work, and changed how we built the planner.

They all ran before the November 2026 data rebuild, so their point figures come from the older data and we have not re-run them.

Differential strategy - failed

We tested two ways of leaning into differentials. The first was a gated multiplier that boosted projections for already-top-25%-rated players who also had low ownership. The second was a bold-differential finder that applied a concentrated 1.5× boost to the top-K candidates per gameweek, ranked by projection × (1 − ownership). Neither beat the baseline. Both lost 64–131 points per season versus the h=4 baseline they were tested against, and 180–248 versus the h=5 no-hits configuration that was Steady at the time.

The diagnosis: ownership is carrying signal we aren’t modelling. The most-owned players are most-owned because they’re the highest-projected ones, and boosting low-ownership picks trades the template players the model already likes for “better-than-projected” alternatives that don’t actually deliver. The idea might still work against a different objective - owning a low-ownership hauler moves you up the league standings even if it doesn’t maximise raw points - but against a points-EV objective, it loses.

Elo seed/retention tuning - failed

We tested lowering the initial Elo for new players and resetting the state more aggressively at the start of each season. Two variants (V4: median seeds plus reset; V6: seeds only) produced visibly better rank correlation with real outcomes, but lost 50–75 points per season in the strategy backtest.

The diagnosis: high seed Elos are load-bearing for how the optimiser thinks about unknowns. A newly transferred-in player needs to look attractive enough for the MILP to take a position on him, even when his underlying Elo is uncertain. Lower seeds give a more honest rating but starve the optimiser of plausible targets. Reverted.

S6 hold-3 + injury-react - superseded

An earlier sweep flagged a hold-3 + injury-react trigger combination as the first robustly winning strategy we’d found - +5% mean, 3/3 wins. Under the current ensemble, the same strategy loses 3.2%. The squad-alerts banner that originally shipped on that finding still exists, but its rationale is stale: the alert is harmless and adds no measurable lift.

Shorter-window h=4 archetype - retired

For most of the model’s evolution we shipped a shorter-window (h=4) variant, marketed as a tighter, more predictable alternative to the longer-window default. The stratified-start sweep retired it: across 16 different starting squads it lost to Climber 15 times, and it couldn’t even back up its “tighter range” pitch - variance was equal or worse. The walk-forward results agree. From a weak start, the best h=4 cell sits at rank 11 of 48; from a strong start the best it manages is rank 29. h=4 simply doesn’t earn a slot.

Climber breakout state machine - failed

When we first identified the Climber profile, an obvious worry was that allowing hits would start leaking points once the squad had actually climbed out of the hole. We built a state machine to flip Climber back to Steady’s parameters once the squad’s projected starting XI total crossed a quality threshold, and tested thresholds from 50 to 70 points per gameweek.

Nothing worked. Low thresholds (50, 55) fired too early and lost 20–30 points by cutting transfers short before they’d paid off. High thresholds (≥65) never fired across three seasons - by the state machine’s measure, Climber’s squad stayed “still climbing” for the entire season. The hits-allowed Climber stays positive-EV all the way through; there’s no point where it locks into bad decisions and needs to be rescued.

Old (h=5) archetypes - superseded by walk-forward redesign

The previous shipped archetypes were Climber = h=5 / T=3.5 / X=2 and Steady = h=5 / T=3.0 / X=0. They were chosen from a 144-run stratified-start sweep using model parameters fit on the same seasons the strategies were being tested on - that’s the “peek at the answer” problem walk-forward is designed to fix.

When we re-ran those exact cells under proper walk-forward, they slipped: old Climber ranks 14 / 48 from a weak start (mean 2162), and old Steady ranks 46 / 48 (mean 2043). Beaten on both ends by H6 T2.5 X2 and H6 T2.5 X0. The redesign moved the horizon from 5 to 6 and the threshold from 3.5/3.0 down to 2.5, which had the bonus of unifying the two archetypes around shared (h, T) values - the hits toggle is now the only thing that differs between them.

Limitations and caveats

Four seasons and 20 squads is still a small sample. Single seasons swing a lot: Steady’s average from realistic squads ran from 2126 to 2321. We read gaps of 20–30 points a season between two strategies as level, and say so where it applies.

Small changes move single results. Re-running a test gives the same answer, but a change that leaves the planner’s aims untouched can still lead the solver to a different, near-equal plan. That alone moved one squad’s season by tens of points, and up to about 110 on the hindsight squads. It is why we average over many squads rather than trusting any one season.

No chip play. None of the numbers on this page include Wildcard, Free Hit, Bench Boost or Triple Captain. The planner plans around a chip you set, but these tests never play one.

Availability flags only from 2024/25. The live planner reads FPL’s injury and availability flags, and we have them stored only from 2024/25, so the first two test seasons may understate it. How much the flags are worth varies: on the squads above they added 8 to 16 points a season to Steady, and in a separate test of Climber on other realistic squads about 85 to 90.

The settings were first picked on these seasons. The grid that chose six weeks of look-ahead and the 2.5 threshold ran over three of the four test seasons, before the data rebuild. The projection model is held out, but the settings are not fully: the re-tests on 2020/21 and 2021/22 above are the check on them.

Rank cut-offs are estimates. The top-100k and top-10k ranges come from third-party estimates, not FPL’s own standings, and they move from season to season.

What we want to test next

  • Chip-aware planning. Letting the planner choose chip weeks, and backtesting chips through a season, is the biggest thing missing from these totals.
  • Real rank cut-offs. Putting each test season’s actual top-10k and top-100k totals next to the planner’s, in place of third-party estimates.
  • More starting squads. More squads a season would narrow the noise around the gaps between strategies.

What we tested over the 2025/26 end of season

These results came before the November 2026 data rebuild; the headline results above replace them.

All the work below ran during the closing weeks of the 2025/26 Premier League season. Pinning each finding to a specific calendar day would oversell how clean the process was - most of these threads ran in parallel for weeks before the numbers settled. Treating them as one cluster of “end-of-season research” is more honest than a tidy timeline.

Walk-forward validation initiative. Built the validation protocol, the strict-past training discipline, and a tag-column separation between the production model fit and the research model fit. Ran an Understat backfill to unlock 2019–2022 as training seasons (the FPL feed only inlined per-match xG from 2022/23 onward), then fixed a data-attribution bug in the historical backfill that had been silently scrambling player identities across seasons. Then fixed a related stub-row issue that had been dropping 923 historic Premier League regulars from training - Kane, KDB, Sterling, Vardy, that whole tier. With those fixes in, we ran a pair of 48-cell strategy grids (bottom-15 and top-15 starts) under strict walk-forward and used the results to redesign the archetypes.

Climber/Steady redesign. The shipped Climber moved from h=5 / T=3.5 / hits ≤ 2 to h=6 / T=2.5 / hits ≤ 2. The shipped Steady moved from h=5 / T=3.0 / no-hits to h=6 / T=2.5 / no-hits. The two archetypes now share horizon and threshold; the hits toggle is the only thing that differs. From a weak start, Climber beats Steady by +94 ± 69 points/season, winning two folds decisively and drawing the third; from a strong start Climber keeps a smaller edge of around 19 points.

Differential sweep. Two differential mechanisms tested - gated multiplier, bold-differential finder. Both lost across all three test seasons. A previously offered “Balanced” middle option was retired off the back of this work.

Component-Elo v2d. The Fantasy Rating gained a fifth Elo component for defensive contribution - the new 2025/26 CBIT rule that earns defenders and midfielders 2 points for hitting the 10-event threshold. Won 3/3 seasons at every threshold we tried in {2.5, 3.0, 3.5, 4.0, 4.5} under production-fit testing. The walk-forward redesign re-tuned the threshold to T = 2.5 alongside the horizon move from h=5 to h=6.

Streak-boost K-multiplier. Low-seed players who are over-performing their cohort now get an accelerated Elo update. Improved the visible player ratings; cost about 10 points per season in the strategy backtest - an accepted trade for more accurate UI numbers.

DNP-weight Elo. Sub-21-minute appearances now apply partial decay rather than a full Elo update, so a player returning from injury for a 12-minute cameo doesn’t get penalised as if he’d played a full 90.

Simulation infrastructure. Wall time on the full 48-cell × 3-fold walk-forward sweep is around 18 minutes on a 12-core machine, after a top-80 player pre-filter and a persistent worker pool. Speed was made a first-class concern so future experiments don’t get discouraged by slow infrastructure.