Lichess.org / CADi Chess / Experiment 1
Does the signal travel?
A fixed cohort. A frozen calibration. A later month held apart. Testing whether a player-relative chess signal carries forward without refitting.
01 / A chronological test
The later month did not choose the model.
Fit
July 2023
Select the 200-player cohort and fit the human-choice reference.
95,300,285 games scannedCalibration
July 2025
Freeze the anomaly cutoffs. No later-month fitting.
93,092,772 games scannedEvaluation
August 2025
Apply the frozen system once. No player replacement or threshold changes.
92,695,519 games scannedThe evaluation archive was absent when the protocol was frozen, then hash-bound before parsing. Later inactivity was observed, not repaired by reselecting players.
02 / What actually ran
A large search. A bounded scoring sample.
281,088,576 games were scanned, not all engine-scored. Stockfish 18 analysed 2,051 retained games at a fixed 10,000-node setting. CADi projected 53,287 eligible decisions into ordered trajectories. Stockfish supplied a move-quality reference, not a verdict about assistance.
| Period | Scanned games | Analysed games | Projected games | Eligible decisions |
|---|---|---|---|---|
| July 2023 | 95,300,285 | 1,200 | 1,016 | 28,511 |
| July 2025 | 93,092,772 | 444 | 441 | 12,742 |
| August 2025 | 92,695,519 | 407 | 401 | 12,034 |
03 / What reached the later month
Coverage is not accuracy.
4.37%
Elevated or high, among scored games
17 of 389 games exceeded the anomaly threshold. These are review signals, not 17 discovered cheating cases. There are no assistance labels against which to measure their correctness.
04 / Did alert volume carry forward?
The threshold stayed frozen.
05 / Fixed players, observed attrition
Missing players stayed missing.
06 / Feature transport
Small measured shifts. A limited conclusion.
| Feature | Standardised mean difference |
|---|---|
| Late top-choice lift | -0.0936 |
| Largest positive rate shift | -0.0350 |
| Longest top-choice run | -0.0343 |
| Contextual-surprisal slope | -0.0199 |
The largest absolute shift was 0.094. This supports a bounded one-month transport finding, not causal identification, superiority to simpler models, enforcement suitability or production economics.
Source and claim boundary
Open games. Bounded claims.
This is CADi research using games from the Lichess open database, whose exports are released under CC0. CADi is not affiliated with or endorsed by Lichess.
An anomaly is not a cheating verdict. Public game records do not establish whether assistance occurred. No cheating probability, accuracy, false-positive rate or causal effect is established by this report. No player names, player-level scores, game links or raw moves are published.
Snapshot: 2 September 2026. Figures are a curated aggregate summary of the completed record, not a new engine run. Later work must receive its own dated report; these results do not update silently.
Aggregate provenance
Website summary SHA-256: 3b4bb539de23ef756e80f7265965214c0a4c346b69fa45115ae63f91d4c63bd6
Completed write-up SHA-256: 3c5df232fdfe636e889c138eeb175ebe4a173167c9513ad469994c2a8145ef79. Values were transcribed from its aggregate tables. The full player-level outputs remain restricted; the website does not claim to reproduce those outputs.