Chess.com
Top-10 rate3 PGNs22.4%
5 PGNs29.5%
10 PGNs42.9%
Scale ends at 50 % so the two platform panels remain directly comparable.
Coverage and accuracy benchmarks
What the current evidence supports
We test whether a known account appears among the strongest style leads when its held-out games are searched. Results are measured separately for Chess.com and Lichess.
Evidence captured from68 independently verified accounts and 816 trials.
Preparation leads, not proof
These figures support platform-scoped style and preparation leads. They do not establish confirmed identity, guaranteed account ownership, or broad 50/70/80 account-finding reliability.
Production-fidelity holdout
Top-10 style-lead rate
| Platform | PGNs | Trials | Top-10 rate |
|---|---|---|---|
| Chess.com | 3 | 156 | 22.4% |
| Chess.com | 5 | 156 | 29.5% |
| Chess.com | 10 | 156 | 42.9% |
| Lichess | 3 | 116 | 30.2% |
| Lichess | 5 | 116 | 36.2% |
| Lichess | 10 | 116 | 47.4% |
Lichess
Top-10 rate3 PGNs30.2%
5 PGNs36.2%
10 PGNs47.4%
Scale ends at 50 % so the two platform panels remain directly comparable.
Top-10
The held-out account appeared among the first ten fully scored style leads.
Trial
One leave-N-out search for a known linked account using games excluded from its indexed training evidence.
Scope
Results are measured separately within each source platform. They support preparation leads, not confirmed identity or guaranteed account ownership.
How to read this result
Higher rates with more PGNs mean that more style evidence was available in this holdout. Results remain platform-scoped and should be read together with their sample sizes.
The original broad reliability target remains unmet, so Chessco does not advertise it. Stronger claims require a separately approved evidence gate.
See the Trust page for data handling and product safeguards.