EXP-1 / completed / evaluation
EXP-1: Untrained policy check
Establish an untrained legal-candidate policy reference before interpreting trained model play.
Non-reasoning next-move model: complete game history from the standard start, three-token causal objective, no thought or search supervision. Follow data → tokenizer → training → harness → bot bridge.
Baseline
Random initialization, not a trained model. No mate helper.
Data
Two recorded local Sunfish games, one per model color. This is a small sanity check, not a matched strength benchmark.
Configuration
Exact recorded configuration
{
"checkpoint_sha256": "711f78adbe1be14172921e5dad5696996889cc1ee31915cacb028ae0337c3396",
"games": 2,
"mate_helper": false
}Compute
Two completed local games; exact aggregate compute was not retained in this public record.
Results
| Metric | Value |
|---|---|
| Games | 2 |
| Wins | 0 |
| Losses | 2 |
What this does not establish
- Only two games; no strength inference.
- Exact historical initialization source is not available in the public snapshot archive. The linked harness is related source, not an exact replay claim.
Provenance
Browse the recorded source snapshot →
Checkpoint SHA-256 711f78adbe1be14172921e5dad5696996889cc1ee31915cacb028ae0337c3396
Download experiment record (JSON)
Evidence: Public random-reference PGNs
Original artifact SHA-256 77a6e33f268e0533cd39f87b6e0f25588acdad20e944b858c0baa2f58e27a00d
Selected public fields; private paths and process/account details excluded.
{
"game_ids": [
"random-baseline-seed20261028-white",
"random-baseline-seed20261028-black"
],
"checkpoint_sha256": "711f78adbe1be14172921e5dad5696996889cc1ee31915cacb028ae0337c3396"
}