EXP-16: CM-7 versus CM-5: stopped partial comparison

This was scheduled for 100 games and stopped by the user. It is not a completed 100-game comparison. The archive includes the 58 completed PGNs and the interrupted PGN marked unfinished; unplayed games have no fabricated records. Do not treat the interrupted game as a draw or a loss.

CM-5 White: 13 wins, 10 draws, 6 losses and one interruption. CM-5 Black: 11 wins, 7 draws and 11 losses. Completed-game points: CM-5 32.5, CM-7 25.5. These partial observations inform the choice to return to CM-5; they do not establish a definitive ranking.

Full-history CM-5 and state-conditioned CM-7 use the recorded comparison protocol and adaptive iterative Rust helper. Exact precision, player checkpoints, decoding, suite and helper settings are in the stopped-run record; both-player provenance stays attached to each game.

These runs compare complete systems on selected fixed openings. They do not isolate input representation, supervision density, curriculum, initialization, inherited training, precision or helper/search configuration. They establish neither causal proof nor universal strength ranking. Unknown bounded searches are not proof of no mate.

All 59 attempted PGNs, including interruption · Comparison and partial audit implementation · Replay.