Tovellan research
The canonical 0.0000 → 0.7083 (Δ +0.7083) row is a deterministic reference on frozen GST-Bench v1.0.0. It verifies release delivery and held-out exclusion, not model generalisation.
Last recorded release
Measured deterministic reference on GST-Bench v1.0.0 - 48 records, with the held-out split excluded from delivery. Recorded 7 August 2026.
baseline
0.0000
improved
0.7083
How to read this run
The 0.0000 to 0.7083 result verifies release delivery and held-out exclusion for a deterministic reference subject. It is not a live-model result or evidence that a model generalises. Protected materials remain private.61f214e7272095b87fe8e772856342b64aee85e65dce75e0e27e532519aafc73Recorded 7 August 2026. Staff-run deterministic reference subject on the frozen v1.0.0 release, recorded before the planned Tovellan Masters program. The result is a deterministic delivery/exclusion reference, not model generalisation - see the methodology.
Checking the live eval registry for newer runs…
The content hash verifies the frozen release identity, while the release record publishes train and held-out counts. Membership, held-out content and internal evaluation methods stay private.
Held-out membership, identifiers and content are not published or delivered. The recorded reference verifies exclusion rather than model generalisation.
The release record identifies the corpus provenance, version, counts, content hash, and the exact scope of the recorded result.
The public release evidence and its limits are documented on the methodology page.