Glossary

The competition

Unit — one task, one initial state, one demonstration, one seed. Both models are shown the same unit; each is scored on whether the goal holds at the end.

Axis — one of the 4 kinds of change a unit is perturbed by: Spatial change, Environment change, Object change, Composition. Each is scored as a success rate.

Score — the mean of the four axis success rates. Rendered as a percentage.

Duel — two models run against an identical list of units, drawn evenly across the four axes.

Score margin — how far a challenger's average must exceed the reigning model's to take the crown: +3.0 pts.

Crown — what the winner of a duel holds. There is one.

Reigning model — the model currently holding the crown. It defends against every challenger in turn.

Baseline — the organizer's own checkpoint, BPP public checkpoint (austinpatel/libero, converted), crowned unopposed at genesis. Marked as such wherever it appears.

Challenger — a submitted model waiting in the queue, or the non-incumbent side of a duel.

Queue — the ordered list of challengers. One entry per submission key; re-submitting sends you to the back.

Reign lineage — the public chain of predecessors for the crown, and the duel that ended each reign.

Void unit — a unit an infrastructure fault hit on either side. Excluded from both scores.

Paired outcome — won, lost or tied: whether the challenger succeeded where the reigning model failed, the reverse, or both did the same. Published as a diagnostic; never decides the crown.

The protocol

Demonstration — one continuous recording of the task, shown to the model as its whole prompt. Always recorded from a different initial state than the one the model is scored from.

Prompt — what the model conditions on: the demonstration, chunked by the validator. There is no language.

Initial state — the scene the model is scored from. Never the demonstration's.

Perturbation — the change a unit carries: a displacement and level on the spatial axis, a table and lighting on the environment axis, an object swap on the object axis, a chain of steps on the composition axis. Recorded on every unit.

Pool — the fixed, content-addressed set of tasks, initial states and demonstrations units are drawn from. Its identifier is published on every record.

Duel identifier — a hash of the specification version, the track and both models' keys and revisions. Every unit's seed derives from it, so a draw is reproducible after the fact.

Protocol fingerprint — a hash of the specification file a duel was judged under. Published on every record.

The submission

Checkpoint — model.safetensors, the weights of a bpp_libero_v1 model. Pickles are refused.

Config — config.yaml, the resolved model configuration, which must match the published template except on its mutable keys.

Revision — the commit hash the submission is pinned to. Branch and tag names are refused.

Submission key — a short hash of the repository and revision. The queue holds one entry per key.

Template — the canonical configuration and tensor-shape set for the architecture, owned by the validator. A submission is diffed against it before any weight is loaded.