Transparency.

The rubric

The scoring rubric is published before any grading happens, and it never changes direction: 1–2 Broken, 3–4 Uncanny, 5–6 Passable in cuts, 7–8 Broadcast-adjacent, 9–10 Indistinguishable. Every take is scored against the same anchors by a blind panel, and results are reported with 95% confidence intervals.

Ties, not false precision

Where the paired-difference confidence intervals of adjacent ranks overlap, the leaderboard reports a statistical tie rather than a ranking the data cannot support.

Data discipline

Numbers are computed, never generated. Nothing on this site renders from hand-typed values: the model roster comes from the edition’s annex, the ratings count comes from the scores table, and the leaderboard renders only when computed results exist. Before grading day the site shows structure — empty frames, row shells — and labels it as structure.

Contamination, in the open

Δ contamination — Difference vs. equivalent secret prompts; unusually large values indicate a possible contamination effect and warrant investigation. The secret canary prompts behind this measure never publish; their results are what make the flag possible.

What this site measures, and what it never does

This site measures its own usage with first-party events only — no third-party trackers, no analytics cookies, no advertising identifiers of any kind. Each page load and each reader action (playing a take, switching takes, stepping frames, downloading data, copying a citation) writes one event: the event name, the page path, and nothing that names you. The session key lives in JavaScript memory and dies with the tab; nothing is written to cookies or local storage. If your browser sends Do Not Track or Global Privacy Control, the site goes silent — no events, no exceptions. Campaign links we publish carry five UTM parameters; they are tab-scoped so a download two clicks later stays attributable, and they are campaign parameters, not identifiers. Event data is stored on our own server and is readable only by the site’s administration.

Nothing fictional

No status lamps, no decorative waveforms, no live indicators, no mock players. If a gauge appears on this site, it is displaying computed data — or it does not exist.

Contact

The project answers in the open on X.