Author login on the source platform.
Grading STOP of the rule, percent. The hit depends on it — a MISS is the hard stop firing before any fixation, so the same (hold, lock, trailing, author) has DIFFERENT hits at different stops. A full part of the rule identity and MUST be on the line, else those rows are indistinguishable.
hits / ideas, 0..1; zero when the author has no known outcomes. A derived convenience kept on the final product (one number per track line, no duplication) so userspace filters by trust directly (jq 'select(.hitRate > 0.5)') instead of the engine baking a threshold into a banned flag.
Number of the author's hits — ideas whose lock or trailing arm fixation fired BEFORE the hard stop inside the rule's window. A miss is the hard stop firing first, the window expiring, or the candles running out before any fixation. The same author has different hit counts under different rules.
Grading window of the rule — the point's holdMinutes. The track depends on it: the same author has different hits in different windows. On every track line so a grep by author shows the rule.
All the author's directional ideas — every idea counts. An idea cut short by the candle data edge is NOT excluded: running out of candles is a loss (a miss), exactly like the hold window timing out.
Grading lock level of the rule, percent (0 = lock off, fixation is the trailing arm alone). Part of the rule identity — the hit depends on it — so it is on the line. NO thresholds: a track is continuous trust, not a 0/1 flag.
Grading TRAILING take of the rule, percent. The hit depends on it — the trailing arm level (entry/(1 - dir*r)) is one of the two fixations raced against the hard stop, so the same (hold, lock, stop, author) has DIFFERENT hits at different trailing values. A full part of the rule identity and MUST be on the line, else those rows are indistinguishable.
One author's TRACK under ONE grading rule (rule = the point's hold x lock x stop x trailing, the four levels the single profit-before-stop outcome depends on). The raw continuous evidence — ideas, hits, hitRate — over the whole simulated range. There is NO ban verdict and NO threshold: the engine grades, and userspace decides who to trust (that is the whole point of cutting the minAuthorTrack/minAuthorHitRate step — it turned a continuous track into a 0/1 flag and lost information). One line per (rule x author), self-contained for grep/jq.