Every level SPECTRE publishes is set in advance and timestamped, then scored against what the tape actually did. This page describes the scoring exactly as it is computed — the tests, the thresholds and the limits — so you can judge the record rather than take our word for it.
A level is scored on intraday bars for the session it serves — the regular trading hours for futures, the whole UTC day for crypto — not on a single end-of-day close. A level counts as reacted if, after price first tagged it, the market moved at least 0.25% in the level's favour at any point in that session (up off a support, down off a resistance).
The reaction rate is simply reacted touches divided by total touches, computed the same way for every family, so the families are directly comparable. That is the headline number on the coverage panel. It is deliberately a reaction measure, not a trade result: it asks whether the level produced an observable move, not whether any particular position would have profited.
Reaction rate is honest but blunt: within one instrument the families look similar. So we also grade every individual line with a symmetric 1:1 test — ±0.25% — and rank the lines best to worst per instrument. Only decided touches count: if a session's bar range decides a winner in either direction off the line, it is a win or a loss; if neither side ever moves the full 0.25%, the touch is excluded rather than scored as a loss.
A line needs at least 80 decided touches before it is published in the scorecard. Below that the sample is too thin to mean anything, so the line simply carries no number. Where a line qualifies, its measured percent (an integer, no decimals) is shown beside the level in the terminal — and it is public, so a free member sees it on the levels they can see. The laggards are shown alongside the leaders on purpose.
The structure families — pivots, value areas, VWAP, prior periods, opening range and SMC — are deterministic from price and volume history, so we can recompute exactly what the board would have shown on any past session and score it on the session it served. With no look-ahead, that turns a few weeks of published sessions into roughly 14 months and on the order of 50,000+ touches, measured the same way as the live numbers.
Gamma is different. The option-chain history behind the flip and the walls is much shorter, so the deep backfill does not cover it — the live scorer owns the gamma categories, over the window it actually has. And golden levels, the confluence constructs re-derived each session, are not given a per-line score at all: a per-type score on a level that changes every session would be meaningless.
Read the full methodology, see the public record on the track-record panel, or start with what gamma levels are.