How we measure ourselves

Every read is receipted, then released

Each market read is published at least two hours after it was written, with its symbol, direction, levels, methodology version and a SHA-256 fingerprint of the exact content. A fingerprint means the read cannot be quietly edited after the market moves — recompute the hash and it either matches or it doesn't.

The charts the model actually looked at are preserved and paired to each read by cycle, so "the AI read the chart" is checkable, not marketing.

A hit rate is not a win rate

A read whose entry trigger never printed is counted as no trade — not dropped. Dropping it would turn a hit rate into a win rate, which is a different and untrue claim. Of everything we publish, most entries never trigger; that is the honest shape of a machine that waits for its level.

The book replays every read on its own printed levels: the entry is the level the card showed, the exit is that card's own target or invalidation. No live account is implied and none is claimed.

The losing number stays on the page

The NOW receipt — the reads we said we would take ourselves — currently shows a negative average. It stays visible until it improves. A record you can only see when it flatters us is not a record.

When we find a defect, we publish the fix reasoning too: in August 2026 a 1,936-trial sweep of our own refused-trade geometry showed only 11% of trades ever reached the old 1.5R first target while 45% reached 0.3R — so most positions never banked anything and rode back to the stop. The first-target floor moved to 0.3R because the data said so, not because it looks better.

Cost is measured before edge is claimed

Round-trip fees and spread are measured per coin and published on the cards. Our own research found transaction cost is the dominant effect on quiet, low-volatility coins — which is why the selection engine now scores opportunity (volatility, volume acceleration, movement) and treats liquidity as a minimum gate rather than a ranking signal.

Engine scores and thresholds are deliberately not published. The stage each coin reached and the reason each read gave are the product; the ranking numbers are the strategy.

Most passes, it finds nothing

Every cycle the whole futures board is priced, a fraction survives deterministic gates, at most a handful get a written read, and most passes nothing is worth real money. The funnel counts are public and free forever. A machine that says yes constantly is selling you activity; ours mostly says no, in writing, with the reason.