Method

A trader's reading of the market, held to a scientist's standard. This page explains how every result on this site is produced and how to read it.

Probability, not prediction

Every read we study is a probability that one level is reached before another. It is never a trade instruction. We do not publish entries, stops, strategies or model settings.

Why a hit rate alone means nothing

A call on the nearer of two levels is right more often by geometry alone. With 3 points to the target and 5 points to the other level, a random walk reaches the target first 62.5% of the time, with zero edge. When the call is wrong, the loss is larger, and in a random walk that exactly cancels the higher hit rate.

So we never show a hit rate on its own. Each one comes with:

• the base rate: how often the outcome happens anyway;
• the distance-and-time reference: what the distances to the levels and the clock alone predict for the same moments;
• the room: how far price still was from the target when the call was made.

The edge we report is the gap between the hit rate and that reference, measured at the moment each call fires.

Rebuild the auction

Years of tick-level trades for ES, NQ and GC, rebuilt minute by minute into the developing profile a trader would have seen at that moment, with nothing from the future.

Define it like a trader

Each situation, such as price outside value or price in the thin gap between two balance areas, is defined by an experienced trader and checked by hand against real sessions before any model sees it.

Rules before results

• Pass criteria are written down before a test is run. Any change after results is logged as an amendment, with its timing.
• Models are fitted on earlier years, checked on later years, and confirmed once on a full year that was held back untouched until the end.
• Since late September 2026, results are logged forward and compared with the confirmed numbers.

Honest inputs

Every input is checked to make sure it uses only what was known at that minute. This check once caught an input that had used the future, and that claim was withdrawn.

Statuses

• ConfirmedPassed rules written in advance, including one check on a held-back year.
• Forward testPassed on the held-back year with a caveat. Forward results decide.
• WithdrawnDid not survive a later check. Kept on the site, with the reason.

Failed ideas are reported as failures, including our own.