How a finding becomes a rule in Apex — and how it falls. Updated 19 September 2026.
A view counts only when it says what would prove it wrong. Every assessment Apex publishes — on the world, on capital rotation, on a market — carries three things: the evidence behind it, a grade of how strong that evidence is, and an invalidation: the observable condition under which Apex changes its mind. A view without an invalidation is an opinion, and Apex does not publish opinions.
Findings live in a register. Today it holds 51 findings; 27 of them are marked FAILED and stay visible on purpose. Each finding is at one stage:
Promotion from one stage to the next is a human decision, made on the record. Nothing is promoted on a story.
Every number Apex shows carries its sample size. Under 10 observations is labelled INSUFFICIENT DATA and no rate is shown at all. Under 30 is labelled SMALL. A hit rate always comes with its 95 % confidence interval, because 100 % on three trades means anywhere between 44 % and 100 %.
“Gold rose the day after” means nothing if gold rises on more than half of all days anyway. Every conditional result is measured against the unconditional baseline of the same instrument over the same period.
The record is cut in two by date — not by row count, so that one heavy day cannot sit in both halves and test itself. A finding passes only if the effect keeps its sign in both halves. Most discoveries fail here: they were one good week.
Every hypothesis Apex has ever tested is logged, and significance is judged against the whole family of tests (a Bonferroni correction), not against a single test in isolation. When 276 tests produce 27 with p < 0.05, that is roughly what chance alone produces, and Apex says so.
An execution rule is measured against its own break-even — the real exchange fee for that instrument plus measured slippage — never against zero. A rule that is right 55 % of the time and moves less than its cost is a losing rule, and is labelled as one.
Before a result is believed, the dataset is split at the dates of known fixes and data outages. A striking pattern that lives entirely on one side of a bug fix is an artefact, not a finding. System-fault closures are excluded from every measurement and marked as such in the record.
Apex runs its own execution engine on a fixed research account, using exchange prices, real fees and measured slippage. The scope is deliberately narrow and the record — every trade, every rule change, every drawdown — starts on 26 August 2026 and is never rewritten. Corrections are applied at read time and documented. The engine exists to test the method in the open, not to trade for anyone: Apex does not manage money and does not give investment advice.
Publicly, Apex shows the state of each read and its definition. Members see the qualitative evidence, Apex's interpretation, and what would change the view. The scoring behind the reads — the inputs, thresholds and weights — stays inside Apex. This is the boundary between evidence, which is shared, and mechanism, which is proprietary.
Apex records its own mistakes in the same journal as its findings: a fee field misread as missing, a Yahoo bar that vanished silently, a rule promoted on a sample that turned out to be one bug. Each is written down with the date, the cause and the fix, because a method that hides its errors cannot be trusted with its successes.
See also: About Apex · Pricing · The Apex read