📊 Published every week

Backtests

Four records, one per indicator per market. Every week we publish what each indicator actually showed, and what price did next, including the calls that went wrong and the sessions it declined to call at all.

Nothing on these pages was in the sample. Each set of tables was measured on history ending before the first week reviewed, so every session published here is out of sample by construction. That is the whole point of writing the record in public rather than showing a curated backtest.

Nasdaq

NQ · MNQ

Tables measured on 5.9 million 1-minute bars, 2009 to 2026. Records run from w/c 3 August 2026.

Atlas IB · NQ

Initial Balance backtests

Every 09:30 window at both durations, scored on the rail it favoured and on the narrower set it was willing to call.

7 weeks published · 70 windows

Open the record →
Atlas Sessions · NQ

Sessions backtests

Every NY AM session captured at the open and again at the close, scored on first sweep, four-way outcome and each level independently.

7 weeks published · 35 sessions

Open the record →

Gold

GC · MGC

Tables measured separately on gold's own history, not transferred from the Nasdaq work. Records run from w/c 10 August 2026.

Atlas IB · Gold

Initial Balance backtests, gold

Every 09:00 window at both durations. Gold's best window is not its busiest one, and the record explains why.

6 weeks published · 60 windows

Open the record →
Atlas Sessions · Gold

Sessions backtests, gold

Every US AM session, open and close. Gold runs on three clocks and its sessions are not the Nasdaq's, so the session map was rebuilt from scratch.

6 weeks published · 30 sessions

Open the record →

How these are scored

  • TranscribedEvery figure on every weekly page is read off the capture of that window or session. Nothing is recomputed afterwards, and the scoring scripts refuse to run if a set of shares does not add up.
  • Out of sampleEach indicator's tables were frozen on history ending before the first published week. No session reviewed was available to the tables when they were built.
  • Abstains countBoth indicators decline to call some sessions. Those are published too, with what happened afterwards, because an abstain that would have been right is as much a part of the record as a call that was.
  • Not a track recordNo entries, exits, position sizing or costs are modelled anywhere. These pages show what the indicator displayed and what price did, not what a trade would have returned.
  • Red folder daysHigh-impact USD releases are marked against the day they fall on, so a week's result can be read with the calendar rather than against it.
A week is not evidence. Ten windows or five sessions cannot confirm a frequency measured over hundreds of thousands, and any single week can flatter or bury a perfectly well-behaved statistic. These records exist so the sample accumulates somewhere you can check it, including the weeks that go badly.
Start free trial See what's inside