Practice & review

Reading time ~5 min · Last updated 2026-07-17

A 90-day contest is roughly 60–200 trades for an active participant. Treated as 200 separate adventures, they teach nothing. Treated as a dataset, they will tell you exactly what your edge is — and more often, exactly where you leak. This article is about building that dataset.

The journal

For every trade, record what the fill data cannot tell you:

• The setup — which pattern/signal from this guide triggered it (flag breakout, funding extreme, sweep-and-reclaim…). One tag per trade forces honesty about whether you had a setup.
• The plan — entry, stop, target, and risk in R (1R = the amount you risked; a trade that made twice its risk is +2R).
• The execution — did you follow the plan? If not, what did you actually do and why?
• The state — one word on your mood: calm, bored, tilted, FOMO. This column becomes embarrassing and then invaluable.

The platform does the bookkeeping half for you: the Transaction Log exports your complete order, trade, and wallet history (including fees and funding) as CSV. Pull it weekly, join it with your setup tags in a spreadsheet, and the analysis below falls out almost for free.

Expectancy — your edge as one number

This is Douglas's probabilities made concrete: a system that loses most of the time and still compounds. Expectancy also diagnoses precisely — a negative number decomposes into "win rate too low" (setup problem) or "losses too big" (discipline problem), which have different cures.

The weekly review

Twenty minutes, same questions every week:

1. Which setup tags made money? Which lost? (Fund the winners, bench the losers.)
2. Was any loss bigger than 1R? Why — gap, slippage, or moved stop?
3. What did the "state" column say on the worst three trades?
4. Rule for next week — one change, not five.

Backtesting mindset

Before risking contest equity on a new idea, scroll back through history and mark every occurrence — entries, stops, outcomes, honestly counted, hindsight resisted (cover the right side of the chart as you go). Twenty historical samples will not prove an edge, but they will kill obviously broken ideas cheaply — and the contest itself is the safest possible forward-test: real market data, real mechanics, zero real capital.