Nonzero Lab Research · Method · 2026-09-05 · 6 min read

How to Run a Blind Chart Review With Any Software

Hide the ticker and date, advance one bar at a time, write each call before the reveal, keep the record. Seven steps, each with its most common mistake.

What exactly is "blind" in a blind review?

Two blinds. The first hides the outcome: until the segment ends, the right side of the chart does not exist. The second hides the origin: you do not know which instrument or which stretch of history you are looking at.

The second blind gets skipped often, and it matters just as much. If you recognize the chart as the March 2020 crash, you already "know" the V-shape that follows; that is recitation, not judgment. With the origin hidden, every call is structurally identical to live trading: the bars that have happened, and nothing else.

Why bother with all this? Because review with a known ending is contaminated by hindsight bias: once people know the outcome, they cannot reconstruct what they thought before knowing, even when explicitly asked to.¹ The mechanism and the experiments behind it are laid out in Why Chart Review Lies to You.

Steps

The whole routine runs on free tools. If your platform can show historical bars and cover the right half of the screen, you have everything. Seven steps, in order, each followed by the way it most commonly goes wrong.

1. Hide the instrument and the date: use your platform's replay mode, or tape paper over the symbol and the time axis, and pick a stretch of history you do not recognize. The common mistake: choosing a famous episode — the moment you recognize the setting, the segment is spent.
2. Advance one bar at a time: step with the arrow keys or the replay step button; never fast-forward, never drag. The common mistake: peeking at the ending "just once" — outcome knowledge cannot be un-seen, and the segment is spent just the same.
3. Write the call before the next bar: at each decision point, record direction, reason, and invalidation — long, short, or flat, because of something visible, wrong if a named level breaks — then advance. The common mistake: writing "looks bullish" — a reason that cannot be checked is not a reason.
4. Keep an uneditable trace: every call, reason, entry, and exit lands on a record you cannot rewrite; paper, a spreadsheet, or a journal app all work. The common mistake: filling in the log from memory afterwards — after the reveal you remember what happened, not why you acted.
5. Review against the record, not against memory: after the reveal, work only from what is written — what you saw, what you missed, whether the reason held. The common mistake: staring at the finished chart and concluding "I saw it coming" — that is narrative, not review.
6. Tag every trade: label each call with a small fixed vocabulary — with or against the trend, planned or impulsive, executed well or poorly. The common mistake: skipping the tags — without dimensions to group by, a hundred sessions are just a diary.
7. Read the sample, not the session: one segment's outcome is mostly luck; read hit rate and payoff only after dozens or a hundred sessions, on a weekly or monthly cadence. The common mistake: sizing up after a few wins, or scrapping the method after a few losses — both verdicts arrive too early.

What does a good written call look like?

The way the call is written decides what can be checked at the reveal. Three elements: direction, reason, invalidation.

An example with hypothetical numbers: "Price 2180, pulled back into the prior low at 2160 and stabilized, long; below 2150 I am wrong, exit." This call can be checked because the direction is explicit, the reason points at something visible on the chart, and the invalidation is a price rather than a mood.

The counterexample is "chart looks strong, bullish." Direction is there, but the reason cannot be checked and no invalidation exists. After the reveal you can neither confirm nor refute it; the session taught nothing. The more specific the call, the less room hindsight has to hide in.

One cheap addition: record your confidence on a one-to-ten scale next to each call. After a month, compare the confidence column against the outcomes. The most confident calls are not always the best calls, and that gap is where calibration starts.

How many sessions before it means anything?

More than intuition suggests. Eighteen wins in thirty sessions still leaves the 95% confidence interval at roughly 42–76%, and separating a true 55% edge from a coin flip takes about 780 sessions.² The arithmetic and the tables behind these numbers are in How Many Trials Do You Need?.

That is not an argument against practicing; it is an argument against early verdicts. While the count is low, the point of the record is to accumulate the sample, not to judge it. Seven straight wins say nothing — a coin flip produces that streak with probability 1 in 128 — and seven straight losses say just as little. Both kinds of single-session emotion are cancelled at the scale of the sample.

What does a weekly rhythm look like?

Run a few sessions a day at a fixed time, and stop when they are done. Stack too many into one day and the calls degrade into clicking. Once a week, sit down with the records and nothing else: hit rate and payoff ratio are the surface, and the more useful cut is by tag — how the with-trend group did, what became of the impulsive ones. Repeat the same tally at the end of the month, reading the trend rather than any single number.

The point of the rhythm is not intensity but continuity: the sample accumulates by the week, and while a missed week costs you no feel, it does break the bookkeeping.