Skip to content
✎ FeedbackNew here?

What being wrong costs

Bottom lineAcross 4,586 resolved trades, the typical one lost a full unit of risk — the median is -1.00 R — and only 21.5% won at all. The top 1% of trades carry 31.7% of every point of positive R, so the average is a story about a handful of outliers, not about a typical trade.
New here? How to read these charts →

A dated study, not a live number. Measured 2026-08-27 over the nifty500 universe — 500 symbols walked, 76,592 rows, 14,474 unresolved rows excluded rather than marked to the last close. It does not update on its own.

Honest boundaryThese are the outcomes of a rule set we have already tested and rejected · descriptive, not a signal.

The shape

patearn.in≤-5 to -4.76R: 1 (0.0%)-3.78 to -3.54R: 4 (0.1%)-3.54 to -3.29R: 1 (0.0%)-3.29 to -3.05R: 4 (0.1%)-3.05 to -2.8R: 1 (0.0%)-2.56 to -2.32R: 4 (0.1%)-2.32 to -2.07R: 6 (0.1%)-2.07 to -1.83R: 15 (0.3%)-1.83 to -1.59R: 23 (0.5%)-1.59 to -1.34R: 38 (0.8%)-1.34 to -1.1R: 80 (1.7%)-1.1 to -0.85R: 3423 (74.6%)-0.37 to -0.12R: 1 (0.0%)-0.12 to 0.12R: 16 (0.3%)0.12 to 0.37R: 45 (1.0%)0.37 to 0.61R: 79 (1.7%)0.61 to 0.85R: 99 (2.2%)0.85 to 1.1R: 95 (2.1%)1.1 to 1.34R: 53 (1.2%)1.34 to 1.59R: 68 (1.5%)1.59 to 1.83R: 71 (1.5%)1.83 to 2.07R: 52 (1.1%)2.07 to 2.32R: 49 (1.1%)2.32 to 2.56R: 30 (0.7%)2.56 to 2.8R: 32 (0.7%)2.8 to 3.05R: 23 (0.5%)3.05 to 3.29R: 18 (0.4%)3.29 to 3.54R: 17 (0.4%)3.54 to 3.78R: 17 (0.4%)3.78 to 4.02R: 12 (0.3%)4.02 to 4.27R: 25 (0.5%)4.27 to 4.51R: 13 (0.3%)4.51 to 4.76R: 14 (0.3%)4.76≥ to 5R: 157 (3.4%)median -1Rmedian -1mean -0.1348Rmean -0.13-5R5Rn=4586 · 145 beyond axis
In plain EnglishEach bar is a count of trades whose result fell in that band, measured in R — multiples of the money put at risk on that trade. A trade that lost exactly what it risked is −1 R. The wall at −1 R is trades stopped out as designed; the bar is tall because that is what usually happens. Mean and median are both marked because they disagree, and the distance between them is the skew.

Why the average hides it

QuantityValueWhat it means
Share that won21.50%Fewer than four trades in ten.
Average winner3.1344 RWhat a win returned, per unit risked.
Average loser-1.0302 RAverage loss per unit risked. Count of losses exceeding 1 R unavailable in this dated study.
Expectancy-0.1348 RThe product of the three above. Negative.
Worst / best-5.5 / 144.9 RThe range the average is drawn from.
In plain EnglishExpectancy is decomposed, never just stated: win rate times average win, plus loss rate times average loss. A rule that widens its stop pushes the win rate up and the average win down, and only the product says whether that was a trade worth making. A win rate quoted on its own is stop width restated.

What the rule change cost

QuantityBefore the ruleAfter it (this run)
Resolved trades14,6264586
Expectancy (R)-0.0797-0.1348
Median (R)-1.0000-1.0000
Share that won (0–1)0.36310.2150
In plain EnglishThese are two different engines, not two attempts at one number. In August 2026 a rule was enforced that had been written down but never implemented: the fifth point of the pattern must actually reach the line drawn through points 1 and 3. That removed about 63% of the setups the older column counted. Read the last row: on the stricter, correct definition the rule set does worse, not better — enforcing the rule did not rescue it. The older column is kept because deleting it would quietly erase that.
Reproduce it. The instrument is in the repository and reads the same archive this site does, read-only:python3 -m scripts.r_distribution --json data/r_distribution.json --checkAbout 40 minutes. Download the 4,586 values as CSV.
RelatedWhat backtests hideCoverageThe same discipline applied to a public scan: