Two kinds are kept apart. Figures over closed trades come from the trades. Figures over a distribution — how returns were spread, how deep the drawdowns went — come from a step-by-step series instead.
The units
Distribution returns are scaled to hundredths of one percent, and the length of a drawdown is turned into days.
The rest are in whatever suits them — money, counts, percentages, plain ratios with no unit at all, and amounts of coin.
Where a figure is missing
Four figures over closed trades are rounded to a fixed step, and those same four come back as zeroes where no trade closed at all.
The four are reached plainly: the mean is the trades' profits summed over their count, each side's volume is its own trades' entry notional summed, and the whole is the two sides added together.
Elsewhere an undefined figure is kept but left empty. A win rate with no trades behind it is still named in the report, holding nothing rather than nought.
A spread in three points
Most distribution figures come back as three points — a low, a middle and a high — so the spread is visible rather than averaged away. One is an exception and comes back as a single number.
Where no real values remain among the inputs, the three are empty rather than zero.
Related
Source evidence
---
row: P3-10
baseline: 49aa659
created: 2026-09-28 11:30 UTC
modified: 2026-11-08 14:48 UTC
evidence:
- claim: "A replay's figures come from the record and from the scenario's own bars"
source: "`build_replay_report.py` 53-57 and 77-83"
- claim: "A paper run is given its starting money separately"
source: "`launcher.py` 3366-3368"
- claim: "Distribution returns are scaled to hundredths of one percent"
source: "`snapshot_result.py` 76-79 with the unit at `metric_conventions.py` 13"
- claim: "Drawdown lengths are turned into days"
source: "`snapshot_metrics.py` 162-165 with the unit at `metric_conventions.py` 15"
- claim: "The mean is the closed trades' profits summed over their count, and each side's volume is its own trades' entry notional summed, the whole being the two added together"
source: "`ledger_metrics.py` 62-65, the whole at 71"
- claim: "Four figures over closed trades are rounded to a fixed step"
source: "`ledger_metrics.py` 67-71 with the step at 20"
- claim: "Those four come back as zeroes where no trade closed"
source: "`ledger_metrics.py` 52-59"
- claim: "An undefined figure is kept in the report as an empty value"
source: "`metrics_serialization.py` 27-45, which names them all regardless"
- claim: "Most distribution figures come back as three points"
source: "`percentiles.py` 62-66, assembled at `snapshot_result.py` 93-102"
- claim: "One comes back as a single number"
source: "`snapshot_result.py` 103"
- claim: "With no real values among the inputs, the three are empty rather than zero"
source: "`percentiles.py` 59-60"
---