A different series
A replay's steps are its bars. A paper run's are the marks its sampler took as the run went — moments paired with a price, sorted and stripped of repeated moments before they are summed up.
This path hands over nothing for the bar-basis figures, so what comes back for those is empty.
What the marks must be
Every mark is checked before any figure is worked out, and a bad one raises rather than being skipped — so one bad mark among the many refuses the whole summary.
A moment carrying no zone refuses it. So does a moment failing to advance on the one before, equal moments included — though a repeat recorded by the sampler has already been stripped out by then. So does a price at or below nought, one that is endless, and one that is not a number.
A gap at or below nought refuses it whatever the marks are, being checked before them.
The gap that is substituted
Stretching the risk-adjusted figure to a year needs to know how far apart the steps are. A replay uses the gap it was given. A paper run works out its own: the whole span over the number of gaps in it, rounded to whole seconds and held at one at least.
A sampler can miss a tick, which leaves the marks unevenly spaced. The gap handed in serves as the stretching gap only where there are fewer than two marks, leaving no span to measure.
Related
Source evidence
---
row: P4-19
baseline: 49aa659
created: 2026-10-03 21:21 UTC
modified: 2026-10-04 11:43 UTC
evidence:
- claim: "A paper run's figures come from the same core a replay's do"
source: "`paper_metrics.py` 80 into `build_replay_report.py` 99-145"
- claim: "Where a replay's steps are its bars, a paper run's are its marks"
source: "`paper_metrics.py` 80 against `build_replay_report.py` 53-57"
- claim: "The marks are the sampler's own, sorted and stripped of repeated moments before they arrive"
source: "`paper_report.py` 42-55, the samples made at `mark_sampler.py` 118-125"
- claim: "This path hands over nothing for the bar-basis figures, so what comes back for them is empty"
source: "`paper_metrics.py` 80 into `build_replay_report.py` 126, 140-142 and 363-380"
- claim: "A gap at or below nought refuses the summary, whatever the marks"
source: "`paper_metrics.py` 56-58"
- claim: "A mark whose moment carries no zone refuses it"
source: "`paper_metrics.py` 64-66"
- claim: "So does one whose moment fails to advance on the one before, equal moments included"
source: "`paper_metrics.py` 68-70"
- claim: "So does a price at or below nought, one that is endless, and one that is not a number"
source: "`paper_metrics.py` 72-74"
- claim: "Every mark is checked before any figure is worked out, and a bad one raises"
source: "`paper_metrics.py` 62-76 ahead of 78-80"
- claim: "The gap is what the risk-adjusted figure is stretched by to reach a year"
source: "`build_replay_report.py` 376 and 427-429"
- claim: "The gap used there is the mean between marks, not the one handed in"
source: "`paper_metrics.py` 99-101, called at 78"
- claim: "It is the whole span over the number of gaps, rounded to whole seconds and held at one at least"
source: "`paper_metrics.py` 99-101"
- claim: "A sampler can miss a tick, leaving the marks unevenly spaced"
source: "`mark_sampler.py` 113-125 and 131-141"
- claim: "The gap handed in serves as the stretching gap only where there are fewer than two marks"
source: "`paper_metrics.py` 96-97, the two set at 24"
---