Redacted excerpt; full evidence withheld.

$ cd <guest-path> && ls -la && id -u && python3 --version && cat README.md 2>/dev/null | head -200
exit: 0
 does not cover some dates lends them the patterns of the same weekday.

## Choosing the segment length

`microsegments tune` scores L ∈ {10, 15, 20, 30, 40, 50, 75, 100} m and picks the smallest L whose
**time lost per vehicle** (day profile) is reproducible: split-half reliability ≥ 0.8 (days split at
random into two halves stratified by weekday, Spearman-Brown corrected) and hotspot localisation spread
≤ 30 m. When no length passes, the default 30 m is kept. The leave-one-day-out Poisson deviance (and its
1-SE rule), the obs/h reliability and the hotspot Jaccard are reported as diagnostics only. 30 m is a good
default for 20 s polling at urban speeds; go shorter only with dense GTFS-RT fixes and many days. `--sensitivity`
[excerpt truncated]