v2.5.0

Release Notes

Version 2.5.0 (September 17, 2026)

New Features:

• Platoon Ratio (platoon_ratio): New aggregation implementing the HCM platoon ratio, Rp = Percent_AOG / (g/C), by phase and bin, plus the HCM arrival type (1–6). The green ratio is measured from phase green intervals (EventId 1 → 8) clipped to bin boundaries, with open intervals at the edges of the data extended to the bin boundary so results are exact under bin-aligned incremental processing. Depends on arrival_on_green (registered in AGGREGATION_DEPENDENCIES). Contributed by @Abimbola08 in #26, closes #6.
• Raw counts in platoon_ratio output: Added Green_Actuations and Green_Seconds columns. Platoon_Ratio is a ratio of two ratios and cannot be averaged across bins, so rolling it up to an hour, a day or a time of day requires summing the raw parts and dividing once. Those two columns plus Total_Actuations make that possible without hardcoding the bin size downstream, and because greens are clipped at bin boundaries, Green_Seconds sums exactly across consecutive bins. See the platoon_ratio schema in the README for the aggregation query.
• Timeline clock update, offset change and ped detector failure: Controller Clock Updated (181) is now a timeline event, Clock Update, shown as a point-in-time event lasting cushion_time with EventValue holding the reported correction in seconds. Because the controller stamps every event with its own clock, any interval overlapping a window from 5 seconds before to 5 seconds after a clock update is marked IsValid=False. Offset change and ped detector failure were added as timeline events as well.

Bug Fixes / Improvements:

• Timeline validity across data gaps in incremental runs: A timeline interval should be invalid when the device has no has_data row for any bin it spans. Single-pass runs checked every bin, but incremental runs only checked the bins the interval started and ended in, so a device that dropped out mid-window and returned hours later produced a valid interval with a duration of hours, and phase_wait counted the Phase Wait as a skipped phase. The opposite case is also fixed: a still-open event had its start bin re-checked against each new run's has_data, where it can never appear, so any interval open across two or more runs was marked invalid even with continuous data.

Version 2.4.0 (August 2026)

Bug Fixes / Improvements:

• Relaxed phase skip threshold when TSP adjusted the cycle (tsp_skip_multiplier): Phase waits overlapping a preempt are dropped outright, but TSP delays service without skipping it, so at a signal calling TSP every cycle the 1.5x multiplier reported skips for phases that were actually served. Rather than dropping those waits, which would blind the check at signals that have TSP, the wait is kept and judged against the new tsp_skip_multiplier, defaulting to 2.0.