Skip to content

Decision story

Why Don't Delays Die?

Where should a fixed budget of recovery time sit along a line?

Rates lift payments. Payments eat buffers. Watch where that pressure concentrates.

Scroll · into the decision

With reduced motion, camera poses snap without smooth interpolation and marks stop breathing. A skip link jumps to the decision section, and the beat narration is announced in a live region.

A delay dies where the timetable leaves room for it to die

This story ships as a directed film. Open /stories/where-should-the-recovery-time-sit/film.

Being late at one stop and being late at the next are two logged times, so survival is measured rather than modelled: 81% at the next stop, decaying to 63% by the fourth. Across 119 legs, the more recovery margin a leg holds, the less often a delay survives it — Spearman −0.79, and it holds inside each service type separately. Margin is also the one thing a planner controls, which is what makes this a decision and not an observation.

Methodology

Directed film scrubbing a seeded field in which one mark is one train at one stop, placed by position in the run against minutes late. Every published figure is read out of the frozen evidence pack: 365 days of Fintraffic Digitraffic arrivals, 395,094 passenger runs and 5,272,722 timetable rows, 91% carrying an actual time. Late means 5+ minutes at a stop on a run that was not cancelled. A leg is (from station, to station, service type) — pooling service types on a shared leg was an error, because the same track is scheduled differently for commuter and long-distance trains. A line is a modal route signature over operated stop sequences, with expresses absorbed as subsequences. Recovery margin is median scheduled leg run time minus the 5th percentile of observed, computed inside one timetable period, because the timetable is re-cut roughly annually and a boundary can move a leg by 8.5 minutes. Survival and seasonality use the whole year. The counterfactual re-lays each line's existing padding toward the legs where delay survives, at five frozen strengths, conserving journey time; the zero-strength replay reproduces measured carry-over exactly. Marks are illustrative and interpolate continuously; printed figures snap to a frozen variant.

Limitations

Cause attribution covers ~1% of rows, so the film can say a delay survived a leg and mostly not why — no beat answers the title with a cause. The technical minimum run time is a 5th-percentile proxy rather than an engineering figure; the finding does not rest on it, since the negative-margin legs beat their schedule on 0% of runs. Samples thin beyond about four stops. Margin predicting survival is an association measured across legs, not a controlled experiment. Margin describes one timetable period. Lines separate when their recovery margin differs and not otherwise, so the picker is recognition and must not rank lines by quality. The counterfactual is arithmetic on observed delays, not an operational plan, and margin sized on a median day is undersized on the days that matter. Finland only; no capacity, rolling-stock, crew or cost model. Not affiliated with, and not endorsed by, Fintraffic or any operator.

Data references

  • [observed] A 5+ minute delay is still there one stop later 81% of the time — Median across 365 days; the daily band is 56–94%.
  • [calculated] Long-distance legs hold 2.27 min of recovery margin, commuter legs 0.24 min — Median scheduled run time minus the 5th percentile of observed, inside the current timetable period. 4% of long-distance legs are negative against 20% of commuter legs.
  • [calculated] More margin, less survival — Spearman −0.79 across 119 legs — −0.66 commuter-only and −0.69 long-distance-only, and it holds in all three timetable periods.
  • [observed] Monthly late share swings 2.7–7.2%; carry-over stays 75–85% every month — Volume is seasonal. What happens to a delay once it exists is close to a constant of the timetable.
  • [hypothetical] Re-laying existing minutes leaves commuter lines at 75–81%, where long-distance already sits today — Modelled replay: one allocation rule at five frozen strengths with journey time conserved to 0.00 min. Not an optimiser — on six of seven lines pushing harder is not better. 3 of 4 commuter lines cannot reach even the worst long-distance line's current figure.

Sources