Fallout — what a cancellation takes with it

When a passenger working is cancelled, the unit that was going to run it does not arrive, and whatever it was booked to form next has nothing to run with. Darwin's NP association is the only thing in any feed here that names that second working, so this page follows it and reports what actually became of it. Measured over the 7 days to 13 September 2026, before this page existed: after a cancelled inbound, the working its unit was booked to form suffered something 62.8% of the time, against a 5.0% baseline for the same links when the inbound ran. And where the second working still ran but lost calls, it started short in 95% of cases — against 0.2% normally. That is the finding this page is built around: a missing unit does not thin a service out in the middle, it eats the front of it.

Links observed

866

2026-10-24 → 2026-11-07

Second working hit

69.4%

460 of 663 settled

Started elsewhere

250

the commonest outcome bar none

Cancelled outright

176

no train at all

Ran anyway

203

a unit was found

Median warning

40m

360 of 663 gave 30m+; 169 came too late

Only workings that ran on at least any 1 day 3 days 5 days of the window. This is a reality check, not a risk score — see the note below.

Current 7 Retrospective 1,268

Cancellations dated 2026-11-07. The figures in the tiles above describe the whole retained window, not this tab — a rate computed over a handful of links at 07:00 would swing forty points between page loads and would be evidence of nothing. Click any headcode to see every impact of that one cancellation; hover it for the UID. explained means the operator published a plain-English alert about it, defect a recorded fault on the set, and notice a disruption incident that passed the three-way test.

Live — cancelled, second working not yet due

The actionable half. Each row is a cancellation that has already happened and a working its unit was booked to form that has not departed yet. We are not forecasting these. The base rates above say what happened to links like them over the last week; this row says only that the link exists and the unit is not coming. Warning time is measured from when the cancellation was detected to the second working's booked departure — a negative figure means the cancellation landed too late to be a warning at all.

Cancelled	From → to	Unit turns at	Next working	From → to	Due	Warning
2J21 06:20	Clitheroe→Rochdale	Rochdale 20m turn	2N53	Rochdale→Clitheroe	08:22	45m

Where a knocked-on working reappears

When the second working runs but starts short, this is where it starts from instead — the first calling point it did not cancel. If you are waiting at a station further back down that route, this is the list that says your train exists but will not be coming to you. Counted over the whole window, so a terminus that turns a lot of units dominates it by arithmetic; read it as "which places absorb a missing unit", not as a league table of blame.

Restarted from	TIPLOC	Times
Manchester Victoria	MNCRVIC	1

Chains — it does not stop at one

Where the second working was cancelled too, it becomes an inbound of its own and the diagram keeps unravelling. Measured: where the second working was cancelled, the third was cancelled 26.7% of the time; where the second one survived, 4.0%. So a chain is worth following, and each hop below is an observed outcome rather than a projection.

No multi-hop chains in the window.

Every link, and what became of it

One row per (cancelled working → the working its unit was booked to form). Ran as booked rows are kept deliberately: they are the honest denominator, and the headline percentage above is meaningless without them. Where a working started somewhere else, the new origin is Darwin's first uncancelled calling point, or — where Darwin flagged nothing and TRUST sent a change-of-origin instead — the location TRUST named.

Date	Op	Cancelled	From → to	Turns at	Next	From → to	Due	Outcome
11-07	NT	2W10 08:00	Hull→Bridlington	Bridlington	2W18	Bridlington→Hull	09:09	Cancelled outright
11-07	NT	2J22 06:49	Clitheroe→Rochdale	Rochdale	2B82	Rochdale→Blackburn	08:52	Started somewhere else — from Manchester Victoria
11-07	NT	2T02 06:07	Ashington→Newcastle	Newcastle	2T07	Newcastle→Ashington	06:55	Ran as booked

By operator

Read the rate, not the count. An operator running dense self-contained diagrams generates far more NP links than a long-distance one, so the left-hand column measures how tightly its units are diagrammed, not how badly it is doing. Operators with only a handful of links in the window are shown but should not be read as a rate at all.

Operator	Links	Second working hit	Rate	Reported at the time
Northern NT	4	2	—	Lines reopened: delays between Ely and Kings Lynn expected until 10:45

The right-hand column is context, not causation. Nothing in the disruption feed names a train — incidents carry operator names and route prose, so the only join available is operator to operator. A notice there means that operator was reporting it while these links were being observed. It does not mean it is why any particular unit went missing.

Operator status right now

The network's own summary, straight from the service-indicator feed. Operators reporting a good service are omitted.

Operator	Status	Detail
Great Northern GN	Custom	Delays between Ely and Kings Lynn
London Overground LO	Custom	Amended service on the Suffragette
Northern NT	Custom	Disruption between Manchester Victoria and Rochdale
South Western Railway SW	Custom	Disruption between Wimbledon and Chessington South / Epsom
c2c CC	Custom	No trains between Barking and Grays via Rainham (Essex)

What this page does not mean

The link is a booking, not an observation of the unit. Darwin says the diagram intended the same unit to form both workings. It does not confirm the unit ever existed, and an operator free to substitute one will do so — that is exactly what the 39% of links whose second working ran as booked represents.

Only services Darwin knows about appear. Roughly half of all TRUST cancellations in the window had no Darwin schedule at all — overwhelmingly freight, which has no NP associations and no passenger fallout to report. This page is passenger-only by construction, not by filter.

"Has a history of running" does not predict anything. It is offered as a filter above and it deserves the demotion: split by how many days the cancelled working had actually run, the second working's cancellation rate was 28.3% / 31.5% / 21.6% at 0 / 1–2 / 3–5 prior days. That is noise. Use it to confirm a working is a real regular service, not to weight the risk.

Turnaround time does not predict anything either over this sample — 26.5% under 30 minutes, 22.3% at 30–59, 18.2% at 60–88, with the longer buckets down to single figures of observations. It is shown on the live table as context and is deliberately not scored.

Cancellation reason codes are not translated. Darwin sends a bare numeric code and we hold no reference table mapping it to text, so rather than guess, the code is simply not shown. Same policy as the operator codes on /defects.

Some cancellations arrive after the fact. TRUST sends retrospective cancellations days later for delay attribution — 126 of them dated 8 September landed on 11 September in the window this was built against. They are real cancellations and they are counted, but they were never a warning: they are the negative warning times above, and they drag the median down rather than flattering it.

The window is short. Darwin associations are not archived and hold about seven days, which bounds everything here. The 62.8% headline rests on 261 links with a cancelled inbound — a large effect on a modest sample, and it should be re-read as the window rolls.