Odd — things in the data that don't look like the rest of it

Non-headcode berth descriptions

A train describer berth normally holds a reporting number — digit, letter, two digits (1S98). Anything else was typed in by a signaller. These are not all possession markers. By far the largest group is four plain digits, which are almost certainly unit or stock numbers under a local convention. The group that does contain possession markers like MC14 is the mixed alphanumeric one — and even there, what the letters mean is not stated by any feed.

Potentially blocked sections

Two or more markers standing on the same line, near each other, with nothing moving between them. That is how a blocked section gets flagged in a describer — but no feed we receive states that a possession exists, so this is never reported as confirmed. Network Rail publishes no possessions data at all: PPS is internal, the Weekly Operating Notice is a weekly PDF, and possessions remain an open-data wish-list item.

Same train in two places at once

One description occupying more than one berth. For a second or two during a step this is normal. Read the spread column: a large spread means a description was never cancelled out of the berth it left, so at least one of those positions is a ghost and the diagram is showing a train that isn't there.

Moving, but in no timetable

TRUST reported these running, but their headcode matches no schedule anywhere in the CIF. Usually a very-short-term-plan special that never reached the daily dump. Most of this list is expected: a Z in the second position means a special by definition, and class 0 (light loco) and 5 (empty stock) are arranged at short notice as a matter of course. The unusual tag marks the ones that are none of those — a revenue-earning passenger or freight working with no timetable entry at all.

On the diagram, never in TRUST

Descriptions stepping through train describers that TRUST has never reported a movement for. Not necessarily wrong — a describer sees everything physically on the ground, while TRUST only sees what an operator activates — but this is precisely the case where the board and the diagram disagree about whether a train exists.

Extreme timekeeping

More than two hours out, either way, in the last 24h. The early end is the more interesting one: a train genuinely running three hours early is close to impossible, so a large negative figure is usually a report matched against the wrong day's schedule slot rather than a real time.

Royal train headcodes

Headcodes matching the configured royal patterns (/^1X0[12]/), anywhere on the network, over the last week. A match is not proof — the pattern is a convention, and an empty list here is the normal state.

TRUST's own exception reports

The five message types TRUST sends for things that deviate from plan. These are the feed stating it, not us deriving it — which is the difference between this section and "moving, but in no timetable" above. 0007 (identity change) is the most likely explanation for the "never in TRUST" section: the describer steps one headcode while TRUST reports another.

Describer areas gone quiet

A subscribed describer area that has gone quiet is invisible in the dashboard's single global "feed age", which is dominated by the 190-odd areas that are fine. An area counts as alive on the later of its CT_MSG heartbeat and its last berth step: the heartbeat alone is too irregular to judge on (some describers go half a day between them while stepping normally), and steps alone can't tell a genuinely quiet 04:00 from a dead feed. Only subscribed areas are judged — berth steps are recorded for those alone, and no other area's silence could affect anything here.

194 live · 0 silent >15 min · 0 awaiting first message · 0 unsubscribed, not judged · 194 areas seen

Barely-seen operators

Operator codes with a handful of movements in the last 48h. Either a genuinely rare visitor — a charter or a minor freight operator — or a code we simply have no name for. Neither is a fault.