Make transit freshness, cancellations and station context consistent across views #401

Activity

1. Medformatik

   added

   enhancement

   needs-triageOpen issue awaiting initial maintainer reviewOpen issue awaiting initial maintainer review

   on Oct 6, 2026
2. Medformatik

   added the

   Feature

   issue type on Oct 6, 2026
3. Medformatik

   updated Priority Medium and Effort Medium on Oct 6, 2026
4. Medformatik commented on Oct 7, 2026

   

   Medformatik

   on Oct 7, 2026

   CollaboratorAuthor

   More actions

   Implemented the bounded presentation and timetable-fetch pass in #425 (fce72ba4, 2c3eb0de, 2b2bb6e4, a2b8e1d1, 47b902d3). The final full suite passed 17,718 tests across 1,614 files; types, lint, policy and docs passed. Five direct PR attachments demonstrate controlled desktop/phone EN/DE states; no screenshot assets were committed.

   This issue remains open: station name-search failures before the timetable-fetch stage can still yield a cold-cache false-empty board, and upstream publication age is unavailable from current wrappers. Missing age stays unknown. The PR lists all review limits and two deferred presentation minors. It preserves linked-board matching, identity contracts and installed-shell query ownership.
5. Medformatik

   removed

   enhancement

   on Oct 18, 2026
6. Medformatik

   removed

   needs-triageOpen issue awaiting initial maintainer reviewOpen issue awaiting initial maintainer review

   on Oct 30, 2026
7. Medformatik

   added a commit that references this issue on Oct 31, 2026

   fix(transit): report failed station lookups instead of empty boards (#…

   ...

   589f00e
8. Medformatik commented on Oct 31, 2026

   

   Medformatik

   on Oct 31, 2026

   CollaboratorAuthor

   More actions

   #694 (589f00e1c) fixes the station discovery gap listed under "Remaining gaps".

   A failed stop-name search no longer turns into an empty board that stays cached:

   • Provider clients report failures. The stop searches of MOTIS (local and Transitous), HAFAS REST, DB, TfL and the dynamic-registry HAFAS adapter used to return [] on upstream errors. They now throw.
   • The orchestrator reports gaps. searchByNameRaw marks its result isPartial when an eligible provider failed or was in health cooldown. It rejects when providers should have answered and none did. Providers excluded by the data-use policy don't count as gaps.
   • Partial discovery is short-lived. It is kept for 60 seconds instead of 24 hours. Partial place responses are sent with no-store, and the linked-stop, route, alert and facility hooks refetch partial answers after a minute.

   The board now shows "may be incomplete", or a failure with Retry, instead of "No departures". So the criterion "Loading/failure/recovery retains useful scheduled information and clear uncertainty" is met for station discovery.

   Still open:

   • Per-stop lookups. Route, alert and facility lookups for linked stops still read a provider failure as "none" (getRoutesForStop, getStopAlerts and getFacilities in the orchestrator return empty results).
   • Upstream data age. No transit provider sets Freshness.dataAsOf yet, so age-based fresh/stale states still occur only in fixtures.
   • Shared stop identity. Departures, journey and navigation views don't yet refer to the same source-aware stop/route identity, so the parent/sibling criterion is still open.
9. Medformatik

   added 3 commits that reference this issue on Oct 31, 2026

   fix(transit): report failed departures, lines, alerts and facilities (#…

   ...

   68c644d

   fix(transit): keep separate departures apart and trip lookups on thei…

   ...

   d64c6e9

   feat(transit): show how old a board's realtime data is where the sour…

   ...

   1ea48b3
10. Medformatik commented on Oct 31, 2026

    

    Medformatik

    on Oct 31, 2026

    CollaboratorAuthor

    More actions

    #702 (68c644dd9), #703 (d64c6e95f) and #705 (1ea48b315) cover the three items left open in the previous comment.

    Per-stop lookups (#702). Departure, arrival, line, alert and facility lookups no longer read a provider failure as "none":

    • The HAFAS REST, DB, iRail, TfL, Transitland and dynamic-registry HAFAS clients throw on upstream errors instead of returning []. So do TfL alerts, Swiss alerts when all three feeds fail, and MOTIS line lookups.
    • The orchestrator tells "no source offers this lookup" apart from "the source is failing or in cooldown". In the second case, getRoutesForStop, getStopAlerts and getFacilities throw.
    • A place board marks its lines, alerts and facilities partial when some linked stops fail, and fails when all do. DB no longer claims stop alerts it doesn't have.

    Shared identity, partly (#703).

    • The merged place board no longer folds the opposite direction of a line, or the next train of a frequent line, into one row. A row absorbs another source's departure only when the line and headsign are compatible. The same source merges only on the same trip.
    • Trip lookups send each source's trip id to its own provider, so one failing source no longer hides the trip.
    • Trip detail opened from a planned journey shows the train's destination, the planned time with its delay, and the platform, cancellation and occupancy, the same way a departure board does.
    • OTP leg ids carry the otp: prefix.

    Upstream data age (#705). Boards now carry Freshness.dataAsOf where the source states it: HAFAS realtimeDataUpdatedAt (dynamic registry and transport.rest; 7–38 s in live samples) and TfL prediction timestamps (44 s). DB passes the same field through, but it is null for now. MOTIS/Transitous, Entur, OTP, MBTA, iRail and SIRI/OJP don't state an age, and neither do Swiss boards, whose realtime comes mostly from OJP. A merged board only has an age when every realtime source on it has one. "Times may have changed" is now judged by how far behind the source was when we fetched, so our own caching and refresh intervals don't count against it.

    Still open, both presentation decisions under criterion 2:

    • Which stop a merged row comes from. A place board merges the departures of every linked stop, such as the main station, its bus stops and its platforms. Rows show a platform when the source provides one, but not which linked stop they came from. At a parent station with nearby siblings, a rider can't tell whether a bus leaves from the forecourt stop or the one around the corner.
    • Opening the exact stop from a journey. Tapping a stop in a planned journey geocodes it to the nearby place and opens that place's merged board (handleStopClick in TransitDetailsView.tsx). The leg's own stopId is only used as a fallback id. So the journey and the board refer to the place, not the specific stop or platform the leg uses.

    A per-row realtime age (instead of one age for the board) is a possible follow-up, but not needed for the criteria.
11. Medformatik

    added 7 commits that reference this issue on Oct 31, 2026

    feat(transit): name the stop a station board departure leaves from (#714

    ...

    ed9ae0f

    feat(transit): open a journey's stop as that stop and keep it as the …

    ...

    1142948

    fix(search): open a famous station as the station, with its departures (

    ...

    5c65562

    fix(transit): show lines by the name riders know and fold repeated li…

    ...

    1a3a220

    fix(transit): link a station's stops by its local name, not a transla…

    ...

    caf4fa5

    fix(transit): show platform changes and the platform riders see (#728)

    ...

    dd7c5e6

    fix(transit): warn of missing departures only for sources that leave …

    ...

    42f1c2b
12. Medformatik commented on Nov 1, 2026

    

    Medformatik

    on Nov 1, 2026

    CollaboratorAuthor

    More actions

    #714 (ed9ae0fae) and #716 (11429489b) cover the two items left open under criterion 2, so this issue is done.

    Which stop a merged row comes from (#714). A row from a stop other than the board's main stop shows that stop's name next to its line and platform. Screen readers announce it as "Stop: …". The stop is the one the source names for the departure, otherwise the linked stop that listed it. Different spellings of one name, such as "Köln Hbf" and "Köln Hauptbahnhof", count as the same stop and get no label.

    Opening the exact stop from a journey (#716). Tapping a stop in a planned journey opens that stop by the leg's source stop id, as search and map taps do. It no longer opens whatever place a name search returns. That stop is the board's main stop and stays so after a reload or in a shared link (stop=). The navigation views don't open stops; they show the itinerary leg's own stop and route.

    Related fixes found while testing live boards:

    • fix(search): open a famous station as the station, with its departures #722: a famous station found in search opens as the station, with its departures.
    • fix(transit): show lines by the name riders know and fold repeated listings #724: lines show the name riders know, and a train listed twice under different spellings shows once.
    • fix(transit): link a station's stops by its local name, not a translation #725: a station's stops are linked by its local name, not a translation ("München Hauptbahnhof", not "Munich Central Station").
    • fix(transit): show platform changes and the platform riders see #728: platform changes show, along with the platform most sources give.
    • fix(transit): warn of missing departures only for sources that leave gaps #730: the "may be incomplete" warning appears only for sources that actually leave gaps.
    • fix(transit): link only stops named like the place to its board #731: a place's board links only stops named like the place. Before, stops that only shared the district were linked too: "Köln Messe/Deutz" showed about 640 departures in 30 minutes, about half from other stops. In a test of 86 places across Germany, links fell from 1,260 to 945.

    Acceptance criteria:

    • Criterion 2 (shared stop identity, parent/sibling stops distinguishable): met by feat(transit): name the stop a station board departure leaves from #714 and feat(transit): open a journey's stop as that stop and keep it as the board's main stop #716.
    • Criterion 5 (loading, failure and recovery): met for station discovery (fix(transit): report failed station lookups instead of empty boards #694) and per-stop lookups (fix(transit): report failed departures, lines, alerts and facilities #702). The stop labels were checked at phone width in feat(transit): name the stop a station board departure leaves from #714.
    • Upstream data age (feat(transit): show how old a board's realtime data is where the source says #705): shown only where the source states it. MOTIS/Transitous, Entur, OTP, MBTA, iRail, SIRI/OJP and DB (whose field is still null) don't state an age, so their boards show none. This is documented.

    Not done, and not needed for the criteria:

    • A realtime age per row instead of one for the whole board.
    • Some bus stops beside a station are no longer on its board when their name doesn't say they belong to it, such as "Köln Bf Deutz/Messeplatz". fix(transit): link only stops named like the place to its board #731 accepts that on purpose, as the safer side of name-based linking.
13. Medformatik

    closed this as completedon Nov 1, 2026
14. Medformatik

    self-assigned this

    on Nov 1, 2026