Corrections

1. 6 October 2026

   Piastri shown "under investigation" after the Bahrain GP in Malaysia

   What was wrong: From race Sunday until Tuesday morning the homepage and the investigations board listed Oscar Piastri as awaiting a ruling. He had been summoned only as the other party in an incident involving Alex Albon, and the stewards ruled on Albon alone (no further action) about three hours later. Separately, ten of the weekend's decisions, among them Lewis Hamilton's impeding ruling, stayed without a plain-English summary for days, and the race result reached our X account more than a day after it was published here.

   What changed: A ruling on one of the cars named in a summons now settles that summons when it concerns the same offence. The document processing was reworked to fit the hosting plan's processing limits, unreadable documents are now retried patiently over hours instead of being given up on within an hour, and the result post no longer waits behind those retries. All ten decisions were queued for processing again.

2. 2 October 2026

   Sprint and practice penalties shown under race results

   What was wrong: Race results listed some penalties that belonged to other sessions. On the Canadian GP, Sergio Pérez carried a 10-second penalty in the race table that was handed down for the Sprint, and race results also showed practice and qualifying reprimands (for example two Azerbaijan GP practice reprimands) as if they related to the race. The session named in each FIA document was being read but never saved, so a fallback guess based on the document title did the sorting.

   What changed: The session is now stored for every new decision, and it was filled in from the original FIA documents for this season's 175 driver penalties. Sprint penalties now sit with the Sprint result; practice and qualifying decisions stay in each Grand Prix's full list of decisions but no longer annotate the race result.

3. 2 October 2026

   Calendar missed the Bahrain Grand Prix in Malaysia

   What was wrong: On the Friday of the relocated Bahrain Grand Prix (re-run at Sepang, 2-4 October, after the April race was cancelled) the homepage still counted down to Singapore practice "in 6 days" and the calendar did not list the race. The site's calendar is a fixed snapshot and had never been updated with the new date. The weekend's stewards' documents were still collected and explained on time; no wrong result was published.

   What changed: The race is on the calendar with its real session times, and the site now compares its calendar against the public timing data every hour - an added, moved or cancelled session raises an internal alarm instead of waiting for a reader to notice.

4. 27 September 2026

   Phantom disqualification on the Azerbaijan GP result

   What was wrong: The corrected-result view showed Pierre Gasly as disqualified (red DSQ) from the Azerbaijan Grand Prix. He was not - he retired on lap 35. A column-alignment fluke in the on-track classification PDF (the previous row's fastest-lap column plus his car number reading like a valid position row) made the parser record him as classified on track, and "classified on track, absent from the final" renders as a disqualification.

   What changed: The on-track classification parser now hard-stops at the NOT CLASSIFIED block exactly like the final-classification parser, the stored result was re-derived from the original FIA PDF, and every stored result since 2025 was audited for the same signature (four other cases found - all genuine, document-backed disqualifications). A regression test carries the literal Baku text.

5. 27 September 2026

   Bottas shown "under investigation" for three days after his ruling

   What was wrong: The live investigations board kept Bottas "awaiting a ruling" from Thursday to Sunday of the Azerbaijan weekend, although the stewards had ruled "no penalty" two hours after the summons. The FIA renamed the offence between the two documents ("yellow flag infringement" vs "overtaking under yellow flags") and the singular/plural difference alone defeated the summons-to-ruling matcher.

   What changed: Offence matching now folds plurals before comparing, with a regression test on this exact document pair. This is the second correction against the same board - a self-check that flags any summons whose weekend has ended without a paired ruling is being added so the next miss is caught by the site, not by a reader.

6. 27 September 2026

   Race weekend results arrived late or not at all (Italian & Spanish GP, then Azerbaijan)

   What was wrong: The corrected results for the Italian and Spanish Grands Prix appeared days late, and the Azerbaijan result almost a day late, while the site's health endpoint claimed all was well. A chain of pipeline defects - documents silently abandoned under CPU pressure, fetch failures during fia.com outages counted against innocent documents, and the health monitor itself scheduled behind the very work that was failing - compounded across three race weekends. Returning browsers could additionally hold a stale copy of "live" pages for up to four hours due to a caching misconfiguration.

   What changed: The pipeline was rebuilt around explicit invariants: a document is only marked processed after its result is stored, one PDF extraction per cycle, source outages pause work instead of discarding it, the health monitor runs before the work it monitors and alarms when it is itself silenced, and pages now tell browsers to revalidate. Each failure carries a visible marker and an automatic retry instead of a silent loss.

7. 30 July 2026

   Stale "under investigation" board

   What was wrong: The investigations board kept a driver marked "awaiting a ruling" for up to three days after the stewards had already published the decision, when the ruling's wording drifted from the summons (e.g. "Starting Procedure" vs "Starting Procedure Infringement").

   What changed: The summons-to-decision pairing now tolerates that drift (matching on the offence prefix plus the car set and Grand Prix), so the board clears the moment a ruling is published. A regression test covers the specific case.

8. 18 July 2026

   Large grid penalties dropped silently

   What was wrong: A safety cap treated grid-drop penalties above 15 places as unclassifiable - which rendered no card and silently skipped the automated post for stacked power-unit penalties (a driver taking several fresh components can be dropped 20-30+ places).

   What changed: The cap was corrected to the real-world range (up to 60 places), and any decision that still cannot be classified now always renders visibly as "unclassified - read the source document" and is surfaced on /health, rather than disappearing.

Spotted something wrong? The source PDF is linked on every decision page - if a summary doesn't match it, that's a bug worth reporting.