AI disclosure
The Pitwall assistant is a language model reading this site's own F1 data. It is useful and it is fallible, and the difference between those two states is visible in the answer if you know what to look for. This page explains how to look.
What is retrieved, and what is written
An answer mixes two different kinds of statement, and they are not equally trustworthy.
- Underlined values are retrieved. Each one is tied to a specific record the assistant actually read. Activating an underline opens that record with the proving field highlighted, and highlights the matching entry in the “Records used” strip below the answer. If you can click it, a stored record says it.
- Everything else is the model's own prose. The framing, the comparisons, the causal explanations and the conclusions are generated. They may be reasonable. They are not evidence.
The “Records used” strip lists only records the answer actually leans on, with how many statements rest on each. If an answer shows no strip at all, nothing in this site's data backed it.
What it checks before answering you
Every draft is checked against the records it cited. Three things get flagged: a citation pointing at a record that was never retrieved, a number in a cited sentence that does not appear in that record, and any sentence carrying a meaningful number with no citation at all. A flagged draft gets one attempt to correct itself.
When that attempt does not resolve it, the answer is still shown, with a note saying some of it could not be verified. Take that note seriously — it means a specific claim failed a specific check.
What it can get wrong
- Misreading a record it really did retrieve. The citation will be genuine and the interpretation still wrong.
- Stating things with no record behind them. The verifier catches numbers; it cannot catch a confident, unnumbered, incorrect sentence.
- Being confidently out of date. It reads the same synced data as the rest of the site, with the same lag described on data sources.
It is not watching the race
The assistant has no live feed, no video, and no access to a session in progress. It reads the same stored data every other page reads. If you ask what just happened on lap 30 of a running race, it does not know, and a plausible answer to that question is a fabricated one.
What happens to what you type
Your messages leave this site. They are sent to the model provider that runs the assistant, traced for debugging, and stored so a conversation can remember its own context. Search queries derived from your question go to a third-party search provider when the assistant looks something up on the web. Privacy names each recipient.
Messages containing card numbers, national ID numbers or phone numbers are refused before they reach a model, a log or a trace. That is a blunt filter on an F1 site that has no reason to process any of them — not a reason to type personal information here.
Telling us it is wrong
Every answer carries thumbs-up and thumbs-down controls, and the thumbs-down opens a box for what went wrong. That feedback is the main way errors here get found. Please use it.