Where things diverge

One row per type or operator, one dot per executed check, coloured by outcome. Sorted worst first, so the clusters are the top of each table.

Over time #

One point per regeneration of the data, oldest first. The history is kept append-only in src/data/history.jsonl, so this chart can be replaced by something richer without losing the record.

Types

Each dot is one example from the official documentation, run on both engines and compared against the output the documentation asserts.

Operators

Each dot is the same expression run on both engines. Every operator that parses is listed, worst first — so the disagreements cluster at the top and the rest runs green.