What broke for other teams
The migration atlas pools anonymous verdict outcomes from teams that chose to share them: for each model switch, how many teams hit a regression and in which kind of case. Updated daily.
0migrations published
No migration has reached 5 teams yet. The atlas fills up as teams run verdicts with swapvera run --share. Until a migration has 5 different teams behind it, we only show how many have shared.
Ask before you run
Before spending a verdict, see what happened to teams with cases like yours. Only counts of your case kinds are sent (for example 25 JSON cases, 5 tool calls), never the cases.
Example of the output format, not real data:
$ swapvera predict
12 teams have migrated openai:gpt-4o -> openai:gpt-5.6.
42% hit at least one regression (95% interval 19% to 68%).
Risk for the kinds of cases in your suite:
JSON output your cases: 25 teams that saw regressions: 50% (24% to 76%, 10 teams)
tool calls your cases: 5 not enough teams yet
The atlas tells you where to look. Only a verdict on your own prompts tells you whether your app breaks.