Half your team switched to the newest top-tier model the week it launched, and never switched back. It is brilliant at untangling a race condition. It is also writing your changelog entries.
The Problem
Engineers pick a model once and keep it. That's reasonable for each of them and expensive for all of you. The price gap between models is large, and the provider bill doesn't say which one ran on which kind of work.
How It Cascades
The most expensive model ends up on renames, test fixes and docs.
Teams argue about model policy from taste, not data.
A cheaper default never gets tried because nobody can show it would hold up.
The Insight
Model choice only matters relative to the work. Put model and kind of work side by side and the expensive mismatches stand out.
The Solution
VibeCheck's Usage page has a model ledger: every model your team ran, grouped by Claude Code, Codex or Cursor, with tokens or estimated cost over the period you pick, plus how much input was served from cache. People pages show each engineer's leading model by spend. In Sessions, filter to one model and one kind of work, say a premium model on bugfix or explore, and sort by cost. Open a few to see whether the extra spend bought anything: did those sessions plan, test and land without rework?
The Outcome
You set a sensible default model by kind of work, keep the premium model where the sessions show it earns its cost, and see the effect in the next month's model ledger.