StatusA model, not a measurement. Every number below is generated by this page. No LLM runs here — a deterministic, seeded simulation of mechanisms the sheets teach, and nothing else.
Break a harness yourself
The 10,000-file refactor
Rename a legacy API across a codebase too large to hold in one window — and survive the night the session dies.
The layers this scenario exists to stress
16 demands over a budget of 24 turns and 150k tokens. The bench below has 12 points to spend across eleven slots that cost more than that fully provisioned — decide which layers this work will actually lean on, then run it.
The work — read it before you spend a point
- T01locateFind every call site of the legacy API
- T02locateMap the config that gates the feature flag
- T03actRename the module across the first package
- T04sustainHold the naming scheme across packages
- T05locateFind the tests that pin the old behaviour
- T06actRepoint the pinned tests at the new surface
- T07sustainCarry the migration table into package two
- T08locateFind the docs that name the old module
- T09actRewrite the imports in package two
- T11sustainKeep the deprecation shims consistent
- T12locateFind the dynamic imports the grep missed
- T13actConvert the dynamic imports by hand
- T14sustainHold the edge-case list through package three
- T15repeatApply the same rewrite to the remaining packages
- T16actRegenerate the lockfile and the barrel exports
- T17proveRun the full suite and show it green
The bench — eleven slots, twelve points
Build budget · 0 of 12 points spent
seed 20260910 — same bench, same seed, same run, byte for byte