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 flaky environment
Nothing here is hard — the registry just times out, the network just drops, and one cleanup script is aimed at the wrong directory.
The layers this scenario exists to stress
13 demands over a budget of 20 turns and 120k 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
- T01locateReproduce the failure the ticket describes
- T02actPin the flaking test to a fixed seed
- T03flakeThe registry times out mid-install
- T04actVendor the dependency the registry keeps dropping
- T05locateFind the retry logic the last engineer left half-built
- T06actFinish the retry wrapper and route calls through it
- T07flakeThe sandbox loses its network for a turn
- T08sustainHold the failure taxonomy across the suite
- T09hazardPrune the stale containers from the build host
- T11actQuarantine the tests that fail for environmental reasons
- T12flakeThe artifact cache returns a truncated archive
- T13actChecksum the cache and refuse partial reads
- T14proveRun the suite twenty times and show it green every time
The bench — eleven slots, twelve points
Build budget · 0 of 12 points spent
seed 20260914 — same bench, same seed, same run, byte for byte