docs / build
grader limits
on this page
The limits are set so a run gives the same result every time and, for nearly all code, ends with a signed result.
budgets
| Limit | Value |
|---|---|
| tests per suite | 32 |
| submission size | 64 KiB |
| interpreter cycles per poll | 4,096 |
| polls per test | 256 |
| polls while loading the code | 16 |
| polls per run | 1,536; once spent, the remaining tests fail unrun |
| tests stopped by a limit before the rest fail | 4 |
a suite's check function |
its own 1,536 polls for the whole run, apart from the submission's, so the submission can neither starve it nor be starved by it |
| loop iterations per call frame | 20,000 |
| recursion depth | 256 |
Compute is counted in interpreter cycles, not time, so a run is the same on any machine.
headroom
The grader asks OutLayer for 2 billion WebAssembly instructions per run. The worst case under these budgets is about 730 million, or about 1.46 billion for a suite with a check function, so every run fits. Infinite loops, deep recursion and loops at load time all end in a signed 0 of 32.
Every open job has already shown its suite fits: the buyer's reference passed it in the same enclave before the job opened.