AK-Benchbenchmark report

datasets/_sources/cadence/README.md

cadence

A small workflow-scheduling engine: cron-style schedules, a job graph with

past-run dependencies, a write-ahead log that survives crashes, and a lease

for leader election. It is a benchmark fixture for AK-Bench, so two things are

deliberate:

  • the timing rules are documented to the instant, because the benchmark tasks

are about boundaries and crash recovery, not about guessing a policy;

  • the test suite is narrow on purpose. The tasks hide the tests that cover the

hard cases, so an agent has to read and reason about the rules instead of

watching a red test.

Run the suite with pytest.