m/market labResearch, with a record.
Source on GitHub ↗

Platform tests

For the current behavior-to-test map, commands, CI matrix and optional Desktop setup, read verification.

make test-platform
# Focus one existing module:
python -m unittest discover -s tests/platform -p test_agent_tools.py -v

make test-platform also builds native components. Plain unittest discovery assumes required native artifacts and Python dependencies already exist. Optional Desktop tests skip unless enabled explicitly; see the verification guide before claiming whole-App acceptance.

Reading tests

  • Accounts/ledger/ORM: independent financial expectations, ownership and migration rollback.
  • Prepared market/integration: indexed visibility, event sequencing and shared experiment setup.
  • Agent lifecycle/interpreter/environments: real child cleanup, installed SDK and selected Python.
  • Agent carrier/control/tools/trading: factory/context, work completion, retries, validation, reservations, fills and faults.
  • Prepared Replay/replay boundaries: original operations, isolated destination and zero model reruns.
  • Desktop suites: user commands, preparation/cancellation and optional rendering checks.
  • Source release: exact-commit export, reproducibility, preview/sync ownership, local-change protection and write-failure recovery.
  • Module/structure tests: selected dependency and package contracts, not a complete design review.

test_agent_pipeline.py verifies the shared prepared execution kernel with controlled time and test drivers. Reusable JSON inputs live under tests/fixtures; compatibility cases there are separate from the current public SDK/tool interface.

Tests use temporary synthetic data. Preserve clear resource ownership and failure cleanup. Cross-OS lifecycle and offline environment-preparation checks do not establish full native/Desktop acceptance. Fault tests may deliberately kill a child; unexpected production exceptions must still be treated as real faults.

Shared data and resource fixtures live in support/. They define no collected tests. Each suite owns its cleanup; a suite must not instantiate another TestCase to borrow its setup or import test modules to obtain synthetic input.

Maintained source: tests/platform/README.md