Test Suites and Runtime¶

The 2026-08-08 profiling baseline covered the then-current 266-test tree: 256 routine tests and 10 slow JAX-based NashConv, satisfaction, and AlphaRank integration checks. The full run completed in 6m56s on the development workstation; a parallel verification matching the CI layout completed the then-current fast shard in 3m46s and the slow shard in 2m10s.

The equilibrium refinement adds 12 tests to the 470-test pre-Nash-Q-rerun tree, covering event-delayed exact evaluation, deployed clock/monitor inputs, nonfinite checkpoint rejection, inactive CER response configuration and isolated campaign admission. The concurrent Stag Hunt rerun contributes additional tests. Coins, native-object-observation, notebook, launcher, CER, and equilibrium coverage was added after the timing baseline, so the historical durations are planning estimates rather than current benchmarks. CI can still execute the two shards in parallel without dropping coverage.

The legacy-numbering primary-tiebreak integration spawned several isolated automaton probes and took 81.47 seconds in an isolated four-CPU check on 2026-10-12, exceeding the routine 60-second timeout. It now belongs to the slow tier and retains all assertions with the existing 240-second integration timeout. This classification changes test scheduling only; checkpoint numbering and re-evaluation behavior are unchanged.

Use the repository runner during development:

# Select non-slow tests from tracked and untracked working-tree changes.
uv run python scripts/run_tests.py changed

# Broad non-slow regression suite.
uv run python scripts/run_tests.py fast

# One component, excluding its slow tests by default.
uv run python scripts/run_tests.py verification

# Include that component's expensive integrations.
uv run python scripts/run_tests.py verification --include-slow

# Expensive tier only, or every test.
uv run python scripts/run_tests.py slow
uv run python scripts/run_tests.py full

changed compares the working tree to HEAD. Pass --base origin/main to include committed branch changes, or repeat --changed-file PATH to inspect an explicit set. Arguments after -- are forwarded to pytest.

Change-to-suite map¶

Every tests/test_*.py module has one primary marker in tests/suite_manifest.py. Collection fails if a new test module has not been classified. Cross-cutting implementation paths can select several components.

Changed code	Routine suite	Include slow when
`src/environments/`	`env`	environment behavior changes affect training integrations
pursuit environment code	`env`, `rl`, `verification`	usually; its integration coverage lives in learner and verification tests
matrix environment code	`env`, `experiments`	experiment construction or training changes
`src/monitor/`	`env`, `monitor`	monitor state or wrapper behavior changes
`src/monitor/counterfactual.py`	`env`, `monitor`, `rl`	replay enumeration or wrapper contracts change
`src/rl/`	`rl`, `verification`	trainer construction, updates, replay loops, or evaluator integration changes
`src/verify/`	`experiments`, `verification`	rollout evaluation, NashConv, AlphaRank, confidence, or driver integration changes
`src/experiments/`	`experiments`	orchestration or artifact persistence changes
pursuit and temporal-game experiment drivers	`experiments`, `verification`	verification stages or real trainer calls change
environment-specific experiment adapters	`env`, `experiments`	wrapper construction or dispatch changes
`src/visualisation/`	`visualisation`	plot exports or presentation behavior changes
`notebooks/`	`experiments`	notebook configuration or launcher behavior changes
`notebooks/utils/regraph.py`	`experiments`, `visualisation`	regraph discovery or export behavior changes
`scripts/run_tests.py`	exact runner test	never
a `tests/test_*.py` file	that exact file	the edited test has the `slow` marker
dependency, pytest, or unknown implementation configuration	`fast`	use `full` if integrations may change
documentation only	no pytest selection	run `uv run zensical build --strict`

The component names are env, monitor, rl, verification, experiments, visualisation, and tooling. They describe primary ownership rather than dependency isolation. Use fast for broad or uncertain changes.

Runtime policy¶

The slow marker is reserved for tests measured at roughly ten seconds or longer. Slow tests receive a 240-second timeout; normal tests retain the 60-second project default. Re-profile before changing the manifest:

uv run python -m pytest -q --durations=50

JAX's persistent compilation cache is stored under .cache/jax-test-compilation, which is already ignored. Cache keys include the compiled program and configuration, so implementation changes create new entries naturally.