Testing: adopt bounded test-acceleration profiles

[edithatogo]

A cross-repository testing audit on 2026-08-26 found:

• 70 Python test/conftest files
• 7 GitHub Actions workflow files
• detected acceleration tooling: testpaths
• path signals: 0 performance/benchmark, 0 database, 0 HTTP/API/client

Recommended bounded work:

• Add a registered slow marker and a fast local profile that excludes it; keep full CI authoritative.
• Trial pytest-xdist with a bounded worker count and isolate resource-sensitive/file-mutating tests into a serial phase.
• Add an opt-in, serial pytest-testmon profile for changed-code feedback; rebuild its local database after dependency or inventory changes.
• Run python -X importtime and profile representative application and test workloads with Scalene before changing import structure.
• Trial pytest-gremlins only on a bounded fail-closed/core-logic target set; retain ordinary tests as the promotion gate.
• On supported Python/Coverage versions, measure the sysmon core; do not force it where branch coverage, plugins, concurrency, or dynamic contexts are incompatible.
• Use Freezegun only where an injectable explicit clock is impractical; avoid global time freezing around asyncio or concurrency.
• Do not add pytest-picked if testmon provides the needed dependency-aware selection; if retained, use it only to order or select directly changed test files.

Acceptance criteria:

• Record before/after wall time and collected/executed test counts.
• Keep a full clean-environment suite in CI; selective profiles are local feedback only.
• Document incompatibilities and exclusions rather than weakening test, coverage, mutation, rights, or publication gates.

Created from the owner-wide test-acceleration audit; implementation should be repository-specific.