Summary
The repository would benefit from an explicit policy for what belongs in PR-gate CI versus periodic / nightly CI for cross-backend consistency tests.
Right now, some jobs appear to carry a mixture of:
- must-not-break smoke/regression coverage
- broader backend consistency checks
- very expensive exhaustive matrix coverage
This makes PR CI slower than necessary.
Motivation
Heavy cross-backend suites are valuable, but not every combination needs to run on every PR.
A clearer split between reduced PR coverage and exhaustive periodic coverage would make CI faster while preserving confidence.
Proposed direction
Define and document coverage tiers such as:
- PR gate: representative smoke/regression coverage on the most relevant backends/configurations
- Periodic/nightly: broader backend coverage and more exhaustive matrices
- Specialized jobs: CPU-only or backend-specific checks where appropriate
Scope
Applies to descriptor/model/fitting consistency suites and other heavy cross-backend tests.
Acceptance criteria
- CI jobs are clearly categorized by coverage level
- heavy exhaustive combinations are moved out of PR-gate jobs where appropriate
- PR CI wall-clock time is reduced while nightly/periodic jobs retain broader coverage
Authored by OpenClaw (model: gpt-5.4)
Summary
The repository would benefit from an explicit policy for what belongs in PR-gate CI versus periodic / nightly CI for cross-backend consistency tests.
Right now, some jobs appear to carry a mixture of:
This makes PR CI slower than necessary.
Motivation
Heavy cross-backend suites are valuable, but not every combination needs to run on every PR.
A clearer split between reduced PR coverage and exhaustive periodic coverage would make CI faster while preserving confidence.
Proposed direction
Define and document coverage tiers such as:
Scope
Applies to descriptor/model/fitting consistency suites and other heavy cross-backend tests.
Acceptance criteria
Authored by OpenClaw (model: gpt-5.4)