Test Offline

MUST use:

• .\test.ps1 -Lane full-offline-gate
• python -m test_kit.workspace_validation only after environment bootstrap sets PYTHONPATH to shared\testing\test-kit\src; prefer .\test.ps1, which already delegates through scripts\validation\invoke-workspace-tests.ps1

Final lane verdict MUST be anchored to canonical full-offline-gate outputs.

Execution Rules

1. MUST run from repository root.
2. MUST execute in this order: test-static -> test-unit -> test-contract -> test-integration -> test-e2e.
3. MUST NOT skip test-static.
4. MUST NOT skip test-contract.
5. IF skip_layer is set, THEN skip_justification is mandatory.
6. IF test-static returns BLOCK, THEN stop immediately and return BLOCK.
7. IF test-contract fails or blocks, THEN test-integration and test-e2e MUST NOT run.
8. IF test-integration fails or blocks, THEN test-e2e MUST NOT run.
9. MUST describe its result as offline closure only.

Evidence Rules

MUST reference:

• tests/reports/static/
• tests/reports/workspace-validation.json
• tests/reports/workspace-validation.md

SHOULD reference:

• case-index.json
• validation-evidence-bundle.json
• evidence-links.md

Blocking Policy

• PASS: all required L0-L4 validations passed with canonical evidence.
• FAIL: a required executed layer failed.
• BLOCK: required evidence is missing, invalid skips were requested, or prerequisite layers blocked progression.
• FAIL or BLOCK from this lane blocks offline-gate approval.

Output Contract

Scope

Entry

Evidence

• <static reports>
• <workspace validation reports>
• <evidence bundle paths>

Verdict

PASS | FAIL | BLOCK

Blocking Impact

none | block-lower | block-lane | block-release

Next Action

Only:

• rerun: .\test.ps1 -Lane full-offline-gate
• escalate: test-limited | test-nightly
• handoff: diagnose-* or implement-*

Examples

Positive:

• Use test-offline after changes that affect boundaries, runtime chains, DXF flow, backend composition, or any offline end-to-end behavior.

Negative:

• Do NOT use test-offline to sign off board-connected HIL behavior or long-run performance thresholds.

Prohibitions

• MUST NOT modify source code.
• MUST NOT relax gates, thresholds, or baselines.
• MUST NOT bypass canonical root-level entry as the source of the final verdict.
• MUST NOT infer higher-layer success from lower-layer success.
• MUST NOT emit ambiguous verdicts such as "mostly pass", "looks OK", or "probably fine".