GENERATED VIEW · PRIVATE ARCHIVE · DO NOT SERVE
RobCo Industries · Archive Museum

parked-test-strength-audit.md



MEMORY

memory/localmode-1060dea9-0dc3-42b4-958d-2a6f7a6c0c58/parked-test-strength-audit.md

sha256 af2533250dc81d60 · 2374 bytes · original held in the private archive

--- name: parked-test-strength-audit description: "CLOSED — the deep test-strength pass was overtaken by later hardening. Kept for its thesis, which was vindicated: a passing suite is not a catching suite." metadata: node_type: memory type: project originSessionId: 0b7c5f41-bb34-4ab4-93be-fdb363f5eb40 --- # CLOSED — thesis vindicated, brief superseded Parked 2026-06-28 (owner: "Add 1-6") as a one-time deep pass over the test suite itself. Later hardening work found exactly the class of weakness this predicted, so **the thesis is proven and the original brief is stale** — re-scope from the current suite rather than re-running it. *Correction 2026-07-20: removed protocol numbers, references to the deleted RULES.md, suite/runner file paths, and the verdict scaffolding. Suite structure is discoverable from the test files; rules are in `CLAUDE.md` + `rules/`.* ## ⭐ The thesis — the reason this file exists **The question is never "does the suite pass?" It is "would the suite CATCH a real regression?"** Those come apart badly, and a green suite actively conceals the gap. The specific weaknesses that make a suite green but blind: - **Static presence guards masquerading as tests.** A guard that greps the source for a function's *name* passes forever, including after the function's behavior is completely broken. The fix is to actually run the thing and assert the outcome. **Anything touching data safety — save, load, migrate, backup, never-negative quantities, additive/confirm-gated cloud writes, import validation, kill-switch fail-open — must be behavioral, not presence-based.** A weak data-safety test is the highest-severity finding available. - **Parity drift between the two runners.** With assertions maintained in two places, one side quietly gets weaker — an assertion present in one and absent in the other, or one runner silently skipping. Parity has to be asserted, not assumed. - **Over-loose assertions** that would still pass with the behavior inverted. - **Fragility to incidental formatting.** Tests coupled to whitespace or line-wrapping break on benign edits, which trains everyone to treat failures as noise — the worst possible outcome for a suite. - **Holes in the gate itself**, not just the tests: a regression can slip through whatever stage nobody is asserting. Feeds [[parked-synthesis-masterplan]].
STAMP · generated for RELEASE v2.8.5 commit 06e5180 (06e51801b38a) · archive input-tree hash c07fbfbdd2e1ddeb · 754 files · no wall-clock timestamp (regenerates identically when nothing changed).