Repository navigation
TEST 118
Run the source-config and assisted-benchmark adapters through an integration fixture with explicit model approval, minimized evidence, budget/deadline policies, and downstream acceptance consumers. Inject a deterministic failure despite model acceptance, a missing mandatory review, a blocker overflow, unreviewed-frame citation, model substitution, expired disclosure, timeout, and an incomplete checkpoint. Resume a valid interrupted synthetic review and inspect the resulting complete receipt. An optional separately authorized live case may use a pinned clone of the public ni/labview-icon-editor project: lv_icon_editor.lvproj, build specification Editor Packed Library, product lv_icon.lvlibp. The PPL is a build output, not the source project to clone; a build receipt independently binds source revision, project/specification, artifact hash, and actual execution environment. Do not execute a new build, mutate a VM, or disclose desktop images merely to run the offline contract tests.
Every accepted review has the approved model identity, exact submitted evidence/policy identities, justified bounded coverage, valid aggregate, and completed lifecycle. Any injected deterministic failure remains blocking despite model approval. Where source-config review is mandatory, its absence or invalidity leaves configuration validity false; advisory does not mean optional. Unknown, failed, cancelled, and partial outcomes never satisfy downstream success or authorize tools/lifecycle effects. Receipts remain private and minimized. The optional PPL case reports only the independently verified build outcome: existing Linux-container evidence is not a new native Windows build, a screenshot is not an artifact, and a positive Ollama assessment cannot create native-build proof. Planned integration cases are not live qualification evidence.