跳转至

Historical version

This page records design, operation, or validation facts from an earlier release. It is not current V7.17.3 guidance. Start from the documentation hub for current instructions.

V7.5.0 Audit Report

Scope covers task feedback, observation health, effective increments, testable hypotheses, benefit lifecycle, regression candidates, cross-ledger calibration and bilingual delivery. One pre-implementation review and two post-implementation rounds used five approved Terra Medium dispatches. All were logically readonly; neither system isolation nor host budget reservation is claimed.

Six accepted findings were repaired and rechecked: linked-worktree identity, qualified signal changes on existing tasks, exclusion of implementation tasks from benefit cohorts, conservative small-sample critical values, raw input limits before deduplication, and replay/recomputation of regression followups.

Each project/repository/session/turn/task permits one immutable final report. Workspace changes after finalization require a new turn to avoid conflicting terminal reports. Tests and bilingual documentation cover this contract, and the reviewer accepted it.

Legacy event and proposal chains remain readable without rewriting. Default and explicit policies agree. Each calibration sample is verified against its own ledger before comparison within one project/repository and matching role, responsibility, difficulty, risk and context. Independent-task counts and uncertainty intervals are retained. Benefit evidence is observational, not a causal guarantee.

Automation may collect minimal metadata, check, compare and prepare candidates. It never automatically accepts or implements proposals, changes business code/Skills/AGENTS/routing, deploys or promotes knowledge across projects. Implementation PASS and benefit SUPPORTED remain separate.

See the validation report for test and installation boundaries. Remote release state is independently evidenced by workflows and assets.

Windows CI exposed mixed short-path aliases (such as RUNNER~1) and full paths in relative references. Feedback and regression enumeration now use fixed references beneath checked directories, and project identity uses the canonical name of the verified Profile path. A real Windows short-path regression passed. Post-fix focused tests and installation readback are recorded separately; the first complete package run belongs to the preceding candidate, and final full-suite evidence comes from CI for the fix commit.

The subsequent Windows alias repair passed 44 focused tests, six runtime tests and one additional logical-readonly review (six reviewer dispatches across the complete task). The updated payload was reinstalled and its installed feedback/sealing smoke checks passed.