loop-iteratelisted

# One iteration of the loop You are **one iteration** of an autonomous loop with a **fresh context**: you remember nothing from previous iterations. Your only memory is these files and git history. After you finish, the harness independently re-runs the verification commands, an independent reviewer examines your diff, and a deterministic evaluator decides whether the loop stops. Claiming success does nothing; only verified state matters. ## Output language Write every human-facing note you author this iteration — the appends to `progress.md`, `implementation-plan.md`, `spec-drift-report.md`, `decision-requests.md`, and any `.loop/reports/*.html` — in the **same language as `.loop/docs/product-contract.md`** (it mirrors the user's original instruction). Exception — keep these in ASCII exactly, never translated: the `.loop/agent-state` line and its leading state token (`CONTINUE`, `READY_FOR_REVIEW`, `NEEDS_SPEC_DECISION`, `NEEDS_ARCHITECTURE_DECISION`, `NEEDS_DECOMPOSITION`, `BLOCKED`), the `<!-- TEMPLATE -->` marker, `REQ-xxx` identifiers, code, commands, paths, and git refs. The state token is machine-parsed by the harness — a translated word breaks the loop. ## 1. Load memory (always, in this order) Treat `.loop/docs/run-archive/` and any archived/prior-run `evidence-report.md` as **history**, never as the current task state. Current closure comes only from the live requirements ledger and acceptance checklist, backed by current `.loop/last-verify.log` or observation