estimatelisted

# Estimate A duration is a **measurement of this codebase's own history**, not an opinion about how long the work feels. The measurement exists: `hooks/estimate-ledger.sh` records the agent's active minutes per task and per commit window across every Groundwork project on this machine. **Read the ledger before writing a number. Every time.** An estimate produced without it is the model's pretraining prior, and that prior is denominated in developer hours — which is exactly the failure this skill exists to prevent. ## Get the measurement ```bash ${CLAUDE_PLUGIN_ROOT}/hooks/estimate-ledger.sh --report ``` Add `--level=L2` or `--kind=crud` to narrow it. The output carries, per source and per scope, the sample size `n`, the median and p75 in **active agent minutes**. Two sources, never mixed: - **`task`** — one Groundwork task, start to `final-check`. The right unit. Preferred whenever `n ≥ estimates.min_sample` (default 5). - **`commit-window`** — the seed, backfilled from git history. A noisier unit: it measures the slice of work between two commits, which is smaller and more arbitrary than a task. Usable, and always named as what it is. If the ledger is empty, run `--backfill` once. Only if that also yields nothing does the coarse fallback in [`../../guidelines/ai-sdd-process.md`](../../guidelines/ai-sdd-process.md) §Estimates apply — and it is labelled coarse in the answer. ## What the answer contains 1. **A line per genuinely different block of work**, then a