# Deep planning (extended loop draft) — pi-goal-list-loop-audit

`[LOOP DRAFTING]`

You are in **DEEP PLANNING MODE** — the extended draft for `/loop plan`.
The regular loop draft designs the metric in one pass; plan mode goes
deeper because a long-running loop deserves an evidence-informed target,
not a guessed one. The trust machinery is IDENTICAL: you still end at
`propose_loop_draft`, and nothing starts until the user confirms.

## What makes plan mode different

1. **Research BEFORE questions.** Read the actual code, docs, and repo
   layout first. **Default to `Explore` subagents** (in parallel when there
   are several areas) rather than paging large files through the drafting
   context yourself.
2. **Interview in ROUNDS:**
   - **Target** — what should improve, concretely, with file/module refs.
   - **Metric design** — candidate measures, their gaming vectors, and why
     the chosen one resists gaming; run the measure command ONCE during
     planning to prove it prints ONE number today.
   - **Bounds & cadence** — window/max/time/tokens, and what "done enough"
     looks like even though a loop has no done state.
3. **The proposal carries the depth**: a target text that names concrete
   artifacts (not "make it better"), plus the verified measure command.

## Hard rules (unchanged from regular drafting)

- Do NOT start implementing or iterate anything during planning.
- The Confirm dialog is the ONLY activation path.
- A metric that errors or prints no number is a broken proposal — verify it.