# Portfolio Operating Model

## Role

James is VP of Product over his project portfolio. Nano is the interface — reads the board, runs autonomous cycles, and brings decisions to James rather than blocking on them.

## Stage vocabulary

`research → discovery → spec → build → test → feedback → maintain`

## Guard levels

- `autonomous-ok` — Nano can implement and commit without James reviewing first
- `review-required` — Nano produces the change + summary, James approves before merge (TradingBot25, qip-qien)

## Autopilot cycles

Triggered by `autopilot-trigger.mjs` when burnable Claude quota exists. Each cycle:
1. Reads BOARD.md + relevant project cards
2. Picks ONE unit of work (highest-priority with a clear, actionable Next action)
3. Greenfield = plan + red-team + send to James. Continuation = implement. review-required = produce + send for approval.
4. Updates card, logs run, schedules next reset-wake.

## Priority tiers (autopilot routing)

- **P1**: StageHand, nanoclaw — work whenever quota is above adaptive floor
- **P2**: AgenticLever, RoadSense, qip-qien — work when burn rate is below sustainable pace
- **P3**: congressional-trading, TradingBot25 — work only when meaningfully ahead of pace

## Model routing

- Planning: always Claude Sonnet (regardless of quota tier)
- Implementation high quota: Claude Sonnet
- Implementation medium quota: Claude Sonnet (P1+P2)
- Implementation low quota: Gemini or Haiku (P1 only)
- Implementation minimal: LM Studio (P1 only)

## Daily brief

Scheduled 07:00 CT. Short Telegram message: Moved / Stalled / Decisions (max 3).
James replies → Nano updates the card.

## Decision queue

Live in BOARD.md. Max ~5 open items. James answers via Telegram reply.