# Crypto trader idea testing and plain-language reporting

Use this reference when the user asks Jarvis to act like a professional crypto trader, bring independent strategy ideas, test them, and decide whether/when to resume live trading.

## User-facing style

- Reports must be short, plain-language, and trader/investor-readable.
- Do not dump raw JSON, process logs, commits, or developer detail unless explicitly asked.
- Lead with: live paused/running, current PnL, open positions/stops, whether the bot is currently profitable, and the next decision.
- Explain critical alerts in one sentence: what it means, whether funds/positions are safe, and whether new entries are blocked.

## Operating stance

- Treat repeated negative closes as an edge failure, not a sizing/leverage problem.
- Pause/kill-switch new live entries before strategy research if the live history is negative.
- Keep existing positions protected by exchange-level reduceOnly stops and reconcile before/after any runtime change.
- Max leverage configured by the user is a cap, not a mandate; keep practical leverage at 1x until paper/shadow evidence is positive.

## Idea generation pattern

Bring independent ideas, but test them side-effect-free first:

1. Market-regime filter: BTC/ETH define bullish, bearish, or choppy regime.
   - bullish -> test long candidates only;
   - bearish -> test short candidates only;
   - choppy -> no trade.
2. Candidate filters: trend alignment across 15m/1h/4h, RSI health, ATR/volatility guard, spread/order-book check, volume confirmation, and expected move versus fees.
3. Strategy archetypes worth testing independently:
   - volatility squeeze breakout;
   - liquidation reversal with volume spike;
   - relative-strength rotation;
   - risk-managed trend following;
   - trend pullback after market confirmation.
4. Avoid promoting fast RSI/scalp churn when history shows short holding-time losses.

## Promotion / live restart gate

Do not restart live from replay alone. Require a fresh paper/shadow sample such as:

- at least 30 closed paper/shadow trades;
- total paper PnL positive;
- last 20 closed paper trades positive;
- winrate >=45% or profit factor >=1.25;
- no dominance by <15m scalps;
- reconcile clean, stops confirmed, and kill-switch deliberately removed.

## Learning loop

- Write compact learnings under `runtime/reports/`.
- Promote findings into pure strategy/risk code only after paper/shadow data supports them.
- Daily reports should include one concise learning and one current optimization, not raw telemetry.
- When paper/shadow trades close, analyse timing around each entry and exit before changing rules: pre/post move, MFE/MAE, RSI, SMA/EMA alignment, candle rejection/reclaim, missed upside, and avoided drawdown.
- Treat chart-analysis findings as hypotheses until repeated across a useful sample; do not overfit one or two trades.

## Supporting references

- `references/trade-timing-and-chart-analysis-feedback-loop.md` — read-only workflow for entry/exit window analysis, chart-pattern feedback, and turning timing evidence into safer entry/exit gates.