Trading Operating Plan

# Trading Operating Plan

Last updated: 2026-07-10

**Dashboard:** Read `docs/dashboard/STRATEGY_AND_SHADOW_GUIDE.md` first for live strategy, shadow tests, and promotion rules.

## Purpose

This is the day-to-day operating plan for the bot. It translates the broader
overhaul roadmap into a practical routine focused on one outcome: build a bot
that can trade slowly, survive bad conditions, learn from results, and improve
without becoming unstable.

This plan is separate from:
- `OVERHAUL_PLAN.md`, which defines the active migration roadmap
- `KRAKEN_LIVE_READINESS_PLAN.md`, which defines Kraken readiness gates
- `docs/archive/plans/WORKER_SPLIT_PLAN.md`, historical live vs worker split notes

## Operating Goal

The bot must:
- place trades only when the evidence is strong enough
- protect capital before trying to maximize profit
- record every decision, fill, miss, stop, and exit
- learn from results without auto-changing live behavior
- keep live trading isolated from heavy research and dashboard work

## Daily Monitoring

Check these items every day:

1. Service health
   - bot active or inactive
   - restart count
   - crash loops or repeated restarts
   - slow-loop warnings

2. Market data quality
   - quote age
   - candle age
   - WebSocket freshness
   - REST fallback usage
   - stale-data blocks

3. Execution quality
   - order submitted
   - order filled
   - order canceled
   - duplicate-order prevention
   - pending order timeout
   - stop placement and reconciliation
   - profit-taking actions

4. Strategy quality
   - decisions made
   - decisions blocked
   - missed opportunities
   - trend alignment delay
   - score threshold behavior
   - regime classification

5. Learning quality
   - decision outcomes count
   - outcome labels freshness
   - matched cycles
   - closed-trade labels
   - recent recommendations

6. Dashboard quality
   - page refresh
   - stale sections
   - wrong provider labels
   - hidden errors
   - stale simulation or shadow data leaking into main views

## Weekly Review

Once per week, review:

- net PnL after fees and slippage
- win rate and loss size
- maximum drawdown
- average trade duration
- average favorable excursion
- average adverse excursion
- fill rate
- missed fill rate
- stop-out rate
- profit-taking rate
- performance by market regime
- performance by signal score band

The weekly review should answer:
- Did the bot trade for the right reasons?
- Did it avoid low-quality conditions?
- Did it hold winners too long?
- Did it cut losers quickly enough?
- Did costs destroy the edge?

## Promotion Criteria

A change may move from research to shadow, or from shadow to live review,
only if it improves evidence on at least one of these:

- net expectancy after costs
- drawdown
- fill quality
- stability
- missed-trade reduction
- exit quality

Promotion must also avoid regressions in:
- restart behavior
- stale data handling
- order duplication
- stop reconciliation
- dashboard correctness
- operator clarity

## What Should Stay Manual

These actions should stay human-approved:

- strategy changes
- risk limit changes
- symbol changes
- provider changes
- live-trading enablement
- leverage changes
- moving from paper to real funds
- auto-promoting a new strategy variant

## What Can Be Automated

These parts can be automated:

- hourly or six-hour health monitoring
- SQLite backups
- status export generation
- decision outcome labeling
- dashboard refresh
- stale-cache invalidation
- alert delivery when critical errors occur

## Simple Rule Set

1. If the bot is unstable, fix stability first.
2. If the bot is stable but unprofitable, improve strategy next.
3. If the bot is profitable in backtest but weak live, inspect execution.
4. If the bot is strong live but sparse, prefer quality over frequency.
5. Never let research workers block live trading.

## Review Questions

When reading the logs or dashboard, ask:

- Did the bot see the market correctly?
- Did it decide for the right reason?
- Did it execute the decision correctly?
- Did it exit correctly?
- Did the result improve or damage capital?
- Would we make the same choice again with the data we have now?

## Success Definition

The bot is moving in the right direction only if:

- it stays up without restart loops
- it keeps data fresh
- it records enough outcomes to learn from
- it places and closes trades correctly
- it improves on net results after costs
- it remains understandable to monitor