Trade Observations
Stop Guessing and Start Observing

AI Exit Warnings: From Narrative to Actionable Signal

March 25, 2026
#systems#risk#research

Converting narrative AI Analysis into a simple, actionable signal.

Overview

This post documents a major upgrade to the Trade Observations system: transforming AI-generated market analysis into a structured, testable signal.

Originally, the AI analysis produced narrative descriptions of price action. While informative, these required subjective interpretation and were difficult to operationalize in a systematic workflow.

The solution implemented here introduces a second-pass classification step that converts narrative analysis into a simple, actionable signal: an AI Exit Warning.

Architecture Update

Before

  • AI generates descriptive analysis (Al Brooks style)
  • Trader interprets context manually
  • No structured signal for automation or analytics

After

  1. First Pass (Analysis)
    • AI generates structured analysis from 5-minute bars
  2. Second Pass (Classifier)
    • AI re-evaluates its own output
    • Produces structured fields:
      • ai_exit_warning (0/1)
      • ai_exit_warning_side
      • ai_exit_warning_strength
      • ai_exit_warning_reason
      • ai_exit_warning_confidence
  3. Storage
    • Stored in ai_snapshots
  4. Visualization
    • Dashboard shows:
      • AI Bias
      • GTO Signal
      • Exit Warning overlay
      • Modal for detailed review

Example Use Case

A recent trade illustrates the value:

  • Market at top of a trading range
  • GTO flips from FLAT → LONG
  • AI warning: > "Late bull channel near resistance; avoid buying high, pullback risk rising"

Interpretation

Instead of chasing the long:

  • Recognized repeated resistance
  • Framed as a fade opportunity
  • Entered a short with defined risk/reward

Result

  • Risk: ~3.5 points\
  • Reward: ~8.5 points\
  • Outcome: profitable

Key Insight

The AI warning did not act as an "exit" signal.

It acted as:

  • Context filter
  • Entry caution
  • Fade opportunity identifier

Reframing the Signal

Although named ai_exit_warning, the signal is better understood as:

  • Trade management warning
  • Context awareness flag
  • "Do not chase" indicator

System Roles

  • GTO → Event / signal
  • AI Warning → Context / caution
  • Trader/System → Execution logic

Metrics Introduced

To evaluate effectiveness, we introduced:

1. Exit Efficiency

Final profit relative to available profit at warning:

ExitEfficiency = realized_pnl / mfe_at_warning

2. Giveback %

Giveback = (MFE_at_warning - realized_pnl) / MFE_at_warning

3. Warning Value (future)

Compare: - Actual exit - Hypothetical exit at warning

Early Observations

  • Warnings cluster near:
    • trading ranges
    • exhaustion moves
    • failed breakouts
  • High-strength warnings often precede reduced follow-through
  • The signal appears more useful for:
    • avoiding bad entries
    • tightening exits
    • identifying fades

Future Improvements

1. Classification Refinement

  • Improve prompt specificity
  • Reduce false positives
  • Calibrate strength levels

2. UI Enhancements

  • Color intensity by strength
  • Overlay warnings on GTO chart
  • Add direct links to snapshot analysis

3. Analytics Expansion

  • Compare:
    • trades with warning vs without
  • Segment by:
    • strength
    • side alignment
  • Track:
    • exit efficiency improvements

4. Machine B Integration

  • Use warning to:
    • tighten stops
    • reduce position size
    • trigger partial exits

5. Trade Tagging

Categorize warning usage: - Exit preservation - Entry caution - Fade setup

Key Takeaway

This upgrade shifts AI from:

Narrative → Signal

From: - subjective interpretation

To: - measurable, testable edge

Next Steps

  • Continue collecting data
  • Validate metrics across trades
  • Gradually integrate into execution logic

The goal is not to replace decision-making, but to augment it with structured context awareness.