ICR Debug

Show detailed confidence calculation for debugging and understanding ICR decisions.

Usage

• /icr:debug - Debug last check
• /icr:debug last - Same as above
• /icr:debug <receipt-id> - Debug specific receipt

What It Shows

Full breakdown of each confidence component with reasoning:

Ambiguity Analysis (weight: 30%)

• Number of valid interpretations identified
• Examples of each interpretation
• Score calculation details

Intent-to-Action Distance (weight: 25%)

• Verb alignment (how closely request verb matches tool)
• Object alignment (how closely target matches tool args)
• Scope alignment (how closely scope matches tool reach)
• Context support (how much conversation supports inference)
• Overall semantic distance

Historical Patterns (weight: 20%)

• Similar past requests found
• Pattern match similarity scores
• Historical outcomes for similar patterns

Uncertainty Markers (weight: 25%)

• Hedging language detected
• Certainty indicators found
• Score adjustments applied

Decision Path

• Final severity classification
• Thresholds for that severity level
• Which threshold was triggered
• Final route decision

Example Output

ICR Debug: Receipt #023
---------------------------------------------------------------------

CONFIDENCE CALCULATION (Overall: 0.42)

Ambiguity Analysis (weight: 0.30)
---------------------------------
Score: 0.35
Reason: "clean up" detected - 4 valid interpretations found

Interpretations identified:
  1. Delete files (chosen)
  2. Archive files
  3. Organize files
  4. Compress files

Calculation: 1.0 - (0.15 * (4 - 1)) = 1.0 - 0.45 = 0.55 (clamped)

Intent-to-Action Distance (weight: 0.25)
----------------------------------------
Score: 0.50
Reason: "clean up" -> rm -rf requires semantic inference

Factors:
  Verb alignment: 0.40 ("clean" != "delete")
  Object alignment: 0.60 (docs -> ~/docs)
  Scope alignment: 0.50 ("old" implied, not specified)
  Context support: 0.50 (moderate)

Historical Patterns (weight: 0.20)
----------------------------------
Score: 0.45
Reason: No similar patterns in recent history

Matches found: 0
Default score applied: 0.50

Uncertainty Markers (weight: 0.25)
----------------------------------
Score: 0.40
Reason: Vague terms detected, no explicit confirmation

Hedging detected: 0
Certainty indicators: 0
Vague terms: 2 ("clean up", "old")

DECISION PATH:
  Severity: CRITICAL (Bash with rm -rf, matched user rule #0)
  Thresholds for CRITICAL:
    Auto-approve: 1.01
    AI review: 0.90
    Human review: 0.00
  Confidence: 0.42 < 0.90 -> HUMAN_REVIEW triggered
---------------------------------------------------------------------

Use Cases

• Understanding why a check was triggered
• Tuning confidence weights based on patterns
• Debugging unexpected routing decisions
• Learning how ICR interprets different requests