Neural Pattern Score

2. Determine Outcome

Classify the outcome:

• Success — the pattern was applied and the task completed correctly, on time, without rework
• Partial success — the pattern helped but required significant modification
• Failure — the pattern was applied but did not solve the problem or caused rework

3. Apply Confidence Adjustment

Update the confidence score based on outcome:

Clamp the result: confidence = max(0.0, min(1.0, confidence))

4. Update Usage Statistics

usage_count += 1
if outcome == success:
    success_count += 1
elif outcome == failure:
    failure_count += 1
last_used = [current ISO timestamp]

5. Write Updated Pattern

Save the updated values back to memory/patterns/pat-[slug].json. Preserve all other fields unchanged.

6. Update the Index

Update the pattern's entry in memory/patterns/index.json with the new confidence and last_used values.

7. Check for Deprecation Threshold

If the updated confidence drops below 0.1:

• Flag the pattern for review by neural-pattern-prune
• Add a warning to the log: "Pattern [name] confidence dropped to [value] — candidate for pruning"

8. Log the Scoring Event

Append to logs/claude_log.md:

### Pattern Scored — [timestamp]
- Pattern: [name] (pat-[slug])
- Outcome: [success/partial/failure]
- Confidence: [old] → [new]
- Usage: [count] total ([successes] success, [failures] failure)

Output

• Updated memory/patterns/pat-[slug].json with new confidence and usage stats
• Updated memory/patterns/index.json
• Log entry in logs/claude_log.md
• Deprecation warning if confidence < 0.1

Review Checklist

• Correct pattern file was identified and loaded
• Outcome was classified accurately (success/partial/failure)
• Confidence adjustment applied the correct delta for the outcome
• Confidence was clamped to the 0.0–1.0 range
• Usage counts were incremented correctly
• Index file was updated to reflect new confidence
• Low-confidence patterns were flagged for pruning