confidence-scoring

Confidence Scoring

Determines whether an observed improvement is real or within measurement noise using Median Absolute Deviation (MAD).

When to Compute

• After 3+ experiment runs in the current segment (including baseline).
• Skip if fewer than 3 runs with positive metric values — report confidence: null.

Algorithm

Given all metric values in the current segment (positive values only):

1. Sorted median: Sort values, take middle element (or average of two middle elements for even count).
2. MAD: For each value, compute |value - median|. Take the sorted median of those absolute deviations.
3. Baseline: The metric value of the first experiment in the current segment.
4. Best kept: The best keep-status metric value (respecting optimization direction).
5. Delta: |best_kept - baseline|
6. Confidence: delta / MAD

Edge Cases

• If MAD = 0 (all values identical): return null — no measurable noise to compare against.
• If no keep results exist yet: return null.
• If best kept equals baseline: return null — no improvement to score.

Interpreting the Score

The confidence score is a multiple of the session's noise floor:

Score	Meaning	Action
≥ 2.0×	Improvement likely real	Safe to trust
1.0×–2.0×	Marginal — could be noise	Consider re-running to confirm
< 1.0×	Within noise floor	Treat as no improvement

How to Apply

When logging an experiment result to autoresearch.jsonl:

1. Collect all positive metric values from the current segment.
2. If count < 3, set "confidence": null in the JSONL record.
3. Otherwise, compute MAD and confidence as above.
4. Record the numeric confidence value in the JSONL entry.
5. When deciding keep vs discard: the confidence score is advisory. It never auto-discards. But flag improvements below 1.0× in your ASI notes as "within noise — may not be real."

Example

Runs: [15200, 15400, 14800, 15100, 14600]
Median: 15100
Deviations: [100, 300, 300, 0, 500] → sorted: [0, 100, 300, 300, 500]
MAD: 300
Baseline: 15200 (first run)
Best kept: 14600
Delta: |14600 - 15200| = 600
Confidence: 600 / 300 = 2.0×  ← improvement is real