---
name: averaging
description: >
  Use this skill when calculating weighted or unweighted averages for ENSDF
  nuclear data using Java_Average.py. Enforces exact transcription of the
  Suggested Adopted Result, minimum-uncertainty rule, and lifetime
  uncertainty limit 99. Suitable for adopting measured values from multiple
  publications.
argument-hint: [VALUE1 UNC1 VALUE2 UNC2 ...]
---

# ENSDF Averaging

ENSDF 80-column data record and field definitions, structural rules, column positions, and uncertainty notation: `.github/agents/ENSDF-Agent.agent.md`. Spot-check policy: `.github/copilot-instructions.md`.

## How

**Numeric mode** — comma after each pair for readability (optional):
```bash
python .github/scripts/Java_Average.py 19.7 1.3, 22 4, 21.5 1.5
```

**Comment mode** — feed the existing cL T$ comment directly:
```bash
python .github/scripts/Java_Average.py --comment "19.7 ps {I13} (1970Br10) and 22 ps {I4} (1975Sm02)"
```

## What to adopt

- Transcribe the printed **Suggested Adopted Result** character-for-character to data fields; never recompute, re-round, or substitute. Comments quote must match the suggestion.
- Use the method the tool labels (Weighted vs Unweighted).
- Adopted uncertainty ≥ smallest input uncertainty (min-uncertainty rule); the tool enforces it — do not override.
- Lifetimes use full precision (limit 99): `197 fs {I50}`. In the record, `T`/`DT` = ln2 × tool τ/unc in the record's unit.

## Uncertainty digits by destination

| Destination | Digits |
| --- | --- |
| Comment line (`cL T$`, `cG RI$`, …) | as printed; 2 digits kept even when >35 |
| Data-record field (`E`/`DE`, `RI`/`DRI`, `T`/`DT`, `2 L BE2=`) | >35 → 1 digit |

When user requests code `Java_Average.py` for calculating averages, follow these rules with absolute precision and zero tolerance for deviation:

- Always use exact Java code `Suggested Adopted Result` value without recalculation or substitution
- Use exact uncertainty value provided by Java code (automatically applies rule: adopted uncertainty ≥ any individual input uncertainty)
- Check whether Java suggests weighted or unweighted average in output comments
- Use whichever method Java code explicitly recommends
- Transcribe all values character-for-character without rounding, adjustment, or omitting units
- Never recalculate averages by yourself
- Never use unrecommended uncertainty results
- Never substitute weighted/unweighted averages contrary to Java's recommendation

## Minimum Uncertainty Rule

**ENSDF-Specific Requirement:** Adopted uncertainty ≥ any individual input uncertainty.

**Rationale:** Prevents averaging from artificially reducing systematic uncertainty below best single measurement. Maintains conservative uncertainty estimates in nuclear data evaluation. Java_Average.py automatically enforces this.

- **Statistical avg < min input uncertainty** → Adopted = min input uncertainty
- **Statistical avg ≥ min input uncertainty** → Adopted = statistical average

**Example:** Averaging 665.56±0.05 and 665.6±0.1 yields statistical uncertainty 0.0447, but adopted uncertainty becomes 0.05 (matches smallest input).

## Gotchas

- **`[critical=X]` is display-only.** The tool decides Weighted vs. Unweighted not just based on the chi-squared critical value.
- **Lifetimes can use full precision** (uncertainty limit 99): write `197 fs {I50}`, not `2.0E2 {I5}`.
- **One value per paper.** Comment mode skips the value before `average of` (it's the old average result) and stops at `Other:`.