---
name: yogsoth-ai-acu-nugget-recall
description: "Tactic: Extract atomic units from one paper and score how much of a caller-supplied summary covers. Use for ACU-style binary or Nugget-style ternary recall checks; cannot run without a target summary."
license: "Apache-2.0"
metadata:
  decimalai-skill-id: "853670be-4448-4b9e-9fd4-28b56e31cb07"
  decimalai-version-id: "dc2d684e-4c45-4fac-a3a6-ebe85df5f129"
  decimalai-version: "1"
  decimalai-body-sha256: "75ccd210e7c7e58e797ad097d4644c69eaebd586df2e1b277e04f5417cd85fb1"
  decimalai-name: "yogsoth-ai/acu-nugget-recall"
  decimalai-format: "1"
  decimalai-invocation: "model"
  decimalai-category: "research"
  decimalai-source: "https://app.decimal.ai/s/yogsoth-ai-acu-nugget-recall@1/SKILL.md"
---
# ACU / Nugget Recall

## Orchestration Pattern

1. Require `target_summary`; never generate the scored summary inside this
   tactic.
2. Fetch the paper and create `acu-nugget-recall/`.
3. Extract units before exposing the summary to the unit-writing step. ACU
   uses extracted/binary units; Nugget uses authored, importance-tagged units.
4. Match all units against `target_summary` in one call.
5. Aggregate the results and write `03-recall-aggregate.json`.

Write `01-atomic-unit-writing.json`, `02-atomic-unit-matching.json`, and
`03-recall-aggregate.json`. Record the summary source. Report every unmatched
unit, not just the score. For a single-summary Nugget run, carry the method's
weak per-topic reliability caveat alongside the number.