# Extraction report: chemla2013experimentingcontextualismnat

## Summary
- Pipeline outputs still used for manuscript structure and wording:
  - Extraction index/status: `papers/chemla2013experimentingcontextualismnat/out/bundle.json`
  - Study design + participants + appendix wording checks: `papers/chemla2013experimentingcontextualismnat/out/fulltext.md`
  - Scenario-text cross-checks and figure/table fragments: `papers/chemla2013experimentingcontextualismnat/out/tei.xml`
- Supplementary archive now used as the primary numeric source:
  - `papers/chemla2013experimentingcontextualismnat/out/external/Analyses-SHARE.zip`
  - Extracted payload: `papers/chemla2013experimentingcontextualismnat/out/external/Analyses-SHARE/`
- Output YAML:
  - `papers/chemla2013experimentingcontextualismnat/chemla2013experimentingcontextualismnat.yaml`
- Reproducible computation notebook:
  - `papers/chemla2013experimentingcontextualismnat/analysis/effect_sizes.qmd`

## What changed
- The previous extraction stored only two aggregate knowledge effects (positive and negative) based on Figure 5.
- The updated extraction replaces those aggregate rows with eight vignette-level knowledge effects:
  - Bank: positive, negative
  - Truck/bridge: positive, negative
  - Train: positive, negative
  - Spelling/typos: positive, negative
- Each effect now includes:
  - Low/high condition means and SDs
  - Paired t-test reconstructed from the raw data
  - Computed SMD (`d`, `v`) using the project's within-subject SMCRP convention

## Manual interventions / non-pipeline sources
- The supplementary `Analyses-SHARE` archive was extracted locally and used directly.
- The original `Analyses.R` script in the archive was followed for:
  - special-case recoding (`SCENARIO == 5`, misc coding fix)
  - inclusion filtering (`Ireponse > -1`, exclude `TYPE == "E"`, English-language respondents only)
- No manual digitization from charts was needed once the archive was available.

## Residual limitations
- The raw archive stores numeric knowledge-scenario ids (`K1`-`K4`) rather than explicit vignette names.
- The mapping `K1->Bank`, `K2->Truck`, `K3->Train`, `K4->Spelling` is inferred from appendix order and preserved as a quality flag in each effect.
- The train vignette text and moderator coding were corrected from the source PDF appendix directly, but the code-to-vignette mapping is still inferred rather than explicitly labeled in the supplementary archive.
- Color, miscellaneous, control, and bias items are not encoded as effects because this extraction targets the knowledge-stakes contrasts relevant to the meta-analysis.

## Validation / follow-up
- The new vignette-level means, SDs, and paired tests are reproducible from:
  - `papers/chemla2013experimentingcontextualismnat/out/external/Analyses-SHARE/raw-results-pro.txt`
  - `papers/chemla2013experimentingcontextualismnat/out/external/Analyses-SHARE/raw-results-anti.txt`
- The new computation path is documented in:
  - `papers/chemla2013experimentingcontextualismnat/analysis/effect_sizes.qmd`
- Moderator audit update:
  - `evidence_reliability` was re-coded per `docs/extraction_instructions.md` Section 6.
  - Final codes: bank = Medium, bridge = Medium, train = Medium, typos = High (applied to both polarity effects for each vignette).
  - The other moderator fields (`scenario`, `skeptical_pressure`, `awareness`, `evidence`, `attribution_person`) were re-checked against the appendix wording and left unchanged.
  - Bank-case provenance/quotes were corrected to the paper's experimental Figure 2 wording on PDF page 14, replacing earlier quotes pulled from the introductory DeRose/Table-2 version.
- Recommended follow-up:
  - re-render the extraction HTML after confirming the YAML (`meta-extract render-extraction chemla2013experimentingcontextualismnat`)