# Extraction report: `dinges2021knowledgeasymmetricloss`

## Pipeline outputs used (preferred order)

- `papers/dinges2021knowledgeasymmetricloss/out/bundle.json` confirmed the extraction run completed but produced **no usable table CSVs** (all Camelot/Tabula candidates were filtered out).
- `papers/dinges2021knowledgeasymmetricloss/out/fulltext.md` used for vignette/prompt wording and context.
- `papers/dinges2021knowledgeasymmetricloss/out/tei.xml` used for DOI and publication date metadata.

## Manual interventions (documented)

### 1) Table extraction produced no outputs

- `papers/dinges2021knowledgeasymmetricloss/out/tables/` is empty (tables were filtered out by the pipeline).
- To verify the reported cell means/SDs, I rendered the PDF pages containing Table 1 / figures:
  - `papers/dinges2021knowledgeasymmetricloss/out/images/pages/p018.png` (Table 1 + ANOVA summary)
  - `papers/dinges2021knowledgeasymmetricloss/out/images/pages/p017.png` (Fig. 3)
  - `papers/dinges2021knowledgeasymmetricloss/out/images/pages/p019.png` (Fig. 4)

### 2) Used OSF data to recover per-cell Ns (and recompute summaries)

- The paper provides an OSF view-only link (footnote) to a complete dataset.
- I downloaded the Qualtrics XML export:
  - `papers/dinges2021knowledgeasymmetricloss/data/data_study_1.xml`
- This was necessary because Table 1 reports **means and SDs but not per-cell Ns**, and group Ns are required to compute `d` and `v` without guessing.

## Computations (reproducible)

- Effect sizes were computed in Quarto (no ad-hoc scripts):
  - `papers/dinges2021knowledgeasymmetricloss/analysis/effect_sizes.qmd`
  - Rendered audit output:
    - `papers/dinges2021knowledgeasymmetricloss/analysis/effect_sizes.html`
- Exclusions implemented per paper:
  - 8 attention-check failures (`check != 1` in the OSF data)
  - outliers: one low-stakes 90% response = 65; two high-stakes 70% responses = 50 and 70
- Note on OSF variable naming:
  - In the Qualtrics export, the condition variables are *suffix-swapped* relative to the paper’s “70% vs 90%” labels (handled explicitly in `analysis/effect_sizes.qmd`).

## Notes / potential issues

- Ingestion produced the directory key `dinges2021knowledgeasymmetricloss` (based on GROBID’s header date); the PDF issue header is `(2023)`. The extraction YAML uses `paper.year: 2023` with the DOI unchanged.
- The DV is a **certainty/evidence-threshold** measure (not a direct knowledge attribution), so `moderators.attribution_person` is coded `null` and justified in `moderators_coding`.

## Open questions (for human review)

None.