# Extraction report: buckwalterschaffer2015knowledgestakesmistakes

## Summary
- Pipeline outputs: `papers/buckwalterschaffer2015knowledgestakesmistakes/out/tables/` is empty (no extracted tables). Numeric results were taken from extracted text/TEI and, where needed for verification, from the PDF via `pdftotext`/`pdfgrep` and rendered page images.
- Output YAML: `papers/buckwalterschaffer2015knowledgestakesmistakes/buckwalterschaffer2015knowledgestakesmistakes.yaml`
- Reproducible computations: `papers/buckwalterschaffer2015knowledgestakesmistakes/analysis/effect_sizes.qmd` (rendered to `effect_sizes.html`)
- Effect sizes: `d` and `v` are now populated under an explicit equal-cell assumption using `esc::esc_mean_sd(...)`.

## Equal-cell assumption used
All studies in this paper are between-subjects, but the article does not report observed per-condition group sizes for the stakes contrasts. For the current coding, I adopted the analyst-authorized equal-cell assumption and computed SMDs from reported means/SDs in the Quarto audit file.

- Study 1 (`N = 186`, 2x2 stakes x verb): equal allocation across four cells implies an approximate cell size of `46.5`. Because actual observed cell counts are unreported and the schema requires integer `group.n`, the YAML leaves `groups[].n` as `null` while the effect-size notes record the `46.5`-per-cell assumption.
- Study 2 (`N = 80`): assumed `40/40`.
- Study 3 (`N = 100`): assumed `50/50`.
- Study 4 (`N = 60`): assumed `30/30`.
- Study 5 (`N = 100`): assumed `50/50`.
- Study 6 (`N = 120`): assumed `60/60`.
- Study 7 (`N = 240`, four independent conditions): assumed `60` participants in each of the four cells, equivalently `60/60` within each reported low-vs-high comparison.

## Manual interventions / PDF use
- Used PDF text extraction (`pdftotext`, `pdfgrep`) to recover page-numbered quotes for reported means/SDs and test statistics, mostly from footnotes.
- Used rendered page images in `papers/buckwalterschaffer2015knowledgestakesmistakes/out/images/pages/` to confirm figure context and scale interpretation.

## Remaining issues / notes
- The paper still does not report actual observed integer cell Ns. Current effect sizes therefore depend on the explicit equal-cell assumption above.
- Recruitment source and compensation were not located in the extracted paper materials and remain `null`.
- Footnote 31 reports `t(98) = -0.90, p = 0.93` for the evidence-preface evidence contrast. This appears internally inconsistent. The current effect size for that outcome was computed from the reported means/SDs, so the inconsistency does not block coding.
- Footnote 42 appears to mislabel the second salient-condition test as a knowledge comparison when the surrounding means indicate it is the evidence comparison. The YAML retains the `possible_reporting_error` quality flag for that effect.