# Extraction report: turrindepistemiccontextualismidle

## Pipeline notes
- Followed the pipeline outputs in `papers/turrindepistemiccontextualismidle/out/` (bundle.json, tables CSVs, fulltext/text, TEI).
- Initial extraction did not consult the raw PDF directly. A follow-up effect-size audit consulted `pdf/paper.pdf` via `pdftotext`
  and rendered pages to verify that Experiment 4 (bank) reports no `d`, while the flight replication reports `d = 0.34`.

## Key extraction choices
- Created one `studies[]` entry per experiment + each conceptual replication (8 total).
- Extracted one primary effect per study (the “Know”/knowledge-related DV for Experiments 1–3 and the evidence-threshold DV for Experiment 4).
- Used `d = mean(low) - mean(high)` as the default sign convention for Low vs High contrasts.
- For Experiment 2 (deferral), stakes are held constant; the contrast is Yes (“I do know”) vs No (“I don’t know”) and uses `d = mean(yes) - mean(no)` (documented in the YAML as `sign_convention: other`).

## Problems / uncertainties (needs human input)

### [FIELD] Paper publication details
The extracted outputs identify this as a “penultimate version … to appear in Australasian Journal of Philosophy” and do not provide the final publication year, volume/issue/pages, or DOI.

Please fill:
- Final published citation (journal details) and DOI (if any)
- Publication year
- Whether to code `paper.published` as `Yes`/`No`

## Effect-size update
- `analysis/effect_sizes.qmd` was rebuilt to use `esc::esc_mean_sd` for between-group computations.
- Experiment 4 (bank; Study 7) reports total `N=99`, group means/SDs, and `t(97)`, but not split condition Ns and not `d`.
  Per analyst instruction, `d`/`v` are now computed using the equal-N assumption `n_low=n_high=99/2`.
- The Study 7 equal-N estimate is `d=-0.161379553952`, `v=0.040535572527`. Checking the two possible integer splits
  (`49/50` and `50/49`) changes `v` by about `0.000004`, so the assumption is practically negligible for the variance.

## Minor extraction issues (handled)
- `tabula_stream_p15_t4.csv` (Study 4; Conceptual replication of Experiment 2) appears to mis-extract the MD value for the “Know” row (listed as `-0.68` despite means 4.54 vs 4.61 and CI centered near -0.07). Effect size was computed from means/SDs + split Ns instead.
- `tabula_stream_p9_t1.csv` (Study 1) shows an implausible df for “Belief” (`1765`); this row was not used for the extracted effect.