# Extraction report: shin2014timeconstraintspragmatic

## Summary

- Extracted four fixed-time stakes contrasts from a paper whose main reported analyses are about time constraints, not stakes.
- Primary evidence used: `paper.yaml`, `out/bundle.json`, `out/fulltext.md`, `out/text.txt`, and `out/tei.xml`.
- Raw PDF was **not** used.
- Reproducible computations are documented in `papers/shin2014timeconstraintspragmatic/analysis/effect_sizes.qmd` and rendered to `papers/shin2014timeconstraintspragmatic/analysis/effect_sizes.html`.

## What was extracted

- Experiment 1 provides a low-stakes time contrast in the body text and a high-stakes time contrast partially recovered through TEI figure caption `fig_1`.
- Experiment 2 provides a high-stakes time contrast in the body text and an additional low-stakes time contrast only in footnote 32.
- To obtain meta-analytic high-vs-low stakes effects, I split the paper into four study records by holding time fixed:
  - Experiment 1, Less-Time
  - Experiment 1, More-Time
  - Experiment 2, Less-Time
  - Experiment 2, More-Time

## Problems and uncertainties

### Issue 1: Stakes effects are reconstructed, not directly reported

- The paper reports time effects within low-stakes and high-stakes variants, but it does not report direct high-vs-low stakes tests at fixed time.
- I therefore reconstructed four between-subjects stakes contrasts from reported group summaries and computed `d`/`v` in `analysis/effect_sizes.qmd`.

Human analyst response:
`[If you disagree with using fixed-time reconstructed stakes contrasts as meta-analytic units, specify which of the four contrasts should be excluded.]`

### Issue 2: Experiment 1 high-stakes summaries come from TEI figure-caption artifact `fig_1`

- The processed outputs did not preserve the high-stakes Experiment 1 numeric summary as clean body text.
- The values `HS-LT: N = 39, Mean = 4.23, SD = 1.55` and `HS-MT: N = 43, Mean = 2.92, SD = 1.98` were recovered from `out/tei.xml` figure caption `fig_1`.
- I marked the relevant group provenance with `table_ref: fig_1`.

Human analyst response:
`[If you want a second-pass confirmation against the PDF for the Experiment 1 high-stakes summary, note it here.]`

### Issue 3: Experiment 2 low-stakes pair is only summarized in footnote 32

- The low-stakes Peter pair is reported only as a short footnote summary:
  - `Peter-Low-Stakes-Less-Time: N = 34, Mean = 1.85, SD = 0.857`
  - `Peter-Low-Stakes-More-Time: N = 35, Mean = 2.71, SD = 1.29`
- The full low-stakes Peter vignette text is not recovered in the pipeline outputs; only the low-stakes comprehension probe survives in the appendix: `It isn't important that Peter has no typos in his paper. Even if there are some typos, it won't matter much.`
- Because of that, I set `effect_size.needs_review: true` for the two Experiment 2 effects and left `moderators.awareness: null` there.

Human analyst response:
`[If you can supply the full low-stakes Peter vignette wording from the PDF or another source, place it here for a second-pass extraction.]`

## Validation

- YAML was validated against `docs/stakes_meta_schema.json`.
- Extraction HTML can be rendered with `invoke render-extraction shin2014timeconstraintspragmatic`.

## Feasibility call

The paper is **conditionally feasible** for inclusion in a stakes meta-analysis, but not as a clean primary stakes study.

- Experiment 1 is usable with moderate confidence because both low- and high-stakes condition summaries are recoverable and the stakes contrasts are numerically unambiguous.
- Experiment 2 is weaker because the low-stakes pair is only summarized in footnote 32 and the full low-stakes vignette wording is not recovered in pipeline outputs.

My recommendation is:

- include the Experiment 1 reconstructed stakes contrasts in the main dataset only if your protocol allows fixed-time post hoc stakes contrasts from factorial condition summaries;
- treat the Experiment 2 contrasts as sensitivity-only unless a human analyst confirms the missing low-stakes vignette wording from the PDF.