source PDF: —
open in tab ↗Your annotation
Notes (optional)
keys: j/→ next · k/← prev · u next unfilled · 1-9 pick pill · 0 no-quote · ctrl+s force save
This annotation site is invitation-only. Open it through the personal link you were sent (it contains your key). If your link stopped working, ask the study coordinator for a new one.
We are auditing how benchmark papers describe what their benchmark measures. For each paper you answer a fixed set of questions (“fields”) — e.g. does the paper claim the benchmark measures a construct? quote the sentence that says so. Your answers come only from the paper itself (left pane), never from prior knowledge about the benchmark, and never from guessing the “desired” answer.
You always code your own copy, blind: nobody else can see your answers, and you cannot see theirs. There is no “correct rate” shown anywhere — honest disagreement is expected and useful.
Everything autosaves ~0.4 s after you stop editing (watch the saved / unsaved indicator).
Some fields only make sense conditional on another: if you answer
construct_claim_present = no, the two quote fields under
it are automatically filled with [none]. You will see an
orange “auto-gated row” banner on such rows — don't fight
it; if you think the value is wrong, go back to the gate field it
names and change that.
The unfilled filter shows only rows you haven't answered; the benchmark dropdown shows your per-paper progress. Your position is remembered, so closing the tab is safe.
Code the practice paper below end-to-end (all fields, in order). Work exactly as you would for real: read the codebook box, search the PDF, quote precisely (copy the paper's own words), use notes when torn between two answers. Expect 45–90 minutes. We will go over everyone's practice answers together before the real assignment starts.
keys: j/→ next · k/← prev · u next unfilled · 1-9 pick pill · 0 no-quote · ctrl+s force save