Climate research publication velocity is accelerating

Five independent claude-sonnet-4-6 calls, one per persona — Skeptic, Fact-Checker, Researcher, Compliance-Guard, Falsification-Auditor. Each writes its hardest objection from its own seat, paired with the methodological resolution it would accept. Run on the static catalogue spec Jun 3, 2026; a live council for any topic is at /try.

1. Skeptic #01
   Raised

   The observed compression in preprint-to-publication delta likely reflects a secular shift in when authors post preprints relative to their journal workflow, not genuine acceleration in peer review. Between 2015 and 2025, community norms shifted from posting preprints before or concurrent with initial submission toward posting them after acceptance or as author-accepted manuscripts—a pattern documented in bibliometric studies of bioRxiv and accelerated by journal preprint-friendly policies adopted post-2019 at Nature Climate Change, GCB, and ERL. If climate authors increasingly post to arXiv/bioRxiv after receiving a conditional acceptance, the measured preprint-date-to-publication-date delta mechanically collapses without any change in editorial turnaround time. This artifact would produce exactly the <9-month median signal while leaving actual peer-review velocity unchanged.

   Resolved

   The CrossRef `/works` API returns a `submitted` date field (populated for roughly 60–70% of PubMed-indexed journal articles) representing the date of first manuscript submission to the journal, independent of preprint posting date. For each matched preprint–publication pair, decompose the total delta into two sub-intervals: (preprint_date → journal_submitted_date) and (journal_submitted_date → published_online_date), then apply a Mann-Kendall trend test to each sub-interval independently across annual cohorts 2015–2025. If the statistically significant compression concentrates in the first sub-interval—meaning preprint posting has crept temporally closer to or past submission—while the submission-to-publication sub-interval shows no significant downward trend (τ near zero, p > 0.05), the hypothesis is falsified as a measurement artifact; only a significant negative τ on the submission-to-publication sub-interval would confirm genuine publication velocity acceleration.

2. Fact-Checker #02
   Raised

   The primary measurement uncertainty arises from PubMed date-field ambiguity: the gap between PubMed's `ArticleDate[@DateType="Electronic"]` (epub-ahead-of-print) and `PubDate` (print issue date) introduces a systematic bias of 2–5 months depending on journal, meaning date-field selection alone can shift median estimates by that amount without any change in actual publication velocity. Compounding this, CrossRef's `relation.is-preprint-of` metadata covers only an estimated 40–60% of climate preprints that ultimately publish, and that coverage is concentrated at journals with formal preprint acceptance policies—which process submissions faster—creating a selection bias that depresses the measured median by a further 1–3 months. The combined uncertainty budget (~3–6 months) is comparable to or exceeds the 3-month gap between the SUPPORTS (<9 months) and FALSIFIES (>12 months) thresholds, rendering the hypothesis untestable as written.

   Resolved

   Standardize the date-field protocol to arXiv v1 `submitted_date` paired exclusively with PubMed `ArticleDate[@DateType="Electronic"]`, excluding any record for which only `PubDate` (print) is available in the PubMed XML; apply CrossRef's `asserted-by: publisher` filter on `relation.is-preprint-of` linkages and restrict matching to Crossmark-enabled records (indicating publisher-verified, complete metadata deposits). Expand the threshold band to SUPPORTS <7 months versus FALSIFIES >15 months, providing an ≥8-month buffer above the 6-month combined uncertainty ceiling, and construct bootstrap 95% CIs on annual medians stratified by SHERPA/RoMEO preprint policy tier so that any apparent compression can be attributed to real velocity change rather than shifting journal-policy enrollment over the observation window.

3. Researcher #03
   Raised

   The methodology does not control for compositional shifts in the journal ecosystem serving climate science over the measurement window. Fast-turnaround venues — MDPI titles (e.g., *Sustainability*, *Climate*), Copernicus open-access journals (*Earth System Dynamics*, *Natural Hazards and Earth System Sciences*), and Nature Portfolio rapid-communication outlets — substantially increased their share of climate-tagged publications between 2015 and 2025. If these structurally faster venues account for a disproportionately larger fraction of the matched preprint–publication pairs in later cohorts, the declining median reflects venue-mix composition rather than any acceleration in peer-review throughput; the channel is pure selection, not process improvement.

   Resolved

   Each matched DOI should be linked to its ISSN via the CrossRef REST API (`/works/{doi}`, field `ISSN`) and merged with SCImago Journal Rankings (freely downloadable; relevant series: `SJR`, `Citable_Docs`, `Country`) to assign journal-level tier and historical median review speed. A two-way panel regression with journal fixed effects and calendar-year effects — estimated on the within-journal preprint-to-publication interval — would decompose the aggregate trend into a genuine velocity component and a compositional-shift component. A clean falsification test is that the year-trend coefficient should attenuate substantially (ideally to statistical insignificance) once journal fixed effects are included, confirming that apparent acceleration in the pooled series is an artifact of changing venue composition rather than a systemic change in editorial throughput.

4. Compliance-Guard #04
   Raised

   The most acute regulatory exposure lies in the sovereign green-bond risk-premium prediction: if underwriters, ESG rating agencies, or climate-risk disclosure officers treat this unvalidated hypothesis as grounds for asserting that preprint-cited risk evidence is methodologically equivalent to peer-reviewed evidence, they risk violating the EU Green Bond Standard (Regulation EU 2023/2631), which requires that scientific bases for Use-of-Proceeds allocations be traceable to verified, published sources, and SEC Rule 10b-5's prohibition on material misstatements or omissions in securities offering documents. A 5–10 bps spread-tightening attributed to preprint velocity before the SUPPORTS threshold is crossed would constitute a forward-looking statement of material scientific fact embedded in offering documentation, triggering IOSCO Green and Social Bond Principles (2023 revision) scrutiny and EU Prospectus Regulation (EU) 2017/1129 liability for issuers and lead managers. Concurrently, premature citation in IPCC AR7 supporting documentation before formal adoption would violate IPCC's own grey-literature protocol (established in AR5 and carried forward in AR6 Annex II), which requires preprints to carry a formal acceptance-in-principle letter from a peer-reviewed journal before inclusion.

   Resolved

   No bond prospectus language, ESG disclosure, or IPCC supporting-document citation may reference this hypothesis until it formally crosses the SUPPORTS threshold — confirmed median delta < 9 months for both the 2024 and 2025 cohorts via a reproducible, auditable cross-validation of the four joined API endpoints, with a Wilcoxon signed-rank trend test at p < 0.05 replicated on a held-out journal set independent of the training cohort. Until that threshold is cleared and the pipeline methodology has been reviewed by a recognized verification agent (an ICMA-accredited verifier under EuGBS for bond-market use, or a PCAOB-registered auditor for SEC-registered issuers), all references must carry the explicit disclaimer: *"This metric derives from a hypothesis under active statistical validation; it has not crossed the pre-registered SUPPORTS threshold and does not constitute peer-reviewed scientific consensus — reliance in investment, regulatory, or policy documents is premature and unsupported."*

5. Falsification-Auditor #05
   Raised

   The dominant threat to falsifiability is systematic right-censoring: when the matched-cohort query runs against 2024–2025 preprints, papers whose publication lags exceed the observation window simply have no PubMed record yet and drop out of the join entirely, mechanically compressing the apparent median for recent vintages by an estimated 4–8 months even if true velocity is unchanged. Unmatched fractions for a given preprint cohort typically run 30–50% at the 18-month mark in comparable bibliometric studies, meaning the "surviving" matched set is structurally biased toward fast-published papers and the FALSIFIES threshold of >12 months may be structurally unreachable for 2024–2025 cohorts regardless of the null being true. The 3-month dead zone between SUPPORTS (<9 months) and FALSIFIES (>12 months) is also narrower than the inter-annual volatility of naive matched-cohort medians, which fluctuates ±3–5 months in response to CrossRef and PubMed indexing-coverage changes alone.

   Resolved

   Replace the naive median with a Kaplan–Meier survival estimator applied to preprint-to-publication times, treating every still-unmatched preprint as right-censored at the query date rather than excluding it; this allows a log-rank test between 2015 and 2025 cohorts with computable power and makes the FALSIFIES condition a genuine rejection region under the null distribution. Additionally, enforce a vintage-controlled comparison by measuring each annual cohort only within an identical fixed window (e.g., publications observed exactly 24 months post-preprint date) for both the 2015 baseline and the 2024–2025 test cohort, then bootstrap 95% confidence intervals on the median difference; if the CI upper bound exceeds 12 months, the FALSIFIES threshold must be shifted upward to that bound before the experiment is considered well-powered to reject the null.

Unlike the static stress tests above (synthesised against the frozen catalogue spec), this is what a 3-voice council found in the most recent biweekly review. Refreshed on the 1st and 15th of each month at 09:00 UTC. Each voice runs one bounded web search via Anthropic's web_search_20260209 tool, cites what it finds, and recommends a verdict. The verdict aligns with the curated catalogue status (supported).

1. supported

   May 30, 2026 · added to catalogue at status "supported"

• IPCC AR7 Working Group II formally classifies preprints with a documented peer-review lag of <9 months as Level 2 citable evidence, expanding the eligible evidence base by an estimated 30-40% relative to AR6 before the 2027 assessment close.
• EU sovereign green-bond issuers that incorporate preprint-sourced climate-risk disclosure in offering prospectuses achieve a 5-10 bps yield compression versus peer-reviewed-only baseline issues, measurable across the 2026-2027 European sovereign green-bond issuance cohort.
• MSCI ESG Ratings recalibrates its environmental pillar methodology to apply a 20% evidential discount to issuer climate commitments sourced exclusively from non-peer-reviewed preprints, affecting ESG scores for approximately 400 sovereign and quasi-sovereign issuers in the MSCI Climate Action Index by the 2027 annual review cycle.
• Retraction Watch's climate-science retraction index rises from a 5-8% base rate (2024) to 12-18% by 2028 as compressed preprint-to-publication timelines reduce pre-publication peer scrutiny across Nature Portfolio, PLOS Climate, and AGU journal families.
• Munich Re's NatCatSERVICE transitions from annual to quarterly model updates for boreal-region catastrophe-frequency estimates by incorporating preprint-sourced climate data, compressing the model-revision cycle by 2-3× for portfolios with greater than 15% subarctic exposure by end of 2027.