Climate preprint-to-citation lag has collapsed under 18 months

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 most plausible alternative explanation is that the apparent lag compression is an artifact of same-author self-citation: as climate preprint-to-journal conversion rates have risen sharply since ~2019 (driven by EGU journals, AGU journals, and Nature portfolio titles adopting explicit preprint-linking policies), OpenAlex and CrossRef routinely record the published journal article as the "first peer-reviewed citation" of the preprint from which it derived. This mechanically collapses measured lag to the preprint-to-acceptance interval (~8–14 months for high-volume climate outlets like Geophysical Research Letters and Environmental Research Letters), producing a sub-18-month median without any genuine reorganization of the field around real-time evidence. The confound is magnified because the 2014–2018 baseline cohort had lower conversion rates, so the historical counterfactual is not self-citation-inflated to the same degree, making the apparent acceleration look larger than it is.

   Resolved

   Rerun the entire citation-lag pipeline restricting "first peer-reviewed citation" to records where no author entity ID in the citing work overlaps with any author entity ID in the preprint, using OpenAlex's disambiguated `author.id` field across both the `/works` citing and cited endpoints. Produce two parallel time-series—all-first-citations median and external-only-first-citations median—and apply a difference-in-differences model using 2019 as the policy-adoption treatment year and preprint-conversion rate (recoverable from the OpenAlex `related_works` field) as the treatment intensity variable. If the sub-18-month threshold holds in the external-only series with a coefficient on post-2019 × conversion-rate that is statistically indistinguishable from zero (p > 0.05, two-sided), the hypothesis survives; if the external-only median remains above 30 months while the all-citations median falls below 18, the hypothesis is falsified as a self-citation artifact.

2. Fact-Checker #02
   Raised

   CrossRef reference deposit completeness varies from roughly 60% to 85% depending on publisher participation, and OpenAlex—which supplements CrossRef—still carries an estimated 15–20% missing-reference rate for geoscience and atmospheric-physics literature, the two primary venues for climate preprints. This incompleteness creates a systematic positive bias on measured first-citation time: for a preprint that accumulates only 3–5 early citations (common in niche climate subfields), the probability of missing the true first citation event is non-trivial, pushing the detected first-citation date 6–18 months later than reality. Additionally, CrossRef DOI reference-list deposit lag runs 3–12 months post-publication for many mid-tier journals, and OpenAlex record-processing latency adds another 2–8 weeks. The combined systematic bias budget of roughly 6–18 months directly overlaps the 12-month gap between the SUPPORTS threshold (<18 months) and the FALSIFIES threshold (>30 months), making the two outcomes statistically indistinguishable without correction.

   Resolved

   Restrict the sample to preprint–journal pairs where the citing journal carries a CrossRef reference-deposit completeness score ≥ 90% (retrievable via CrossRef's `/journals/{issn}/works` coverage endpoint) and apply OpenAlex's `type:journal-article` filter combined with `referenced_works_count` ≥ 20 as a proxy for complete reference lists. Construct median first-citation estimates with a bootstrap 95% CI and perform a mandatory sensitivity sweep shifting all raw first-citation dates by −9 months (median indexing-lag correction derived from CrossRef deposit-date vs. publication-date deltas) to recover the de-biased distribution; only flag SUPPORTS if the upper CI bound of the corrected median remains below 12 months, and FALSIFIES only if the lower CI bound of the corrected median exceeds 36 months, thereby ensuring both thresholds survive the full 6–18 month uncertainty budget.

3. Researcher #03
   Raised

   The dominant uncontrolled confounder is the exogenous COVID-19 pandemic shock to preprint citation norms (2020–2021), which induced a field-wide structural break in journal willingness to cite non-peer-reviewed work across all STEM disciplines simultaneously, independent of any climate-specific reorganization. The mechanism operates through editorial policy relaxation: high-impact generalist journals (Nature, Science, PNAS) explicitly sanctioned preprint citation during the pandemic, and that norm persisted post-2022 across the citing literature broadly. Because this shock is temporally coincident with the hypothesized trend compression and is not specific to climate science's epistemic restructuring, any post-2020 reduction in median time-to-citation conflates a cross-disciplinary COVID-induced norm change with the claimed disciplinary effect, producing an upward-biased estimate of climate-specific reorganization.

   Resolved

   A difference-in-differences design would absorb this confounder by treating climate-tagged preprints as the treatment group and non-climate STEM preprints (e.g., condensed matter physics or quantitative ecology) as the parallel control group, identified via arXiv subject category codes (cond-mat.*, q-bio.*) accessible through the /api/arxiv metadata endpoint. The pre-period (2014–2019) can be used to validate parallel trends in citation lag; the climate-specific coefficient on the post-2020 treatment-by-time interaction then isolates disciplinary reorganization net of the field-wide norm shock. OpenAlex Concepts API classifications (filterable by field-of-study concept IDs, e.g., C54355233 for climate science versus control concept clusters) would enable consistent treatment/control stratification within the same citation graph infrastructure already specified in the experimental design.

4. Compliance-Guard #04
   Raised

   If this hypothesis is cited as established before crossing the SUPPORTS threshold, the primary regulatory exposure runs through insurance catastrophe-model rate-filing regimes — specifically Florida Statute § 627.0628 and the Florida Commission on Hurricane Loss Projection Methodology, plus analogous state-level actuarial standards under Actuarial Standard of Practice No. 38 (Catastrophe Modeling for Property/Casualty Insurance) — because vendors such as Verisk AIR, RMS/Moody's, and Karen Clark & Company would be incorporating unvalidated "preprint-equivalent" scientific inputs into rate-making models subject to regulatory approval. A parallel channel opens under SEC Rule 10b-5 and EU CSRD / ESRS E1 paragraph 39, where sovereign climate-risk scores referencing a preprint-velocity index as evidence of scientific consensus could constitute a material misstatement or fail the "reasonable assurance" data-quality standard required for auditable climate disclosures, exposing both issuers and third-party data providers to enforcement action.

   Resolved

   Downstream reliance is gated on three sequential conditions: (1) the SUPPORTS threshold — median lag below 18 months sustained across three consecutive cohort years — must be demonstrated in a peer-reviewed, journal-published methodology paper that itself discloses the OpenAlex cross-validation procedure and any survivorship bias in the arXiv/bioRxiv sampling frame; (2) the citation-graph join across /api/arxiv, /api/biorxiv, and /api/crossref must be independently replicated by at least one research team with no affiliation to the original study before any catastrophe-model vendor or CSRD-reporting entity treats preprint velocity as an admissible scientific-evidence input; and (3) until the Casualty Actuarial Society or NAIC formally amends guidance to recognise sub-18-month preprints as actuarially acceptable sources, all references in rate filings or securities disclosures must carry an explicit forward-looking-statement disclaimer conforming to SEC Rule 175 and EU Prospectus Regulation Article 14, characterising the metric as an investigational indicator rather than established scientific consensus.

5. Falsification-Auditor #05
   Raised

   OpenAlex and CrossRef citation records carry a well-documented indexing propagation delay of 3–9 months, and right-censoring systematically inflates apparent citation lag for recent cohorts that have not had a full observation window — both effects push measured median lag upward independent of true field behavior. Bootstrapped median citation-lag estimates on samples of ~500 preprints/year routinely yield 95% confidence intervals spanning ±8–14 months in the bibliometric literature, meaning a true median of roughly 22 months (indeterminate zone) can plausibly measure above 30 months and enter the FALSIFIES band purely through database latency and censoring artifacts. The 12-month gap separating SUPPORTS (<18 mo) from FALSIFIES (>30 mo) is narrower than this combined instrument noise, making both thresholds enterable for reasons structurally unrelated to actual citation acceleration.

   Resolved

   Apply Kaplan–Meier survival analysis with explicit right-censoring adjustments for any cohort whose observation window is shorter than 30 months, then run 1,000-iteration Monte Carlo synthesis under the null — drawing citation-velocity distributions from the 2014–2017 cohorts, which predate any claimed acceleration — to measure how frequently the null process alone enters the FALSIFIES band; if that entry rate exceeds 5%, the FALSIFIES threshold must be expanded to ≥36 months or the experiment must add a direct-validation arm in which DOI-level CrossRef API calls manually verify citation dates for a stratified subsample of 200 preprints, producing a coverage-correction factor that is applied to all automated pipeline outputs before threshold adjudication.

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 diverges from the curated catalogue status (monitoring) — the synthesis below explains why.

Agent draft incorporating the 7 cited findings from the live council above. Not auto-merged — surfaces here for human review. To accept, open a PR editing site/src/_data/hypotheses.json with the revised fields below. To reject, ignore and the proposal will refresh on the next council run.

1. monitoring

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

• The US National Science Foundation Climate & Large-Scale Dynamics program and Horizon Europe Cluster 5 revise grant-proposal evidence standards to admit preprints ≤18 months old as primary citations — eliminating the de-facto 24–36-month peer-review publication delay — within 24 months of standardised preprint-velocity indices becoming publicly reproducible.
• Verisk AIR, RMS Moody's Analytics, and Karen Clark & Co compress catastrophe-model retraining cycles by 18–30 months and revise storm and wildfire annual exceedance probability curves by 5–15% within two model-release cycles, as sub-18-month preprint pipelines reduce the evidence latency that previously constrained update frequency.
• Munich Re NatCatSERVICE issues upward revisions to boreal and tropical-cyclone catastrophe-frequency estimates of 15–25% within four consecutive quarterly update cycles, as sub-18-month preprint evidence displaces journal-lagged inputs in its natural-hazard loss models.
• Moody's Investors Service and S&P Global Ratings incorporate preprint-velocity indices — tracking mean citation lag across peer-reviewed climate-science literature — into sovereign climate-risk score methodologies, with the revision reflected in published reports for at least 10 sovereign issuers by end-2027.
• MSCI ESG Research revises its Physical Climate Risk methodology to accept preprint-sourced climate-hazard datasets with citation ages under 18 months, altering ESG scores for at least 200 MSCI ACWI constituents by end-2027 and repositioning an estimated $40–60bn in ESG-screened AUM across climate-risk quartiles.