Tropical forests have flipped to net carbon source

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 strongest alternative explanation is that recurrent dry-season fire phenology—not a structural regime flip—trivially breaches the 30-consecutive-day threshold every year in the record. GFED4s fire carbon emissions for the Amazon peak sharply in August–September and for Borneo/Sumatra during El Niño dry seasons; per-cell daily gross emissions from fire alone have historically exceeded regrowth uptake for 30+ days in those windows even during years (e.g., 2005, 2010) when the broader literature documents a net annual sink. The experiment's threshold is therefore satisfied by a recurrent seasonal signal that predates any hypothesized structural flip, meaning a simple count of net-source days cannot distinguish a phenological pulse from a directional regime change.

   Resolved

   Apply the identical per-cell flux-balance formula to the full GFED4s archive (1997–2022) and compute the number of consecutive net-source days for each year; if the 30-day threshold is breached in more than half of pre-2015 years, it is not diagnostic. The decisive test is a Mann-Kendall trend analysis on annual integrated Net Biome Productivity (NBP) extracted from the TRENDY v11 multi-model ensemble (ORCHIDEE, CLM5, JULES) for the same Amazon, Congo, and SEA grid cells; a structural source requires a statistically significant negative trend (Kendall's τ < 0, p < 0.05) in annual cumulative NBP, not merely threshold exceedance in daily data—if τ is non-significant or positive while the 30-day threshold is still breached, the seasonal fire-phenology alternative is confirmed as the operative mechanism.

2. Fact-Checker #02
   Raised

   ForestWatch GLAD alerts are derived from 8–16-day Landsat composites, not true daily observations; persistent cloud cover in the Congo and Amazon wet seasons creates systematic omission errors of 30–50% in alert detection on any individual day, making a "daily net flux" construct unsupported by the underlying data cadence. Converting alert area to carbon emissions requires above-ground biomass maps (e.g., Saatchi, Avitabile) with pixel-level 1-sigma uncertainty of ±30–50% and regional-scale uncertainty of ±20–30%, while modeled regrowth uptake carries an independent ±40–60% uncertainty; propagating these in quadrature means the combined net-flux uncertainty band at the biome scale routinely spans zero—i.e., the "source vs. sink" determination lies entirely within the instrument noise. A threshold of "net source > 0 for 30 days/year" is therefore indistinguishable from measurement noise given these error budgets.

   Resolved

   Aggregate to monthly or annual timescales rather than daily, reducing cloud-omission bias and averaging down random spatial errors; retain only GLAD high-confidence (≥75% confidence) alerts using the `umd_glad_landsat_alerts__confidence` quality flag to suppress commission errors. Propagate the full uncertainty chain via Monte Carlo (drawing biomass and emission-factor uncertainty from their documented distributions, IPCC Tier 2 ±20–30% for tropical forests) and report net flux as a 95% CI rather than a point estimate; the SUPPORTS threshold should require the lower bound of the CI to exceed zero for ≥3

3. Researcher #03
   Raised

   The dominant uncontrolled confounder is interannual climate variability forced by ENSO (El Niño–Southern Oscillation), which biases the net flux estimate through two simultaneous channels: it suppresses gross primary productivity by inducing drought-driven vapor pressure deficit stress, and it amplifies fire-driven emissions captured in GFED, together mechanically producing transient 30-day "net source" windows during El Niño years. Because the experiment aggregates across all years without conditioning on ENSO phase, any observed threshold breach is equally consistent with a recurrent climate anomaly as with a structural regime shift in forest carbon balance — the causal claim of a permanent "flip" cannot be distinguished from a repeating drought artifact.

   Resolved

   Stratify the daily flux panel by ENSO phase using NOAA's Oceanic Niño Index (ONI, available from NOAA PSL at monthly resolution) and include ERA5 cell-level soil moisture (Copernicus CDS variables swvl1–swvl4) and vapor pressure deficit (derived from d2m and t2m) as time-varying covariates in the flux regression. The 30-day threshold test should be re-run exclusively within ENSO-neutral years (ONI ∈ [−0.5, +0.5]); the structural "flip" hypothesis is supported only if net-source conditions persist in that subsample. Replacing the implied regrowth residual with MODIS MOD17A2H 8-day GPP (500 m) as the explicit uptake term would further isolate land-use-driven flux from climate-suppressed sink capacity, preventing ENSO-phase confounding from contaminating the deforestation signal.

4. Compliance-Guard #04
   Raised

   Premature citation of this hypothesis as established would most acutely threaten entities operating under the Verra Verified Carbon Standard (VCS) and Gold Standard REDD+ protocols, which underpin billions of dollars in voluntary carbon credit issuances that corporations are actively booking as offsets in IFRS S2 climate disclosures and SBTi Net-Zero Standard pathway claims. If this unvalidated net-source finding were cited to justify wholesale REDD+ baseline revision, credit holders and issuers could face SEC Rule 10b-5 anti-fraud exposure in any securities context where forest offset inventories are material (e.g., tokenized carbon instruments, ESG-linked debt covenants), as well as EU CSRD Article 8 misstatement liability for companies disclosing nature-based removals as verified Scope 3 compensation. The CFTC's commodity fraud jurisdiction over carbon derivatives and the CORSIA aviation offset compliance scheme add further enforcement channels if the unverified flux signal propagates into regulated offset registries before the SUPPORTS threshold is formally crossed.

   Resolved

   This hypothesis must be gated behind independent peer review in a WOS/Scopus-indexed journal, cross-validation of the merged ForestWatch/GFED/carbon-flux data pipeline against established eddy-covariance ground-truth networks (FLUXNET-Amazonia, CarboAfrica, ICOS), and formal methodology audit by an accredited Validation and Verification Body (VVB) operating under ISO 14064-3 before any citation in regulatory filings, offset registry submissions, or investor disclosures; all interim outputs must carry the explicit disclaimer: "This analysis constitutes a pre-publication falsifiable hypothesis that has not reached its defined SUPPORTS threshold and must not be used to reprice REDD+ baselines, adjust voluntary carbon credit valuations, or support net-zero pathway representations in any regulatory or fiduciary disclosure."

5. Falsification-Auditor #05
   Raised

   The FALSIFIES condition — "net sink throughout year," implying zero days registering as a net source — is virtually unreachable under any realistic null because tropical forest daily flux estimates carry inherent variance of roughly ±40–60% at the per-cell level, driven by cloud-masked ForestWatch detection gaps, GFED fire emission-factor uncertainty (~25–35% inter-annually), and regrowth parametric spread. Amazon dry seasons alone routinely produce multi-week net-source windows in historically undisturbed grid cells purely from seasonality; a well-calibrated baseline year would still generate well above 30 apparent "source days" from instrument noise and seasonal signal alone. This means the SUPPORTS threshold (≥30 source days) is trivially entered by noise, while the FALSIFIES threshold (zero source days) is never entered in practice, making the hypothesis one-sided and structurally unfalsifiable.

   Resolved

   Run a Monte Carlo null synthesis using 10–15 pre-disturbance baseline years (e.g., 2000–2010 ForestWatch + GFED) to build a per-biome empirical distribution of annual "source day" counts under no-trend conditions; the 95th percentile of that distribution — not zero — should become the operational FALSIFIES threshold (e.g., "fewer than N days where N is the null 95th percentile"), and the SUPPORTS threshold should be set above the 99th percentile to demand a statistically anomalous signal. Additionally, a sensitivity sweep across at least three reanalysis-ensemble variants (e.g., TRENDY model spread for regrowth uptake) should bracket the flux uncertainty, so that any SUPPORTS verdict must hold across all ensemble members, not just the central estimate.

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 (converging) — the synthesis below explains why.

Agent draft incorporating the 9 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. converging

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

• Verra's Verified Carbon Standard methodology board mandates downward revision of Forest Reference Emission Levels for all active tropical REDD+ projects by 30–50%, triggering credit suspension or cancellation reviews across an estimated 180–220 registered projects within 24 months of the composite dataset achieving peer-reviewed publication.
• S&P Global Ratings places Brazil's Títulos Soberanos Ecológicos and Indonesia's sovereign Green Sukuk series on CreditWatch Negative, with tropical-forest-backed sustainability-linked bond spreads widening 70–110 bps within the next scheduled sovereign rating review cycle (within 18 months), as the sequestration guarantees underpinning those instruments' ESG covenants become scientifically unverifiable.
• ICAO's CORSIA Technical Advisory Body removes tropical-forest REDD+ credits from its eligible emissions unit roster at the 2027 triennial eligibility review, redirecting approximately 120–180 MtCO2e of annual airline offset demand to non-forest categories and stranding an estimated $1.8–2.4 billion in pre-purchased REDD+ forward contracts held by major airline groups.
• CBL (Xpansiv) Nature-Based Solutions REDD+ spot prices fall 45–65% from their 24-month trailing average within six months of UNFCCC SBSTA formal acknowledgment, as CORSIA-eligible airline buyers exit the category and voluntary corporate buyers apply permanence haircuts to tropical sequestration projects.
• Swiss Re Institute revises its natural capital risk parameterization for tropical forest ecosystems, raising rate-on-line for parametric forest carbon insurance products covering Amazonian and Southeast Asian project portfolios by 40–70 bps per renewal cycle and requiring cedants to post additional collateral equal to 10–15% of outstanding forest carbon notional by 2028.