Governance as a Treatment

1 · The reversal experiment

Land-use policy is almost never a clean experiment. Interventions arrive bundled with commodity price shocks, elections and droughts; they are rarely reversed; and when they are reversed, it is usually discovered afterwards. The Amazon Soy Moratorium is the exception on all three counts.

For twenty years it barred the purchase of soy from land cleared after 2008. In monitored municipalities, deforestation fell 69% between 2009 and 2022. [DATA] On 5 January 2026, traders representing roughly 45% of Brazil's soy exports withdrew from it through Abiove — Cargill, Bunge, ADM and Amaggi — days after a Mato Grosso state law took effect on 1 January permitting the state to strip tax benefits from firms applying environmental standards beyond federal law. IPAM projects Amazon deforestation up to 30% higher by 2045 as a consequence. [DATA]

So the instrument has all four properties a natural experiment needs: a measured pre-treatment effect, a dated imposition, a dated withdrawal, and an uninterrupted observational record spanning both. The land-surface response separates from interannual climate variability over roughly 2027–2030, which is to say it has not happened yet. That is unusual. Most policy evaluation is archaeology.

2 · The counterfactual next door

WP66 relocated the gate K from an ice-nucleation aerosol to the Andes–Amazon moisture corridor, and argued that the corridor must be treated as one hydrological system across Brazil, Bolivia, Peru and Paraguay. That argument has a methodological consequence it did not draw: Bolivia is a control group.

Bolivia lost 596,000 ha of primary forest in 2022, third globally; its clearing rate rose 259% over eight years, roughly 75% concentrated in Santa Cruz and the Chiquitano dry forest. [OPEN — inherited from WP66, not traced to primary source] Published supply-chain modelling finds Brazilian restriction displaced 43–53% of avoided clearing domestically, largely to the Cerrado, while cross-border leakage to Bolivia, Argentina and Paraguay stayed under 3%. [OPEN — same]

That last number is what licenses the design. If cross-border leakage were large, Bolivia's frontier would be Brazil's displaced demand and the two systems would not be independent. At under 3% it is not displacement: it is the same crop, on the same corridor, under the same climate forcing, governed differently. WP66's own phrasing — what crosses the border is the frontier-clearing model, not the soybeans — is precisely the statement that the treatment is governance and not trade.

3 · What is observable

The design is a matched-pair difference in differences, in space and in time. Municipality-scale units on either side of the frontier are paired on precipitation regime, soil, slope and pre-2008 forest cover, and differ in governance regime. The 2009–2022 moratorium period is the pre-treatment baseline; January 2026 is the treatment date.

Across each pair we compute standard resilience indicators — lag-1 autocorrelation and variance of detrended anomalies — on four physically distinct observables, so that no conclusion rests on a single retrieval pathway:

IMERG and reanalysis supply the climate covariates that must be removed before any governance signal may be claimed. A.15's Subelement 2.3 asks in its own words for optical, thermal, radar, and lidar across space and time; this is that, with the sensor families chosen so their failure modes do not coincide.

4 · What this paper does not claim

Whether the Amazon as a whole approaches a tipping point is an open and actively contested question. Resilience-loss findings from vegetation optical depth and NDVI have been published, disputed in the same journal, and defended. This design does not enter that dispute and does not need to.

Resilience indicators are used here comparatively, between matched landscapes under a policy contrast — not as absolute evidence of proximity to a threshold. The difference matters: a comparative claim survives a systematic bias in the indicator provided the bias is common to both members of a pair, and an absolute claim does not. It is the weaker statement, and it is the one the data can carry.

5 · Why the series machinery is excluded

The obvious move from inside this corpus is to read the corridor as a bistable system, identify a fold, and calibrate ε₀ against the observed onset of critical slowing down. WP66 already proposes exactly that, correctly tagged [prospective]. It is deliberately absent here, and the reason is internal.

A proposal whose subject is bistability and threshold proximity walks directly into that finding. Importing r* or ε₀ = 1/3 here would take the corpus's most exposed claim and place it where a reviewer is already looking — and it would buy nothing, because the resilience-indicator apparatus is standard and requires none of it.

The same exclusion applies to WP41–43's intervention doctrine: ice-nucleation deployment costing, planetary triage and relocation scenarios are not observational Earth science, and WP66 has already withdrawn the aerosol identification of K on which they rest.

Recording this is the point of putting the paper in the book. The corpus has a habit — documented repeatedly in its own correction ledger — of extending a framework into a neighbouring field on the strength of shared vocabulary. Here the framework is held back on purpose, and the reason is a finding from within the framework itself.

6 · Falsifiability

If governance leaves a land-surface signature, matched pairs separate after 2026 in at least one observable, with an identifiable lead time and a scale at which detection becomes possible. If they do not separate — or if apparent divergence tracks precipitation rather than policy once covariates are removed — that is a reportable negative result, and a consequential one: it would bear directly on whether satellite monitoring can serve as an independent check on corporate supply-chain commitments, which is a claim currently made without that evidence.