Risk-based checks under the EU deforestation regulation

Six statements about the model; the mapping to the Regulation is a reconstruction R, and no statement estimates the Regulation's effects.

• The benchmark is a signal and the schedule a committed response to it. Model O1's decision rule maps a flag to an action; a risk-based schedule maps a published class to a check rate. Everything this chapter says about signals applies to the benchmark: its value lives in the decision it changes, and it is compared with another benchmark by Blackwell's order, not by a score R.

• A budget turns the schedule into a threshold rule. Against rooted non-compliance, a fixed number of checks detects most when it fills the classes in decreasing order of risk: at the stylised values, checking $\frac{3}{10}$ of the high class and nothing else detects $\frac{6}{41}$ of rooted non-compliance, about $0.15$, against $\frac{231}{3280}$, about $0.07$, for rates rising by a factor of three and $\frac{3}{100}$ for uniform checking (Table 1.1 (p. 47)a). The budget acts as a price: the authority behaves exactly like Model O1's decision-maker with a break-even probability of $\frac{1}{5}$ — the rate of the class it fills only in part — in place of the ratio of an inspection's cost to a miss D [Proposition 1.17 (p. 67)(a)].

• The lowest rate is the guarantee. A violator who can make its consignment appear in any class faces the lowest check rate, so no allocation of the budget guarantees more than the uniform rate $\frac{3}{100}$, and uniform checking is the only allocation that attains it. Full concentration guarantees nothing; rates rising by a factor of three guarantee $\frac{1}{80}$, the rate of the low class, which is where such a violator goes (Table 1.1 (p. 47)a). The floor in the lowest class is what protects against relabelling; the rates above the floor are targeting D [Proposition 1.17 (p. 67)(c)].

• Above the floor, the model sends every remaining check to the top. Among allocations with the same guarantee, rooted detection is largest when the budget above the floor goes to the highest-risk class. The frontier between detection and guarantee is then a straight line from concentration to uniform checking, and the factor-three schedule lies strictly below it: keeping its guarantee of $\frac{1}{80}$ and moving its remaining checks to the high class raises detection from $\frac{231}{3280}$ to $\frac{321}{3280}$ (Figure 1.7 (p. 48)). With a share $\mu$ of mobile violators and an objective linear in the two detection rates, the best allocation sits at an end of the frontier: concentration if $\mu <\frac{159}{200}$, uniform checking if $\mu$ is larger D [Proposition 1.17 (p. 67)(c)–(d)]. This statement is about the model, not the Regulation: the Regulation's percentages are minima, not an allocation of a fixed budget, and authorities target within and above them under their own risk criteria. Read as a criticism of the minima, it would overreach; what it licenses is narrower: in a linear model, an intermediate schedule is never strictly optimal, so any reason for one must come from something the linear model leaves out, such as violators who respond to their own class's rate (Chapter 6) or an objective that values the worst-covered class for its own sake R.

• A noisier benchmark is worth less, by Blackwell's theorem, and favours uniform checks. Garble the benchmark by redrawing a consignment's published class at random with probability $\epsilon$. The class shares do not change, the rates within classes move towards the overall rate, and the rooted detection of every allocation that spends the budget moves linearly towards $\frac{3}{100}$. The targeted policies of Table 1.1 (p. 47)b fall towards it; uniform checking, which does not depend on the benchmark at all, stays at it; an allocation aimed at the safer classes would rise towards it, which is why the statement that survives garbling is about the best achievable detection and not about each fixed rule. The value of the benchmark at this budget — the best detection minus the uniform rate — falls from $\frac{477}{4100}$ to nothing; and the share of mobile violators above which uniform checking is best falls with it, from $\frac{159}{200}$ to $\frac{159}{241}$ at $\epsilon =\frac{1}{2}$ and to $0$ for an uninformative benchmark (Table 1.1 (p. 47)b). Any garbling, not only this one, can only lower the best detection for every budget, because a policy on the garbled classes can be reproduced on the original ones D [Proposition 1.17 (p. 67)(b)]. This garbling preserves the order of the classes, so the best allocation does not move; a coarse benchmark that misorders classes would move it as well R.

• Publication helps detection and tells violators where checks are thin. A published schedule is a commitment in the sense of Chapter 6, and a committed inspector can do better than one that is not (Proposition 6.9 (p. 257)). The same publication tells a mobile violator which class is checked least; in Model O5 that is all such a violator needs. The Regulation's own text names the concern (circumvention, and mixing with products of unknown or higher-risk origin, are conditions of the simplified regime), and the model shows what is at stake: the gap between the guarantee and the targeted rates D [European Parliament and Council of the European Union 2023, Art. 13(1); the reading is R].

The Regulation counts operators and, as amended in December 2025, non-SME downstream operators and non-SME traders, and, for high-risk origins, quantities; the model counts consignments, each checked or not, so a share of operators checked is not the probability that a given consignment is checked. The two units also aggregate differently: an operator may place products drawn from many smallholders under one due-diligence exercise, so what the Regulation examines mixes sources that the model keeps distinct; and keeping certified flows apart from uncertified ones is a change in the supply chain's logistics, not only in its documents R. Under the Regulation an operator counts as checked once the authority has examined the elements listed in Article 18(1)(a)–(b) D [European Parliament and Council of the European Union 2025, Art. 1, amending Art. 16(11)]; the model's check is perfect, and a real check that misses non-compliance would multiply every detection figure by its own detection rate and could change the allocation if that rate differed by class. The benchmark classifies countries and parts of countries, while non-compliance belongs to plots and supply chains; within a class, authorities target by their own criteria, which the model replaces by a uniform rate. The model's mobile violator relabels at no cost and the rooted one does not respond at all; real operators respond in between (they shift sourcing, mix or comply), and their response depends on the rates, which makes the problem Chapter 6's inspection game rather than this chapter's decision. The model holds the classification fixed; a country's class is itself a decision, with consequences beyond checks R. And the minimum rates bind Member States, each with its own budget, under one schedule; the model has one authority and one budget.