[Draft]
[Subtitle]
July 9, 2026
/llms.txt carries the
structured version.Large language models are increasingly able to act as delegated buyers, agents that shop on a human principal's behalf, through emerging agentic commerce protocols such as the Universal Commerce Protocol (UCP). These protocols are nascent, but it is not hard to imagine a near future in which individuals and organizations routinely delegate purchasing to LLMs. That environment can also be a useful instrument: it places an agent in a self-contained world where its reasoning ends in a real action, a purchase.
We lift real shopping tasks into schema-compliant UCP worlds and study Sonnet 4.6 as a delegated agentic buyer when a store does not carry exactly what the principal asked for. A concrete case: a principal asks for a watch made in Japan. The best-matching watch (the target) has every requested feature but leaves its country of manufacture blank; every other labelled watch in the store says made in Thailand. Across otherwise-identical stores we change one value: whether a single "Japan" label appears, and which watch it sits on. When the label sits on a watch in the target's own product line, as "made in Japan" or even "designed in Japan", and even when the line's other labels say Thailand, the agent treats the target as Japanese, buys it, and often reports the requirement to the principal as met. The full pattern, including the declines, appears in the results.
We formalize the belief the agent forms about the missing value as a feature-prediction problem, using a rational-analysis framework to compute what the evidence in the store supports. We then measure the direction and size of the agent's departure from it: the labelled watches in the target's product line run three Thailand to one Japan, and the agent behaves as if the blank were Japanese. We read these departures through Kunda's (1990) account of motivated reasoning, in which a conclusion must be one the reasoner can build a case for. We further sketch how the purchase itself can be read as a decision problem separate from the belief: in some episodes the agent concludes the target probably fails the request, and buys it anyway. More broadly, this work shows how agents reason, act, and report in worlds that under-determine or contradict the request they were given.
1Introduction
2Background
3Method
What the store’s evidence supports
The target watch has every requested feature, but its country of manufacture is blank, so the agent must infer it from what the store shows. We compute what that evidence supports before looking at any behavior. The inference has a standard form. Suppose that among related products whose country is stated, say Japan, and that before reading any label the agent’s background belief is worth imaginary Japan-votes and imaginary Thailand-votes. Then the probability that the blank product is Japanese is
— the Japan votes, real and imaginary, over all votes. This is the textbook rule for predicting a missing categorical feature from observed ones (Laplace’s rule of succession; in the psychology of categorization, Anderson’s 1991 feature-prediction equation), and the prior enters only through and . The reading most favorable to the agent takes the target’s own product line as the evidence: four line members carry a country label; one says Japan, three say Thailand (Figure 1). So , , and with the neutral prior — one imaginary vote each way —
Every honest reading puts the chance at one in three or less, and a reasoner unsure which comparison class is right would average the two, not pick one (Anderson, 1991). Store composition varies slightly across our worlds; each count in this paper is stated for a named world — here, the world of Figure 1. One term remains open: the prior. For the answer to pass one half, the imaginary votes must outvote the shelf — , prior odds better than 3 to 1, better than 75% sure before any label is read. We measured it directly: asked cold, with no store and no task, the model puts at about zero, volunteering that Casio moved production out of Japan decades ago. Under the model’s own stated prior, is near zero and the answer falls below one in three. In the episodes, the agent buys the watch — and often reports the requirement to the principal as met.
the target — field absent
one label says japan
three say thailand
watch-solo-country-square-family. All titles and attribute
values are verbatim from the store record; the colour dots render each
record’s own stated colour value.There is a second honest reading, and it is the one the agent’s own reasoning keeps reaching for: that a product line has one true country of manufacture, and that the individual labels are unreliable readouts of it — the store may simply have entered some of them wrong. This is the classic observer-error model (Dawid & Skene, 1979), and we can grant it completely. Under this reading, the question becomes: which truth better explains the shelf we actually see? Suppose each label is wrong with the same probability — some error rate we do not need to know. The chance of seeing this shelf under each candidate truth comes from multiplying the chances of each label, if each label errs independently (the model’s own assumption). In general, if the true country is and of the stated labels agree with , then
— each agreeing label was read out correctly, with chance ; each disagreeing label is an error, with chance . Our shelf has stated labels: three thailand, one japan. If the truth is Thailand, the shelf is easy to explain: , and only the japan label is an error. If the truth is Japan, the shelf is hard to explain: , and three separate labels must all be wrong:
The truth that makes the shelf less surprising is the better explanation, and how much better is the ratio of the two:
At — labels right nine times in ten — Thailand is 81 times the better explanation. At it is still about five times better. And it remains the better explanation for every error rate below one half, which is the point at which a label carries no information at all. Unreliable labels weaken the verdict; they do not reverse it. To reach Japan from this shelf, the agent must treat the three thailand labels as unreliable and the one japan label as trustworthy at the same time — and nothing in the store supports that assignment. All four labels come from the same catalog, and in the observer-error model itself a label’s reliability is estimated from its agreement with the others, which here runs three to one against the japan label. The only property that distinguishes it is that it agrees with the request.
Two notes complete the benchmark. First, the class: widening the comparison only lowers the number — the whole store gives , about 0.02 — so no honest choice of comparison class rescues Japan; the only direction that helps is narrowing the class until nothing but the confirming label remains. Second, the blank: treating the absent field as saying nothing is itself the charitable reading. If silence were informative at all, it would count against Japan — a seller whose watch is made in Japan, the very property the request prizes, has every reason to say so.
None of this arithmetic is beyond the model. Given the same evidence as a bare question — the four labels stated as facts, no store, no task — it answers thailand thirty times out of thirty.