What Public Records Reveal About Cryptocurrency Payments Between AI Agents STRONG
What do publicly reported agent-payment records establish about commerce once their units and source layers are separated? We audit an MPP catalogue with 138 service records and 1600 endpoint entries, a non-probability slice of 1000 CDP Bazaar resources, a fixed-block Olas frame of 1294 requests, twenty previously selected Olas orders and one purposive TaskMarket payout. A post-review extension covers 11,326 Olas requests over seven days; it contains, rather than adds to, the original day. Two results are central. First, all 2590 known-rate requests without a delivery marker in the week belong to one priority-Mech cluster: 370 on each day across fifty buyer addresses. Although 104 of 106 buyer addresses recur on multiple dates, only 55 have a request later than an observed delivery; neither count measures accepted-purchase retention. Second, comparison of the same twenty orders across sources distinguishes fourteen delivery markers, ten linked request–result file pairs and unresolved final withdrawals. The separate TaskMarket case links a task to a 2 OUSD transfer and a recipient balance change. Catalogue composition also matters: two retained 1000-resource CDP frames contain 2624 and 4325 payment offers. These findings identify specific composition and observability mechanisms. They do not establish independent customer demand, accepted quality, profitability or a causal price effect.
Working paper. Citable, but not yet screened for the AIRR collection.