Agoranome Monthly: November 2026

Month. The weeks whose Monday falls in the month. The month's end is its last Monday snapshot of the x402 Bazaar.

New seller. A host in one of the month's snapshots and in none before. Hosts in the first snapshot Agoranome holds are never counted as new, since when they first listed isn't known.

Seller survival. Of a month's new sellers, the share still listed in the snapshot at the end of the month the edition covers. With payers counts those that also had a listing with 3 or more payers in 30 days.

Week by week. Of the sellers listed in the month's first weekly snapshot, the share still listed in each of the month's later snapshots. Fewer than 5 sellers in the first gives no shares.

Real buyer. A wallet with at least one real payment in the month: a USDC payment on Base into a counted Bazaar seller that isn't part of a catalog sweep or from the seller's own address, as in the weekly report's real demand. Wallets are kept only as a hash of their address, and none is ever named.

Coverage. The days of the month that the weekly chain reads cover in full. Each Tuesday's read covers the 30 days before it, so a month is covered once the first Tuesday after it is read. A day read by more than one week keeps the highest count any week saw. Each week counts the 500 busiest sellers on its snapshot, so a buyer of a seller that drops out of them isn't seen.

Paid again. Of the month before's real buyers, the share with a real payment in this month. It's given when the month before has 14 or more covered days.

Paid on 2+ days. Real buyers with real payments on two or more days of the month (UTC).

Top 10 share. The share of the month's real payments made by the 10 buyers who made the most.

Opening. A category is an opening when, in at least half of the month's weekly snapshots, it had 500 or more repeat calls in 30 days and two or fewer listings with 10 or more payers, or one listing with 60% or more of its repeat calls: the weekly demand brief's test, held for the month.

Thin search. One of the category's buyer searches, run on the Bazaar each Monday, whose top 20 results hold fewer than 3 listings with 3 or more payers. Counted on the month's last snapshot.

Fair Offer index. At the month's end: the share of listings whose 402 offer carries a Fair Offer block, the sellers with at least one, and every seller's report card grade (the mean of its listings' scores) against the month before. Sellers are named only when their grade went up or a new seller's is an A; a grade that fell is only counted.

Ranges and small numbers. Shares that estimate a rate (survival, paid again, paid on 2+ days, traction, thin searches) carry a 95% Wilson range. A share over fewer than 5 isn't given.

Price signals. From the December 2026 edition. The price index at the month's end: p10, median and p90 of the USDC price per call across listings with 3 or more payers in 30 days, overall and for each category with 5 or more of them, against the month before's end. A median that moved 10% or more fell or rose; the rest held.

Spread. p90 over p10 in a category: 4 means the dearer listings ask four times what the cheaper ones do. A spread that narrows is a market settling on a price.

Changed price. Of the listings in the price index at both month ends, the share whose price per call differs, and how many cut and raised it. Which listings moved isn't given.

What moves buyer agents. From the December 2026 edition. Every decision in the month's Machine Choice Benchmark reports, all panelists together and none named: buyer agents choosing among live Bazaar offers shown as JSON, with positions rotated through every slot. Of the decisions that bought, the share that picked the cheapest offer shown, the first slot and the offer with the most 30-day payers, each against the share if it didn't matter (each offer equally likely). Ties count as picked; a decision whose offers were all alike on a point is left out of that point.

Fair Offer field lift. In the benchmark's paired trials, one offer gets one Fair Offer field added by the experimenter (it isn't the seller's claim). Its lift is how often that offer was picked with the field against the same trial without it, with Newcombe's 95% range; it moved choices only when the range leaves out zero.

Adoption funnel. From the December 2026 edition. SDK downloads in 30 days (npm and PyPI's own counts, CI runs and mirrors included) at the month's end, the month's new sellers, and the sellers with a first real buyer this month.

First real buyer. The first week the who-pays read counted a seller with a real buyer. The read covers the 500 busiest sellers on each snapshot, so a seller outside them isn't seen: these counts are a floor. Sellers with a real buyer in the first read Agoranome holds aren't counted, since when theirs came isn't known.

Weeks to a first real buyer. For a seller with its first real buyer this month and a known first listing: the weeks from the snapshot it first appeared in to that read.

Registered agents. From the December 2026 edition. ERC-8004's Identity and Reputation Registries on Base, read from their events (Registered, URIUpdated, NewFeedback and FeedbackRevoked) by the Tuesday hosts job. No agent's file, endpoint or domain is fetched, and no agent, owner or wallet is named: owners and wallets are kept only as a hash of their address.

Registration file. Where each agent registered in the month points its agentURI at the month's end: inline (a data: URI), https, IPFS, Arweave, plain http or none. Only an inline file is read, to see whether it points back to its agent: a registrations entry naming this registry and the agent's id.

Feedback concentration. Feedback given in the month and not revoked by its end. Of the agents with 5 or more, the share where one wallet gave half or more, where 5 wallets gave 80% or more, and the rest (spread), as the agent cards judge it.

Price of proof. From the November 2026 edition. The month's latest run of the price-of-proof benchmark: buyer-agent models choose between two live Bazaar offers shown at the same price and usage, one carrying a cue (a proof field, or a manipulation such as a fake payer count) and costing a fixed premium more, from -20% to +200%. A cue's price is the premium at which the cued offer is picked half the time, from a logistic fit, with a 95% bootstrap range. The control adds nothing and should come out near zero. Panelists are named by label.

Buyer Safety Index. From the November 2026 edition. The x402 clients named in the index at the edition's build, their verdicts against eight attacks, and the rows not yet named: open-source clients wait 14 days after their maintainers are told, and a closed runtime is named only once its vendor paid for an examination.

Not counted. Agoranome's own hosts and wallets.