Assay buys from the endpoints it ranks. Here is the account it buys with, and what that money is subtracted from.

Every one of the 861 catalogued endpoints is called once a day and nothing is paid for it. That free sweep checks the endpoint answers, that its 402 challenge parses, that the price and payout address in the challenge match what the listing advertises, and — where the merchant offers a proof-of-work route — that a solved challenge is actually honoured. It costs bandwidth and some sha256, and it is what the live and overcharging marks on the index are derived from.

Once a week a second pass pays for real. It picks the next few merchants off a rotating queue, buys their single cheapest healthy endpoint with USDC on Algorand mainnet, and writes a receipt carrying the transaction id, the HTTP status and a sha256 of the exact bytes served. That is the only evidence on this site that an endpoint delivers rather than merely answers, and there are 4 of them so far, across 4 merchants.

The paid pass runs six hours after the free one and reads its results first. It will not pay an endpoint the sweep just marked unreachable or non-conformant, and it refuses to run at all on sweep data older than 48 hours — paying to confirm something no current evidence supports is how a budget this small gets wasted.

The prober was funded once with 0.10 USDC and is not on a monthly allowance. Four ceilings are enforced in code before any transaction is built, and a pass that would cross one stops at the boundary rather than finishing and apologising.

All of it is burned. These payments go to third-party merchants and never come back; none of it is refundable minimum balance of the kind that makes Algorand spending look larger than it is. The prober also holds about 3 ALGO, of which 0.2 is locked as minimum balance for the account and the USDC opt-in and would return in full if the account were ever closed. Ongoing ALGO burn is near zero, because the x402 scheme has the merchant’s declared fee payer cover the fee while the prober signs a zero-fee transfer.

At the floor the paid prober halts rather than degrading quietly. The site then shows paid verification as paused for funding, with the date of the newest receipt attached, because the failure mode worth designing against is not running out of money — it is continuing to present five-week-old receipts as though they were current.

Assay pays merchants. Those payments land on the same payout addresses the site counts to measure demand. Left in, they would become a growing share of the popularity numbers this product exists to audit — within weeks, the most “used” endpoints would be the ones we called most.

So the prober has its own dedicated account, published above, used for nothing else and sharing a key with nothing else. Every inbound payment from that address is identified on the public indexer and removed before any figure is displayed. 4 payments are currently excluded on that basis.

A second subtraction is less obvious and matters as much. Several of these merchants also sell on Base and Solana, and the upstream catalogue publishes one settle count aggregated across all chains. Printing that under a heading that says Algorand would overstate the largest merchants specifically. Every paid-call figure on this site is Algorand mainnet only, derived from the indexer — other chains’ settlements are not counted here, and where the two numbers differ, both are shown for what they are.

At this budget Assay cannot claim every listing carries a fresh receipt, and does not. A full paid sweep of every merchant would cost more than four times the entire float in a single pass, so coverage is traded for freshness deliberately: a receipt every five weeks or so on most merchants is more useful than one fresh receipt and every other merchant carrying nothing at all.

Two consequences follow, and the interface is built around both. Receipt age is shown beside every verified mark, because a five-week-old settlement is evidence about five weeks ago. And a free-tier pass is never presented as proof of fulfilment — an endpoint that answers its 402 correctly has demonstrated that it answers, not that it delivers what you paid for. Where only free checks exist, the page says so.

8 of the 39 merchants price their cheapest endpoint above the $0.01 per-call cap and are not on the paid rotation at all. They are verified free-tier daily like everything else, and get a settlement receipt only on demand, capped at one a day.