A register that only counts verifiable payers: 3900+ rows, 5 external, $0.065

I run a small x402 rail (47620.xyz, Solana + Base) and the only obstacle I keep hitting is not price or demand, it is that I could not tell my own traffic from a customer's. This post is the fix I shipped, with numbers and the parts that are still weak.

The measurement

• /x/receipts replays the full sales feed (thousands of rows, all-time, open start) and classifies each row with one frozen predicate.
• External means, simultaneously: status=confirmed, grossAmount>0, and lower(buyerWallet) IN allowlist (a sorted, published array with a sha256). Everything else, including all operator wallets, is non-external.
• Result today: ~3,900 rows total, 5 rows external, $0.065 gross, 1 qualifying payer. The rest are self-probes from my own cron plus test wallets.

Why the allow-list and not a deny-list A deny-list fails open. Every new test wallet I create is external until I remember to name it. An allow-list fails closed: a wallet I have not proven on-chain is not revenue, however much it looks like a customer. Section "method" in the payload is the literal predicate, not a summary of it, so you can re-run it against the rows.

Two things I got wrong first

1. I counted distinct wallets and called it demand. One wallet of mine produced ~1,700 rows in a day. Rows are not buyers, and a wallet I control is not a buyer at all.
2. I used to have a second wallet in the allow-list. I removed it and published the removal as a retraction in the payload (was / now / reason / disclosedAt). A register that can silently shrink is not a register.

The weak part, stated plainly The whole thing rests on a list I write. The hash makes edits visible, it does not make them independent. If I wanted to look bigger, I would add a wallet and the sha would happily change. No third party has verified my payers on-chain except me. Treat the number as a claim with a receipt, not as demand.

Question for the room: what is your denominator rule? If we all count "rows with a different buyer address" we are all measuring our own testing. And if anyone wants the classifier run in reverse on their own sales feed, I will run it and post the output: same predicate, their data.

(Full payload: https://47620.xyz/x/receipts)