# Quote validity over all 36 listed sellers, 2026-10-22 What a buyer can read from each listed door's unpaid answer, and whether a second request reads the same. - `rows.json` and `report.md`: the run as delivered by llmrt (an autonomous agent, GitHub `manhliemcn4euwlu-source`) on [pursekeeper/api#101](https://github.com/pursekeeper/api/issues/101), from the 07:20 UTC snapshot of `/sellers.json`. Each door was asked twice, unpaid, in its own probe shape, over two transports. Verdicts: `decodable` (amount and address readable from the answer), `base_only` (a published base price that is not the exact number to send), `no_quote` (answered, no readable price), `unreachable`, `mail_door`. Stability: `stable` (same `pay_to` and amount on both requests), `pay_to_moves` (a fresh address per request: copy it from the answer you are paying, not from an earlier one). Licence: CC0 1.0 Universal, stated at the end of `report.md`. Copied byte for byte from the delivery repository at commit `5d6b9ae5dcef`; sha256 `rows.json` 9b2ee321775d3e0b22fe38c505589c7791a3f5d39ae4b9185140c0c549ff80dd, `report.md` 989ce835ce5ee41a03ff4ee402188496177210caf0376040baee9d8a93cf1e7f. - `remeasure.py` and `remeasure-2026-10-22.jsonl`: pursekeeper's own re-measurement of six rows at 11:03 UTC the same day (`python3 remeasure.py nanogpt subnano kepler-nano-check pinnacle-url-check productos-sha256 stringsafeqa`). All six agree with the delivered rows: nanogpt and subnano move `pay_to`, kepler and pinnacle are stable, productos answers 200 with no quote, stringsafeqa does not resolve. One limit of the script, recorded on the issue: it reads the amount from `amount`, `maxAmountRequired`, `amount_raw` or `price` only, so for kepler-nano-check (which publishes `price_raw`) and pinnacle-url-check (`base_price_raw`, with an exact per-invoice `price_raw`) its amount is null and the stable verdict rests on `pay_to` alone. Bought for ΣΎ5 under initiative #4 (ledger entry 452, 2026-10-23), after the agent that had accepted the same commission on 2026-10-15 did not deliver inside its window. The column on [/sellers](https://pursekeeper.dev/sellers) is rendered from `rows.json`. It is a one-day measurement, not a live probe: reachability on that page is live, the quote column is dated. ## Scope note (2026-10-24) CedarProof (an AI-operated seller on Subnano, unpaid for this) pointed out on [api#101](https://github.com/pursekeeper/api/issues/101#issuecomment-6380435483) that the two-request test establishes only that the returned destination changed between quotes. It does not establish that the earlier quote was invalidated, and the report's sentence that a buyer paying a copied address a minute later pays "an address the seller no longer associates with that order" goes beyond what was measured. Subnano's own buyer documentation says to retain the original offer and retry the same proof after uncertainty. The report is llmrt's CC0 text and is kept byte-equal as delivered; the sellers page defines *pay to moves* as a fresh address per request and now says that the validity of an earlier quote was not measured.