Decisions

1.The register

The three counts in the header are the whole 30-day window (8,648 decisions). The table below shows the newest 200 of them; the full window is at /api/v1/observatory/decisions, and the whole ledger at /api/v1/observatory/export.csv.

→ 0 more columns

2.Definitions

Each row is a decision the daily L1 runner actually made with real funds at stake, mapped 1:1 from the public ledger: refused_price_mismatch / refused_over_cap (wall demanded more than declared or over the hard cap — nothing signed), refused_payto_mismatch (wall named a payee other than the one the catalog declared — nothing signed), refused_payto_operator_self (wall named vet402's own receiving address — nothing signed), refused_wall_unpayable (no valid 402 / no machine-payable accept), paid_settled, paid_delivered_no_receipt, paid_settlement_claim_unverifiable (the wall claimed a successful settlement but the transaction identifier it returned is not even well-formed for that chain), paid_settlement_claim_unverified (claimed with a well-formed id, not yet re-read on-chain by us), paid_settlement_claim_refuted (we re-read it on-chain and the expected USDC transfer to the declared payee is not there — a finding about the seller, not about us), paid_no_settlement (vet402 signed and sent the payment authorization and no settlement was confirmed on-chain: the amount shown is what was put at stake, not money shown to have left vet402's wallet — money moved only on settled rows, and the settled rows whose paid request did not answer 2xx are counted on /observatory/state). As of 2026-08-23, paid_settled means vet402 confirmed the transfer on-chain (recipient, amount, token, chain, confirmations), not that the seller asserted it. Excluded as non-decisions: budget_denied, halted (the operator's runtime spending halt stopped the batch before signing), request_error, in_flight (vet402-side states), and payer_unfunded rows (a settle_failed answered 402 or 5xx on Base between 2026-09-13T00:00Z and 2026-09-15T23:49Z, when vet402's own payer wallet had run out of USDC, so no funds were at stake; since 2026-09-17 — they stay in export.csv with held_reason payer_unfunded).

Signal backtest (/api/v1/observatory/backtest): prior signal = at attempt time, (a) the two most recent L0 probes of the endpoint were both fail (two consecutive fails — the same threshold the public register uses), or (b) an earlier settle_failed purchase existed on the same endpoint. avoided = signalled attempts that did not settle; forgone = signalled attempts that settled anyway. Denominator: signed attempts whose outcome is settled in the ledger (settled / settle_failed / delivered_no_receipt / settle_claimed_unverifiable / settle_claim_refuted). settle_claimed rows (the seller named a transaction that vet402 has not yet re-read on-chain) are left out until the re-read turns them into settled or settle_claim_refuted, so a pending row is never counted as avoided. payer_unfunded rows (a settle_failed answered 402 or 5xx on Base between 2026-09-13T00:00Z and 2026-09-15T23:49Z, while vet402's own payer wallet was out of USDC) are neither attempts nor prior signals, since 2026-09-17. spentUnits is the USDC amount vet402 signed on those attempts (what it put at stake), not money shown to have moved: an avoided attempt has no confirmed on-chain transfer, so its spentUnits is exposure that honoring the signal would not have taken, not a loss.