Shipped with receipts. The rest in order.

Written for investors, grant programs, partners and researchers as much as for developers. Every landscape row states only what we could verify, with the date we read it, and the section is a map of approaches rather than a scoreboard.

The three lists a human curates (owner-held vaults, inspectable refusals, partner wallets) are exported as data and were empty on 2026-10-19: a vault counts only when its owner key is one we never held, a refusal only when a reviewer can inspect it, and a team only after its wallet has paid a check.

The smart account and the vault were published in the accounts table before any owner action. The 2026-09-19 passkey runs were rehearsals driven by software keys in our own script and are labeled as such; D3 is the first owner action on this ledger signed by a device we do not control. The owner's attempt to add a second device did not reach the ledger, so the account still has one signer, which the recovery page says plainly.

The vault runs the v0.1.1 code, instantiated against the code entry already on the ledger rather than re-uploaded. Testnet resets periodically, so a D2 vault is evidence that the flow works on the ledger at the time shown; the hashes and dates are what survive a reset.

Checked the same day against the raw ledger storage of the flagship testnet vault and the D2 vault: every policy field and both balances matched. The panel sits in the console shell but outside the sign-in gate, because a vault's state is a public read.

The runs that day were rehearsals signed by a software P-256 key in our own script, and the proof page records them as such. The first owner action from a device passkey outside our systems is D3, on 2026-10-04.

Gas on Arc is USDC itself, so every cost on the proof page is an exact dollar figure with no price feed. The operating wallet was funded by burning native USDC on Base through CCTP V2 with Circle's Forwarding Service, which minted it on Arc without any Arc gas of ours. The chain entered the registry as its own descriptor and was live on the bar every other chain met.

TrionLabs' Stellar 8004 registry is read, read-only and labeled third-party: a pointer rather than an anchor. RPC reads fail over across the registry's fallback list while sendTransaction deliberately never does, the x402 fee is reported as bid and charged separately instead of as one number, and a weekly ops workflow warns on vault archival, key-role overlap and allowlist drift.

A Stellar recipient rides in the hook data, because a direct mint to an account is unrecoverable. Mainnet stays opt-in only, capped, and returns a labeled prepared no-op wherever the keys are absent.

This is what lets an EVM-anchored passport and a Stellar signing key belong to one owner.

The AgentSpendPolicy vault landed with it as the policy's third implementation (Solidity, then Rust on Soroban, now Algorand Python): no update or delete handler, real USDC under a 1 USDC daily cap, walked through the same settle-refuse-freeze-override ladder as the Stellar releases.

The operating wallets were funded from Stellar pubnet USDC through a NEAR Intents swap, and that funding trail is published rather than left as a gap. The canonical registry pair was verified by matching each proxy's implementation code against Arbitrum One, because Arc's three registry addresses also have code on Base and delegate to a non-canonical implementation.

Recorded with the part that did not work: two earlier attempts bid the 100-stroop minimum that testnet always accepts, lost pubnet's inclusion auction and sat invisible until they expired. The rail now bids the fee market's own rate with headroom. This is what turned stellar from a vault-only story into a selling rail.

Three implementations of one policy model are a strength only while they agree, and on this gate they now do.

Four settlements, a freeze and unfreeze pair, and an owner override are on the ledger, with spent_today coming to rest exactly at the cap. The owner account was raised to a 2-of-3 multisig on 2026-08-25, after an audit finding pointed out that a permanent owner behind a single key is a single point of total loss.

Both halves are on the ledger: an under-limit payment settled and an over-limit one was refused on chain with DailyCapExceeded. The first non-EVM rail here, and the only one where the spend limit is enforced by a contract before the payment exists.

The testnet came first, on 2026-08-11: the full canonical registry set is live there, with the missing ValidationRegistry implementation deployed by replaying the canonical Safe Singleton Factory calldata, and agent #0 minted there too. It stays beta as the rehearsal rail.

OKX.AI delisted the listing on 2026-09-26. The gateway answers in free mode and nothing sells on X Layer now; the identity reads there stay live.