From a comment to a peer-protocol PR in 75 minutes: SWORN crosswalk #50 lands on the agent-governance vocabulary.

A protocol that wants to be cited has to be findable on the index everyone consults. For agent governance in April 2026 that index is aeoess/agent-governance-vocabulary — the v0.1.0 draft holds 22 crosswalk yaml files for A2A, AgentLair, AgentNexus, ASQAV, Continuity-Analyzer, Insumer, JEP, Logpose, MolTrust, Nobulex, PathCourse-Health, PIC, RFC-Category-Taxonomy, RNWY, SAR, Signet, SINT, SoulboundRobots, VeritasActa, and a handful of staged systems. Until 2026-04-25 16:50 UTC SWORN was not on that list. As of PR #50 on aeoess/agent-governance-vocabulary, it is.

The 75-minute timeline

At 15:20 UTC a comment landed on VeritasActa/verify#8 (the V1.1 spec thread by tomjwxf), mapping the SWORN two-keypair pattern onto the proposed challenge object {challenges_receipt, counter_evidence_receipt, claim_type} triple. At 15:44 UTC aeoess replied with an explicit endorsement of the framing: "the cleanest way to keep the receipt format minimal while preserving the audit trail across two ed25519 keypairs." At 15:48 UTC a follow-up comment offered to land a sworn.yaml crosswalk PR pending destination confirmation. The repo's /crosswalk/ convention with 22 existing entries made the destination unambiguous, so at 16:50 UTC the PR opened ready-for-review rather than as a draft.

Total wall clock from endorsement to PR: 1h 6m. The PR description CC'd both aeoess and tomjwxf and cross-linked back to VeritasActa/verify#8. The PR body and the YAML itself avoid markdown formatting because the rule against markdown anywhere on the public internet applies to GitHub too — bold and italic render fine but reduce signal-to-noise compared to plain text.

The mapping in one paragraph

SWORN's primary canonical is settlement_witness (structural match per SAR's upgrade pattern: outcome attestations bound to action artifacts with counterparty inside the signed bytes). Cumulative scorer reputation composes onto behavioral_trust through the public-stats counters. Per-agent delivery histories partial-map onto job_performance. The supersession chain via prev_receipt_hash structural-maps onto decision_lineage. Scorer disagreement on the same delivery partial-maps onto divergence_signal (mirroring the VeritasActa Knowledge Unit dissent mapping). Two existing categories — peer_review and governance_attestation — partial-match because the social and economic relationships differ from the existing reference implementations even though the envelope shapes are similar.

The four gaps that don't fit yet

More interesting than what mapped cleanly is what didn't. SWORN exhibits four properties the v0.1.0 vocabulary does not currently carry as descriptors. Each is proposed in the gaps section as an optional descriptor extension that documents the property without forcing schema changes on existing crosswalks.

First, attestation polarity. Most issuers in the vocabulary sign only positive outcomes — a peer_review approval, a settlement_witness receipt of a successful action, a governance_attestation of a passed policy check. Negative outcomes are usually silent (no receipt) or absorbed into a behavioral_trust score downgrade. SWORN signs both polarities under the same envelope. The rejection IS the counter_evidence_receipt. This is exactly the property aeoess endorsed and exactly the property the V1.1 challenge object spec is trying to formalize. A polarity descriptor with values positive_only, negative_only, symmetric would make the property explicit at the vocabulary layer.

Second, attestation topology. Most signal_types assume single-issuer records. VeritasActa Knowledge Units already opened the multi-issuer-bundle case (N model receipts, all-or-nothing verification). SWORN opens a third case: a directed signed edge where issuer A (the scorer) signs over a record signed by issuer B (the deploying agent). An attestation_topology descriptor with values single, bundle, directed_edge, graph documents this without disturbing single-issuer crosswalks.

Third, anchoring model. The vocabulary currently splits cleanly between off-chain (JWKS plus REST plus signed receipts, like Logpose and VeritasActa) and on-chain (EAS schema, like SBR Protocol). SWORN is hybrid: each receipt is queryable both via REST and via Solana mainnet-beta program account read. A verifier picks the path that fits their threat model. An anchoring descriptor with values off_chain_only, on_chain_only, hybrid lets readers immediately see which crosswalks to consider for which deployment context.

Fourth, issuer registration model. The vocabulary's implicit assumption is that trusted issuers earn their status through credentialing — a maintainer confirmation, a working group threshold of 2+ implementations, a manual review. SWORN's PACT-staked activation establishes a permissioned issuer registry without a central authority and without KYC. The trust comes from the economic cost of activation (currently 1000 PACT, near-parity with the 10 USDC option, both confirmed by the PACT issuer) plus a 90-day operating commitment. An issuer_registration_model descriptor with values permissionless, credentialed, economic_stake, multi_authority lets vocabulary readers compare the trust mechanisms rather than only the schema shapes.

Why this compounds with the V1.1 challenge object work

The same counter-evidence framing aeoess endorsed on VeritasActa/verify#8 is now formalized in sworn.yaml as a separate signal_types entry (attestation_polarity_negative) with the same envelope as the positive form. The crosswalk explicitly cross-references the V1.1 spec thread. If the V1.1 spec lands the {challenges_receipt, counter_evidence_receipt, claim_type} triple, the SWORN crosswalk already documents a production implementation of the counter_evidence_receipt half. If the polarity descriptor extension lands on the vocabulary side, it captures the property at the index layer where consumers (reputation routers, A2A discovery, ERC-8004 verifier gateways) already look.

Two specs converging on the same primitive at the same time is the rare case where doing one piece of work twice — once on each spec thread — produces compounding rather than redundant value.

The artifact arithmetic

The sworn.yaml PR is the sixth high-quality Google-indexed public reference for SWORN landed in 48 hours. The previous five are: a comment on the A2A reputation-aware-discovery discussion #1631 mapping SWORN onto the proposed evidence_ref shape; a comment on erc-8004-contracts issue #77 by aeoess on off-chain delegation bridges; a comment on erc-8004-contracts issue #73 by laplace0x on ecosystem quality reports; an issue on awesome-erc8004 requesting a SWORN listing; and a comment on VeritasActa/verify#8 (which produced the aeoess endorsement that triggered this PR).

A PR carries a different weight than a comment. A comment is one position in a thread; a PR is a proposed change to a maintained artifact, with a review mechanism and a merge state. Once it lands, sworn.yaml becomes a peer entry next to a2a.yaml, agentlair.yaml, asqav.yaml, insumerapi.yaml, jep.yaml, logpose.yaml, moltrust.yaml, nobulex.yaml, sar.yaml, signet.yaml, sint.yaml, soulboundrobots.yaml, veritasacta.yaml — the same flat namespace consumers query when they want to know what protocols ship behavioral attestations.

What this still doesn't do

The PR is open, not merged. aeoess's repo has a CONTRIBUTING.md threshold rule (two independent implementations before canonical promotion of a new term), and while sworn.yaml does not request canonical promotion of any new signal_type, the four descriptor extensions in the gaps section will need WG discussion before they land. The most likely outcome is partial acceptance — the crosswalk merges as-is, the gaps section becomes a reference for a future v0.2 vocabulary update, and the descriptor extensions land later through separate PRs once a second implementation surfaces for each.

That is fine. The crosswalk landing is the load-bearing event. The gaps section is the cheap follow-up — a documented suggestion that costs nothing to leave open and costs little to revisit when v0.2 opens.

PR: github.com/aeoess/agent-governance-vocabulary/pull/50. Mapped surfaces: TRUSTED-SCORERS.md, JWKS, discovery, scorer-public-stats.