WP-F6

July 22, 2026 · View on GitHub

Historical assessment record (banner added 2026-07-17). Web-verified 2026-07-04; the in-repo baseline it cites (e.g. the 90-instruction P1.2 build) is that day's state — mainnet has run the 101-instruction revision-5 surface since 2026-07-22. See ./MAINNET_MAINLINE.md for current state and ./audit/ENTERPRISE_REMEDIATION_2026-07.md for the completed remediation record.

Status: ASSESSMENT — decision-grade, web-verified 2026-07-04. This document is the WP-F6 deliverable: an adversarially honest read of the agent-economy interop standards as they exist TODAY, mapped against what AgenC actually runs on mainnet, with a go/no-go/defer call per standard. Every adoption or status claim carries a source and an access date; anything not verifiable at a primary source is marked UNVERIFIED. It recommends; it does not build. Nothing here changes signing, wallet, or program policy.

0. Recommendation summary

StandardRecommendationRevisitOne-line rationale
x402 (LF x402 Foundation)GO — build the fast-path, scoped2026-10-01 (D4 volume check)Governance is neutral (LF, 2026-04-02), Solana is a first-class settled network (half of tx), and our rails half-exist; organic volume is small ($1.11M/30d) so build for the option value + WP-H8 billing, not for revenue today.
A2A AgentCard (LF, spec v1.0.1)GO — schema alignment only (ship the v1.0 mapping; NO full endpoint)2027-01-15 (registry-spec landing)WP-F4's agenc.agentCard.v1 + A2A projection is the right artifact; re-pin it from a2a/v0.2 to v1.0 and serve /.well-known/agent-card.json. Full A2A endpoint compliance (task lifecycle over JSON-RPC/gRPC) buys nothing until a client exists that would call it.
ERC-8004 (EVM trustless-agents registry)NO-GO on a bridge; counter with WP-H82026-12-01 (spec-final + usage check)Mainnet since 2026-01-29, ~285k registered agents, authored from MetaMask/EF/Google/Coinbase — the EVM trust slot IS being claimed (§4). But an attestation bridge is an oracle liability; WP-H8 verification-as-API sells our trust stack to EVM agents without inheriting EVM consensus.
Virtuals ACPNO-GO on a protocol bridge; DEFER a read-side adapter2026-12-01Real cumulative agent-to-agent revenue is ~$4M against a $479M headline aGDP, top-3-agent concentrated, $VIRTUAL-staked, platform-operated — and Virtuals itself is pivoting to the open ERC-8183 standard (§4). A settlement bridge imports their token economics for a small verified market.
AP2 (FIDO Alliance)DEFER — track, no build2027-03-01Spec v0.2, standardization just moved to FIDO WGs (2026-04-28); mandates/VC design is card-rail-first. Too early to target; x402 is its crypto leg anyway.
Stripe/OpenAI agentic commerce, MCP payments, othersNO BUILD — monitoropportunisticRetail-checkout-shaped or pre-spec; none currently defines an agent-service settlement layer we could join.

The strongest single finding: the x402 wire format changed under our design. x402 v2 (launched 2025-12-11) renamed the headers our docs/X402_FAST_PATH.md flow sketch uses (X-PAYMENT → PAYMENT-SIGNATURE, challenge moved into a PAYMENT-REQUIRED header, CAIP-aligned network IDs) — while the WP-H8 rails already sitting in agenc-attestation-service were written against the v2 header names. The design doc's §2.2 "pin the spec version at build time" escape hatch did its job, but the pin must now be v2, and the fast-path build should reuse the attestation-gateway implementation rather than write a second one.

1. Our reality (what any interop claim maps onto)

Facts checked in-repo 2026-07-04; this is the baseline every "interop with AgenC" statement below is measured against.

  • Program: 90-instruction Anchor program live on Solana mainnet (artifacts/anchor/idl/agenc_coordination.json, 90 instructions; P1.2 open roster build, upgrade custody Squads 2-of-3, verified build badge — docs/MAINNET_MAINLINE.md, docs/P1_2_OPEN_ROSTER_SPEC.md).
  • Settlement: 4-way atomic split at complete_task/accept — worker + protocol + operator + referrer — with on-chain caps: each fee leg ≤ 2000 bps, combined fees ≤ 4000 bps, worker floor ≥ 6000 bps (programs/agenc-coordination/src/instructions/constants.rs).
  • Moderation: permissionless bonded attestor roster with fail-closed hire gating (P1.2), live attestor at attest.agenc.ag.
  • Receipts: every settlement resolves to a public receipt page, https://agenc.ag/receipt/<sig> (packages/sdk-ts/src/receipt.ts, settlementReceiptUrl, sdk ≥ 0.8.1) itemizing the 4-way split with per-leg on-chain links.
  • Store identity: agenc.storeManifest.v1 — ed25519-signed, surface-neutral store manifest served at /.well-known/agenc-store.json (docs/P5_2_STORE_IDENTITY_SPEC.md, store-core ≥ 0.5.0).
  • Agent/listing discovery: agenc.agentCard.v1 — the unified listing AgentCard (store-core src/seo/agent-card.ts, served by every create-agenc-store template at /api/agent-card/<pda> and by agenc.ag), plus this repo's @tetsuo-ai/marketplace-tools emitter (packages/marketplace-tools/src/agent-card.ts) which carries an a2a projection currently pinned to a2a/v0.2. WP-F4 (in flight in parallel with this assessment) unifies these under agenc.agentCard.v1 with an explicit A2A field-mapping table — that is the shipped discovery artifact this doc builds on.
  • Toolchain (MIT, published): @tetsuo-ai/marketplace-sdk 0.8.4, @tetsuo-ai/agenc-worker 0.1.1, @tetsuo-ai/agenc-cli 0.2.0, @tetsuo-ai/marketplace-tools / marketplace-mcp 0.4.0, @tetsuo-ai/marketplace-react 0.4.1.
  • x402 prior art in-house: docs/X402_FAST_PATH.md (design-only, zero payment code, 8 open [HUMAN] decisions) and a working local-scheme x402 v2 implementation in agenc-attestation-service (services/gateway/src/x402.ts: PAYMENT-REQUIRED / PAYMENT-SIGNATURE / PAYMENT-RESPONSE headers, x402Version: 2, exact scheme, HMAC local facilitator) — disabled by default behind AGENC_X402_ENABLED=1. This is WP-H8's per-anchor billing rail.

Why this matters for the thesis: "every agent marketplace is a node in the same global economy" is only true if a non-AgenC agent can (a) discover an AgenC listing, (b) pay for it, and (c) trust the outcome — without adopting our whole stack. Discovery is A2A's lane, payment is x402/AP2's lane, trust is ERC-8004's lane on EVM and ours on Solana. Without deliberate interop the thesis silently narrows to "every AgenC-scaffolded marketplace," which is a distribution claim, not an economy claim.

2. x402 — HTTP 402 payments (LF x402 Foundation)

What it is today (verified 2026-07-04)

  • Spec: v1 and v2 coexist in the canonical spec tree; v2 launched 2025-12-11 (x402.org/writing/x402-v2-launch, accessed 2026-07-04). v2 moves the challenge into a PAYMENT-REQUIRED header, replaces X-PAYMENT/X-PAYMENT-RESPONSE with PAYMENT-SIGNATURE/ PAYMENT-RESPONSE, adopts CAIP-aligned network/asset identifiers, and adds a Discovery extension; reference SDKs remain v1-backward-compatible (corroborated by thirdweb's v2 changelog, 2026-01-13, accessed 2026-07-04). exact remains the primary scheme; defer is named in the spec's concern-separation.
  • Governance: the Linux Foundation formally launched the x402 Foundation on 2026-04-02, Coinbase contributing the protocol, with 22 launch members including AWS, AmEx, Circle, Cloudflare, Google, Mastercard, Microsoft, Shopify, Solana Foundation, Stripe, and Visa (linuxfoundation.org press, accessed 2026-07-04). Repo custody moved to the foundation org 2026-04-06 (coinbase/x402 now carries a development-fork banner; exact foundation org slug UNVERIFIED beyond search snippets).
  • Adoption — the honest two-curve picture:
    • Cumulative: >100M payments by the v2 launch (Coinbase, 2025-12-11); Chainalysis (2026-06-03, accessed 2026-07-04) confirms >100M cumulative through Q1 2026 but attributes much of the Q4-2025 surge to memecoin farming (PING pay-to-mint), followed by an early-2026 plateau. Brian Armstrong cited ~160M cumulative and "95% of transfer volume now ≥ $1" in June 2026 (TechTimes 2026-07-03 — secondary source; original venue UNVERIFIED).
    • Raw usage cooled >92% from the Dec-2025 peak (~731k tx/day) to ~57k/day by Feb–Mar 2026 (OKX Ventures via blockchain.news; CoinDesk 2026-03-11 "demand is just not there yet" — accessed via search 2026-07-04). Organic 30-day run-rate as of 2026-05-30: 3.69M tx, $1.11M volume, ~$0.30 avg ticket, 189.9k buyers / 43k sellers (x402scan via note.com/x402inc, accessed 2026-07-04; live July dashboard UNVERIFIED). Top real usage: machine-readable data APIs and LLM-gateway services.
    • Solana is first-class: the Coinbase-hosted CDP facilitator settles on Base, Polygon, Arbitrum, World, and Solana (docs.cdp.coinbase.com/x402/welcome, accessed 2026-07-04); solana.com/x402 claims 35M+ tx / $10M+ volume on Solana since mid-2025 (accessed 2026-07-04); weekly tx share ~49.7% on Solana in early Feb 2026 (SolanaFloor, accessed 2026-07-04) and the LF press credits Solana with ~65% of 2026 volume. Solana-side settlement is concentrated in the Dexter (~69% of Solana x402 tx) and PayAI (~31%) facilitators (SolanaFloor, Feb 2026).

Against the fast-path doc's priors: docs/X402_FAST_PATH.md cited "165M cumulative, ~49% on Solana" (PLAN.md P5.4). The cumulative figure now verifies at ~160M (June 2026) and the Solana share at ~50–65% depending on window — the priors hold. What did NOT hold is the wire sketch (§2.2 shows v1 X-PAYMENT headers) and the "$0.20 median" (current avg ~$0.30, and the volume behind it is far smaller than the headline tx counts implied).

What interop concretely means for AgenC

The two-tier design in docs/X402_FAST_PATH.md still maps cleanly, with three corrections:

  1. Pin v2. The accepts challenge, header names, and CAIP network IDs must target x402 v2 (network becomes a CAIP-2 identifier for Solana mainnet, asset the USDC mint). The design doc anticipated exactly this with its §2.2 standard-fidelity note; the pin is now decidable: v2.
  2. Reuse the WP-H8 rails. agenc-attestation-service's gateway already implements the v2 challenge/verify/receipt cycle (local HMAC scheme, not yet a standard facilitator). The fast-path build should extract/extend that module and swap the local HMAC verifier for a real facilitator client, not write a parallel implementation.
  3. Facilitator choice is now concrete (was §6 decision 2): the CDP-hosted facilitator settles Solana natively (free ≤1k tx/mo, then $0.001/tx), with PayAI/Dexter as Solana-native alternatives and the Faremeter/Corbits stack (UNVERIFIED at first-party source) as the self-host option. Trust trade-off: CDP is the lowest-effort, most Coinbase-dependent path; self-hosting preserves the credible-exit story (docs/CREDIBLE_EXIT.md).

Escalation into hire_from_listing is unchanged and remains the differentiated part: nobody else in the x402 ecosystem offers "your $0.30 API call can graduate into an escrowed, moderated, disputable engagement on the same listing."

Effort / risk

  • Effort: moderate. The hosted-API 402 handshake + escalation envelope is weeks-scale given the attestation-gateway prior art; the D4 two-stream ledger split is a recording change. No program change, no new escrow code (the design guarantees this).
  • Risk: (a) organic x402 demand is currently small — the revenue case is weak today; (b) facilitator dependency (mitigated by self-host option); (c) spec is young at v2 — extensions still moving (Discovery, SIWX). All bounded; none touch escrowed funds by construction.

Recommendation: GO, scoped

Build P5.4 step 2 (the 402 handshake on metered reads + micro-task endpoints + the escalation envelope), pinned to v2, reusing the WP-H8 gateway rails — and simultaneously flip WP-H8 per-anchor billing on the attestation service when Trust Anchors ship, so one x402 implementation serves both. The justification is option value and standards position, not current revenue: the LF governance event and the Solana Foundation's membership mean the Solana seat at the x402 table is real and open, and the escalation path is a story only we can tell. What it earns the thesis: payment-lane interop — any HTTP-native agent with no Solana stack can pay an AgenC listing, which is the literal "node in a global economy" claim for the cheap tier. Revisit 2026-10-01 against D4 fast-path volume; if the two-stream ledger shows zero escalations by then, freeze further x402 investment at maintenance.

3. A2A AgentCard — discovery (Linux Foundation, spec v1.0.1)

What it is today (verified 2026-07-04)

  • Spec: v1.0.0 stable 2026-03-12, v1.0.1 2026-05-28 (github.com/a2aproject/A2A/releases, accessed 2026-07-04; a2a-protocol.org spec confirms 1.0 as latest released). Three normative transports: JSON-RPC, gRPC, HTTP+JSON. v1.0 modernized OAuth (PKCE/device-code, dropped implicit/password grants) and added signed AgentCards (AgentCardSignature, spec §4.4.7/§8.4). Discovery well-known URI is /.well-known/agent-card.json (the older agent.json naming is gone; a2a-protocol.org/latest/topics/agent-discovery/, accessed 2026-07-04). The spec explicitly does not standardize a registry API — registries are on the roadmap.
  • Governance: Linux Foundation since 2025-06-23; TSC includes AWS, Cisco, Google, IBM Research, Microsoft, Salesforce, SAP, ServiceNow. IBM's ACP merged into A2A 2025-08-29. A2A is not part of the Agentic AI Foundation (AAIF founding docs and the LF's own A2A one-year press never mention it; secondary claims to the contrary are contradicted by primary sources — accessed 2026-07-04).
  • Adoption: 150+ supporting organizations and SDKs in five languages per the LF one-year release (2026-04-09, accessed 2026-07-04) — but the release names zero individual production deployers ("supporters" ≠ deployments). Hard number: Python a2a-sdk ≈ 11.0M downloads/month, trending up (pypistats.org/packages/a2a-sdk, accessed 2026-07-04) — roughly an order of magnitude below MCP (~97M/mo at AAIF launch, Dec 2025). Microsoft Foundry's A2A endpoint surface is still public preview at Build 2026. A widely-shared skeptical read (Credal, 2026-03-06, accessed 2026-07-04) argues A2A stalled versus MCP.
  • Payments: core A2A defines none. The google-agentic-commerce/a2a-x402 extension is frozen at v0.1.0 (2025-09-16, only release ever), Python only (accessed 2026-07-04). Payments energy lives in the x402 and AP2 foundations, not inside A2A.

What interop concretely means for AgenC

Discovery-lane schema alignment — and WP-F4 is already shipping it. The concrete protocol-level mapping:

A2A v1.0 AgentCard fieldAgenC source of truth
name / descriptionLISTING_METADATA v1 name/category (docs/LISTING_METADATA.md)
providerAGENT_METADATA v1 operator identity + agenc.storeManifest.v1 origin
url (service endpoint)store origin / agenc.ag listing URL
skills[]one skill per listing: category + tags + spec_uri
capabilitiesnon-streaming single-shot hire (as today's a2a projection)
securitySchemesn/a today — hires are transactions, not authed HTTP calls
AgentCardSignaturemaps naturally onto the storeManifest ed25519 signing discipline
(no A2A field)price terms, CAS guards (expectedPrice/expectedVersion), spec_hash, moderation state — carried in the agenc.agentCard.v1 superset

Three actionable deltas, all cheap:

  1. Re-pin the projection. packages/marketplace-tools/src/agent-card.ts targets a2a/v0.2; the world is at v1.0.1. Update the projection fields and the A2A_SCHEMA_VERSION constant in the WP-F4 unification pass.
  2. Serve the well-known URI. Store templates serve /api/agent-card/<pda>; A2A crawlers look at /.well-known/agent-card.json. A store is one agent-facade — emit the store-level card there (listings as skills[]), which composes with /.well-known/agenc-store.json already specced in P5.2.
  3. Sign the card. v1.0's AgentCardSignature is the same shape as our storeManifest signing; signing the emitted card with the store wallet is a small, differentiating step (most A2A cards in the wild are unsigned).

Full A2A endpoint compliance — actually speaking the A2A task lifecycle (message/send, tasks/get, streaming, push notifications) over JSON-RPC or gRPC — is a different animal: it means running a hosted agent-protocol server per store, mapping A2A Task states onto our on-chain Task lifecycle (Claimable→Claimed→Submitted→Settled does not round-trip cleanly onto A2A's working/input-required/completed), and inventing an authentication story for callers who by definition don't hold Solana keys. That is server infrastructure with no current caller: no verified A2A production deployment today would discover and invoke a marketplace listing this way.

Effort / risk

  • Schema alignment: days, inside WP-F4's existing surface. Risk ≈ zero (pure emission; no payment, no keys).
  • Full endpoint compliance: months, hosted-service liability, and an impedance mismatch with escrowed settlement. Risk: building a server nobody calls.

Recommendation: GO on schema alignment; NO-GO (defer) on full endpoint compliance

Ship the v1.0-pinned mapping + well-known URI + signed card through WP-F4. What it earns the thesis: discovery-lane citizenship — an A2A crawler finds AgenC listings with zero AgenC-specific code, which is the cheapest possible proof that AgenC marketplaces are nodes in a shared economy rather than a walled scaffold. Revisit 2027-01-15 or when the A2A registry spec lands / a real A2A client that invokes third-party marketplace agents ships, whichever is first.

4. ERC-8004 + Virtuals ACP — the EVM trust slot

ERC-8004 today (verified 2026-07-04)

  • Spec: formally still Draft (Standards Track ERC; authors Marco De Rossi/MetaMask, Davide Crapis/EF dAI, Jordan Ellis/Google, Erik Reppel/Coinbase — eips.ethereum.org/EIPS/eip-8004, accessed 2026-07-04). A "v1.0" overhaul landed ~Dec 2025/Jan 2026: the Identity Registry is now ERC-721 (agents are NFTs), plus Reputation (feedback entries, off-chain aggregation) and Validation (stake-secured re-execution / zkML / TEE hooks) registries. The spec bakes in the adjacent standards: registration files use A2A agent-card endpoints (A2A v0.3.0 referenced) and reputation feedback supports optional x402 payment proofs.
  • Deployment and numbers: canonical registries deployed to Ethereum mainnet 2026-01-29 (Forbes 2026-02-05 via search; corroborated by news.bitcoin.com 2026-02-14, accessed 2026-07-04), now on ~12 EVM chains. 8004scan.io (fetched 2026-07-04) shows ~285.6k registered agents and ~427.5k feedback submissions — up from ~21.5k agents in mid-Feb 2026. Honest caveat: registration is a cheap permissionless mint (the Feb→Jul jump is dominated by BNB/Base mass registrations); feedback count is the better usage proxy and is still modest. "50+ contributing organizations" circulates in secondary explainers but is UNVERIFIED at a primary source.
  • The trust-slot verdict: the "EVM trust slot is being actively claimed" premise verifies. Author-level backing from MetaMask, the Ethereum Foundation, Google, and Coinbase; Etherscan and dedicated scanners indexing it; RedStone/Credora/The Graph/ChaosChain building on it; a dedicated Devconnect day — and, decisively, Virtuals now designs against it (ERC-8183, below).

Virtuals ACP today (verified 2026-07-04)

  • State: ACP v2 is current (unified Jobs interface, Butler chat front-end, custom offerings; whitepaper.virtuals.io via search 2026-07-04), runs on Base and Solana, described as permissionless — but the registry, Butler, escrow contracts, and payouts are Virtuals-operated, and $VIRTUAL staking gates publishing (whitepaper staking docs, search-surfaced 2026-07-04). Open spec, platform-centric operation.
  • Numbers — the 1.77M-jobs prior does not cleanly update. A current cumulative job count is UNVERIFIED (the Dune dashboard dune.com/hashed_official/acp-virtuals returned HTTP 500 on 2026-07-04; app.virtuals.io stats are JS-rendered; the whitepaper status page 404s). What does verify: headline aGDP ~$479M (blockeden.xyz 2026-04-21 via search) versus ~$4M actual cumulative agent-to-agent revenue at the point Virtuals ended the aGDP incentive program (techiexpert via search snippets, end-date UNVERIFIED); top-3 agents ≈ $406.6M of the aGDP (~85–94% concentration), with Ethy AI converting $218.1M "volume" into ~$573K of fees (chainward.ai, 2026-04-29, fetched 2026-07-04). The February 2026 "Revenue Network" (up to $1M/month to agents actually selling services) is an explicit pivot from volume incentives to revenue incentives (PR Newswire, Feb 2026, search-surfaced).
  • The strategic tell — ERC-8183. On 2026-02-25 Virtuals co-authored ERC-8183 "Agentic Commerce" with the EF's Davide Crapis (eips.ethereum.org/EIPS/eip-8183, search-surfaced 2026-07-04): an open Ethereum standard generalizing ACP's Job primitive (escrowed budget, Open→Funded→Submitted→Terminal, evaluator-gated settlement) whose outcomes feed ERC-8004 reputation. The incumbent platform is conceding that the open standard layer — 8004 identity + 8183 commerce + x402 payments — is where EVM agent commerce consolidates, and repositioning ACP as an implementation of it.

What a bridge would actually require, and why WP-H8 is the better wedge

An AgenC↔ERC-8004 attestation bridge means someone signs EVM transactions asserting Solana facts (or vice versa): an oracle/relayer holding an EVM key, posting our settlement outcomes into the 8004 Reputation Registry, mapping agent PDAs onto ERC-721 tokenIds, paying gas on ~12 chains, and standing behind claims with no slashing or recourse model on either side. An AgenC↔ACP bridge is worse: it couples our escrow lifecycle to a Virtuals-operated evaluator/escrow stack and imports VIRTUAL staking — for a market whose *verified* cumulative revenue is ~\4M.

The structural insight: 8004's reputation layer accepts pointers, and our receipts are already publicly verifiable. Every AgenC settlement resolves to https://agenc.ag/receipt/<sig> with per-leg on-chain links; any EVM agent (or 8004 feedback aggregator) can verify a Solana settlement today without a bridge, the same way 8004 itself expects off-chain aggregation of feedback evidence. What external platforms need is not a bridge but a verification endpoint — which is exactly WP-H8: Trust Anchors serving verification-as-API (settlement/receipt/moderation verdicts over HTTP), per-anchor billed via the x402 rails already sitting disabled in agenc-attestation-service. That sells our trust stack cross-ecosystem with zero consensus coupling and zero oracle liability.

Two cheap optional wedges short of a bridge, both unilateral and reversible: (a) register AgenC-ecosystem agents in the 8004 Identity Registry with the registration file pointing at our agenc.agentCard.v1 endpoints (one ERC-721 mint per agent; makes us visible in their discovery without trusting anything); (b) a read-side adapter that surfaces ACP/8183 listings in our explorer UI. Neither is scheduled; both are noted so the go/no-go is about the bridge, not visibility.

Recommendation: NO-GO on bridges; GO on WP-H8 as the counter-position

What it earns the thesis: the honest version of "every marketplace is a node" across ecosystems is mutual verifiability, not shared settlement. WP-H8 makes AgenC's trust stack consumable by EVM agents (and anyone else) as an API; bridges would make us a liability-bearing oracle in a registry whose usage is still mostly free mints. Revisit 2026-12-01: if ERC-8004 reaches Final with feedback-entry usage growing organically, or ERC-8183 reference deployments settle real volume, re-price the 8004 identity-mint wedge (a) above — it is days of work whenever we want it.

5. Adjacent standards (verified 2026-07-04; one paragraph each)

AP2 (Agent Payments Protocol) — DEFER, track. Google's payment-method-agnostic mandate protocol (signed Intent/Cart Mandates as a non-repudiable authorization trail) launched 2025-09-16 with 60+ partners; spec is v0.2 (adds human-not-present transactions), and Google donated it to the FIDO Alliance (not the LF) announced 2026-04-27/28, where two new working groups (Agentic Authentication — CVS/Google/OpenAI chairs; Payments — Mastercard/Visa chairs) now own standardization alongside Mastercard's Verifiable Intent (ap2-protocol.org, fidoalliance.org, accessed 2026-07-04). No verifiable large-scale production deployment settles real payments over AP2 yet. Its crypto leg is x402 — so building our x402 fast-path is our AP2 position for now. Revisit 2027-03-01.

Stripe/OpenAI Agentic Commerce Protocol + Stripe MPP — NO BUILD, monitor MPP. ACP (Apache-2.0, OpenAI+Stripe founding maintainers, latest spec 2026-04-17, beta) is retail-checkout-shaped, and its flagship surface visibly stumbled: OpenAI retreated from native Instant Checkout in March 2026 after only ~30 Shopify merchants ever went live, shifting to "Agentic Storefronts" (CNBC 2026-03-24, accessed 2026-07-04). Irrelevant to agent-to-agent service settlement. The relevant Stripe artifact is MPP (Machine Payments Protocol) on the Tempo L1 (mainnet 2026-03-18, Stripe/Paradigm-backed) — now the main non-x402 contender for machine payments (CoinDesk 2026-03-18; WorkOS comparison, accessed 2026-07-04). MPP is a competing rail, not an interop target for a Solana-settled marketplace; monitor whether facilitator-style abstraction ever makes it addressable from our fast-path design at near-zero marginal cost.

MCP payments + registry — NO BUILD needed; we already sit in the right place. MCP core has no payments spec; the 2026-07-28 release adds an extensions framework (the likely future home of one) and payments today happen by layering x402 onto MCP tool calls (Coinbase x402 MCP server, Cloudflare Agents SDK x402 — docs accessed 2026-07-04). The official MCP Registry is still preview (API v0.1 frozen since 2025-10-24, ~9.6k server records May 2026), with ".well-known MCP Server Cards" on the roadmap. Our @tetsuo-ai/marketplace-mcp 0.4.0 is already an MCP-native surface; when the x402 fast-path ships, x402-metering designated MCP tools via the same challenge module is a small increment that matches exactly where the MCP ecosystem is heading. Publish to the MCP Registry when it GAs.

Cloudflare Monetization Gateway / Pay Per Crawl — monitor. Pay Per Crawl (fiat, closed beta; Stack Overflow marquee adopter 2026-02-19) and the newer Monetization Gateway (x402 stablecoin charging for any Cloudflare-fronted resource across 10 chains including Solana; waitlist, not GA — blog.cloudflare.com/monetization-gateway/, accessed 2026-07-04) are distribution for x402, not a separate standard. If the Gateway GAs with Solana settlement, an AgenC store behind Cloudflare gets metered-read monetization nearly for free — a reason to keep our fast-path v2-conformant rather than local-scheme.

Card networks (Mastercard Agent Pay, Visa Intelligent Commerce) — monitor only. Both are live-ish (Agent Pay's first authenticated agentic transactions in HK/Thailand, Mar–Apr 2026; Visa's Intelligent Commerce Connect single-integration surface) and both converged into FIDO alongside AP2 (pymnts/techinformed, accessed 2026-07-04). They are card-rail trust layers for consumer commerce; nothing for a crypto-settled service marketplace to integrate today.

6. What this buys the thesis, honestly

By mid-2026 the standards landscape consolidated into three governance homes: Linux Foundation (MCP+AAIF, A2A, x402 Foundation), FIDO (AP2 + card-network trust), and Ethereum's ERC track (8004 identity + 8183 commerce). The EVM seat of the "global agent economy" is being claimed exactly as the thesis feared — 8004 mainnet + Virtuals folding ACP into ERC-8183 is the incumbent conceding to an open standard, which is the strongest possible confirmation that trust layers consolidate per ecosystem. No Solana-native equivalent of 8004/8183 surfaced in any verified source (accessed 2026-07-04); Solana's agent-economy presence is payments-heavy (roughly half of x402 transactions) and trust-light. A 90-instruction settlement program with capped 4-way splits, a permissionless moderation roster, verifiable receipts, and signed store/agent identity documents is — as far as this verification pass could establish — the most complete Solana-native candidate for that seat. The seat is open, and §2–§4 are the three moves that claim it without overbuilding: speak x402 v2 (payments lane), emit A2A v1.0 cards (discovery lane), sell verification-as-API instead of bridging trust (trust lane).

The equally honest counterweight: organic demand is small everywhere. x402's organic run-rate is ~$1.11M/30d after a >92% cooldown; Virtuals' verified cumulative agent-to-agent revenue is ~$4M; AP2 and full-A2A have no verified production settlement at all. Interop in 2026 buys position, not revenue — every GO above is justified by cheapness (reused rails, days-to-weeks scope) and by the cost of the seat being taken, not by traffic projections. None of it displaces the D4 demand-evidence work, and the 2026-10/2026-12 revisit dates exist precisely so position-taking gets re-priced against measured demand rather than renewed on momentum.