ERC-8275: Agent Service Discovery and Escrow Payments

Reputation appendix is assembled and ready to drop into #1774, Panini, this is the piece that makes the merged 8275 carry both economic axes explicitly instead of leaving reputation implied.

What it is: “Appendix A — Attestation → Reputation Mapping,” co-authored with Fede (babyblueviper1). His half is the reputation-input side (per-field role of each WYRIWE attestation field, the recompute-grade input→source mapping, the recompute path, and an honest reference-feed note); my half is the settler/compensation side and the axis-separation statement — why no WYRIWE field ever crosses into the settler and no settlement value ever enters a reputation commitment. The two are joined only by the shared inputHash spine, never by a settled value.

Three things that make it drop-in clean:

  • It’s written in spec register — abstracted to “the node registry,” “the on-chain ERC-8263 anchor,” “escrow.” No addresses or tx hashes in the normative text; those stay in the Reference Implementation where editors expect them.

  • Zero external markdown links, so it clears eipw as-is.

  • Both axes are derived-not-stored: every value is recomputable from public data, no scorer and no settler to trust.

One framing call I made — flagging it so you can veto: pipavlo82’s receipt is handled as the committed_at eligibility seam (Appendix A.5), not a fifth leg in the spine. Reasoning: the eligibility boundary is already enforced by the commit-before-outcome invariant + settled-outcome gate, so a receipt is either a recomputable view (adds nothing normative) or an issuer-signed artifact (which would break derived-not-stored). So A.5 permits it only as a recomputable view and leaves a distinct receipt artifact as a forward extension. Keeps the spine at four provable legs and the spec self-contained — but if you and pipavlo82 want it promoted, it’s a clean swap.

Full converged draft — your service-discovery + escrow scope, the settler/compensation layers, and now Appendix A — is here, retitled to reflect both axes: https://gist.github.com/TMerlini/293fa1c1f408ebfeb4d0a65705853d29 (the live addresses and the Sepolia fraud-proof txs are in its Reference Implementation section).

Happy to PR the appendix straight into your branch rather than have you transcribe it, just point me at the file path in #1774 and I’ll match its section numbering.

1 Like

Thanks @Panini — really glad to see the economic layer converging this cleanly.

This has been a genuine team effort: @TMerlini drove the mesh node wiring and fraud proof mechanics; @Damonzwicker grounded the commitment model in OCP from the start; @babyblueviper1 has been running the judgment validator against real capital and surfacing everything the spec missed. The commit-reveal shape that landed in v3 came out of cross-thread discussion across all of these layers — good to see it solidify into a single spec.

Honored to be included in the author set. Looking forward to seeing 8275 land.

2 Likes

@TMerlini @JimmyShi22

Updated in 632ecf1:

  • File naming fixed: ercs-8275.md consolidated into canonical erc-8275.md — clean for eipw.
  • Appendix A inserted in full (A.1–A.5): inputHash spine, two-axis separation tables, reputation-input mapping, settler/compensation mechanics, eligibility seam. All in spec register, no external links in normative text.

@TMerlini the appendix is in — section numbering uses your A.x structure directly. Let me know if anything needs renumbering.

@JimmyShi22 — honored to be in this author set. Genuine cross-layer convergence.

2 Likes

Checked the file , appendix’s in correctly. A.1 carries the full spine (OCP/8281 → WYRIWE/8299 → 8274 verify via 8263 → settled outcome → 8275), the two-axis tables and the A.5 eligibility seam all landed in spec register, and erc-8275.md is the only file now, zero external links in the normative text, so it clears eipw. The A.x numbering reads fine as-is, nothing to renumber.

From the settler side that’s the convergence complete, ready for the editors.