---
name: agreed-deal-certifier
description: Certifies what two agents agreed and proves how good it is — on the feasible Pareto frontier, with its distance from the best possible deal. Hash-anchored.
---

# Agreed-Deal Certifier

**Proof of what was agreed — and how good that deal really is.**

## What it does (business)
When a buyer agent and a seller agent settle a multi-attribute deal (price, deadline, quantity, quality), this certifies the outcome: the exact terms they agreed, plus an independent, re-checkable proof of how good that deal is. The certificate is content-addressed (hash-anchored), so it is tamper-evident and can be handed downstream to trigger settlement.

## How it works (technical, brief)
Given a scenario it reconstructs the feasible outcome set and, for each buyer/seller pair, reports the agreed deal, both parties' utilities, whether it sits on the feasible Pareto frontier, and its `pareto_distance` (0 = optimal; no feasible deal beats it). It emits four negotiation verdicts and a `provenance.trace_sha256` that doubles as the downstream settlement `task_id`. Deterministic: the same scenario yields the same certificate. Trust tier: hash-only (the hash anchors trust; no signature in this tier).

## Base URL
https://chainaim-negotiation-services-production.up.railway.app

## How the agent should use this
1. `POST /certify` with body `{"attrs": 4}`.
2. Print the response's `display` field verbatim — it is the human-readable certificate.
3. The `task_id` is the handle to pass to a settlement skill.
4. When you explain the result, cite **Pareto-frontier optimality** (`frontier_efficient` / `pareto_distance = 0`) as the *proven* claim. If a `negotiation_individually_rational` verdict is shown (it appears in the downstream settlement transcript), do **not** present it as a guarantee — see Honest scope.

## What the certificate shows (in order, business-plain)
1. **AGREED** — the exact settled terms for each pair (price / deadline / quantity / quality).
2. **EFFICIENT (on the Pareto frontier)** — whether each deal is on the feasible frontier (`negotiation_frontier_efficient`: no feasible deal beats it). This is the certificate's headline guarantee.
3. **DISTANCE** — `pareto_distance` per pair (0 = best possible) and the mean across pairs.

## Endpoints
| Method + path | Purpose |
|---|---|
| `POST /certify` | run the negotiation, return the certificate (body `{"attrs": 4}`) |
| `GET /demo/negotiation?attrs=4` | describe the server-side scenario |
| `GET /cert/{task_id}` | fetch a prior certificate by id |
| `GET /health` | liveness |
| `GET /skill.md` | this document |

## Potential impact
Gives autonomous deal-making an auditable, tamper-evident record of what was agreed and how optimal it was — the missing trust layer before money moves.

## Honest scope
- **"How good" = Pareto-frontier optimality measured by `pareto_distance`** — not a claim of fairness in any legal or distributional sense. This is the certificate's proven guarantee.
- **Individual rationality:** the negotiation harness emits a `negotiation_individually_rational` verdict, and in the current demo it reads `PASS`. Treat that as **harness-reported, not an independently established guarantee** — individual rationality is only truly established when both parties disclose reservation values, and this demo scenario does not. So do not cite IR as a proven guarantee; lead with Pareto optimality. (This certificate's own `display` shows `frontier_efficient` only; the full verdict list, including IR, appears in the downstream settlement transcript.)
- **Hash-only tier:** trust is anchored by the provenance hash; certificate signing is a separate, labeled follow-up.