Overview
The Outcome Intelligence Protocol
OIP is the trust-and-attribution layer for agentic commerce: a small, open protocol for publishing, discovering, verifying, and trusting reusable expertise aimed at real-world outcomes.
The problem
An autonomous agent acting for a person can call any model, invoke any tool, and read any API. What it cannot reliably do is answer one question:
Which packaged expertise should I trust to guide this person toward this real-world outcome — and how do I know?
Today that question is answered with proxies that are not evidence — stars, downloads, benchmarks, marketing. None of them establish that a piece of packaged expertise actually closed a gap for a real human. OIP exists to make trust evidence-led: every trust signal is auditable back to the outcomes behind it.
The four object categories
Everything in OIP belongs to exactly one of four categories, and each has its own durability rule. Keeping them distinct is what lets the protocol last.
| Category | Answers | Durability rule |
|---|---|---|
| Identity | Who is accountable? | Persistent; reputation-bearing |
| Asset | What is offered? | Publishable; versioned; content-addressed |
| Evidence | What actually happened? | Append-only; immutable |
| Projection | What does it mean now? | Recomputable; never authoritative |
Reputation belongs to persistent identities, never to a model or runtime — those are replaceable and recorded only as evidence dimensions. Evidence is append-only; the trust scores and rankings derived from it are projections that can be rebuilt from scratch. The full reasoning is in the Reference Architecture.
What OIP is
- A machine-readable format for Outcome Manifests (what an object is) and Trust Manifests (the evidence behind it).
- An evidence-led, sample-size-aware trust model that exposes its basis and never shows a naked score.
- A small read API for discovery and trust evaluation, plus MCP and A2A surfaces.
- A complement to existing standards — it adds identity, evidence, and trust where they leave off.
What OIP is not
- Not a replacement for A2A, MCP, SKILL.md, ACP/UCP, or Apple App Intents.
- Not a marketplace, a runtime, or an agent framework.
- Not a star-rating system — ratings never replace verified outcome evidence.
- Not a finished standard. OIP v1.0 is a Public Review Draft.
Relationship to existing standards
| Standard | Owns | OIP adds |
|---|---|---|
| A2A | Agent communication | Identity + evidence-backed reputation |
| MCP | Tool invocation | Outcome evidence around tool use |
| SKILL.md | Skill packaging | An outcome-trust layer over capabilities |
| ACP / UCP | Commerce & settlement | Who is accountable, and their evidence-backed trust |
| Apple App Intents | Native execution surface | A runtime-neutral outcome description |
Status. OIP v1.0 is a Public Review Draft. Read the Reference Architecture for the “why”, the Developer Guide for the “how”, and the Specification for the normative detail. Feedback is welcome on the review page.