Open Decision Evidence Standard · v0.2 draft
The decision travels. The evidence should too.
ODES is a vendor-neutral, portable decision-evidence record format that lets AI-influenced decisions travel with structured governance metadata across organizational, jurisdictional, and system boundaries — without requiring shared infrastructure, proprietary trust networks, or disclosure of confidential source data.
Issuing party
Makes or influences a decision — model, human reviewer, or a mix of both.
Relying party
Verifies the record and decides whether to rely, ask for more, or escalate.
The problem
Decisions cross boundaries. Their evidence gets left behind.
An AI model might help approve a claim, clear a transaction, review a vendor, or triage a case. That decision then moves — to a customer, a regulator, an auditor, a partner. The outcome travels, but the structured evidence behind it stays trapped in the issuer's systems.
The receiving party often can't answer basic questions about what they're being asked to rely on:
- Q1
Was AI involved — and was it retrieving, recommending, executing, or escalating?
- Q2
Did a human actually review the output, modify it, override it, or just accept it?
- Q3
Was that human authorized to make the decision at that time?
- Q4
What model version, tool access, and runtime conditions shaped it?
- Q5
Is the evidence still current — or has it expired, been revoked, or been superseded?
What ODES is
A decision-evidence container: coordinates, not conclusions.
ODES defines one thing well: a portable record format and shared vocabulary that lets a decision carry its governance coordinates across a boundary. It is intentionally minimal, but not simplistic: not a platform, not a ledger, not a governance suite — the portable decision-evidence object those systems can emit, exchange, and verify.
Authority & roles
Who or what decided, under whose authority, the machine's role, and what the human actually did.
Basis, not raw data
References, hashes, attestations, verification metadata, policy identifiers, and risk coordinates — the confidential evidence can stay home.
Freshness
Expiration, revocation, supersession, and consumption conditions are first-class parts of the record.
Why it works
It works by staying narrow.
ODES does not try to govern all of AI. Its strength is that it does one job in the open: carrying decision evidence across a boundary without forcing shared infrastructure, platform lock-in, or raw evidence disclosure.
Portable, not internal
Governance frameworks manage a firm internally. ODES focuses on the decision evidence that has to leave the firm and remain evaluable by another party.
Vendor-neutral
No single platform, ledger, cloud, or proprietary trust network. Counterparties keep their own tools and still interoperate.
Human role made explicit
It distinguishes machine role, human disposition, and reliance level — so nominal approval reads differently from real human judgment.
Privacy by coordinates
The raw evidence doesn't have to move. Hashes, references, and risk coordinates let the relying party apply its own rules.
Scope
Honest about what it is — and isn't.
ODES treats decisions as governed artifacts rather than bare outcomes. It is a candidate interoperability primitive, not a silver bullet. Keeping the scope clear is exactly what makes it credible.
ODES does
- Define a portable decision evidence record
- Give parties a shared vocabulary and profiles
- Let a relying party verify structure independently
- Make freshness and revocation checkable
- Sit between existing governance and provenance tools
ODES does not
- Prove a decision was correct or compliant
- Replace legal, audit, or regulatory review
- Act as a governance platform or model registry
- Move or expose the raw underlying evidence
- Force anyone onto a single vendor or ledger
In practice
Six steps, one boundary.
Implementations differ by industry, but the logic is consistent from issue to reliance.
A decision is made or influenced by an AI system, a human reviewer, or a human-machine workflow.
The issuer creates a record using the ODES field vocabulary and a schema profile.
The record captures the essentials — authority, model state, machine role, human disposition, evidence commitment, policy basis, risk coordinates, freshness.
It travels with the outcome, while confidential underlying evidence can remain with the issuer.
The relying party validates it — structure, profile, freshness, revocation, supersession, and consumption conditions — then applies its own reliance rules.
A reliance decision is made: rely, request more evidence, reject, or escalate for human review.
The missing link
ODES doesn't replace the stack. It connects it.
Governance systems show that an organization has controls. AIBOMs describe what an AI system is. Provenance tools show lineage. ODES describes what a specific AI-influenced decision was — the layer that lets the rest interoperate.
AIBOM
Describes what the AI system is.
ODES
Describes what the AI-influenced decision was, and what evidence another party can evaluate before deciding whether to rely.
Get started
An open standard is only as strong as who shows up.
ODES is an early, open discussion draft. Read the draft, run a pilot on one boundary, or help shape the vocabulary.