Per-answer route facts
Route receipts describe the runtime path that served a specific AI response, not only the model family named in the UI.
Draft v0.1
Route Receipt Specification
A compact event record for the model, tier, tools, fallback, safety, and location route that served an AI answer.
Open draft specification proposal. Not a formal standard.
route_receipt.json
v0.1
- model
- provider/resolved_id
- tools
- retrieval, search
- fallback
- attempted=false
- view
- user | auditor
Portable JSON Schema
The v0.1 schema defines required fields, redaction markers, and extension points for provider-specific metadata.
Practical disclosure boundary
The record is designed to expose consequential route facts without exposing hidden reasoning, safety bypass details, or proprietary router logic.
Route Receipt Specification
Route Receipt specifies how to record the runtime route behind an AI answer. It gives developers, users, and auditors a short record of which provider, resolved model, tool path, fallback state, safety decision, and processing scope served a response.
The goal is to keep private chain-of-thought, router weights, safety prompts, and datacenter-level routing out of view while making the facts that already affect trust visible enough to debug, compare, and audit.
Available artifacts
The v0.1 package includes a human-readable draft spec, a JSON Schema that can validate receipt documents, and implementation notes for product teams that want to add receipts to APIs, logs, or dashboards.
The schema is hosted at /schemas/route-receipt/v0.1/schema.json. The URL is stable for the paper draft. Future releases should keep the same pattern and use semantic versioning.
Draft status
This is an open draft proposal maintained by Vincent Schmalbach. It is not a formal standards body, not a schema.org extension, and not an official W3C artifact. The site is meant to make the idea concrete enough for review, implementation experiments, and citation.