What Is This?

Most people encounter AI products through a chat box, a button, or an API call. Behind that simple interface, the product may make several runtime choices before an answer comes back. It may choose a model, resolve a model alias, call search, retrieve files, apply a safety layer, use a cheaper or faster tier, or fall back to a different provider when something is unavailable.

This is the same kind of software infrastructure web apps already use: load balancers, queues, caches, regions, feature flags, and fallback systems. AI products are now using that routing logic too. The difference is that the route can change the answer itself, not just the speed of delivery.

A casual user may not care. For a developer, customer, buyer, researcher, or auditor, it often does. If an answer changes because a model alias moved, a fallback ran, retrieval failed, search was used, or a different data-handling path applied, the answer alone does not explain why it changed.

Route Receipt is a small public format for showing what route this AI answer took before it reached me.

It makes consequential route choices visible in a compact record attached to a response, log entry, dashboard event, or audit export. It leaves out private reasoning and does not require providers to publish router code. It is a practical disclosure layer: enough route information to support trust, debugging, billing review, reproducibility, and governance.

Core idea

A route receipt presents these fields in structured form:

In practice, the record can include the provider and served model, whether the named model was an alias and, when available, which resolved version served the answer, which tools, retrieval systems, or external calls were used, whether fallback, safety intervention, or policy routing affected the request, what processing scope or data residency information can be disclosed, and which fields are withheld, redacted, unavailable, or not applicable.

The record can be tailored to each audience. Normal users may only need a short route summary, developers may need a debugging view, and auditors may need a more complete export. The schema supports those views without forcing every field into every interface.

What Route Receipt excludes

Route Receipt is a route record that excludes chain-of-thought disclosure. It leaves out hidden reasoning traces, safety classifier prompts, proprietary router weights, exact failover thresholds, and datacenter-level routing. Those details can create security, privacy, or competitive problems.

It is a route record that does not decide which model a product should use, which fallback is acceptable, or which jurisdiction should process data. It records the route facts that let people inspect those decisions after they happen.

Who uses route receipts in practice

Product teams use route receipts to show customers and internal reviewers how model routing works.

Developers use them to debug routing failures when a model alias changes, a retrieval path fails, or a fallback path changes output quality.

Enterprise buyers use them to verify that a product’s runtime behavior matches procurement, compliance, and data-handling promises.

Researchers and auditors can use them to compare which route-level facts adaptive AI systems expose and which they hide.

Why route receipts matter today

AI systems are becoming adaptive infrastructure, and they choose among providers, model versions, tools, tiers, safety systems, and regional processing options at runtime. That flexibility is useful, but it can hide the path an answer actually took behind a simple label.

Route Receipt closes that gap with a versioned record. The v0.1 format is built for route transparency for one response, keeps sensitive internals out of scope, and gives implementers a stable JSON Schema they can validate against today.

How to read the rest of the site

Use the specification for field definitions, the schema for machine validation, and the implementation notes when you are adding route receipts to an API, log pipeline, dashboard, or export flow. The companion paper lays out the research argument and situates the proposal in the transparency and provenance literature.