# exo evidence boundary

## Snapshot

- Repository: https://github.com/exo-explore/exo
- Pinned revision: `b5375f8cee4368d09e1ce96a56b9f81fb0bc81aa`
- Verified: 2026-07-21

## What the evidence supports

1. README directly supports the project's stated discovery, cluster, API, dashboard, parallelism, and self-reported benchmark claims.
2. `AGENTS.md` directly supports the current Router/Worker/Master/Election/API architecture and event-sourcing model.
3. `placement.py` and `topology.py` directly support the documented constraint filtering and graph operations.
4. The Zenoh migration commit supports that the networking layer recently changed from libp2p.
5. The Kimi K2.7-Code commit is a maintainer-reported single-configuration distributed test, not a broad benchmark.

## What remains blocked

- No universal linear scaling claim.
- No production-grade high-availability claim.
- No assumption that all model-card backends are equally mature end to end.
- No assumption that README/App environment-variable names are fully synchronized after the network migration.

## Upgrade rule

A stronger conclusion requires a pinned public artifact or reproducible measurement with hardware, OS, network, model hash, context, raw logs, repetitions, and failure conditions.