# Flow

> The first full evidence-to-reviewed-truth implementation and the source of the original portable grammar.

## Assessment identity

- Repository: `kinra-ai/flow`
- Revision: `4cc53682493850098dd0ef1ff59f57cb8c5044ef`
- Assessment date: 2026-08-09
- Evidence class: bounded maintainer attestation; product and customer data
  remain private

## Domain

Flow coordinates a manufacturing account whose customer intent, commercial
facts, and production reality arrive through differently authoritative and
sometimes conflicting sources.

## Mechanisms observed

- outside arrivals remain immutable evidence until a reviewed change advances
  the account;
- the account carries a monotonically advancing revision;
- proposed changes cite evidence, exact before and after values, rationale,
  uncertainty, and the observed revision;
- application refuses stale revision, incomplete provenance, or an invalid
  mutation and advances the coherent account atomically;
- current fields retain evidence and change ancestry;
- source authority is scoped rather than inherited from a system's importance;
- generated, sent, acknowledged, and incorporated publication states remain
  separate;
- returned outside contributions enter as evidence; and
- deterministic reads expose current truth and provenance to the operator and
  AI Peer without requiring storage reconstruction.

## What Flow paid to teach

Flow made the delta the durable product output. The system can answer three
different questions without collapsing them:

1. What is currently believed?
2. What changed and why?
3. What exact account crossed an outside boundary at that time?

It also demonstrated that an interface is not the reconciliation boundary. A
general operator reconciliation client was built and removed after use showed
that the operating AI Peer needed governed reads, while outside people needed
only focused projections.

## Limits

Flow's customer-account schema, command catalog, role names, repository ledger,
and specific review policy are not portable requirements. The implementation
does not prove that Git is the right storage substrate for a high-concurrency
or large-scale truth system. Outside adoption and business value remain
separate claims from mechanical correctness.