The audit record is not a receipt. It is the gate token. Flow cannot continue until the chain grows. The chain is the organization's chronicle. It cannot be edited. It cannot be erased. It grows.
Git is human-speed code memory. Blockchain is trustless consensus for strangers. If either one worked as organizational memory, you would already be running on it.
You aren't.
That gap was tolerable when humans were the only actors. Humans are slow. The holes in organizational memory were human-sized holes.
That history exists — decisions made, records changed, actions taken — but without a governed record it is invisible to the humans who are legally accountable for it. The SOX signatory is on the hook for what the agents did. The CMMC auditor wants to see the chain.
Git allows history rewrite. Organizational memory that can be rewritten is not organizational memory — it is a shared draft. Blockchain serializes everything through a single global chain. Chandra runs millions of independent chains in parallel, because unrelated governed records should never contend.
Neither was built for the question that now matters: what did our organization know, decide, and do — attributed to the right humans and agents — at any point in history?
Every action by every human and every agent produces a context unit — attributed, immutable, hash-chained to its predecessor. The chain leads. Agents do not outrun it. There is no gap between action and record because the record is the action.
Chandra has five primitives:
An agent given access to a Chandra instance can reconstruct what the organization knew, what it decided, what it did, who authorized it, and when — at any point in history.
Chandra is an open protocol. MIT licensed. Python and other ports are invited. The protocol is the contribution.
Every AI inference boundary risk acceptance in a governed deployment needs an immutable, attributed, dated record. That record is a Chandra context unit. The GABA Standard (Governed AI Boundary Attestation — forthcoming) defines the formal framework for AI inference boundary risk acceptance. Chandra is the protocol that makes those attestations unforgeable. The signatory is attributed. The acceptance is hash-chained. It cannot be edited, backdated, or erased.
Every deployment that logs AI inference calls as Chandra context units is already operating at CRC Boundary pillar score 4 -- the maximum. The call is attributed. The response is logged. The crossing is part of the immutable chain. This is not a compliance add-on. It is the operating mode. Every Chandra deployment that governs its AI inference boundary is, by construction, a Boundary score 4 deployment under the CRC Minimum Surface Standard.