Autogenous turns spatial state into bounded, derived observations

What Autogenous is

Governed evolutionary software: a runtime failure becomes a candidate adaptation, the adaptation is verified, and it can be reversed. The repository labels itself a research prototype, which is the right label.

What changed

The radio-moe 0.3.1 release checkpoints the Cognitum Spaces adapter. It uses a strict HTTPS client, typed pagination with opaque cursors, and explicit workspace identifiers — and it recursively rejects raw sensing fields, enforcing anonymity, hierarchy, timestamp, confidence and response size contracts.

In September the envelope gained a regression candidate kind, recorded as ADR-404, so a regression can be treated as a first class evolution input.

Get started

A Rust workspace, not a one line install. Clone, build, and read the ADR index before wiring it to a live system.

git clone https://github.com/ruvnet/autogenous.git
cd autogenous
cargo build

Expected result: a built workspace and a readable ADR history explaining every decision.

No documented npx entry point. This is a research prototype; the repository says so in its own status badge.

Use it today

Practical case: converting sensing state into observations an agent may see without leaking the underlying signal. Input is authorized spatial state; the workflow is envelope, derive, bound; the output is an observation that cannot be reversed into raw sensing.

Acceptance test: feed the adapter a payload containing a raw sensing field. It should be rejected, not sanitised silently.

Push it further

Experimental commentary. Software that adapts itself to its own failures is the long bet. The dependency is reversibility: every adaptation must be portable and undoable, or you have built a system nobody can debug.

Falsifiable test: apply an adaptation, then reverse it, and diff the runtime behaviour against the original. Any residue is a bug in the reversal path.