Fractalish public laboratory

AI

We are investigating whether an artificial system can become computationally cheaper, more selective, more recoverable, and differently disposed to reason because of what has happened to it.

The first distinction

Capability is not formation.

A capable host may already know how to write, calculate, code, simulate, reason, and use tools. The developmental question is different: can experience leave executable consequence that changes what the same capable machine can notice, reject, trust, retrieve, test, recover, or reach next?

CAPABILITY != HOST POLICY != ACQUIRED FORMATION

Nothing here requires accepting Consequential Formation in advance. The architecture must survive matched baselines, causal intervention, ablation, recovery, and independent review.

Status
ACTIVE RESEARCH PROGRAM
Evidence class
Mixed: frozen narrow results, implementations, specifications, failures
Current question
Can retained consequence become causally executable formation?
Current falsifier
Native baselines explain the gains at equal quality and lifecycle cost
Next test
Reversal, recovery, transplant, and ablation on a profiled host
Source artifacts
Open the artifact office
Code / data
Linked per project; absence remains explicit
Current orientation
CF v0.4 RC1, not frozen

Working architecture

The record is not presumed to be the runtime.

Formation Ledger

Audit, provenance, contradiction, transfer, debugging, and causal credit.

CANDIDATE REPRESENTATION

Formative Field

Indexed susceptibilities recruit a sparse coalition under current conditions.

BUILD-TEST TARGET

Minimum Formative Machine

The smallest causal experiment that can distinguish acquired formation from host capability or policy.

SPECIFICATION

The wider machine

Reasoning sits inside sensing, energy, body, and evidence.

Boundary

This is not an achieved intelligence.

MFM and the Formative Field remain build-test targets. Distributed embodiment does not establish identity across bodies. CNTM is optional. Energy availability is not intelligence. Multi-model agreement is not evidence.