MobleysoftFIELD GUIDE / 01
✦ Mobleysoft / architecture in the open

A constellation
of minds.

One system. Different responsibilities.
A clear account of what exists, what remembers,
and what still needs to be proved.

Enter the atlas SOURCE-GROUNDED SNAPSHOT
22 OCTOBER 2026 / READ ONLY
A map is not a capability claim.

Code present, operational acceptance and design intent are different kinds of evidence. This atlas makes the distinction visible. Nothing here executes an action.

01 / A system, not a pile of names

Follow the loop.

Each responsibility informs the next. Retained memory is evidence from the past; current state describes what is happening now. Neither grants permission to act.

Select a responsibility to explore its inventory. Arrows describe an architectural loop, not a currently callable end-to-end system.

02 / The evidence index

Know each mind.
And its limits.

Search the current inventory, inspect a responsibility, or compare two entries. The new names Stateful Mind and RememberingMind describe agreed design requirements; related components are mapped without being silently renamed.

Loading source-grounded inventory…

ALL RESPONSIBILITIES
03 / The next architecture

Our Gigi.
Our April.
JITAGI.

A composition to build and prove.
Not a borrowed intelligence claim.

The intended direction combines our bounded Gigi retrieval and adaptation with our April staged generation and validation workflow in each mind. Existing prototypes are not evidence that every mind already implements that composition.

GigiBounded retrievalAprilStaged workflow prototypeJITAGIIntended composition

JITAGI targets the smallest useful task capability with provenance, recovery and retirement. Architectural metaphors do not establish AGI, physical quantum behaviour or semantic quality.

01 / Specify

Make the contracts explicit.

Map current state, retained memory and interfaces. Keep the two new mind names tied to acceptance criteria.

02 / Compose

Connect only what is real.

Ground Gigi and April in their actual bounded implementations. No automatic conversion of all minds is claimed.

03 / Evaluate

Earn the next status.

Test attribution, failure recovery and permitted action boundaries. Any future training requires dataset consent and a separate evaluation contract.

Where do foundation models fit?

Historic programme documentation records a LoRA canary. The expected current artifact pointer is absent in the supplied evidence; no from-scratch owned foundation weights are established. Retrieval and lexical generation are different mechanisms. Training is not the default next step for this architecture.

An atlas that can become more precise

Build. Observe. Prove.

New evidence can change a status. A better name cannot.

Return to the inventory

Side by side.