[Epic] Build the shared human-agent understanding environment

[DG-creative-lab]

Product outcome

Learning Foundry is a shared understanding environment that keeps humans and agents able to think, create, and make changes together as systems evolve.

It converts approved sources and practical activity into a traceable Living Theory, then uses explanations, interactive exploration, retrieval, prediction, transfer, capability evaluation, and feedback to keep human understanding and agent capability evolving together.

Problem

AI can increase production velocity faster than humans can build and maintain the mental models needed to participate. Learning Foundry should make the resulting understanding gaps visible and provide useful interventions without claiming to measure cognition or “solve” cognitive debt.

Product principles

• Understanding exists to support participation, not only verification.
• Sources remain canonical; generated synthesis is a traceable, revisable lens.
• Human confidence, human understanding, agent synthesis, source fact, and validated behavior are distinct epistemic states.
• Memory is necessary but not equivalent to understanding.
• Practical prediction and transfer matter more than content completion.
• Dynamic media is used where interaction reveals a causal mechanism.
• Capabilities remain inspectable, versioned, evaluated, and approval-gated.
• The prepared journey is reliable offline; live Codex execution is optional.
• The core product remains domain-independent. Design density is the first sample journey.

Backlog

• [Architecture] Introduce the Living Theory as the shared product model #1 [Architecture] Introduce the Living Theory as the shared product model
• [Sources] Synthesize approved sources into the Living Theory #2 [Sources] Synthesize approved sources into the Living Theory
• [Learning] Generate source-grounded explainer artifacts #3 [Learning] Generate source-grounded explainer artifacts
• [Learning] Add recall, prediction, teach-back, and transfer checks #4 [Learning] Add recall, prediction, teach-back, and transfer checks
• [Learning] Define and render interactive micro-world artifacts #5 [Learning] Define and render interactive micro-world artifacts
• [Memory] Replace mastery scores with human, agent, and shared theory projections #6 [Memory] Replace mastery scores with human, agent, and shared theory projections
• [Architecture] Detect evidence-backed understanding gaps and cognitive-debt signals #7 [Architecture] Detect evidence-backed understanding gaps and cognitive-debt signals
• [UX] Make Understanding and the Living Theory the central workspace #8 [UX] Make Understanding and the Living Theory the central workspace
• [Foundry] Gate capability activation with evaluation and understanding evidence #9 [Foundry] Gate capability activation with evaluation and understanding evidence
• [Architecture] Consolidate practical feedback into traceable theory and capability revisions #10 [Architecture] Consolidate practical feedback into traceable theory and capability revisions
• [Foundry] Add an optional live Codex execution adapter #11 [Foundry] Add an optional live Codex execution adapter
• [Demo] Ship the design-density cognitive-continuity journey #12 [Demo] Ship the design-density cognitive-continuity journey

Delivery sequence

Phase 1: shared model

Start with #1, then #2 and #6. This establishes the stable theory and projection contracts before UI work depends on them.

Phase 2: understanding loop

Build #3 and #4, followed by the constrained micro-world in #5.

Phase 3: operational product

Use those contracts to implement #8, #9, and #10.

Phase 4: integrated demonstration

Complete #12 as the acceptance test for the full prepared journey.

Phase 5: extensions

Add bounded cognitive-debt signals in #7 and the optional live adapter in #11 once the deterministic path is stable.

Epic acceptance criteria

• The prepared journey runs from source ingestion through theory, human understanding, agent capability, practical application, and revision.
• Every artifact and decision exposes provenance.
• Human and agent learning states remain visibly distinct.
• At least one prediction and one transfer task produce evidence.
• At least one interactive micro-world reveals a causal relationship.
• At least one practical correction revises both the learning path and capability proposal.
• No capability activates without explicit approval.
• The complete demo works offline and can be reset.