Back to work

Earlier questions.
Artifacts worth keeping.

This archive preserves earlier explorations without presenting them as current commitments.

Personal Agent Runtime

A model can propose an action without being the authority that permits it. This earlier engineering exploration puts a registered capability and policy boundary between a proposed plan and its execution, then checks the outcome.

The architecture separates planning, permission, execution, verification, and bounded recovery. Its offline acceptance path uses a fake client; this is software evidence rather than a demonstrated live-network run.

  1. Model proposes a plan
  2. Runtime checks capability and policy
  3. Execute within the boundary
  4. Verify, then finish or recover

Phase Reversal Hypothesis

An earlier macro research note on how AI might change which constraints bind capital allocation. These documents preserve the original hypothesis; they are not evidence of a validated economic model.

RIZHICHI

The existing secondary page remains available. Current project states and evidence are maintained on the individual case pages.

Open RIZHICHI