Preserves the human's working map across long AI-assisted workflows—direction, corrections, source state, branch state, permissions, unresolved questions and next movement—so work can continue without rebuilding context or transferring decision ownership to the model.
A system can retain facts and still lose the relationships that make them usable: what the human already settled, which branch is live, which correction controls, what source is missing, what authority is held, and what should happen next.
Human Signal & State Preservation carries direction, correction history, source demand and return, branch state, permissions and unresolved questions as governed state rather than generic chat memory. A portable Decision State Passport can project only the state a receiver is permitted to inherit; the receiver does not become owner of the underlying decision authority.
Current-world movement can record what source is missing, what was returned, what route is being considered, and whether an independent challenge supports movement before a bounded next action is carried forward.
Each item below is deliberately narrower than a product-readiness or superiority claim.
Field-level lifecycle and permission state can be projected into a portable handoff while hash-bound receipts preserve what was carried, changed or withheld.
Database work exercises governed state, permissions and row-level policy behavior rather than leaving continuity as prompt-only memory.
Python state and handoff surfaces have been exercised across host/container equivalence, GitHub Actions and bounded Azure/AWS runtime tests with artifact custody preserved.
Paired testing measured state preservation and Answer-Key Burden under a frozen protocol. The preregistered Q1 criterion was not met; later dependency testing also exposed stale next-movement invalidation that remains part of the product evidence.
It does not claim that memory is solved, that the model owns governing state, that repeated human re-explanation has been eliminated, or that controlled studies have demonstrated a generalized efficacy benefit.
The executed product body already includes state-map, Passport, database, receipt, transfer, container and bounded cloud surfaces. Current evaluation work is focused on dependency invalidation, blinded measurement, evaluator calibration and determining which state-preservation mechanisms earn promotion without leaking the answer key.