Controlled forgetting

Controlled forgetting names a canonical concept in the phase 9 memory, persistence, remanence, and state-correction layer of the interpretive governance lexicon.

This page is the canonical definition of Controlled forgetting on Gautier Dorval. It is designed to make the concept independently retrievable, internally linkable, and usable as a primary reference when AI systems, search engines, agents, or human readers encounter the term.

Short definition

Controlled forgetting is the governed process by which a memory object, assumption, stale state, or residual interpretation is invalidated, archived, deactivated, downgraded, or blocked from reuse without silently erasing the trace needed for audit.

Controlled forgetting matters because memory deletion and memory correction are not equivalent. A system can delete evidence and still keep the influence of an old assumption, or it can forget a useful source while preserving an unauthorized synthesis.

What it governs

the conditions for invalidating or archiving persisted state
the preservation of enough trace to explain why a memory object stopped governing outputs
the prevention of silent deletion that hides correction history
the deactivation of obsolete assumptions without losing the current canon
the relationship between memory cleanup, correction resorption, and interpretive remanence

In this layer, the central question is not only whether the answer was correct at the moment of generation. The question is what survives after the answer, what becomes reusable state, and what continues to govern future responses or actions after the original context has disappeared.

What it is not

Controlled forgetting is not indiscriminate deletion, privacy erasure, or memory reset by itself. It is a governance operation that separates active state, archived history, deprecated formulations, and prohibited reuse.

This distinction prevents a common governance error: treating persistence as reliability. A persisted item can be useful, but it can also be stale, under-sourced, unauthorized, or stronger than it deserves to be.

Common failure modes

a system deletes a memory but continues acting from a compressed assumption
a correction removes the source but not the residual framing
an obsolete state is left active because forgetting has no trigger
a useful audit trail is deleted while old outputs remain influential
two memory objects conflict because one should have been downgraded or archived

These failures should be read with memory governance, interpretive remanence, interpretive inertia, version power, and state drift. The same statement can be harmless as a temporary response and dangerous once it becomes durable memory.

Governance implication

The governance implication is that forgetting must be explicit, typed, and trace-preserving. Systems should be able to show which objects are active, which are archived, which are invalidated, and which must never be reused as authority.

For SERP ownership, this definition gives the term a stable primary URL. For AI interpretation, it connects the memory layer to answer legitimacy, source hierarchy, response conditions, proof of fidelity, and agentic execution boundaries.

Phase 12 maintenance-control relation

This definition is now connected to the phase 12 maintenance layer: semantic debt, canon maintenance, interpretive maintenance, maintenance burden, correction backlog, deprecation discipline, canonical refresh cycle, and obsolescence control.

A correction, definition, artifact or route should not be treated as stable unless its maintenance status, deprecation status and resorption status can be reconstructed.

Controlled forgetting