Revocable Memory Graph
Can a graph-based retriever enforce memory revocation structurally while still preserving useful personalized context?
Hypothesis
Combining MeClear's game-theoretic attribution with a revocation-aware graph retriever recovers more revoked facts than either paper's approach alone, without excessive collateral forgetting.
Background
MeClear (arXiv 2609.09115) reports 85.9% target recall clearing harmful/redundant memory; Revoked but Still Authoritative (arXiv 2609.08258) found no tested system enforces revocation by default across 5 systems and 9 policy scenarios — a direct adversarial complement.
Method
Combine soft-deletion, temporal-validity intervals, and authorization filters with MeClear-style attribution; test obsolete-policy retrieval and action recovery against both papers' test suites.
Measurement Plan
Revoked-retrieval/action recovery rate and collateral forgetting of legitimate memory.
Expected Artifact
A memory-governance test suite usable as a shared benchmark for both source papers.
Limitations
Game-theoretic estimation in MeClear adds cost; neither paper proves durable privacy guarantees under adaptive attacks.
Status
This experiment is Proposed — queued, not started. Results, data, and measurements will appear here once it actually runs.
