Problem
Two related issues with the pattern system:
1. Stale patterns persist at full strength
Patterns like "Redundant Memory Logging" (about PR #132, months old) and "Full-Stack Scientific AI Infrastructure Development" (molecular generation platform — not even this project) surface at strength 1.0 in every recall. They add noise to every response and there's no mechanism to decay, archive, or dismiss them.
2. Cross-project bleed
Recall scoped to project: mnemonic still surfaces patterns from Petra-PB, CannaSage, felixlm, and other projects. Pattern matching doesn't respect the project filter that memories correctly honor.
Expected
- Patterns should decay over time if their constituent memories are archived/fading
- Patterns should respect project scoping on recall
- There should be a way to archive/dismiss a pattern (MCP tool or API)
Impact
Every single recall response includes 5 patterns, and typically 2-3 are irrelevant noise from other projects or stale observations. This is the highest-noise component of recall output.
Problem
Two related issues with the pattern system:
1. Stale patterns persist at full strength
Patterns like "Redundant Memory Logging" (about PR #132, months old) and "Full-Stack Scientific AI Infrastructure Development" (molecular generation platform — not even this project) surface at strength 1.0 in every recall. They add noise to every response and there's no mechanism to decay, archive, or dismiss them.
2. Cross-project bleed
Recall scoped to
project: mnemonicstill surfaces patterns from Petra-PB, CannaSage, felixlm, and other projects. Pattern matching doesn't respect the project filter that memories correctly honor.Expected
Impact
Every single recall response includes 5 patterns, and typically 2-3 are irrelevant noise from other projects or stale observations. This is the highest-noise component of recall output.