Memory importance scoring and de-duplication #3

New issue

Closed

opened 2026-01-21 20:45:24 +00:00 by jack · 1 comment

jack commented

2026-01-21 20:45:24 +00:00

Owner

Problem

Over time, similar/repeated observations are stored with the same weight → noise, reduced relevance.

Solution

1. Importance Scoring

Weighting: decisions > writes > reads > informational

function scoreObservation(obs: Observation): number {
  if (obs.type === 'decision') return 1.0;
  if (obs.type === 'bugfix') return 0.9;
  if (obs.type === 'feature') return 0.8;
  if (obs.type === 'refactor') return 0.7;
  if (obs.type === 'discovery') return 0.5;
  return 0.3; // informational
}

2. De-duplication

Hash-based for exact duplicates
Similarity-based for near-duplicates
Merge or skip duplicates

Migrated-From: jack/claude-mem-fork#80

## Problem Over time, similar/repeated observations are stored with the same weight → noise, reduced relevance. ## Solution ### 1. Importance Scoring Weighting: decisions > writes > reads > informational ```typescript function scoreObservation(obs: Observation): number { if (obs.type === 'decision') return 1.0; if (obs.type === 'bugfix') return 0.9; if (obs.type === 'feature') return 0.8; if (obs.type === 'refactor') return 0.7; if (obs.type === 'discovery') return 0.5; return 0.3; // informational } ``` ### 2. De-duplication - Hash-based for exact duplicates - Similarity-based for near-duplicates - Merge or skip duplicates -------- Migrated-From: jack/claude-mem-fork#80