Quality Score

Every memory has a quality score from 0–100. Mindr uses this score to rank memories for context injection, surface the most relevant decisions first, and sort memory list output.

The score is entirely deterministic — no ML, no embeddings, no external calls.

Components

Component	Max points	Condition
Recency	40	Decays linearly over 30 days from creation
Commit association	25	Memory has a git_commit tag
Manual capture	20	Memory was created with role: 'user' (i.e. mindragent remember)
Retrieval frequency	10	metadata.retrieveCount increments each time the memory is served in context
Contradiction penalty	−25	Memory is tagged as a reversed decision

Recency formula:

recencyScore = max(0, 40 × (1 − daysSinceCreation / 30))

A memory created today scores 40. A memory created 15 days ago scores 20. A memory older than 30 days scores 0 from this component.

Total:

total = clamp(recency + commitAssociation + manualCapture + retrievalFrequency + contradiction, 0, 100)

Why this matters

Context injection is token-limited. When Mindr has more information than fits in the budget, it ranks memories by quality score and drops the lowest-scoring ones first. A high-quality memory is one that is:

• Recent (still relevant)
• Associated with a git commit (grounded in real code change)
• Manually captured (the developer considered it worth storing)
• Frequently retrieved (agents found it useful before)

A reversed decision scores negatively to ensure it is excluded before any active memory.

Inspecting a score

mindragent memory inspect <id>

Output includes qualityScore and the full qualityBreakdown:

{
  "id": "mem-abc123",
  "content": "Decision: switch internal APIs to tRPC",
  "qualityScore": 85,
  "qualityBreakdown": {
    "recency": 40,
    "commitAssociation": 25,
    "manualCapture": 0,
    "retrievalFrequency": 10,
    "contradiction": 0,
    "total": 85
  }
}

Sorting by quality

mindragent memory list --sort quality

const memories = await mindr.query({ type: 'decision' });
// Already sorted by quality descending
memories.forEach(m => console.log(m.qualityScore, m.content));