Skip to content

RETRIEVAL PRIORITY

Retrieval Priority Logic

You already feel this, but here it is clean so you can wire it into SDKs.

Priority Stack

  1. ECHO Layer – “recent memory / cache”
  2. Fast, cheap, high-hit-rate for recent topics

  3. Good for: “What did I decide yesterday?” / “What did Mobius already say about X this week?”

  4. Encyclopedia – “long-term canon”

  5. Slower than ECHO but much more stable, GI-filtered

  6. Good for: “What is MIC?” / “What is the HIVE game?” / “What is Thought Broker?”

  7. Thought Broker – “live consensus”

  8. Most expensive, uses all engines
  9. Good for: new questions, changing facts, or when canon doesn’t exist / is out of date.

Simple Decision Tree

  1. Derive or detect topicId (from question, or explicit param).
  2. Ask ECHO:
  3. If hit with giScore ≥ 0.93 and timestamp < freshness window:
    • Use ECHO result.
  4. Ask Encyclopedia:
  5. If canonical entry exists with giScore ≥ 0.95:
    • Use Encyclopedia result.
  6. Else:
  7. Use Thought Broker; if giScore ≥ 0.95 → optionally enqueue for Encyclopedia ingest.

That’s the “cache → canon → compute” pattern.