Summary
In a self-hosted Honcho deployment, we observed that:

• session writes succeed
• direct session context retrieval succeeds
• dialectic queries can correctly answer about newly written information

…but:

• peer card retrieval may remain empty
• semantic-search-style context retrieval may be sparse or empty immediately after writes

Observed behavior
Using Hermes’ Honcho integration:

• messages are written successfully
• dialectic_query() can correctly recall newly written phrases
• get_peer_card() often returns []
• search_context() is much more useful after explicit conclusions exist than after fresh raw messages alone

Why this is confusing
From an integrator perspective, it can look like:

• “memory works” via dialectic
  but
• “search/profile doesn’t work”

Potential interpretations
We’re not yet sure which is intended:

1. peer-card generation is expected to lag and is working as designed
2. search/profile retrieval over raw recent messages is weaker than expected
3. integration assumptions are wrong about when peer_card / peer_representation should be available

What we tested

• direct writes to session
• direct session context reads
• dialectic Q&A over fresh writes
• search/card retrieval over the same session
• explicit conclusion writes

Interesting result
After writing an explicit conclusion, representation-based search became more useful.
That suggests the issue may be specifically about:

• raw-message-to-representation timing
• or peer-card generation thresholds/selectivity

Question
What should an integrator expect here?

• Is an empty peer_card immediately after writes expected?
• Is search_context over peer representation intended to exclude recent raw-message content unless it has been derived into observations/card state?

Why this matters
Integrators may want to know whether they should:

• rely on dialectic for continuity
• wait before using search/profile
• or implement a fallback to raw session excerpts when representation/card are empty