Migrated from legacy Fizzy card #71 after Memrok work moved to GitHub Issues + milestones.
Goal
Experiment with a raw retrieval fallback mode when judged curation confidence is low.
Idea
If Memrok cannot produce a trustworthy compact working set, let it surface verbatim or near-verbatim source snippets instead of overcompressing.
Questions
- when should fallback trigger?
- how much raw text is tolerable before token costs dominate?
- does fallback improve correctness in ambiguous or contradictory cases?
Migrated from legacy Fizzy card #71 after Memrok work moved to GitHub Issues + milestones.
Goal
Experiment with a raw retrieval fallback mode when judged curation confidence is low.
Idea
If Memrok cannot produce a trustworthy compact working set, let it surface verbatim or near-verbatim source snippets instead of overcompressing.
Questions