design: synthesis proposals never applied — dead-end feedback loop #368
Description
Problem
26 synthesis result files exist. 4 have non-zero proposals. Zero have been applied to anything. Proposals sit in .kata/synthesis/ and no process reads them to take action.
What are synthesis proposals?
They are LLM-generated recommendations from the cooldown synthesis step:
- Archive stale/duplicate learnings
- Promote a learning to a higher permanence tier
- Suggest a methodology change
- Propose bets for the next Keiko (NextKeikoProposalGenerator output)
The gap
There is no apply step. After cooldown --yolo runs synthesis and generates proposals, nothing:
- Archives the suggested learnings
- Creates the proposed next-keiko bets in the staging area
- Surfaces the methodology recommendations to the user
Design questions (needs-human)
- Should high-confidence proposals auto-apply during cooldown --yolo? Or require explicit review?
- Should next-keiko bet proposals automatically populate the staged cycle? Or just be surfaced as suggestions?
- What is the 'apply' surface — a kata cooldown apply command? Part of cooldown complete?
- What counts as 'high confidence' for auto-apply?
Impact
The synthesis loop is entirely decorative right now. Cooldown runs, proposals generate, nothing changes.