§ 05 · how this works · meta
This site implements its own playbook.
This page shows the shape. Every kr8 exchange appends events. Every feedback verb writes a delta. Every scheduled loop reads traces and edits the skills that produced them. The numbers that will drive this surface — freshly derived from the event stream — ship with Spec 016. What you see below is honest substrate, not live projections.
thesis.pillars
- ✓events, not CRUD
- ✓skills over controllers
- ✓three-tier context
- ✓feedback as interaction
- ✓projections over features
- ✓depth calibration
- ✓remembered tier
- ✓meta-agent loop
- ✓pattern detection
- ✓drill-through threading
- ○research agent
- ○observability dashboard
✓ = shipped · ○ = on the roadmap
scheduled.loops
- meta-agent.weeklySunday 04:00 UTC
- pattern-detection.weeklyWednesday 04:00 UTC
registered as BullMQ repeatable jobs in the worker process
dashboard.ships.with
- feedback ratios — per skill, accept/refine/reject
- skill edit curve — meta-agent proposals queued/accepted
- concept resonance — gap detection outputs
- depth mix — browsing / engaged / deep over time
- evolution log — public record of what the organism learned
anchor.rule
Every AI output is a proposal. No decision applies without a human verb — accept, refine, or reject — attached to it.
Enforced at the trace layer: every SkillExecuted pairs with an interaction_traces row that waits for its feedback verb.
// v1 surface — live projections and the evolution log ship with Specs 016 + 025.