Workshop
Tika Experiments
Working notes from the bench. AI workflows, orchestration experiments, prototypes, and the failures that taught more than the wins. Posted as they stabilise — never polished into case studies.
- Experiment · 01
Butler-orchestrated retro automation
Driving a cross-Cell retro through Butler. Found three Capability candidates; promoted one to L2.
- Experiment · 02
Token Guard with adversarial prompts
Stress-testing the budget perimeter. The interesting failures were not the obvious ones.
- Experiment · 03
Inspector traces over a week of Cells
What 14,000 traces look like when you tag them by Agent Contract surface.
- Experiment · 04
Promotion proposal in plain prose
Asking an inspector to write its own L1 → L2 case. Useful. Not yet trusted.
- Failure · 05
Why our first Kernel grew
We let convenience into the Kernel. It stopped being defensible in eleven weeks.