Open components. Explicit contracts. Operational freedom.
THE OPENNESS TEST / SCORECARD
Score your stack.
Six tests decide whether an agentic architecture is open in practice, whatever its licenses say. Give each one 0 (absent), 1 (partial), or 2 (demonstrated), and write down the evidence. The total is out of 12. Score the system you run, not the one on the roadmap.
Score every test to see the result. Partial answers are saved in this browser only.
Scoring runs in your browser; nothing is sent to a server. Share links carry the six scores only, not the stack name or evidence.
How to read the score
0 to 4: Closed in practice. Most exits would mean a rebuild. Start with the lowest-scoring test that blocks the next change you already know is coming.
5 to 8: Partly open. Some seams are real and some are assumed. Fix the weakest test first; a single 0 usually costs more than several 1s.
9 to 11: Open with gaps. The architecture holds up. Turn each remaining 1 into a 2 by exercising it: swap, export, or replay something for real.
12: Open by evidence. Every test is demonstrated. Keep the evidence current and repeat the score when a major component changes.
Using it well
Evidence beats opinion. A 2 means someone exercised the property: swapped the component, exported and reloaded the profile, replayed the decision. Claimed support is a 1.
Score the relationships. A stack of open-source parts can still score low if every part assumes the others.
Repeat it. Score again when a major component changes, and keep the JSON next to the architecture decision it informed.