Loop design
Studio-level axioms for closing loops — how Round Online decides when AI runs autonomously, when humans stand at gates, and how to tell the difference.
Overview
The loop-design axioms will name how Round Online decides which workflows close (AI runs autonomously) and which keep humans at gates. The system is currently proposed — the framework is being explored in the Journal essay Open Loop, Closed Loop before any principle locks.
This is the difference between an axiom system and an essay: an essay argues; an axiom locks. We don’t lock until we’ve operated by the principle long enough to trust it.
Working hypothesis (paraphrased — not yet formally numbered)
- Every workflow is a loop of four points. Input, AI action, outcome, feedback. If any point is missing, the loop is open.
- Humans belong only where judgment is irreducible. Every human-in-loop position must answer: why must a person be at this gate? “Habit” is not an answer.
- Loop closure needs three things. Data must be queryable. Action must be executable. Outcome must be measurable. Whichever is missing is the next thing to fix.
What this page becomes
Once the framework has been tested against Round Online’s own operations and a few external engagements, the principles lock as L1, L2, … with statements, derivations, and version history. Some will be wrong — those get superseded with a public changelog.
The Journal essay remains as the record of how the principle was first proposed.
Coming soon — derivation continues through 2026.