The working paper above deliberately leaves the audit function abstract — it defines what a decision record must contain, and what a verdict must accomplish, without committing to one specific scoring method. That abstraction is intentional: different domains may need different ways of weighing evidence.
SYCDS-STD-001 is the current working implementation of that abstract framework — it defines a concrete evaluation model (the TAC structure: Alignment, Tension, Convergence) and a specific verdict classification used across sycds tools today. As the standard matures, its scoring approach may evolve without changing the underlying Decision Audit definition above.
Before there was an API, before there was a company, and before "AI agent" meant anything to me, there was a fairly specific fascination: why does naming the parts of a geometric figure make it easier to reason about?
Years ago, working through problems involving intersecting circles, I noticed that a diagram only becomes usable for deduction once its parts acquire names with semantic weight — not arbitrary labels, but roles. A point isn't just "P"; it's the upper intersection. A line isn't just a segment; it's a radius, a tangent, a bisector. Once a structure has been named this way, reasoning about it stops being guesswork and starts being something closer to syntax: you can manipulate the relationships because you know what each part is for.
I didn't think of this as a method at the time — it was just how I liked to work through problems. But the habit persisted, and over the years it kept reappearing in different forms, applied to different kinds of structure.
Decision Audit is the most recent expression of that same instinct, applied to a very different object: a decision. Instead of naming the parts of a diagram, it names the parts of a commitment someone (or some AI system) is about to make — the decision itself, the evidence behind it, the assumptions it rests on, the alternatives it passed over, its expected impact, how reversible it is, and what remains uncertain. Once those parts have names, the decision stops being something you either trust or don't — it becomes something you can actually look at.
It's the same move I made with circles and tangent points, years before I had any reason to think it would end up here.
— Sean Yang, Founder