Comment by q00

"Garbage In, Garbage Out" – but what if users don't know their requirements are garbage?

The real problem isn't AI execution. It's human unknown unknowns.

Users can't specify what they don't know they don't know. So AI agents receive incomplete requirements and execute them faithfully – producing exactly what was asked, not what was needed.

Ouroboros solves this with Socratic questioning + ontology modeling before any code runs.

The Big Bang phase:

1. Socratic interview – surfaces your unknown unknowns

2. Extracts explicit ontology:

- Entities (what exists?)

- Relations (how connected?)

- Constraints (what rules?)

- Invariants (what never changes?)

3. Double Diamond: Diverge (explore) → Converge (crystallize)

4. Outputs immutable "seed" – the agent's north star

Example:

- User: "Build me a movie tracker"

- Agent: "Track watched movies, wishlist, or both? Web or mobile? TMDB API or manual? Core value – stats, social, or logging?"

→ User realizes they hadn't thought about half of this

→ Ontology generated, seed locked, autonomous execution begins

Result: No more "that's not what I meant" loops.

pip install ouroboros-ai github.com/Q00/ouroboros

Is "clarity-first" worth the upfront friction? Curious what HN thinks.