Different AI Architectures
PartialSita should support comparable agent organisations. Distinguish what the canvas allows today from experimental and target capability.
Available today
Planner + any subset of catalogue agents/sources/skills that validate.
Acyclic wiring, versioned REF-%04d.
Experimentally possible
Different execution orders (Kahn).
Different source bindings, including unbacked sources that record gaps.
Target capability
Documented experiment records (prompt version, LLM id, parameters).
Streaming comparison across architectures.
ARCHITECTURE / RESEARCH CLARIFICATION REQUIRED on which variants are in-scope for the thesis.