The adaptive "what-next" seam (plan §9.1 — "internal LLM inference is used
only for choices"). Given the results so far, decide which experiments to run
in the next wave. The default implementation is fully deterministic
(PlanOrderWaveStrategist) — it just iterates the plan's
ProposedExperiments in priority order, in waves. An optional LLM-backed
strategist can be injected later to swap in adaptive "given these results,
what to try next" behavior without touching the deterministic loop.
The adaptive "what-next" seam (plan §9.1 — "internal LLM inference is used only for choices"). Given the results so far, decide which experiments to run in the next wave. The default implementation is fully deterministic (PlanOrderWaveStrategist) — it just iterates the plan's
ProposedExperimentsin priority order, in waves. An optional LLM-backed strategist can be injected later to swap in adaptive "given these results, what to try next" behavior without touching the deterministic loop.