…s itself)

A backlog modeled as a dependency graph that schedules itself: deterministic and
hand-checkable, no AI. Readiness (deps met) + Phi (value/cost) over the ready set +
the longest-cost critical path + sensitivity for what unblocks the most, via a cmg
ModelGraph (assess -> classify -> prioritize). The assess stage is the seam where a
model later judges state from evidence. TS and Python output byte-identical; self-check
PASS; expected-output.txt included.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…tectors, real model calls) (#8)

The AI version of example 09: each task carries a done_when + simulated detector
evidence; the assess stage is now a real model-call transform (LiteLLM model-string
routing) that judges {status, value, cost}, and the example-09 scheduler consumes it
unchanged. Offline deterministic stub (expected-output.txt) + live via MODEL (Vertex
MaaS / OpenRouter / OpenAI / Gemini); the LLM adapter lives in the example, not core.
Includes a real MiniMax sample run, a Takeaways section, and a pointer from example 09.

Co-authored-by: cartpanda-boateng <boateng@cartpanda.com>
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>