ADE

ADE's scoring flow is option_scoring → softmax → argmax → margin, where margin is the difference between the top-2 scores and represents confidence.

ADE (Agent Development Environment) is a runtime environment designed around agents writing code, emphasizing pane management, state visibility, inter-agent messaging and lifecycle control, and is distinct from traditional IDEs designed around humans writing code.

In the ADE scenario, the value of worktrees is that they let multiple agents modify different copies of the same project in parallel, avoiding overwriting each other's files or causing lock conflicts; without worktrees, multi-agent parallelism degrades into fighting over files in the same directory.

ADE's option scoring stage uses Tetris heuristic scoring, with features including max_height, holes, aggregate_height and wellness, combined with the number of lines cleared.

ADE's softmax normalization uses a temperature parameter T=0.5, controlling the balance between exploration and exploitation.

ADE sets margin < 0.3 as the uncertainty marking threshold, and margin < 0.1 triggers VPU parallel exploration.

In ADE's self-test, because the O piece has only one legal placement, the softmax degenerates into a degenerate distribution, and the confidence and margin are naturally 1.0 rather than overconfidence.

When the full ADE engine is unavailable (wrong path or missing dependencies), the agent degrades to pure heuristic mode; the formula chain and decision structure are the same, but advanced scoring is missing, and the API interface is completely identical.

The VPU (Vector Processing Unit) is used in ADE for high-dimensional parallel exploration when margin < 0.1, returning a convergence metric based on Shannon entropy; the lower the entropy, the more it converges on a certain strategy.

ADE's formula-traceable mode can be transferred to fields such as autonomous driving, financial trading and medical diagnosis, sharing the decision framework of 'scoring → normalization → selection → confidence → audit'.

ADE's rule evaluation stage includes margin gating, safety and stability rules, used respectively to flag uncertain decisions, reject placements that create holes or exceed the height limit, and prefer placements that form flat surfaces.

ADE's self-test benchmark shows: average latency 0.18 ms/decision, latency range 0.07–0.29 ms, average confidence 0.533, average margin 0.275, and 1–2 triggered rules per decision.

ADE (Aether Decision Engine) is an explainable decision framework designed for Tetris AI, whose decision pipeline consists of five stages — option_scoring → softmax → argmax → margin → rules — leaving an auditable record at each step.

Each ADE decision generates an ADEDecision data class containing the piece type, rotation, column, confidence, margin, softmax probabilities, formula chain, triggered rules, audit trail, decision ID and latency.

ADE (Aether Decision Engine) is a Tetris decision system integrated with VPU, and every decision is auditable.

Orca is an open-source project from stablyai that positions itself as an ADE (Agent Development Environment) for managing and coordinating fleets of agents running in parallel.