Maple AI OS Course

Model routing, evals, and promotion gates

How Maple decides when a Hermon adapter is ready to serve a real domain.

Domain hubEvaluation snapshot

Learning Objectives

What this lesson should make precise

01

Name the core abstraction and its failure modes.

02

Translate the concept into a Maple/Hermon proposal contract.

03

Define at least one evaluation case that can fail the model safely.

Tutorial Flow

How this lesson becomes a demo and training target

Each tutorial is written as a user education path and a model-improvement artifact. The diagram shows how the idea moves into a lab, a typed contract, an eval gate, and then a Hermon/MapleAI demo route.

01

Concept

Routing is an OS decision

02

Applied Lab

Applied lab: Model routing, evals, and promotion gates

03

Output Contract

eval_score, required_keys, promotion_gate, health_check

04

Eval Gate

Does the answer separate model proposal from deterministic execution?

05

Demo Route

Maple AI OS training and public model checks

01

Routing is an OS decision

A user asks for work, but the OS chooses a model route based on domain, latency, privacy, cost, safety, and eval fitness. The model should not choose its own authority path.

  • Route by domain.
  • Route by capability.
  • Route by eval status.

02

Eval-directed training

The training loop should read failed evals, generate targeted examples, train the adapter, probe for regressions, and promote only if the relevant contract passes.

  • Do not train blindly.
  • Use failure fields as data.
  • Probe unsafe requests.

03

Production promotion

Promotion should update the live symlink, restart or hot-load the adapter, run health checks, run demo probes, and publish the new snapshot. If any gate fails, rollback is the default.

  • Promote atomically.
  • Health check before publish.
  • Rollback on gate failure.

Lab

Applied lab: Model routing, evals, and promotion gates

Define a promotion policy for Hermon OS, including required eval score, missing-key tolerance, health checks, and rollback trigger.

Expected result

  • A typed JSON-style proposal rather than free-form advice.
  • Clear authority boundaries and denied operations.
  • A test or rubric that decides whether the proposal is deployable.

Evaluation

How Maple would grade this work

Rubric

  • Does the answer expose assumptions instead of hiding them?
  • Does the answer separate model proposal from deterministic execution?
  • Does the answer produce artifacts that can be tested, reviewed, and rolled back?

Output contract

eval_score, required_keys, promotion_gate, health_check, rollback_trigger, published_snapshot

Use this lesson as training direction

A strong lesson gives users a mental model and gives Hermon a sharper target for examples, probes, and demo prompts.

View evalsDomain hub