Canonical flow

Document routing

Non-generative classification pipeline for routing documents.

Abstract pattern

Non-generative classification pipeline

The substrate-agnostic proof that PAA can govern a small classifier and still rely on live gating, monitoring, and promotion evidence.

  1. Assisted Human-led with automation support.
  2. HITL Human approves each action.
  3. HOTL Human samples or monitors.
  4. Autonomous Automation acts within guardrails.

Task contract

Document routing

The substrate-agnostic proof that PAA can govern a small classifier and still rely on live gating, monitoring, and promotion evidence.

Boundary
Input is the document plus metadata. Output is a label and a routing destination.
Evidence log
Document features, predicted label, evaluator verdict, downstream correction when a human reroutes the item, and the handling outcome.
Evaluator
Target
Input and output
Technique
Schema validation, out-of-distribution detection, and low-confidence escalation in flight, with downstream reroute monitoring behind the scenes.
Oracle
Invariants for the inline checks and human reroutes as the live signal.
Position policy
autonomous
monitoring โ€” in-distribution high-confidence decisions route directly; live reroute rate and drift monitors run continuously
hitl
blocking โ€” out-of-distribution and low-confidence documents are held for human triage
Promotion rule
Held-out precision and recall meet the bar and live reroute rate stays low, so the pipeline can route directly.
Demotion
Rule
Reroute rate climbs above threshold or a drift monitor fires, expanding the escalation scope and returning uncertain documents to human handling.
Fallback
Out-of-distribution and low-confidence documents route to human handling as a permanent design constraint.

Teaching point

What this flow proves

The live evaluator and the promotion evidence are different things, and both are required if the automation is going to stay safe.

Teaching context

Why this flow is governed this way

Worker context
A routing classifier with no generative model in the loop.
Without PAA
Routing errors surface only as downstream complaints; there is no live signal for drift or confidence degradation; a bad batch ships in full before anyone knows.
Never full-auto
Out-of-distribution documents โ€” the classifier declares uncertainty on these and routes them to human handling as a permanent design constraint, not a temporary state.

This page is linked from the canonical card set on the flows index.