Network/Register/Ai Proposed Mappings Model Choice And Cost Envelope
Gap Register
G-046Public

AI-proposed mappings: model choice and cost envelope

Tier 2 — Structurally thin, not launch-blocking
Status
Open — not started
Owner
Creator
Why now
Affects ingest economics directly; cost per provider onboarding compounds as the provider list grows.

Decide LLM(s) for mapping generation, deployment topology (in-VPC vs. provider API), per-onboarding cost envelope, and replay caching.

Why this matters

Mapping proposals are LLM-generated. Open decisions:

  • Which model(s) — single best-in-class, model committee, or tiered (cheaper for routine, premium for ambiguous)?
  • Deployment — provider API (lower op cost, data leaves VPC) or self-hosted in-VPC (higher op cost, data stays put, license-friendlier for sensitive providers)?
  • Cost envelope per provider onboarding — bounded so ingest economics stay predictable as the provider list grows.
  • Caching — mapping generation should be cached and replayable so a re-run doesn't re-pay for the same inference.