Availability: Arranged through sales
The honest page. A backlog engagement goes well when both sides know what automation carries and what it does not, before processing starts.
What automation does not carry
The description-level limits are the same ones documented for the API: referential calls that depend on geometry outside the document, lot and block descriptions without their recorded plats, internal angles, spiral curves, and width transitions referenced to stations. In a backlog these are not failures, they are scoping facts: discovery and the pilot surface the share of your documents each affects, and plat-dependent records can be scoped into the engagement where plat integration is warranted.
What review is for
Automated placement quality varies with document quality, description complexity, and the reference parcel data available. That variance is designed for rather than denied: every result carries a confidence score, and human review runs in confidence order, riskiest plots first. Review time never reaches zero on a real backlog, and a proposal that assumed otherwise would not survive your own audit.
One measured result, from one engagement: in the strongest bulk engagement to date, 90 percent of plots located at a confidence score of 95 percent or higher. That is a single engagement’s outcome, not a specification, and a confidence score is the model’s own estimate rather than a verification of correctness.
What to expect of the process
Scope and rate are set together on a call, a pilot on your own documents is available where the use case needs proving, and the report per batch accounts for every document either way. Where results disappoint on a document class, that class gets named, not buried.