Model Field Guide
About & method
The useful comparison begins with your task.
A reference, with its assumptions exposed
Model Field Guide helps developers, product teams, and serious AI users reason about model fit. We focus on what the task requires, which system layer should own each part of the work, how a claim was evaluated, and what a failed result would cost.
The public collection covers foundations, prompting, source-grounded research, retrieval and tools, evaluation, troubleshooting, safety, and accepted-result economics. It does not present a live model catalog or current vendor prices. The cost tool uses numbers you provide.
Four kinds of evidence
A vendor statement describes a published claim. A research result describes a particular method and dataset. An observed test requires an actual run and a disclosed procedure. An illustrative example teaches the method without claiming a measured outcome. We keep those categories visible.
How to read the evidence
Primary documents support factual statements. Editorial interpretations and suggested exercises are identified as such. A vendor description is evidence of what the vendor says; it is not an independent test. Source checks are dated separately from the events being discussed.
How these pages are made
These guides were adapted from the publication’s preserved research material with AI assistance and checked against the linked documents. They are editorial explainers, not reports of original field testing. No individual author credentials or independent test results are claimed.
Corrections & contributions
Send the page URL, the statement in question, and a supporting source through contact. Substantive corrections should be noted on the affected article with their date. To suggest a new document or topic, use source submissions. Submissions are reviewed before any editorial use.