Task Classes Under Assessment
Every order submitted to The Agentic Foresight Registry is classified into exactly one of seven task classes before assessment begins. The classification is a select field on the order form. It is not inferred. It is not guessed from the wording of the task, and it is not revised afterwards by anything reading that wording — classification is deterministic input, supplied by the buyer, and the buyer's choice is the one the method uses.
Each of the seven classes carries its own fixed set of 4 tracked risk factors, with fixed severities, documented in full on that class's own page. 28 risk factors in total. 4 per class. None shared between classes. None inferred at runtime.
We call the line between the seven the classification boundary, and it is drawn once, by the buyer, on the order form. Nothing downstream redraws it — not the method, not the certificate, and not any reading of the task description, which is hashed and discarded before the computation begins.
These pages publish risk profile and mechanics. They do not publish a verdict, a recommended execution window, or the hourly figures a certificate carries — that material exists only against a specific order, and delving through the directory does not reconstruct it.
The seven task classes
The seven classes are exhaustive by construction: six describe a recognisable operational shape, and the last exists so that a task which fits none of the others is assessed as what it is rather than forced into a category it does not belong to.
| Identifier | Name | Covers | Risk-factor page |
|---|---|---|---|
retrieval | Information retrieval | Search, information gathering, research, summarisation | View task class → |
code | Code modification | Writing, editing, refactoring, and testing code | View task class → |
communication | Communication | Drafting and sending messages, emails, and documents | View task class → |
transaction | Transaction | Purchases, bookings, payments, and orders | View task class → |
analysis | Data processing | Data handling, transformation, analysis, and computation | View task class → |
generation | Media generation | Images, video, audio, speech, slides, and other generated artefacts | View task class → |
other | Unclassified operation | Everything else | View task class → |
Why classification is a select field
Classification is a select field because a select field cannot drift. A free-text task description is hashed and discarded; the class, by contrast, is one of seven fixed strings, and the same string produces the same class every time it is submitted — from any environment, in any month, on any run. Think of it as the one input the buyer controls outright: nothing between the order form and the certificate reinterprets it.
Related surfaces
Task class is one of two assessed axes. The model configuration is the other. /models holds the maintained model list and their computed profiles; /method holds the AFR-1 derivation in full, published openly rather than described vaguely; /pricing.json holds structured pricing for every tier, unauthenticated, machine-readable, and identical to the prices printed on this page.
Get an assessment
In conclusion: seven classes, 28 tracked risk factors, one select field, and a price that is printed rather than quoted.
An assessment costs EUR 1.90 at the Standard tier, EUR 4.90 at the Extended tier, and EUR 14.90 at the Full Chart tier, which enters the permanent chart itself onto the certificate. Assessments from EUR 1.90. The task class is chosen on the order form, in one field, before anything is computed.