← All documentation

Assessments

An assessment is a structured walk through the questions of your framework packs. There are two kinds: the program assessment, which asks the organization-level questions about your AI governance program, and a system assessment, started from an AI system's own page, which asks the questions about that system. Each question is answered Yes, No, Partial, or N/A, with optional notes and supporting evidence.

Running an assessment

  1. Create an assessment and name it, and choose the frameworks it reports against (for example NIST AI RMF 1.0, or a custom framework of your own). You can change the frameworks later.
  2. Work through the questions, answering Yes / No / Partial / N/A with optional notes. Answers save as you go — you don't need to finish in one sitting. Marking a question **N/A requires a short justification** (e.g. "we operate no on-prem servers") so an auditor can tell a deliberate scoping decision from a skipped question — the note saves automatically once you enter it. You can also mark a whole section N/A with one shared justification.
  3. Attach evidence to questions that an auditor would expect to see documented.
  4. Optionally mark the assessment Under Review — an interim state that signals it's ready for an internal check while answers stay editable and no findings are generated yet. Then Complete when finished: completing locks answers and turns every gap (a "no" or "partial") into a tracked finding. An assessment with nothing answered cannot be completed. If some questions are still unanswered, completing asks you to confirm how many will be skipped: no finding is raised for an unanswered question, and the report covers only what was answered. You can reopen a completed assessment later.

Assignment

Org admins can assign whole sections or individual questions to a teammate. Assigned users can edit only the questions assigned to them; everyone else sees the assessment read-only. The "Assigned to me" filter narrows the list to your work.

Delegating a section to an outside helper

Some sections are best answered by someone outside your organization, such as the vendor that supplies a system or a consultant. Invite external helper (on the assessment view, admins only) emails them a private link to answer just that section, no account needed. The link covers every question in that section of this assessment (on an AI system's assessment, that includes the system-level questions), expires after 14 days, and can be revoked at any time; their answers are labeled with their name on the question ("answered by … — external helper") and in the audit trail. Guest links come with a plan; they are not in the demo.

Renewal & comparison

Completed assessments can be renewed into a new run that carries prior answers and section assignments forward. On a renewal, use Compare to prior run to see exactly what improved or regressed since last time.

While answering, each question shows a Prior assessment panel when a previous answer exists — the verdict (Yes/No/Partial/N/A), the date, and the prior note — with a Use this answer button that carries it forward in one click. The source is the renewal parent if set, otherwise your most recent completed assessment of the same type. The panel opens automatically on questions you haven't answered yet, so a year-over-year reassessment is mostly review-and-confirm.

AI help on a question (when AI is enabled)

Two Watney helpers appear on each question:

  • Explain this requirement — a plain-language explanation of what the cited control asks for, examples of evidence that satisfy it, and the common pitfalls. It is grounded strictly in the requirement's stored framework text and recommended solution, and is cached per question so your whole team reuses one generation.
  • Draft answer — a non-binding starting point: a suggested answer value, editable draft notes, and the evidence you'd need to support it. It reflects an honest reading of what you've shown (it won't claim compliance you haven't demonstrated). You decide how to use it:
    • Fill the fields drops the draft into the answer box for you to edit before saving — nothing is written until you click an answer.
    • Apply to this answer saves the suggested answer and notes in one step. It asks you to confirm first, is recorded in the activity log, and can be edited or re-evaluated afterward — it's a faster path to the same save, not an automatic one.

What is sent is the requirement text, your own prior notes, and the AI summaries of evidence already attached to that control (so the draft reflects what's actually on file). Your notes are free text you typed, so they may contain personal or confidential information; keep it out of notes where you can. Watney's system card names the model provider and how it handles what is sent.

After an AI evaluation, if the answer already has a finding (findings are generated from "no"/"partial" answers), the result offers a Draft remediation for this finding → link that takes you to the finding, where Watney can draft the remediation tasks to close it.