gtmjosh

Interactive guide

Show the Reasoning and Let People Overrule It

A decision can be ignored when people cannot inspect it, or over-trusted when its explanation only sounds convincing.

In either case, a person needs enough evidence to understand what the system consulted, what it did not, and how to disagree or correct the result. You will build a Decision Explanation and Review Policy. The lab saves your progress in your browser.

Make a recommendation reviewable with Maya.

Open the source evidence, separate facts from inference, and route uncertainty to a person. Keep the original recommendation visible when a reviewer changes the decision.

Maya’s decision workshop. A planning wall, pinned request cards, an open notebook, and a warm wooden worktable.
Maya, decision designer, welcome pose
Maya · Decision designer

Start with a broken answer

Show the Facts Behind the Result

Chapter 1 of 8 · 14 min

Start guide →

The learning path

An explanation has a job, and it is not to make the answer look reasonable. Most of this guide decides what a person has to be shown. The rest decides what happens once they have looked and disagree.

The explanation

  1. 01Show the Facts Behind the ResultShow what was actually consulted, not a story written after the answer.~14 min · exercise
  2. 02Show Which Rule MatteredShow the decisive rule or jointly required conditions, and distinguish checked from skipped steps.~14 min · exercise
  3. 03Separate Facts From InferenceLabel observed, inferred, human-assessed, and absent evidence with its source and basis.~14 min · exercise
  4. 04Communicate Unknowns and ConfidenceMake "not enough to say" visibly different from "no".~14 min · exercise

What a person does next

  1. 05Route Ambiguous Cases for ReviewSend a human only what a human can act on, and size the queue to the people who have to clear it.~16 min · exercise
  2. 06Support Overrides Without Erasing HistoryKeep the recommendation and the decision, so you can find out which was right.~16 min · exercise
  3. 07Protect Human-Owned DecisionsDefine a human-owned field boundary and show what a separate advisory output may write.~16 min · exercise
  4. 08Capstone: Review and Resolve Three DecisionsReview three prepared examples, validate the latest saved policy, and approve that version.~22 min · exercise