The Thinking Box

A product-development journey with human judgement at every door.

  • Design prototype
  • Think stage
  • Not user validated

Version 23 is a fictional design demonstration captured on 4 September 2026. It is not a production application, a live AI system or evidence of learning outcomes.

02

Ten stages. One non-linear journey.

The route can move forward, back, pause or stop. A stage opens only when the evidence and a named human decision justify the next move.

You are here: ThinkThe later stages are conceptual, not completed product screens.

03—06

Evidence earns the next move.

The prototype explores one simple discipline: ask, examine, interpret, decide.

01

Choose a proof question.

Start with the question most likely to change the direction of the work—not the easiest box to tick.
Version 23 Think workbench showing a question list, a current task and a selected question preview.
Think workbench · fictional School Club Finder example

02

Inspect the evidence.

Read the sources yourself. The AI can compare and summarise them, but its reading is not your review.
Expanded proof question showing its reading sequence, rationale and fictional evidence sources.
Expanded question · source-reading state

03

Interpret what it means.

Separate what the evidence establishes, what it merely suggests and what is still unknown.
Gate review showing a research summary and Brendan as the human decision owner.
Gate review · human decision owner named explicitly

04

Make the human call.

Move forward, go back, pause or stop—and record why. Progress is an option, not the objective.
Gate decision view with options to move forward, return, pause or end and a required rationale field.
Gate options · no decision submitted in this capture

07—08

Different work. Different authority.

AI does the legwork.

AI specialists can investigate permitted sources, compare readings, expose gaps and draft a possible next step.

  • Aiko guides product framing and explains trade-offs.
  • Marcus examines sources and distinguishes evidence from inference.
  • Both roles are simulated in the current prototype.

People make the call.

The product lead defines the intent, reads the evidence, writes the interpretation and owns every consequential gate decision.

  • AI advice can be challenged or ignored.
  • Saving a draft is not approving a gate.
  • Movement requires a human reason, not a completion score.

09

Prototype, openly.

This page shows a design hypothesis and the reasoning behind it. It does not present the interface as a finished or proven product.

Shown in the prototype

  • A Think-stage question and evidence flow.
  • Separate AI guidance and human interpretation.
  • An explicit, human-owned gate decision.
  • Tutor setup and a read-only student preview.

Still unproven

  • No live AI.
  • No real student data.
  • No user validation.
  • No learning-effect evidence.

The School Club Finder scenario, people, sources and team data are fictional. Accounts, durable persistence, permission enforcement and a production tutor system have not been demonstrated.

10

Interested in how the system thinks?

GitHub access may be required. The live Figma file may have changed since this Version 23 capture.