How the tool works

A whole session, start to finish.

It takes about five minutes. You describe what you observe, answer 6 to 10 questions and get hypotheses with their grade of support — plus a follow-up plan with a date on it.

The full path, step by step

Each step exists to prevent one specific error — and the error it prevents is written next to it.

01

You describe what you observe

Free text, in your own words, with no category form. The tool treats what you write as data, never as instruction — a request embedded in the text is ignored and handed back for rewording.

Error this step prevents: starting the diagnosis from an interpretation (“the team is resistant”) instead of an observable behaviour.
Observed situation

The tracking spreadsheets keep being sent by email every week, instead of being uploaded to the platform.

label: procurement team, South project

02

Triage hands it back, if it needs to

If the description interprets instead of describing, the tool does not diagnose: it asks a question and shows the difference with examples. Your original text stays there — nothing is discarded.

Error this step prevents: diagnosing, with apparent precision, a sentence that describes no behaviour at all.
interpretation

“my team resists change”

“there is no commitment”

observable behaviour

“the spreadsheets keep going out by email”

“the material only appears once the stage is over”

03

6 to 10 questions that separate hypotheses

Each question is chosen to distinguish two competing explanations — not to confirm the first one. The theory under test never appears in the question: asking “is this territoriality?” contaminates the answer.

Error this step prevents: confirming the hypothesis you already held when you opened the tool.
question 5 · target 10

Suppose there were free time and nobody chasing anyone. Under that condition, would the work be done the new way?

No — they wouldn't know how
They'd know how, but the routine or the group wouldn't allow it
They'd know how and could, but the current way would still be the one chosen
04

Hypotheses with a grade of support

Each hypothesis comes with the signals that support it — and with the signals against it, shown with the same weight. The ruler has three notches; there is no percentage, because there is no measurement behind one.

Error this step prevents: mistaking a plausible reading for a measurement — and acting as if there were certainty.
C9 Structural incentive for the status quo
high

+ Whoever delivers fast gets praised, with no look at how they delivered

+ Publishing generates questions and extra work for whoever published

Doing it through the platform is no slower than through the old channel

05

An intervention with its “when not to use it” in plain sight

Steps, effort, risk, a suggested indicator — and the contraindication, which is usually the most useful part. When the reading points to legitimate withholding or a matter of policy, the recommendation is do not intervene.

Error this step prevents: applying a psychological intervention when the incentive points the other way — the most expensive error in the method.

Make publishing give something back to whoever publishes

medium effort · low risk · incentive redesign

  1. Identify a demand the team repeatedly receives
  2. Make publishing eliminate that demand
  3. Show the effect with the number before and after
When not to use it: as a one-off prize disconnected from the work.
06

Follow-up — and the tool learns

You write down what you expect to observe and set a date. At closing, a negative result lowers the support of the hypothesis that motivated the intervention — and it stops being recommended for that team.

This is where the tool stops being an oracle and becomes a record of what you have learned about your operation.

Aug 2026 · C9 · high

Publishing gives something back to whoever publishes

✕ no change

Sep 2026 · C9 · moderate

Recognise the process publicly

✓ improved

The intervention applied produced no change — evidence against the hypothesis. C9's support fell from high to moderate.

What the tool does — and what it does not

Does
  • Forms hypotheses about situations and mechanisms
  • Shows the signals for and against, with equal weight
  • States the grade of scientific evidence for each intervention
  • Recommends not intervening when that is the case
  • Records outcomes, negative ones included
  • Produces a one-page report to take into a meeting
Does not
  • Does not assess, score or classify people
  • Does not accept an individual's name as a team label
  • Does not produce a percentage of fit or of risk
  • Does not monitor activity, messages or productivity
  • Does not replace the conversation with the team
  • Does not promise to eliminate bias — neither yours nor its own
Where AI comes in

Reads the free-text description and structures it

Chooses the next question within the method

Writes the result and the report in plain language

What it does not do: decide the diagnosis. The rule for support belongs to the method, is auditable, and does not change from session to session.

Questions that always come up

Who can see my sessions?

Only you. No other manager reaches your session by any path — not even by direct link. Support access only exists with a log entry you can see, and that log cannot be erased.

Will HR use this to evaluate managers?

There is no way to. The Culture team's panel only shows aggregates of five sessions or more and has no path back to the individual session — by design, not by an unchecked permission.

Why is there no percentage?

Because there is no measurement behind one. “C9 · 87%” would look more precise than “C9 · high” without being so. The three-notch ruler is honest about what the tool actually knows.

What if the tool gets it wrong?

It does — they are hypotheses. That is why every session ends with an expected result written down and a date. The error is recorded and lowers the hypothesis next time. A tool that is never wrong is one that never tests itself.

Do I need to know behavioural science?

No. The questions are about what you observe at work. The theory only appears in the result, with the mechanism's card in plain language — and always with its “how to tell it apart from”.

How long until the first result?

About five minutes per session. The real value shows up in the second cycle, when history starts to contradict or confirm earlier readings.