Research · Diagnose

Issue tree

Breaks one question nobody can answer into sub-questions that evidence can settle. Every other tool in the bank is chosen against this tree, or chosen at random.

The printable canvas and the handoff into Claude Code are part of a paid plan. See what a plan includes. It needs the Claude desktop app on this machine, and your team and the prompt library already installed in that project — we cannot see your disk, so open it there.

Stage
01 Diagnose
Works at
All levels
Maturity
Early startup → Large organisation
Time
Half a day, and the first version will be wrong

What it is

A decomposition of one large question into the sub-questions that together answer it, and those into the analyses that settle each one. The discipline is that branches at any level are mutually exclusive and collectively exhaustive — no overlap, no gap — so answering all of them answers the parent, and answering fewer does not.

The question at the top
One sentence, and it has to be something a person could answer yes, no or a number to. "Growth strategy" is a topic; "where should the next two million of capital go" is a question.
MECE branches
Mutually exclusive, collectively exhaustive. Overlap means two people do the same work; a gap is a hole in the answer that nobody notices until the room does.
The cut
How you split at each level — by customer, by geography, by line of the P&L, by stage of the funnel. The cut you choose decides what the analysis is capable of seeing.
Answerable leaves
The bottom of the tree is where a specific piece of evidence settles a specific question. If a leaf still needs judgement rather than evidence, it is not a leaf yet.
The load-bearing branches
Two or three branches usually decide the answer and the rest confirm it. Finding those early is the difference between a fortnight and a quarter.
The mistake people makeBuilding the tree by category rather than by question. A branch labelled "Customers" is a folder, not a sub-question, and nobody can tell when it is finished.
What it’s forYou have a question big enough that nobody can answer it directly, and work is about to start.
What it’s not forThe question itself is not agreed. Decompose the wrong question and every branch is wasted.

How you run it

  1. Write the question as a questionIf it does not end in a question mark and imply a shape of answer, everything beneath it will be a filing system rather than an argument.
  2. Choose the cut deliberatelySplitting by product, by market or by cost line produces three different trees. Say which you chose and why, because it decides what the work can find.
  3. Test every level for MECETwo branches that overlap duplicate a fortnight of somebody's work. A missing branch is the question you get asked in the meeting and cannot answer.
  4. Push down until evidence settles itA leaf is where a named analysis produces a yes, a no or a number. Anything above that is still a question about questions.
  5. Mark the branches that carry the answerTwo or three usually do. Resource those properly and sample the rest, or you will research everything to the same shallow depth and finish none of it.

The prompt

Two ways to run it

Run this tool in your own Claude

The short prompt starts your partner against the library on your disk. The long one carries everything with it and needs nothing installed.

🔒 Prompt locked

Everything above is free — what this tool is, what it is for, what it is not for, and how to run it as a workshop. Three tool prompts a month are free with an account; beyond three, and for the printable canvas, it is the paid part.

See plans — from £19 a month