LLI Wiki — My Digital Knowledge Base

Mind maps for technical analysis: exploring a problem with Freeplane

Structure of a good analysis map (context, constraints, options, risks, decision) and turning the map into tasks.

Before writing a single line of script, I put the problem on a mind map. Linear text forces you to choose an order too early; a map lets you throw ideas in loose, group them afterwards, and see at a glance what is missing. I use Freeplane (free, version-friendly XML files), but any mind-mapping tool will do.

Why a map rather than a document

Limit: a map does not replace text for deciding. It is for exploring and structuring; the final decision is written afterwards (see the article on analysis before coding).

The structure of a good analysis map

I always start from the same five-branch skeleton:

flowchart LR
    R["Problem to analyse"] --> C["Context"]
    R --> K["Constraints"]
    R --> O["Options"]
    R --> X["Risks"]
    R --> D["Decision"]
    C --> C1["Current state"]
    C --> C2["Who is affected"]
    K --> K1["Technical"]
    K --> K2["Deadline, rights, cost"]
    O --> O1["Option A"]
    O --> O2["Option B"]
    O --> O3["Do nothing"]
    X --> X1["Per option"]
    D --> D1["Choice and reasons"]
    D --> D2["Tasks"]
Branch What it holds Question it asks
Context Current state, volumes, who uses what What exactly are we talking about?
Constraints Technical (service limits, permissions), organisational (deadline, budget) What is not negotiable?
Options At least two solutions, plus "do nothing" Which paths exist?
Risks Per option: what can break, how to detect it What can go wrong?
Decision Choice, reasons, open points What do we do, and why?

Tip: always include the "do nothing" option. It is the baseline, and sometimes it wins.

Running a session

  1. Empty your head: all facts, doubts and ideas loose under the root node, no sorting.
  2. Group: move each node under one of the five branches. What fits nowhere is often an open question: I mark it with an icon.
  3. Verify facts: every context node must be confirmed (documentation, read-only query, rereading the code). I visually separate what is verified from what is assumed.
  4. Compare options under their branches: benefits, costs, risks.
  5. Decide and write the decision in its branch, with the reasons.

From map to tasks

An analysis that produces no actions is useless. Under "Decision" I create a "Tasks" sub-branch: one node per task, phrased with a verb, with a check ("test passes", "view updated"). Those nodes then become items in a task list or tickets. Freeplane can also export the map as indented text, which gives a first plan to copy.

Map nodes can point to the working folder through a relative link: the map also becomes an index of the code concerned.

And AI?

An assistant can help fill branches ("which risks for this option?") or reread the map to find gaps. But it readily proposes generic risks or features that do not exist: every technical claim must be verified before it stays.

Lessons learned

Mastodon LLIMastodon SocialFosstodon