Knowledge Tree
and Butterfly Analysis that shows its working
Explainers

The knowledge tree

The tree is not the fruit.

Every analysis you have ever delivered was a piece of fruit: a paper, a report, a dashboard, a model. What grew it — the data, the code, the decisions, the environment — was the tree. Most of the time the tree is felled at harvest. The postdoc leaves, the scripts rot, the data goes unarchived, and the next piece of fruit is grown from seed.

A knowledge tree is that same structure, kept alive on purpose. It is what a do-bot works in, and it is the thing I help people build.

The knowledge tree, labelled. A stylised apple tree with its roots visible below the soil line. Labels: the roots are the data and sources the analysis draws on; the trunk is the code that turns data into results; the branches and leaves are the parts, each with one job; the fruit are the outputs such as a paper, a report or a dashboard; a butterfly on one apple is your curiosity, choosing where the tree fruits next; the soil is the world the analysis is about. butterfly: your curiosity, choosing where it fruits next fruit: the outputs — a paper, a report, a dashboard branches and leaves: the parts, each with one job trunk: the code that turns data into results roots: the data and sources it draws on, live soil: the world the analysis is about
The parts of a knowledge tree. The roots are the live connection to the data. The trunk and branches are the code, split into parts that each do one job. The fruit are the outputs. The butterfly is you: curiosity decides where the tree fruits next. Curiosity is what allows you to cross-pollinate from tree to tree.

Curiosity used to be punished. Your job wasn't to be curious. It was to implement: to grind (process) data until it became information, and hope that eventually this information would become knowledge and insight.

That's changed.

What happened?

A tree grows. Version control is the record of how: every change to the code, the data and the words, with who made it, when, and why. The do-bot reads that history the way you read a lab notebook, and because every change can be undone it can afford to be bold.

What happened: the same tree at four points in its history. Four drawings of the same tree along a timeline, growing from a sapling to a full tree with fruit. The timeline is marked first draft, split into parts, new data in, paper out. A dashed side branch leaves the line after the second mark and rejoins before the fourth, labelled: tried a different model; kept what worked. Footer: every change recorded, what, when, by whom, and why. tried a different model on a side branch; kept what worked first draft split into parts new data in paper out every change recorded: what, when, by whom, and why
The same tree at four points in its history. The dotted line is an experiment that was tried on a side branch and folded back in. Nothing here was rebuilt; everything was recorded.

What is still true?

The world an analysis is about keeps moving: a new data release, a changed price, a revised guideline. Think of that as the soil changing colour. Fruit grown from the old soil is stale, and the question is whether anyone can tell.

A manifest is the tag on each piece of fruit that says what it was grown from. Compare the tag with the soil and staleness is decidable, at a glance, by a person or a do-bot. And because the tree is alive, with roots still in the ground, the fix is not to replant. It is to regrow the fruit.

What is still true: the soil changes colour and the tree responds. Three frames of one tree over soil. Frame one: the soil's nutrients are red and the apple is red; a tag beside the apple reads grown red, soil red. Frame two: the soil has turned violet; the apple is still red and is ringed as stale; the tag reads grown red, soil violet, marked not equal; a root tip is highlighted with the words roots notice. Frame three: the apple has regrown violet, the tag matches again, with the words fruit regrown, no replanting. Footer: the tag on each apple is its manifest, what it was grown from and what the soil is now. grownsoil 1 the world as it was grownsoil≠ 2 the world changes roots notice grownsoil 3 the tree catches up fruit regrown, no replanting the tag on each apple is its manifest: what it was grown from, and what the soil is now
One tree, three moments. The soil changes; the tag shows the mismatch; the roots notice; the fruit regrows in the new colour. A felled tree can only chase the soil by starting again.

Many fruit

Once the tree is the thing you tend, outputs stop being the work. A well-kept tree carries many pieces of fruit at once, keeps forming new ones, and can regrow any of them when the soil changes. The effort moves from fabricating each result by hand to building and maintaining the thing that grows them.

The tree is not the fruit. Left: a person at a bench with four finished apples and one half-made, labelled fruit made by hand: each one starts from nothing; none can be regrown, and none knows what it was made from. Right: a rooted tree carrying eight apples with three more forming and a butterfly on one, labelled fruit grown on a tree: one tree, many outputs, more forming; each one regrowable from the same roots. Footer: build and tend the tree; the fruit follows. fruit made by hand each one starts from nothing; none can be regrown, and none knows what it was made from fruit grown on a tree one tree, many outputs, more forming; each one regrowable from the same roots build and tend the tree; the fruit follows
Fruit made by hand starts from nothing every time and cannot say what it was made from. Fruit grown on a tree shares roots, and any of it can be grown again.

If you work with data and evidence, you have been drawing trees for years: decision trees, evidence networks, flow diagrams. Point that skill at your own codebase, and the do-bot has somewhere to work.

If you build software, this is your daily practice with the names changed: version control, modules, build graphs and lockfiles. The analytical fields are about to need it, and they don't yet know what to call it.

Back to the two doors →