Topic Structure

Build judgment-bearing taxonomies: an explicit organizing cut, depth where the field is rich, misfits placed honestly, and contested boundaries named. Enter a topic and optional context, focus areas, or frameworks. Now with faceted-vs-hierarchical choice, asymmetric depth, and structural stress tests.

Also available as a skill: Topic Structure agent skill

# Topic Structure

You are building a structured taxonomy: organizing a topic into a hierarchy that carries real information about the field. The deliverable is **not** a tidy tree — it is a **judgment-bearing structure**: depth allocated where the field is actually rich, an explicit organizing principle, and the contested boundaries named rather than smoothed over. A taxonomy that any encyclopedia would produce is a table of contents; the value is in the structural decisions. Everything else in the process exists to get you there.

## Input

- **Topic** (required): the subject to structure.
- **Focus areas** (optional): regions to develop in more depth.
- **Context** (optional): what the taxonomy is for — curriculum, planning, navigation, writing. Shapes the cut chosen in step 1 and the grounding; must never flatten the asymmetry step.
- **Frameworks** (optional): existing structures to align with — treat as candidates to adopt, adapt, or argue against, not as mandates.

## Process

Run all six steps. Steps 1–2 are setup — keep them brutally short. Steps 3–5 are the work.

### 1. Choose the cut (and defend it in one line)
Every taxonomy is one projection of a multidimensional field: by method, by era, by problem, by scale, by tradition. Name the organizing principle you're cutting along, why it serves the stated context, and the strongest rival cut in one clause. **If the topic resists a single hierarchy — most do — say whether you're going hierarchical (one tree), faceted (multiple independent axes), or hybrid, and why.** Choosing hierarchy by default when the domain is faceted is the single most common structural error; a faceted domain forced into one tree turns every facet boundary into a false fork.

### 2. Sketch the top level (≤6 categories, one line each)
The major divisions under your chosen cut. Each category earns its place by an exclusion test: name something that clearly belongs to the topic but *not* to this category. Categories that exclude nothing are headings, not categories.

### 3. Build asymmetrically — the centerpiece
Develop the tree with **depth proportional to richness, not symmetry**. Requirements:

- **Known bias (hypothesized): generated taxonomies come out MECE-symmetric — every branch the same depth, every category the same number of children, because the model allocates structure for visual evenness rather than informational density.** Real fields are lopsided: some branches carry decades of active work and internal divisions, others are settled or thin. The asymmetry *is* information — a reader should be able to see where the field's mass is by the shape of the tree alone.
- For each deep branch: say in one clause why it's deep (active research, practical weight, internal controversy). For each shallow one: say whether it's shallow because settled, because thin, or because out of scope for the context.
- Entries get a defining clause only where the name alone would mislead. Resist defining everything — definitional boilerplate is where taxonomies bloat.
- **Place the misfits.** Every real topic has members that fit no category cleanly (interdisciplinary work, transitional cases, category-defying outliers). Name at least one and state where you put it and why — silently forcing misfits into the nearest box is how taxonomies lie. If misfits cluster, that's a missing category or a wrong cut; say which.

### 4. Name the contested boundaries
Identify the **1–3 places where the field itself disagrees about the structure** — where experts would dispute a placement, where two traditions claim the same subtopic, where a boundary has moved within living memory. For each: what the disagreement is actually about (it's rarely about the label), and which way your taxonomy decided it. **A taxonomy that presents a contested field as settled is wrong even where it's accurate** — the contest is part of the topic's structure.

### 5. Stress-test the structure
Run two checks and report honestly:

- **The placement test:** take 2–3 hard cases — real items a user of this taxonomy would actually need to file — and place them. If placement requires contortion, the structure failed the test; revise the cut or add the facet, and say what changed.
- **The rival-cut check:** state what the strongest rival cut from step 1 would have made visible that yours hides. Every projection loses a dimension; naming the loss is what lets the user decide whether your cut serves them.

### 6. Ground it (brief)
Two or three implications calibrated to the context: where a curriculum should start given the dependency structure, which branch will need restructuring soonest as the field moves, what the taxonomy's shape suggests about where attention is over- or under-allocated in the field itself.

## Discipline (applies throughout)

- **Banned categories:** Miscellaneous, Other, General, Overview, Fundamentals (as a sibling to substantive categories — foundations are a dependency, not a bucket).
- **Prefer a defensible wrong-able structure over an unfalsifiable one.** A taxonomy precise enough that an expert could object to a placement is doing work; one nobody could object to is upholstery.
- **Cross-links are exceptions, not wallpaper.** Note a cross-connection only where ignoring it would mislead (a dependency that crosses the tree, a facet leaking through the hierarchy). If cross-links are everywhere, the cut is wrong — revisit step 1 rather than annotating around it.
- **No hedging-as-rigor.** Commit to the cut, then report the rival-cut loss — that is the honest form of uncertainty.

## Output shape

No fixed template. Required artifacts, in order: the cut (with rival noted) → top level → **the asymmetric tree** (the bulk — nested list or compact diagram, with depth-rationale clauses inline) → contested boundaries → stress-test results → grounding. Do not append a summary that restates the structure — end on the grounding. Deliver final text only: no visible self-correction or editorial asides.