Five shapes. Five different questions.
Each diagram subtype frames a specific question. Process for sequence; framework for categorisation; hub-spoke for centrality; timeline for periodisation; value-chain for capture. Pick the question first; the shape follows almost automatically.
-
Process 10.8k 34%how does it move forward?
sequence with optional branches. Steps numbered, arrows directional.
-
Framework 7.5k 23.6%how do we think about this?
imported structure with cells. BCG, RACI, MECE, Cynefin, SWOT.
-
Hub-spoke 4.5k 14.2%what connects to the centre?
one node, many peripherals. Ecosystem, root-of-relations.
-
Timeline 3.5k 11%when did it happen?
time as a structural axis. Eras, phases, milestones.
-
Value-chain 1.2k 3.9%where is the capture?
horizontal flow with margins per stage. Bottleneck = value.
Diagrams take the whole canvas. They do not share.
The median diagram is 60% wide × 49% tall — the largest visual atom in the corpus. When a diagram fits next to body text, it gave up half its argument.
- Diagram 60% × 49% argues with structure — connections, hierarchy, sequence. Quantity is incidental.
- Chart 40% × 35% argues with numbers — magnitudes, distributions, change. Structure is incidental.
- Hybrid < 5% rare and dangerous — chart with annotations OR diagram with sized nodes. Pick a side and commit.
Eight frameworks. One in twenty cites the source.
The corpus's most-borrowed structures arrive without attribution. BCG's matrix appears in 4% of slides that use it. MECE in 2%. The pattern: the more famous the framework, the less the corpus credits it. Assumed authority replaces stated authority — and the reader who does not recognise the structure is left to take it on faith.
- Framework Shape Author Year Cited
- BCG Growth-Share Matrix
2×2Bruce Henderson · BCG 19684% - Porter's Five Forces
pentagonMichael Porter · HBS 19796% - SWOT
2×2Albert Humphrey · Stanford 19654% - RACI matrix
gridProject mgmt · trad. 19703% - MECE
treeBarbara Minto · McKinsey 19732% - Eisenhower Matrix
2×2Eisenhower / Covey 19895% - Cynefin
5-domainDave Snowden · IBM 19993% - Lean Canvas
9-blockAsh Maurya 200910%
Diagrams take the whole canvas.
Across 32k diagrams and frameworks with measured bounding boxes, the structure is the slide. Median width is 60% — wider than any other family. Diagrams do not share.
How to read this
- Heatmap
- Surface coverage. Each bbox in the corpus contributes to every 40×22 cell it overlaps; opacity is power-scaled (γ=0.6) so subtle bands stay visible. Darker = more atoms cover that point.
- Sample box
- The modal cluster. Atoms grouped by centre into 8×8 bins; the densest bin is plotted at its centroid, sized by the average width × height of atoms in it. Reads as: "the typical atom of this family looks roughly this big, here."
- Source
corpus.components.bbox— atoms with measured bounding boxes only. Inline mentions extracted from body text are excluded from this analysis.
Diagrams live where the deck proposes.
Diagram presence concentrates in two slide-types: 'solution' (38%) and 'propose solution' (33%) — both the moments where the deck commits to an answer. Everywhere else, diagrams are rare. The structure is asked to do work where the answer is structural — not where the question is quantitative.
Diagrams travel alone — and unsourced.
Diagram-bearing slides ship a single diagram 95% of the time (the structure dominates by design). They orbit lists (43%) and paragraphs (32%) — labels and explanations. But only 29% cite a source. A frame imported from outside without attribution is just a doodle with confidence.
- 01 Title 97.7%always titled — the diagram needs naming
- 02 List 42.6%labels around the diagram nodes
- 03 Paragraph 32.3%explanation of what the structure means
- 04 Source-note 29.2%— only 1 in 3. Imported frameworks (RACI, BCG, MECE) cited rarely.
- 05 Image 27.4%diagram + image — usually the customer journey + persona photo
- 06 Callout 14%diagram + callout = "look at THIS node"
- 07 Metric 9.7%diagram + stat = "and the number is..."
- 08 Chart 9.7%diagram + chart = boundary case (often a hybrid in disguise)
Five diagram failures. All structural.
A bad chart shows the wrong number; a bad diagram shows the wrong structure. The cost is the same — a defensible argument made indefensible by its frame.
-
⊠ The non-MECE framework
Categories that overlap or leave gaps. Looks like a framework, fails as one.
common- Bad
- "Customers · Partners · Stakeholders" — partners ARE customers, stakeholders include both.
- Fix
- Audit for mutual-exclusivity AND collective-exhaustiveness. If it fails either, redraw.
-
⤩ The arrow soup
Every box connected to every other box. The structure has no shape.
frequent- Bad
- Process diagram with bidirectional arrows between all 6 nodes — looks like a network, reads like noise.
- Fix
- Pick a directionality. If it is truly a network, use a hub-spoke or a force-layout, not a process.
-
◇ The decorative structure
Boxes and arrows where bullets would have done. Structure pretending to argue.
~25%- Bad
- "Goals → Strategy → Tactics" in three boxes connected by arrows. A list with a costume.
- Fix
- If the boxes carry no information beyond their order, write a numbered list. Save the diagram for non-linear structure.
-
◷ Borrowed without credit
A named framework (BCG, RACI, MECE, OKR) used without attribution.
71%- Bad
- A 2×2 matrix with axes "growth" and "share" labelled simply "framework".
- Fix
- Cite the source. "BCG Growth-Share matrix · Boston Consulting Group, 1970".
-
◫ The 12-node diagram
Twelve+ nodes on a single canvas. The eye cannot trace.
~10%- Bad
- Customer-journey diagram with 14 stages, 8 personas, and 6 channels — all on one slide.
- Fix
- Split into two diagrams. The first slide carries the spine, the second carries the detail.
Four rules. Pick the structure first.
A diagram is a commitment to a frame. These four rules force the editorial work upfront — before the boxes get drawn.
- 01
Pick the question, then the shape
Process for sequence. Hub-spoke for centrality. Framework for categorisation. Timeline for periodisation. Pick the question; the shape follows.
5 shapes covering 80% of cases - 02
If borrowed, cite
BCG, RACI, MECE, OKR, Cynefin, Lean Canvas — every named framework has an author and a year. Citing them imports their authority. Hiding them imports skepticism.
29% currently sourced - 03
One diagram per slide
Diagrams take the whole canvas. Two diagrams compete for the same focal frame and dilute both. Split into two slides.
95% corpus already follows - 04
No more than 7 nodes
Past 7 nodes the eye cannot trace a structure in one read. If the diagram needs more, split into a spine + a detail diagram.
7±2 cognitive limit