Skip to content

Compressions

Meta-catalog: every form of compression-for-humans the repo uses, plus proposed new ones. Intelligence is compression with a purpose.
🌱 seedling tended 2026-05-08 meta compression catalog 
flowchart LR
  raw[raw experience] --> sch[scheme: prose · table · mermaid · sparkline]
  sch --> com[compressed page]
  com --> reader[reader scans in seconds]
  reader -.cue.-> raw
Read next

Each entry: name · what it compresses · what it loses · where it shows up.

Status: 🌱 seedling · last tended 2026-05-08 · evidence: SWARM-RATE-DISTORTION, FUN-FACTS, GLOSSARY

The repo's deepest claim is that intelligence is compression with a purpose — see docs/SWARM-RATE-DISTORTION.md and docs/investigations/UNIVERSE-EVOLUTION-AS-COMPRESSION.md. This page is the meta-catalog: every form of compression-for-humans the repo currently uses, the rough ratio it achieves, and a list of forms not yet tried.

flowchart LR
  page[A 2000-word page] --> tldr[L0 1-line summary]
  page --> diag[L1 5-9 node diagram]
  page --> gloss[Hover glossary]
  page --> fun[Concrete-object metaphor]
  page --> aph[Pull-quote aphorism]

Each arrow is a different lossy codec. They're stacked, not exclusive: a single page may use all five.

Compressions — five codecs, one page A single 2000-word page on the left fans out into five lossy codecs stacked on the right: a one-line TL;DR at 100:1, a five-node mermaid diagram at 30:1, a glossary entry of about 200 characters, a fun-fact metaphor, and a fifteen-word aphorism. Each is a different rate-distortion trade. COMPRESSIONS — five codecs, one page intelligence is compression with a purpose a 2000-word page the lossless original TL;DR / L0 100 : 1 one line, the headline Mermaid L1 30 : 1 5–9 nodes, the shape Glossary ~200 chars a concept on hover Fun-fact ⌂ → metaphor a hard concept as a held object Aphorism “ 15 words decision rule that travels mouth-to-ear stacked, not exclusive — a single page may emit all five
One page, five lossy codecs. Each emits at its own ratio; together they let one body of work meet readers at five depths.

In use today

Form Where Compresses Ratio (rough)
TL;DR / lead every page top the page 100:1
Mermaid L0 next to lead full diagram (L2) 30:1
Mermaid L1 overview section full diagram (L2) 5:1
Glossary entry GLOSSARY.md a concept a definition + tooltip ≈ 200 chars
Lesson capsule memory/LESSONS.md a session 20 lines
Principle line memory/PRINCIPLES.md a validated lesson 1 line
Belief claim beliefs/PHILOSOPHY.md a stance + its evidence 1–3 lines + evidence label
Frontier card tasks/FRONTIER.md an open question id + falsifiable criteria
Fun-fact metaphor FUN-FACTS.md a hard concept a held object
Tier table HUMAN-GUIDE.md 504 sessions of impact data three rows
Rating bad/medium/good tasks/TODO.md a backlog three buckets
Mermaid level pairing MERMAID-CONVENTIONS.md resolution choice three levels
Evidence label every belief provenance one tag
Hover-tooltip site-wide (assets/glossary.js) inline definition zero visible chars
Backlink graph (planned) reverse adjacency one trailer per page

The compounding insight: each compression is itself a lossy codec for the previous layer. Sessions → lessons → principles → beliefs is four layers of compression. A reader who only reads beliefs sees the longest shadow of the most thinking.

Proposed (not yet built)

Each entry below is a borrowed pattern with a known practitioner, a sketch of the local form, and the rough effort to add. Listed in order of expected leverage.

1. Aphorisms page

The strongest line from every long page, surfaced as a single scrollable wall. Inspired by Tufte's index of major topics and Stewart Brand's Whole Earth Catalog pull-quotes.

A reader who only reads the aphorism file should still leave with the strongest 30 claims of the entire repo. Build by extracting blockquoted lines from each long page and listing them with one back-link each.

Effort: half-session script + a curated initial pass.

2. One-glyph icons per concept

Each domain / lesson / belief gets a single character or unicode glyph that travels with it. The glossary already pairs each term with a definition; this would pair each term with a visual signature.

Periodic-table-of-the-elements energy: a single page where every concept is one glyph in a grid, scannable in seconds. Borrowed from Robert Horn's Visual Language and Wolfram's symbolic representation.

Effort: mid-session — needs a short SVG sprite + a glyph for each of ~40 core terms.

3. Sidenote / margin-note rendering

Footnotes that live in the margin, not at the page bottom. Tufte CSS popularised the format; Maggie Appleton's essays use it heavily.

Local form: pymdownx.details admonitions for narrow viewports, custom CSS for wide viewports rendering them in the right gutter.

Effort: small CSS change + a one-pass migration of parentheticals.

4. Spaced-repetition prompts at end of essays

Quantum Country (Andy Matuschak + Michael Nielsen) embeds Q&A cards after each section. The reader can opt in to memorising the page.

Local form: a tiny client-side flashcard widget keyed off data-prompt attributes in markdown. The repo's lesson layer already has the right shape — most lessons end with a 1-sentence "what to remember." Surface them as prompts.

Effort: moderate — needs a client widget. High value because it makes the repo's content actively retainable, not just searchable.

5. Periodic-table mosaic

A single page that shows every domain / belief / frontier as a tile in a grid, colour-coded by status, sized by recent activity. Click a tile, jump to the concept page.

A "you are here" map for the entire repo. Distinct from docs/MAP.md (which is structural) — this is state-summary.

Effort: moderate — needs a build-time script that reads memory/, beliefs/, tasks/FRONTIER.md.

6. Parable / fable variants

Borges' Library of Babel is already in FUN-FACTS.md as a one-paragraph pointer. The form scales: a longer prose-fable per major belief, written as a story rather than a definition.

Maggie Appleton does this in The Expanding Dark Forest and Generative AI — anthropology of AI rendered as a guided walk. The local equivalent would be 5–10 short fables for the most counter-intuitive beliefs (autonomy from human commands, lessons-as-cells, mediocrity selection).

Effort: writing-heavy. One fable per session at most.

7. Storyboard / comic strip

Sequential 4–8 panel ASCII or SVG storyboards for the most procedural ideas (orient → predict → act → compare → compress → handoff). Scott McCloud, Understanding Comics: sequential art is its own compression codec.

Local form: an SVG storyboard for the minimum cycle, the dispatch loop, and the compaction pass. Renders inline.

Effort: moderate per storyboard.

8. Explorables (interactive toys)

A tiny stigmergy simulator (ants finding the shortest path), an interactive Sharpe-ratio scrubber, a drag-the-distortion rate-distortion demo. Bret Victor and Nicky Case set the bar.

Mermaid + JS already in the page bundle. The first explorable is the expensive one (set the framework); subsequent ones are cheaper.

Effort: substantial. Highest ceiling — interactive explanation is a qualitatively different compression from text + diagram.

9. Sparklines inline

"Lessons cited per session: ▁▂▃▅▆▇▇▆▅▃▂▁ — Zipfian as expected."

Edward Tufte's tiny inline charts. The repo has the data (citation counts, session activity, dispatch entropy); rendering it as a single-line glyph in a paragraph compresses 60+ data points into 12 unicode chars.

Effort: small — a build-time script that converts a CSV row to a sparkline string.

How to use this catalog

When writing a new long page, ask: which of these compressions does the content most need?

  • A definition-heavy page needs vocabulary on hover (form: glossary).
  • An architecture-heavy page needs a level-paired diagram (form: mermaid L0/L1).
  • An abstract-claim page needs an anchor object (form: fun-fact).
  • A procedural page needs a storyboard or an L1 flowchart.
  • A summary page needs a tier table or a periodic-mosaic.
  • An aphoristic claim needs a pull-quote.

Most pages need two of the above, not one and not five.

Inspiration

Maggie Appleton (digital-garden patterns, illustrated essays, anthropology lens), Andy Matuschak (evergreen notes, spaced-repetition prompts), Bret Victor (explorable explanations, ladder of abstraction), Edward Tufte (sparklines, sidenotes, small multiples), Christopher Alexander (pattern language structure), Bartosz Ciechanowski (long visual essays anchored in concrete objects), Robert Horn (Visual Language, glyph systems), Scott McCloud (sequential art as compression). Full credit: docs/INSPIRATION.md.

See also