HyperZettel app

Introducing a wik{A}i that looks for what none of its cards say

Books 1151Paper 114Articles 265Media 13Doc 202Connections 18Authors 424Wiki 523

HyperZettel is an app that turns a reading slips archive into an active tool for thought. It reads directly from a markdown wiki vault with thousands of single units of knowledge ("cards") covering books, academic papers, essays, and relational notes that map connections across cards.
The raw markdown notes remain the single source of truth.

Read

Every card is read whole down to the quotes

Extract

Each card falls into four to six core concepts

Generate

Distant ideas meet in theories that no single source holds

Compute

Novelty is an empty intersection, built by math every night

Inhabit

Hypotheses are read from where one subject stands

Evaluate

Keep, discard or object. The machine reads the verdicts and never edits them

Philosophy

A library should argue back

HyperZettel treats a structured library as a universe of atoms: every card (a source) reduced to its cores, every core set against the others, until a novel idea emerges

Is this idea original?
Does any card already say all of this?

The first question asks for an opinion. The second can be answered

Atom A · 3 cardsAtom B · 12 cardsAtom C · 4 cards ∅
Each circle is the set of cards holding one atom. A combination counts as new only when the centre is empty

Only what the sources say

Every generated text rests on the cards alone and is presented as a hypothesis and each carry a limit: one sentence stating what would disprove them

Novelty is measured

Whether a claim is new is decided by the set of cards behind it, never by the taste of whoever writes it. The check runs again on every construct

Small steps, in the dark

Work happens overnight, in logged passes with backups and validity checks. The human in the loop steers with explicit judgments

933

Raw sources

Documents, reviews and notes stay where they are. Copies are converted to Markdown and never edited1

2,486

Wiki pages

Markdown files, read in Obsidian. This is the master copy of every card, and corrections happen here1

1

Derived index

A folder of JSON files holding every construct. It is rebuilt by routines and read live by the app, which needs no recompiling

Top ↑
Constructs

Seven levels. One way up

Atoms, Molecules, Bodies, Umwelt and finally Partitur: each card is distilled into core claims. Identical claims coalesce into Atoms. When distant atoms intersect, they form Molecules—theories that no single card articulates on its own. Three or more molecules assemble into Bodies, representing higher-order hypotheses that form Umwelten, "surrounding world".
Finally these worlds build an entire single Partitur, a genuinely novel "system"

Bars on a logarithmic scale. Counts of 15 Nov 2026 for the main chain12

One idea, followed all the way up

A real line of descent from the archive

Cards

Four books that never meet

When AI meets desire on luxury retail. How Fashion Brands Get AI Visibility Right on assistants. The Cashless Revolution on Chinese finance. The State Strikes Back on the party and the economy. Eight cards in all feed this line, and none holds all of it.

Units
  • “For decades luxury strategy organised itself around destinations controlled by the brand, such as store, site and wholesale partner.”
  • “An assistant recommends brands from associations learned in training and from sources fetched by live search, and position cannot be bought.”
  • “China went from cash to super-apps that bring together payments, credit, investments, shopping, travel and messaging.”
  • “Under Xi the platforms were hit and their stock of technology and data was absorbed by the public side.”
Atoms
  • Luxury and channels · 3 units
  • Being understood by AI and being chosen are different things · 12 units
  • Super-apps bring together payments, credit and shopping · 2 units
  • Under Xi the State takes back control of the economy · 4 units
Molecule

The meeting point with the client becomes contested

Hypothesis: luxury strategy organised itself around brand-controlled destinations while discovery moves to intermediaries the brand does not govern. Where platform data are absorbed by the state, the intermediary answers to a power outside the market. The meeting point becomes a contested resource, and the brand must reckon with whoever governs the intermediary as well as with the client.

Would be disproved if luxury brands kept control of discovery even when search and purchase pass through assistants and super-apps subject to public power

Body

Intermediated growth holds only with written rules

Joined with two other molecules: fixed costs and the stock market force growth toward agents that compare what is comparable, and in agent systems full price survives only as a written constraint. The maison would have to carry written rules into the contested point itself.

Umwelt

The sales director must grow where he does not decide

Three bodies read from one chair. He perceives budget, cost pressure and volumes from outside agents. The fading of rank and the assistants' reading of the client reach him late. He can still write price and distribution as constraints agents can read, and choose early which intermediaries to admit.

Partitur

Written constraints and the memory to reopen

It sets all seven habitats side by side. Each subject acts on what is written, counted or quoted, while the lever often lies outside its perimeter. The voices diverge on where to act: in the masters, in the invariant, in the history of the pieces, in the rules for agents.

Four constructs beside the ladder

They do not stack. They feed the daily practice of reading

44

Concepts

One each morning, from two or three linked cards: 70 to 90 words, a question, a verified artwork, and an experiment to try

18

Connections

Chosen by the human: two or more cards and two lines of instructions. Claude compiles them, citing the card behind every claim

233

Notes

Cards that start from a working document instead of other cards. Anything marked confidential stays out of the construct chain

26

Evenings of suggestions

A book, an article and a report each night, verified on the web, new to the wiki, and tied to named cards

Top ↑
Environments

The same archive. Seen from every height

Each screen shows one level of the structure, and each links to the others: an atom leads to its cards, a card to the atoms it shares, a molecule to the atoms that form it. The toolbars below work: they jump between screens3

Cards

The archive at a glance: what came in, what talks to what, and which cards still talk to nothing

HyperZettel
Search title, author, publisher, wiki entry
Cards · the archive at a glance
Cards · total
1561
Books
1151
Paper
114
Articles
265
Last 7 days
+312
Isolated
258
Shelf by genre
Philosophy 550AI and technology 382Psychology and psychoanalysis 157Luxury and business 155Other 103History 76Economics 58Anthropology and sociology 50Politics 46Language and semiotics 34Logic and mathematics 22Fiction 13
By original year · 1405–2026
Most in dialogue
When AI meets desire48
The AI operating system38
The agentic transformation office37
Co-Existence32
Scaling the Agentic Enterprise28
Not yet in dialogue · 258 cards share no atom
A Thousand Plateaus→
A return to seriousness→
Adorned in Dreams→
An Inside Look At How To Tame Unstoppable AI→
An Introduction to Functional Grammar→

The card

A card is a working page. Beyond summary, thesis and context, it lists the atoms drawn from it and every other card that holds the same idea

Paper · Deloitte AI Institute
State of AI in the Enterprise 2026
The untapped edge
Open the PDF
Summary
Annual report of the Deloitte AI Institute, January 2026, built on a survey of 3,235 leaders in 24 countries and six industries. It argues that AI is moving from pilots to scale as worker access widens, but that only a third of companies use it to transform their model, that work is not being redesigned, and that agents are spreading faster than the rules meant to govern them.
The thesis
In a year the share of workers with approved AI tools rose from under 40% to about 60%, yet fewer than 60% of those with access use them daily. 74% of leaders hope AI will grow revenue and only 20% already see it, while 84% have not redesigned roles around it.
Conceptual units
6 atoms · on average 5.0 per card across 1736 cards
Adopting AI is not the same as getting value from it
“Approved access to AI tools rises in a year to about 60% of workers, but fewer than 60% of those who have it use them every day.”
Also in: Power and Prediction · The AI operating system · The AI value curve · Global Powers of Luxury Goods 2026 · +8 more
ART-035
Publication
Deloitte AI Institute
Date
2026-01
In the wiki
The Formula for Agentic AI Value
Scaling the Agentic Enterprise
The AI Dossier
Open the card in the wiki
Top ↑

Wanderkarte

Every card is a node, in two or three dimensions. Each morning a new concept appears and the cards behind it light up in the graph

HyperZettel
Search title, author, publisher, wiki entry
Wanderkarte
1561 cards · 278 authors · 523 wiki entries · 202 docs · 10,970 links
Concepts 19/28 · At the meeting of 2 sources
Transparency Tests Prestige
Song
9 to 5
Dolly Parton, 1980
In knowledge work AI touches professional identity, and firms that reach scale say early how roles will change. In fashion employers are chosen for reputation, yet only 28% see it confirmed each day. Announcing the change frankly tests the prestige people built their identity on.
Can a maison tell the truth about changing roles without spending the prestige that brings it talent?
Explore · Suggested by the cards · 2026-11-02
Book · Lisa Gitelman · 2014
Paper Knowledge
OpenAdd as card
Collision
Umwelt 7/7 · From at least three bodies
The sales director must grow where he does not decide
LinkITENPinPNG
Unprecedented · 3 bodies, 6 molecules
Bodies 13/13 · From at least three molecules
What the model makes canonical reopens from outside
Unprecedented · 3 molecules, 10 atoms, 42 cards
Molecules 19/19
The trained archive behaves like a canon
Top ↑

Concepts

The morning's thought, and the place where the reader answers back. Keeping a concept turns it into a connection. Discarding teaches the night what to avoid. Objecting asks for the opposite thesis on the same sources

HyperZettel
Search title, author, publisher, wiki entry
Concepts · 19 of 28
Concept · 2026-10-27 · At the meeting of 2 sources
Transparency Tests Prestige
Song
9 to 5
Dolly Parton, 1980
Written for the film of the same name, the song follows an office worker and the distance between what work promises and what it gives back, the same distance between reputation and daily life that the survey measures.
In knowledge work AI touches professional identity, and firms that reach scale say early how roles will change (Scaling the Agentic Enterprise). In fashion employers are chosen for reputation, yet only 28% see it confirmed each day, and those who see it contradicted think of leaving (How AI Is Reshaping the Battle for Fashion and Beauty Talent). It follows that in a maison, announcing frankly the changes AI brings tests the prestige on which people have built their identity.
Can a maison tell the truth about changing roles without spending the prestige that brings it talent?
Experiment
Before announcing an AI programme in a maison, ask three functions a single question on how far the brand's outside reputation matches their daily work. Start the first pilot where the match is highest, and after three months compare adoption and intention to leave with the other two.
  • Scaling the Agentic Enterprise · Andrew Pearce, Alex Kouptsov, Tim Nolan, Jordan Luft
  • How AI Is Reshaping the Battle for Fashion and Beauty Talent · Dan Hastings, Eoghan O'Donnell
Keep itDiscardObject
Quick note
What does it make you think?
Concept · 2026-10-26 · At the meeting of 2 sources
The Alien Space Has a Tacit Side
Top ↑

Explore

What to explore next. Each suggestion is checked on the web, searched for in the vault to prove it is new, and tied to named cards it would argue with

HyperZettel
Search title, author, publisher, wiki entry
Explore · 13 evenings, most recent first
Suggested by the cards · 2026-11-02
Book · Lisa Gitelman · 2014
Paper Knowledge. Toward a Media History of Documents
A history of the document as a genre of knowledge, from printed forms to PDF. It speaks to Lecture 47 and Ferraris's documentality, which reasons in the abstract, by carrying the problem onto concrete media, and to The Social Life of Things, where the biography of objects never asks what format it is written in.
OpenAdd as cardLecture 47The Social Life of Things
Article · Michael K. Buckland · 1997
What Is a “Document”?
A Journal of the American Society for Information Science article on how information science widened the notion of document beyond texts. It offers From Product Passports to Smart Products a criterion for deciding when an object or a product datum counts as a document.
OpenAdd as card
Report / paper · Shayne Longpre, Robert Mahari, Anka Reuel and others · 2024
A large-scale audit of dataset licensing and attribution in AI
The Data Provenance Initiative's audit in Nature Machine Intelligence of licences and attribution in training datasets. It brings data to a question Lecture 47 leaves to social ontology: who answers for the documents a model is trained on.
OpenAdd as cardThe AI Mirror
Top ↑

Planes

The whole hierarchy pulled apart, one plane per level. Drag the sliders, then click any dot: its line lights up from the Umwelt down to the cards. This one is live, drawn from the real links between constructs2

HyperZettel
Search title, author, publisher, wiki entry
PLANESDaring 3 · Standard

Daring, on a slider

The same toolbar in the app holds a slider from Cautious to Daring, which every night pass reads before it starts

Every state, kept

The app logs each change to its data with a reason such as “+1 molecule · +1 body” and the counts of the moment

Go back, undoably

A history panel restores any earlier state. The current state stays in the list, so a restore can itself be undone

Top ↑

Atoms

A matrix of the whole library: each row a card, each column an idea. A dot means the card holds that atom, and a long thread means one idea crosses distant parts of the archive

HyperZettel
Search title, author, publisher, wiki entry
Atoms
1736 cards · 6245 atoms · on average 5.0 atoms per card
Search atom, card or sentence
Atoosa KasirzadehA New Dawn For Mathematical Progress
Andrew Pearce, Alex KouptsovScaling the Agentic Enterprise
Dan Hastings, Eoghan O'DonnellHow AI Is Reshaping the Battle
Lieven Van der Veken, Sandra DurthThe AI operating system
Antoine Montard, Dago DiedrichWhere AI will create value
Deloitte AI InstituteState of AI in the Enterprise 2026
Chase Covington, David PralongThe agentic transformation office
Francis Buttle e Stan MaklanCustomer Relationship Management
David LoshinMaster Data Management
Ida Palombella, Katie WeirGlobal Power of Fashion
Ralph Kimball e Margy RossThe Data Warehouse Toolkit
Asutosh Padhi, Brian ElliottThe Formula for Agentic AI Value
LectraHow Lectra helps you lead the way
TEHA GroupJust Fashion Transition 2025
Joe Ngai e Nick LeungThe Next China Is Still China
The Business of Fashion & McKinseyThe State of Fashion
Annika A. CulverDemocratizing Luxury
Pierre-Yves Donzé e Rika Fujioka (eds)Global Luxury
Teri AginsHijacking the Runway
Rosalind H. WilliamsDream Worlds
Michel Chevalier e Gérald MazzalovoLuxury Brand Management
Mark TungateLuxury World
Martin Kihn e Christopher B. O'HaraCustomer Data Platforms
Silvio Palumbo, Mark Abraham, Yun LimHow the Composable Shelf
Each row is a card, each column an idea atom. A dot means the card contains that atom; an empty cell means it doesn't.
BooksPaperArticlesConnections3623 atoms in a single card
Top ↑

Molecules

Three or four atoms that no card holds together, turned into a theory with a stated way to fail. Each one shows its proof of novelty in green

HyperZettel
Search title, author, publisher, wiki entry
Molecules · 19 unprecedented combinations of atoms
The trained archive behaves like a canon
LinkITENPinPNG
The winners rewrite the history of doctrineThe canon generates commentaryFrequency does not make identityManuscripts give back the defeated voices
Unprecedented · None of the 3 cards involved contains all 4 atoms
Hypothesis: a model trained on a maison's archive could behave like a canon. It learns what the house did most often, while identity lies in what survived the changes of hand. A canon makes a text untouchable and generates commentary: whoever queries the model glosses an already fixed version instead of disputing it. In religious histories the winners rewrite the past by presenting the alternatives as deviations, and the defeated voices become audible again only through recovered manuscripts. It would follow that a house's rare choices turn into deviations through statistical weight, and that hearing them again requires documents kept out of training.
What would disprove it
The theory falls if people working with models trained on an archive recovered rare pieces and abandoned lines as well as people who consult the archive without a model.
Source cards
  • Lost Christianities
  • Cultural Memory
  • The Early History of God
Created 2 Nov 2026 · Daring 3 · Standard
Top ↑

Bodies

At least three molecules joined in a higher-order hypothesis. Three tests stand between a candidate and the file, and the screen prints the result

HyperZettel
Search title, author, publisher, wiki entry
Bodies · 13 hypotheses from at least three molecules
What the model makes canonical reopens from outside
LinkITENPinPNG
The trained archive behaves like a canonThe misplaced concreteness of AI metricsThe passport as a feedback organ
Solid coral line: the two molecules share atoms (the thicker, the more atoms in common) · dashed: no atom in common
Unprecedented · 3 molecules, 10 atoms, 42 cards: no card and no molecule brings them all together
Combined molecules
  • The trained archive behaves like a canon
  • The misplaced concreteness of AI metrics
  • The passport as a feedback organ
A plausible hypothesis to discuss: what a model makes canonical might reopen only from outside the model. One molecule observes that a trained archive fixes what is frequent and turns rare choices into deviations. Another notes that a visible, partial layer such as adoption metrics takes the place of what actually operates. The third reads the digital product passport as a return channel that lets a brand notice its own errors after the sale. Taken together, they suggest that whoever consults the synthetic memory of their own house risks mistaking it for the house, and that correction would have to come from events recorded outside training: the life of the pieces.
What would disprove it
The theory falls if brands pairing the model with life-cycle data corrected their reading of the archive no more than brands using the model alone.
Top ↑

Umwelt

The turn from theory to situation. Three bodies are read as the habitat of one subject: a chief financial officer, a workshop head, a shopping agent. What it perceives, what it can act on, and what follows

HyperZettel
Search title, author, publisher, wiki entry
Umwelt · 7 habitats from at least three bodies
The sales director must grow where he does not decide
LinkITENPinPNG
Intermediated growth holds only with…Luxury writes down what it once…Access to the client is governed…
Unprecedented · 3 bodies, 6 molecules: no body contains them all
Subject
The sales director of a listed luxury group, who answers for volumes and chooses which channels and intermediaries to open
Merkwelt · What it perceives
The budget, the pressure of fixed costs and the stock market, the volumes arriving from external agents and super-apps. Less reaches him, and late: the signs of rank wearing away and the client as read by the assistants.
Wirkwelt · What it acts on
He can write price and distribution as constraints agents can read, choose early which intermediaries to admit and on what terms, and decide what information about the pieces stays in house.
Combined bodies
  • Intermediated growth holds only with written rules
  • Luxury writes down what it once kept in places
  • Access to the client is governed like a supply chain
What would disprove it
The thesis would fall if groups opening most to intermediated channels, without written rules, kept full price and desirability better than the others.
Top ↑

Partitur

Uexküll imagined nature as a score in which the worlds of living beings sound together. Here there is one score only. It is never replaced, only rewritten in numbered versions whenever something changes below it

HyperZettel
Search title, author, publisher, wiki entry
Partitur · The single score of all the Umwelten
Partitur 007 · 2 Nov 2026 · 7 Umwelt · Daring 3 · Standard
Written constraints and the memory to reopen
Hypothesis: the seven Umwelten describe subjects who act on what is written, counted or quoted, while the lever often lies outside their perimeter. The brand receives the client already interpreted, the top team sees adoption early, the agent reads rules and data, the designer inherits a quotable archive, the finance chief measures the fast component, the workshop head loses sight of judgment passed from person to person, the sales director must grow through channels where he does not decide. The voices diverge on where to act: in the masters, in the invariant, in the history of the pieces, in the rules for agents. The night's body adds a caution for designer and brand: the model can make canonical what is frequent, and correction would come from outside. It would follow that written constraints need a return channel able to reopen what has been fixed.
What would disprove it
It would be disproved if maisons using archive-trained models and writing their own constraints, with no return channel from pieces and markets, corrected their errors as well as those that have one.
Atoms
5201
Molecules
25
Bodies
19
Umwelt
13
The score · One voice per Umwelt, one bar per version
001002003004005006007The brand in the assistants' habitatThe top team that decides before seeingThe agent that sees only what is writtenThe new designer inherits an archiveThe CFO measures what runsThe workshop head fixes the mastersThe sales director must grow
Top ↑

Graph

Where the other screens show the cards, this one shows how they are classified and what the nightly pass has stated about them: the taxonomy, the matrix of links between themes, and the graph of typed links

HyperZettel
Search title, author, publisher, wiki entry
Taxonomy · group → theme → theme page → people and authors
Filosofia e teoriaAI e tecnologiaLusso e modaLavoro e politicaLavoro di MarcoMetafisica e storia della filosofiaPsicoanalisiMarxismo e critica dell'economia politicaConsumo e modernitàStrategia di marca e di lusso
Filosofia e teoriaAI e tecnologiaLusso e modaLavoro e politicaLavoro di Marco
Ontology · typed links between the 36 themes
  • causes 0 links
  • precedes 23 links
  • developed in 259 links
  • popularised in 15 links
  • confirms 9 links
  • opposes 24 links
  • part of 11 links
  • resembles 4 links
Link graph · 460 cards, 345 typed links, 10 bridge hypotheses (dashed)
causes 0precedes 23developed in 259 +2 hyp.popularised in 15confirms 9 +2 hyp.opposes 24 +2 hyp.part of 11resembles 4 +4 hyp.
36

Themes, one list

A controlled vocabulary in five groups gives every one of the 1,647 cards a primary theme and up to two more. A sample check of 350 cards found about 6% of primary themes wrong, and they are corrected one by one

8

Types of link

Causes, precedes, developed in, popularised in, confirms, opposes, part of, resembles. A link is kept only when a text states it, with the sentence beside it

44

Bridge hypotheses

Where two distant themes seem to touch, the nightly pass writes a hypothesis and keeps it apart from the links. It becomes a link only when a text says so outright

Top ↑
Dephasing

A laboratory. With a red triangle on the door

Twelve experiments, twelve Focus structures and eight Complexity environments use everything in the index. A click on the red triangle opens the three groups, each on its own page. Those that generate produce candidates only: nothing enters the wiki until the human presses a button

0

External services for vectors and sums

Embeddings, linear algebra and on-device language models all run on the Mac

12

Experiments

Collider, Hopeful Monster, Distiller, Pachinko, Aria and Ramon generate. Atom Space, Latent Spaces, Ficciones, Grid and Latent space describe. Shaker forces links. Twelve Focus structures arrange the cards, and eight Complexity environments read the index after the theorists of complexity

(?)

A declared method

Most numbers on screen sit beside a question mark that explains how they are computed: weights, thresholds, formulas

Collider

Two elements of any level run round a ring and collide. The detector reports how improbable the pair was, and a local model writes what the crash produced. Press Collide

HyperZettel
Search title, author, publisher, wiki entry
COLLIDER Two distant elements of the wiki, accelerated until they collide
Beam A ·
Beam B ·
Improbability
–
Shared cards
–
Energy
–
Level jump
–
Press Collide now
What the collision produced
The summary is written on the Mac by Apple Intelligence, from the two texts and the question.
Write with ClaudeCollide againAnother pair

Both pairs and their written outputs come from real runs of the app

Hopeful Monster

Named after Richard Goldschmidt's 1940 theory of evolution by leaps. A construct is treated as a genome and pushed along eight steps of time. Most offspring die. Rarely one finds a free niche5

HOPEFUL MONSTER Macromutations along a timeline: most offspring die, rarely one finds a free niche
The longer the horizon, the bigger each jump: more genes replaced at every step, fewer survivors.

Macromutation

Each branch bears 7 offspring per step. A share of genes is swapped from the level's pool: 25%, plus 5 points a year, up to 80%

Vitality

An offspring lives if every gene links to the others through a shared card or wiki entry, and if no element already holds that combination

Selection

Fitness is the share of linked gene pairs. The fittest continues the line, the runner-up opens a timeline with probability 0.55, to a maximum of five

Atom Space

Every atom placed twice: by what it says, and by the cards it lives in. The gap between the two maps is where connections hide that nobody has made6

ATOM SPACE
Atoms
5201
Dimensions
1024 / 2984
Never met
60
Too broad
3
Near = live in the same cards and wiki entries
PhilosophyLuxury and businessAI and technologyHistoryOther
Near by meaning, never met
Exploitation ↔ Exploitation and legal equality78%
Concentration of spending ↔ Sensitivity to AI cost75%
Desire of the other ↔ Wandering and mystery74%
Too broad to be one atom
People keep judgment, goals and exceptions0.54 · 19 cards
The problem of universals between individual and species0.52 · 7 cards
Limits of the theory0.50 · 10 cards
1,024

Dimensions of meaning

Title and first three units pass through BGE-M3, a multilingual embedding model that runs on the Mac. Vectors are centred, scaled to length 1 and compared by cosine. Apple's Italian model stays as a fallback

60/30/10

Weights of the card map

Same card 60%, measured with the Jaccard index. Same wiki entries 30%, cosine weighted by rarity. Same genre 10%, used only to fill gaps

≈140

Dimensions that really count

Half the variance fits in 59 dimensions and 90% in 270. A panel under the map shows how much the flat picture distorts: 31% of true neighbours survive on average

Latent Spaces

The main threads of the library, found without reading a word. Only one thing is measured: which atoms travel together across cards7

LATENT SPACES The main threads of the library, found by an LSA of cards × atoms
Latent themes · 12 directions from 959 cards × 1059 shared atoms
AI AND PEOPLEDATA, PEOPLE, SYSTEMSPRESENCE AND RITUALDESIRE AND LANGUAGETRUTH IN FRAGMENTATIONPOWER AND TRUTHPOWER KNOWLEDGE SUBJECTIVITYCRITIQUE OF ENJOYMENTINTERSECTIONS OF POWERTHE TURN TO THE TRUTH OF BEING
Theme 1 · 0.6% of the structure
AI and People
Atoms
AI adoption does not coincide with value
People keep judgment, goals and exceptions
An agent applies only what the firm has written
The future of AI lies in collaboration
Cards
Scaling the Agentic Enterprise
The AI operating system
Co-Existence
The AI value curve
Latent Semantic Analysis (LSA) builds a table of cards by shared atoms, weights rare atoms up, centres the columns and takes the 12 largest eigenvectors with an exact Singular Value Decomposition (SVD). Apple Intelligence then names each direction. An atom or card that loads on two themes sits between them on the map, and those bridges are the most useful nodes in the picture.

Distiller

The whole ladder in one sitting, steered by hand at every stage. Pick cards, set how novel or how tense the combinations should be, and distil column by column up to a score of its own

CARDS · 12
Source · live
The Scientific ImageBas C. van Fraassen
The Postmodern ConditionJean-François Lyotard
Steer atoms
Cards12
ATOMS
Steer molecules
Novelty60%
Atoms per molecule3
Molecules6
MOLECULES
Steer bodies
Shared ↔ tensionbalanced
Bodies4
BODIES
Steer Umwelten
Merkwelt ↔ Wirkweltboth
Umwelten2
UMWELTEN
Steer Partitur
Focus: what the score should listen for. Voice weights in the score.
Partitur D001 · 29 Oct 2026, 15:17
Anomalies Point to a Summit, or Dissolve It
Two voices of equal weight share one ground. Each knower sits at one level of a graded order and receives only effects of the level that founds it. The whole tells that the anomaly is both the shared method and the undecided verdict: the same residue either points toward a first principle or dissolves it into plural efficacy.
Distilled constructs live in their own file with their own identifiers, apart from the main chain, and appear in Planes and in the Partitur tabs as a side branch. One full run exists so far. Reading it beside the main score shows which theses survive a different choice of atoms.

Shaker

Take one card. Force its atoms into cards that share nothing with it, at a chosen level of recklessness, and read the bridges that come out

SHAKER Forces connections, even daring ones, between the atoms of a card and cards they are not linked to
Card
Recursivity and Contingency
An essay in the philosophy of technology in which Yuk Hui traces, across some two hundred and fifty years of thought, the relation between organism and machine, from Kant's critical philosophy to cybernetics.
6 atoms · 1147 cards with no atom in common
Daring: recklessLinks: 4Shake
Bridge · From Overcoming Metaphysics into Labor and Monopoly Capital
The idea of the man as shepherd of being must surrender the sacred duty of preserving the mystery to the quantified control of managerial knowledge, where the execution of labor becomes a fragmented function of a class-based apparatus.
A shaken pair can graduate. The connection card “Luxury creates markets where none existed × Second Thoughts” began here, as an atom of Global Luxury forced into a book by Wilfred Bion with no atom in common. Then is compiled under the usual rules, and it now stands in the wiki with its limits written down.

Ficciones

The ladder drawn as a garden of forking paths. Atoms, molecules, bodies, Umwelten and the versions of the Partitur stand in five columns, joined by a line wherever one is made of another. Point at a title and its whole path lights up

FICCIONES The forking paths from atoms to molecules, bodies, Umwelten and the Partitur: what each one is made of, and what grows from it
Search every level by title or text
Only this pathAtoms
Un agente applica solo ciò…La costituzione del modell…L'addestramento assorbe ar…Le regole colpiscono il fo…ATOMS · 120of 5784Il setting come ripetizione dell'ambie…Il lusso è ciò che non si scriveLiving monads and the break from repre…Overdetermination excludes a simple fi…No order can certify its own originLa prima interpretazione come nuova co…The uncaused cause as obstacle in pena…Univocal Being, Graded Foundation of E…Accidental chains of power, essential…Il desiderio come circuitoIl passaporto come organo di retroazio…Le scelte irreversibili cadono prima d…Il punto d'incontro con il cliente div…Il margine sta nel componente più lentoIl lusso diffuso cerca valore nella bi…La concretezza mal posta delle metrich…La maison è ciò che resta quando cambi…La moda incontra una macchina di citaz…Il prezzo pieno regge solo se scrittoL'archivio addestrato si comporta come…Il genio della maison si crede, la fil…La crescita obbligata spinge il lusso…Chi ha simulato l'autentico non può di…Il giudizio diventa raro quando tutti…Il lusso si difende dove verificare co…I registri leggibili lasciano il poter…Ciò che la maison scrive si ferma sopr…Il gesto si protegge legandolo al corp…Senza traccia scritta si controlla ciò…L'immagine sintetica rimanda il lusso…Il lusso difende il prezzo limitando l…La gradazione del lusso regge solo nel…La deferenza spetta a chi risponde di…La creazione resta vuota finché la mai…La fiducia negli agenti si ancora a un…Il lutto collettivo come rito sostitut…Cultural stasis as real opposition ope…MOLECULES · 372 unusedLa qualità del canale di ritor…Il lusso come esenzione dalla…Breaking the Uncaused Cause in…Graded Knowers Learn Their Fou…Truth is reached only through…L'accesso al cliente si govern…Al marchio resta la storia del…First Cause Grounds Truth, Obs…Il lusso scrive ciò che prima…Il cruscotto guarda il compone…Fissare l'invariante, lasciare…Quando il nuovo abbonda, vale…Il valore sta dove la misura n…La crescita intermediata regge…Ciò che il modello rende canon…La continuità della maison sta…Dove la macchina interpreta e…Il racconto della maison vive…Il controllo passa dal leggere…Il valore passa dalla storia a…L'identità della maison si dec…Il prezzo del lusso si riancor…Il controllo degli agenti si c…L'abbondanza generativa string…Nel lusso agentico la parte sc…Il lusso sceglie i propri vinc…La maison governa il sapere de…Nel lusso agentico si crede a…BODIES · 281 unusedIl marchio nell'habitat degli…Il vertice che decide prima di…L'agente che vede solo ciò che…The Knower Who Learns Through…The Graded Knower Reading Its…Il nuovo stilista eredita un a…Il direttore finanziario misur…Il capo laboratorio fissa i ma…Il direttore commerciale deve…L'archivista deve tenere il ca…Il responsabile dell'autentica…Il responsabile del prezzo scr…Chi firma il racconto assegna…Chi addestra l'assistente deci…Il direttore dell'immagine pag…Chi nomina le anomalie scrive…Il consulente di boutique regg…Il maestro nominato porta il v…L'ufficio stile decide che cos…UMWELTEN · 19001 · Ogni habitat d…002 · Gli habitat si…003 · Negli habitat…D001 · Anomalies Poi…004 · Il valore rest…005 · Ciò che il reg…006 · Le mani da fis…007 · I vincoli scri…008 · Il ritorno da…009 · Ciò che si può…010 · Ciò che si scr…011 · Ogni scrittura…012 · Le voci cercan…013 · Le voci si spo…014 · Le voci si acc…015 · Le voci si acc…016 · Le voci si acc…017 · Le voci si acc…018 · Le voci si acc…PARTITUR · 19
Molecule4 atoms · in 1 body · 3 Umwelten
Ciò che la maison scrive si ferma sopra il modello
Hypothesis: what a maison writes for its own assistants may stop at a high layer of the machine.

Real constructs of 12 Nov 2026, the Distiller's side branch included. Titles as they appear in the archive. Here the atoms are marks; in the app each has its title

Made of

Every line is a fact of the index: a molecule lists its atoms, a body its molecules, an Umwelt its bodies, a Partitur version its Umwelten. No model draws or weighs anything here

120 atoms on the page

Only the atoms that at least one molecule combines are shown, out of 5,784. A button hides the column and gives the width back to the other four

One path at a time

A click holds a path and Only this path hides everything else. The search lights every element that mentions a word, with its paths. Past a minimum row height the page scrolls

First entry

An Umwelt is tied to the Partitur version where it first enters. Choosing a version shows every Umwelt it gathers, the earlier ones with a dashed line

Still unused

Each column counts what the next level has not picked up yet: today 2 molecules no body combines and 1 body no Umwelt gathers

Grid

Every card on one montage table. Covers come from Open Library, articles and videos bring their preview image, and a card with no picture becomes a tile of type in the colour of its kind. Narrow by type, dialogue, decade and genre

GRID Every card on one montage table: covers, previews and titles side by side, narrowed by type, dialogue, decade and genre
1522 cardsTypeDialogueDecadeGenre
Newest firstDense
book · 2026World Models for AI AgentsVolodymyr Pavlyshyn
article · 2026World Models HandbookMichael Brenndoerfer
article · 2018World Models (Ha e Schmidhuber)David Ha, Jürgen Schmidhuber
article · 2026The spirit of the leaderManish Chopra
book · 1967The Nature of ExplanationKenneth Craik
book · 2024The Experience Machine (Clark)Andy Clark
article · 2026The Case for a Bubble at Life’s StartPerri Thaler
book · 2026The Age of World ModelsBruno Bianchini
book · 2019Surfing UncertaintyAndy Clark
article · 2021Stoics as activistsMassimo Pigliucci
article · 2026Serbia’s Experiment- Combining Democracy & MeritocracyNathan Gardels
article · 2026Rural healthcare and America’s $340 billion workforce earn…Bonnie Dowling, Katherine Linzer…
article · 2026Reinventing business planning with agentic AIHolger Hürtgen, Denis Lebedev, J…
article · 2026How CFOs can make AI pay offCristina Catania, Davide Grande,…
articleGeoffrey Hinton- AI is learning more like Humans than we r…Geoffrey Hinton
article · 2025From Words to WorldsFei-Fei Li
article · 2026Democrats Should Not Revert To The Status Quo AnteNathan Gardels
article · 2026CEO Talks- Titan’s Kuruvilla Markose on the Long Climb Up …Lily Templeton
book · 2022Being YouAnil Seth
book · 2024Bayesian Models of CognitionThomas L. Griffiths, Nick Chater…
article · 2026Bally’s Italian Units Face Liquidation as Milan Stock Sale…Martino Carrera
book · 2025Active InferenceThomas Parr, Giovanni Pezzulo e …
book · 2022A Thousand BrainsJeff Hawkins
book · 2025A Brief History of IntelligenceMax Bennett

The 24 public cards added most recently, as tiles of type: here covers are left out. Books in green, articles in amber

Latent space

The cards laid out on a plane by one chosen characteristic, and when the characteristic changes they slide to their new places. Meaning, principal axes, time and dialogue, colour of the cover

LATENT SPACE The cards laid out on a plane by a chosen characteristic: meaning, principal axes, time and dialogue, colour. Switch and watch them slide
MeaningPrincipal axesTime × dialogueColour
Philosophy · 683AI and technology · 289Other · 215Luxury and business · 152History · 75
The Savage MindThe God TestChina's Last EmpireLoïc Prigent, con(mp)teur de rub…Il mondo magico

Real layout of 1415 public cards, Meaning view: t-SNE of the average BGE-M3 vector of each card's atoms. Coloured by genre, one card named per main genre

Meaning

Each card is the average of the vectors of its atoms. t-SNE keeps neighbours together, so cards that say similar things land close, whatever their genre

Principal axes

The same space cut along its two main directions, with the atoms that pull hardest towards each end written at the edges

Time and colour

The year across and how many cards a card reaches through its atoms upwards. Or the hue and lightness of the cover, for a library seen as a palette

Aria

One card unfolded like Bach's Goldberg Variations. Its atoms are the fixed bass. Above them run thirty variations in groups of three, then a Quodlibet and the aria da capo. Each text is written only when its variation is opened, and nothing is saved

ARIA A card unfolded like the Goldberg Variations: its atoms are the fixed bass, thirty variations build on it, then the Quodlibet and the aria da capo
Card
Being You
Fixed bass · 5 atoms
Ariathe card itself
Var. 1genre pieceread through Focus
Var. 2two manualscrosses a molecule or a body
Var. 3canon at the unisonnearest card
Var. 4genre pieceread through Focus
Var. 5two manualscrosses a molecule or a body
Var. 6canon at the seconda card further away
Var. 7Al tempo di gigaread through Focus
Var. 8two manualscrosses a molecule or a body
Var. 9canon at the thirda card further away
Var. 10Fughettaread through Focus
Var. 11two manualscrosses a molecule or a body
Var. 12canon at the fourtha card further away
Var. 13Ornate ariaread through Focus
Var. 14two manualscrosses a molecule or a body
Var. 15 · minorcanon at the fiftha card further away
Var. 16French overturethe skeleton read from a higher level
Var. 17two manualscrosses a molecule or a body
Var. 18canon at the sixtha card further away
Var. 19genre pieceread through Focus
Var. 20two manualscrosses a molecule or a body
Var. 21 · minorcanon at the seventha card further away
Var. 22Alla breveread through Focus
Var. 23two manualscrosses a molecule or a body
Var. 24canon at the octavea card further away
Var. 25 · minorAdagioread through Focus
Var. 26two manualscrosses a molecule or a body
Var. 27canon at the ninthfarthest card, no bass
Var. 28genre pieceread through Focus
Var. 29two manualscrosses a molecule or a body
Var. 30Quodlibettwo popular atoms over the bass
Da capothe aria again

The real plan of the cycle for this card. The texts are not shown: they exist only while a variation is open

In threes

The first of each three is a genre piece read through one of the twelve Focus structures. The second crosses into another level, a molecule or a body that shares atoms with the card. The third is a canon

Canons at growing distance

The canon at the unison answers with the nearest card, the canon at the ninth with the farthest, and there the bass falls silent

Quodlibet

Variation 30 lays the two most widespread atoms of the library, which the card does not hold, over its bass. Then the aria returns, unchanged and heard differently

Ramon

The Ars of Ramon Llull turned on the archive. Nine latent themes become the alphabet B to K. Three rings turn until two themes and one of Llull's nine questions sit under the index: that alignment is a chamber, and Claude answers it by reading the strongest cards of both themes in full

RAMON Llull's Ars on the archive: nine latent themes as the alphabet B–K, three rings that align two themes and one of the nine questions, Figure T measured between them
BCDEFGHIKBCDEFGHIKBCDEFGHIKT
The alphabet · nine themes of 2026-11-14
BAI Value Through Human Collaboration
CHuman Judgment in Agentic Enterprise
DIntersectional Feminist Standpoint Theory
EPower and Knowledge
FBuilding and Understanding Language Models
GKnowledge, Power and Discourse
HLogic and Philosophy of Science
IAgentic AI in enterprise
KFeminist Theory of Gender
The nine questions, inner ring
B · whetherC · whatD · of whatE · whyF · how muchG · of what kindH · whenI · whereK · how

The nine strongest latent themes of the archive as computed on 2026-11-14, with the names the app gave them

Figure T is measured

Llull's figure of relations is not chosen: concordance, difference or contrariety is read from how the cards of the two themes overlap

A card to validate

The answer to a chamber is a new card, kept or discarded by hand like everything Dephasing generates

The Tabula

A nine by nine table, Llull's Figure A laid flat, marks which pairs of letters have already been asked

Focus

Twelve structures, once called lenses: the same cards arranged in one shape at a time. Each one says what its shape brings out and what it hides, and a suggester proposes the one that fits a question

01
Timeline
Reveals sequence and changeHides simultaneous causes
02
Hierarchy
Reveals order and dependencyHides reciprocity and ambiguity
03
Cluster
Reveals association and similarityHides causal direction
04
Sequence
Reveals stages and transitionsHides feedback and interruption
05
Comparison
Reveals difference and proportionHides shared context
06
Causal chain
Reveals directional influenceHides multiple interacting causes
07
Feedback loop
Reveals recursion and reinforcementHides external conditions
08
Constellation
Reveals distributed relationshipsHides priority and sequence
09
Spectrum
Reveals gradual variationHides category breaks
10
Tension
Reveals opposition and tradeoffHides possible synthesis
11
Layer
Reveals depth and accumulationHides movement across levels
12
Open field
Reveals plurality and emergenceHides stable boundaries

Complexity

Eight environments after the theorists of complexity: criticality, catastrophes, the adjacent possible, levels of learning, canalisation and requisite variety, applied to the index. For now only their interface: the cards and dates are real, the cascades, jumps and distances are illustrative

After Per Bak
Sandpile
Every new card is a grain on the index. Most stay put; now and then one sets off an avalanche of links
After René Thom
Cusp
A theme followed through time on two controls. Where the surface folds it does not drift: it jumps
After Stuart Kauffman
Adjacent possible
The unread books one or two steps from the active themes. The only environment that looks forward
After Gregory Bateson
Learning III
Three planes: the card that adds a fact, the one that changes how a set is read, the one that moves the frame
After Gregory Bateson
Pattern that connects
Two atoms with no card in common, and the form they share all the same
After C. H. Waddington
Epigenetic landscape
The archive as terrain: genres are valleys as deep as their cards, and each new card rolls into one
After Gregory Bateson
Schismogenesis
Pairs of active themes that push each other: both growing, or one rising as the other falls
After W. Ross Ashby
Homeostat
How much variety of genres, authors and types the last week brings, against the whole archive
Top ↑
AI at work

Three models. One division of labour

Who may write what is the central design decision. What stays in the archive is written by one model under strict rules. What gets thrown away may be written by another. What is counted is left to math

Writes what stays

Claude

Reads the full cards and writes everything that enters the wiki and the index

Where
Scheduled routines, and buttons in the app that open Claude with a prepared request
Writes
Cards, units, atoms, molecules, bodies, Umwelten, Partitur, concepts, suggestions
Writes what is thrown away

Apple Intelligence

The on-device model, reached through Foundation Models. Its texts are read and discarded, or become a request to Claude

Where
On the Mac, in Dephasing
Writes
Collision summaries, monster descriptions, theme names
Reads and never writes

BGE-M3

A multilingual embedding model turns atoms into vectors, with Apple's NaturalLanguage as a fallback. Accelerate does the algebra

Where
On the Mac, with vectors cached under a hash of each text
Produces
Maps, neighbours, never-met pairs, broad-atom lists

A day in the system

Two night passes, with separate jobs so that neither writes over the other. A push notification reports anything new

Wiki pass
23:47
  • Absorb new documents and reviews
  • Compile and link wiki entries
  • Write the concept and three suggestions
  • Compile up to three connections
  • Consider one new Umwelt
Construct pass
00:40
  • Process up to 100 cards, in blocks of 25
  • Update units and atoms
  • Try one molecule and one body
  • Propose merges and splits of atoms
  • Write a Partitur version if anything changed
  • Recheck novelty across the board
Co-intelligence
07:00
  • The human reads the concepts
  • Keeps, discards or objects
  • Leaves a note that can seed the next concept
  • Sets the Daring level
  • Approves or rejects atom proposals

How each construct is formed

The rules are short, written down, and enforced twice: once as an instruction to the model, once as a check on what comes back

ConstructModel's taskInputOutput contractMachine check
CardSummarise a source, on request or overnightThe source text, plus web checks for bibliographic dataSummary, Thesis, Structure, Context. Unverified data stay emptyBroken links and orphan entries reported
CoreReduce a whole card to coresThe entire card, never the summary alone. Bibliography, links and sources excluded4 to 6 core of one sentence, 30 words at most. No outside knowledge. No claims about receptionSHA-256 fingerprint of the text read: a date change alone does not re-queue a card
AtomDecide whether a unit says the same as an existing atomThe new unit and the current atomsSame idea, in different words. Same topic is not enough. Title of 9 words at most. A card falls in about 5 atomsValid JSON on re-read, every card must exist. Identity itself cannot be counted
MoleculeFind 3 or 4 atoms no card joins, and theoriseShared atoms and their units. Needs at least 20 shared atomsTheory of 90–110 words as hypothesis, plus a limit. One a nightIntersection of card sets must be empty, combination must be unused
BodyJoin at least 3 molecules in a higher hypothesisMolecule theories and atom unitsTheory of 110–130 words, plus a limit. One a nightNo card or molecule holds the union of atoms. Each molecule adds an atom. Combination unused
UmweltRead at least 3 bodies as one subject's habitatBodies and their molecules. Needs a new body since the last UmweltSubject, Merkwelt, Wirkwelt, theory of 110–130 words, limitNo body holds all the molecules. Each body adds one. Combination unused
PartiturRewrite the single scoreAll Umwelten and what changed that nightA new numbered version: title, 120–150 words, limit, counts. Old versions untouchedWritten only if counts or identifiers changed
ConceptDerive what two or three linked cards imply togetherSummary and thesis of each card, plus the reader's verdicts and notes70–90 words, a question, a web-verified artwork, an experimentSource identifiers must exist, one concept per date
After writing, novelty is recalculated for every molecule and body. A molecule that no longer qualifies is marked “not unprecedented” and stays in the file. Nothing is deleted, because a growing archive can overtake an idea that was new last week.

Daring

One slider sets how far the night may reach. The evidence rules stay fixed at every level: only what the sources say, always a limit, never more than one new element of each kind

The math is on show

The Collider's score is a weighted sum of three distances over data the index already holds4

improbability = 0.62 · (1 − Jaccard of card sets) + 0.20 · (1 − overlap of title words) + 0.18 · (level jump ÷ 4) atom → molecule, no shared cards, no shared words, jump 1: 0.62 + 0.20 + 0.18 · ¼ = 0.865 the app reads 86.5% energy = 0.1 TeV per card behind the two particles; (teraelectronvolt, for flavour:-)

The human stays in the loop

The routines read the human's files and have no permission to edit them

Keep it

The concept becomes a connection, and its themes get deepened

Discard

Its theme and combination of cards are avoided from then on

Object

AI writes the counterpart: the opposite thesis on the same sources

New with Claude

Any level can be requested by hand, under the same rules as the night

Top ↑
Limits

What it does not know

A molecule says that a combination of ideas is missing from the wiki and proposes what would follow. Its worth is settled outside the index, in a conversation or a project

23.0%

Atoms shared by two or more cards

1,435 of 6,245. Most ideas appear in one card only, so the upper levels work on a fraction of the material

1

Step that cannot be counted

Deciding that two units say the same thing is the model's judgment. It is the one link in the chain with no arithmetic behind it

0.7%

Structure in the largest theme

The library is very dispersed and no direction dominates. Themes read as threads, never as categories8

31%

Of true neighbours kept by the map

The flat picture is a rough guide. Reliable neighbours are the ones computed in the full space

Local

Novelty stops at the wiki

An empty intersection says nothing about the wider literature. An unprecedented molecule may be well known elsewhere

258

Cards still silent

258 cards share no atom with any other card, so no combination can reach them yet

Top ↑
Specs

Tech specs

Platform
macOS, native app written in Swift with SwiftUI. Foundation Models requires macOS 26.
Library, 15 Nov 2026
1,561 cards: 1,151 books, 114 papers, 265 articles, 13 media, 18 connections. 9,251 units. 6,245 atoms.
Constructs
33 molecules, 27 bodies, 17 Umwelten, Partitur, 44 concepts, 40 atom proposals awaiting a decision.2
Storage
Markdown vault read in Obsidian. JSON files in a derived index. The app rereads files when they change and writes Markdown pages for each construct.
On-device frameworks
Foundation Models. BGE-M3 through Python (sentence-transformers). NaturalLanguage (NLContextualEmbedding), as fallback. WebKit (the Graph screen). Accelerate (BLAS, Basic Linear Algebra Subprograms; LAPACK, Linear Algebra PACKage; vDSP). Grand Central Dispatch. SwiftUI Canvas.
Methods
Jaccard index. Cosine similarity on centred vectors. t-SNE (t-distributed Stochastic Neighbor Embedding), perplexity 30 for meaning and 15 for cards, 1,000 iterations. LSA with an exact eigen-decomposition via dsyev. FNV-1a cache keys.
Embedding
BAAI BGE-M3, a multilingual model with 1,024 dimensions, run on the Mac through a Python script the app starts when vectors are missing. Each atom is embedded from its title and its first three units (700 characters at most, 512 tokens at most); vectors are normalised, stored as 16-bit floats in one file of the vault, and keyed by an FNV-1a hash of the text, so an unchanged atom is never recomputed. They are then centred on the mean of all atoms and rescaled to length 1. If the script or the model is unavailable, the app falls back to Apple's NLContextualEmbedding, Italian, 512 dimensions. On a trial with 5,201 atoms and 251 typed links, the share of links whose two cards have atoms among each other's eight nearest neighbours was 0.35 with BGE-M3 against 0.24 with Apple's model.
Card neighbours
Cards inherit the embedding of their atoms. For each atom the eight nearest neighbours pass their weight 1/rank to the cards that hold them, and the weights add up per card. A card keeps its six heaviest neighbours that share no atom with it, after leaving out the 30 widest atoms (the largest mean distance of their units from their centre) and the connection cards. The result is one file, rewritten when Atom Space is recomputed, that Propose links and the nightly routine read as a source of candidates.
Night throughput
Up to 100 cards in blocks of 25. At most one molecule, one body, one Umwelt and one Partitur version.
Safety
Copy before every write, re-read after it. State registry with restore. Failed validation restores the backup.
Output
Deep links through a hyperzettel:// scheme. English view through the system translator. PNG export at 1920 × 1080 pixels.
Screens
Cards, Graph, Wanderkarte, Concepts, Explore, Planes, Atoms, Molecules, Bodies, Umwelt, Partitur, Dephasing, plus Doc, a read-only list of source files.

Yours from the latent space,
m{arco|achine}

  1. Counts read from the app's files every night, most recently on 15 November 2026. Raw sources are the converted files in the vault; wiki pages include the card files.
  2. The app counts 40 molecules, 31 bodies and 19 Umwelten, because it includes 7, 4 and 2 distilled ones. This page gives the main chain, as Planes does here.
  3. All screens on this page are re-creations in HTML of the app as it stood on 2 November 2026, except the Graph screen, drawn from the data of 8 November 2026. The counters and the archive totals on the screens are refreshed every night.
  4. The weights of 62, 20 and 18 are design choices meant to make results readable, and they are not estimates. Summaries are Apple Intelligence output and vary in quality.
  5. This demo runs the app's stated rules on an invented pool of genes. In the app the genes are real atoms, molecules or bodies, and at the upper levels the small pool often produces stasis.
  6. The map shown is rebuilt for this page: 2,400 sampled atoms, projected from their cards and genres by SVD and t-SNE. The app computes its own maps from full data, wiki entries included. Figures around the map are the app's.
  7. Theme names and figures come from the app. The positions on this map are drawn for this page.
  8. Share of structure is an eigenvalue over total variance, from the app's run of 8 November 2026.

HyperZettel is a personal project by Marco Formento. Apple Intelligence, macOS and Swift are trademarks of Apple Inc. Claude is a product of Anthropic

Top ↑