Foundation

Zur Sprache dieser Seite. Die Foundation ist der englisch-primaere Kanon des Projekts; sie steht hier im Original. Die deutschen Entsprechungen der Begriffe stehen im Text. Die kurze deutsche Fassung derselben Position: Idee.

Status: Canonical reference. Supersedes earlier AI-language framings of the registry. Scope: All frontend copy, seal axis semantics, and external communication must conform to this document. Languages: Primary English. German equivalents given where the institution operates bilingually. This page carries the canonical English text in both language versions of the site; a German rendering would be a second text, and the canon is one.

1. Core position

The pragmatic level. The question of authorship is not settled on the level where it is sought. It is sought in the wording and the content — what a text consists of and what it means. It is settled one level above: in who does what with a text and what they stand behind. Two texts can agree word for word and differ in everything that follows from them — whether anyone can be questioned, whether anyone retracts, whether anyone carries the consequence. Naming this level does not add a fifth foundation to the linguistic, legal, speech-act, and anthropological layers of this document. It orders them: it states which level is load-bearing.

The shift has a historical occasion. Wording and meaning were always scarce; whoever could produce correct, idiomatic, clearly meant sentences held something not everyone held. That scarcity is gone — such language is now available at no marginal cost, in any quantity, in any register. What is not supplied along with it is the party who means something, addresses someone, retracts, and bears the consequence. Where one level becomes freely available, the weight moves to the next.

On this level, what the substrate lacks is not understanding but commitment. It enters into no commitment. It asserts nothing, because to assert is to expose oneself to accountability, and nothing that has nothing to lose can do so. What accrues without this act is not speech but material — language awaiting an act that is not its own. This formulation is deliberately not a claim about the interior of the model. It holds whether or not language models turn out to possess semantics, world models, or internal representations: a model that understood perfectly would still be a model that asserts nothing. The anti-anthropomorphism of this document therefore takes the form it enters into no commitment, not it has no will.

A language model is accordingly not an artificial intelligence in the anthropomorphic sense. It operates on the surface of language — what Saussure called the signifiant — and in Benveniste’s terms it produces énoncés without performing énonciation, which is the same finding stated on the linguistic level: what is absent is the act, not the meaning. It is a statistically condensed echo of the entire written corpus, made available for re-actualization. That it has no intentionality, no standpoint, no will to utterance is the vivid form of this position and stays serviceable as illustration; it is no longer what carries the position.

This is not a philosophical preference. It is a description of a relation, and it is testable in the only way that bears on the institution: by the practical difference it makes. Where no difference follows in what can be asked, retracted, or answered for, none is claimed. The institution of Maschinenschrift is built on this description and rejects all framings that smuggle a second subject into the writing relation.

The model is therefore not a collaborator, assistant, or co-author. It is a language substrate (Sprachsubstrat): a medium through which the author writes. The relation is closer to that of a writer to a printing press than that of a writer to an editor. The word carries the position more exactly than its choosing knew: a substrate is that on which someone else works — a ground, not an action.

The name Maschinenschrift carries this position by historical inheritance. It denoted, in the twentieth century, the product of the typewriter — text that had passed through a technical apparatus, written by a human, mediated by a machine. The registry extends this lineage. It does not name a new phenomenon. It names the next chapter of an old one.

The diagnostic formula. The project’s working diagnosis of what changes with the substrate is condensed in a single German phrase: Geist aus der Maschine — Geist (spirit, intellectual presence, mind in the Hegelian sense) from the Machine. It inverts the ancient deus ex machina (the external, artificial resolution dropped into a play) and refits it to the present situation: language now arrives with a spirit-effect whose origin is not the person at the desk but a statistical condensation of all prior writing. The seal answers the question the formula poses: in this work, how much of the Geist is from the Machine? An empty seal — the default — answers none. The dots count what was imported.

The delegable mode. This follows from the pragmatic level and is its consequence for where the border runs. The border between unmediated and machine-mediated writing does not run through the material of speech. It runs through its mode. Speech whose effect is settled before its wording — the mode Habermas calls success-oriented, and whose limit case Frankfurt describes as indifference toward what is said with full attention to its effect — requires no one who means anything. That is why it, and only it, can be handed to an apparatus without remainder. The substrate does not calculate: it pursues no interest and orients itself by no success. Calculation is a human mode, and its delegability follows from its emptiness, not from any capability of the machine. What is not delegable is the point at which someone stands behind a statement. The seal records the first and presupposes the second. This is also the anthropological ground of the refusal to detect: a procedure that machine-verifies whether a human was at work has already ceded the question to the mode it opposes.

2. What this replaces

The following vocabulary is rejected because it imports an anthropology the substrate cannot bear:

All occurrences of these terms in frontend and documentation are replaced according to §3.

3. New vocabulary

The institution’s working vocabulary, in both languages:

4. The three sense dimensions

Following Luhmann, every sense-processing operates in three dimensions. The substrate alters each.

Time (Zeitdimension)

Writing was historically a slow art. Brooding, incubation, layered revision over weeks. With the substrate, the genesis-time of text contracts toward zero, while reception-time remains unchanged. This produces a new asymmetry: rapid production, slow reading. The substrate itself carries a frozen past — its training corpus has a fixed cut-off, and what it mirrors is the written record up to that date. The author writes in the present, against an echo from a closed past.

Implication for the registry: the registry’s construction is a deliberate counter-movement. Where the substrate accelerates production, the registry is built to be hard to lose — plain-text data, no database anyone must operate, no accounts anyone must maintain, licenses that let another party take it over. This is not a technical detail. It is a temporal stance.

The stance is stated as a construction, never as a term of years. An earlier formulation promised the registry for fifty years; that promise is withdrawn. No institution of one person can honour a term it will not outlive, and the architecture is checkable where the promise was only assertable. Duration is claimed by how the thing is built, not by how long it says it will last.

Social (Sozialdimension)

Writing has always been socially structured: author and reader, author and editor, author and publisher. Each role had a position and a name. The substrate fits none of these. It represents an aggregated anonymity — the statistical condensation of all authors in its training corpus, none of them individually identifiable. The reader can no longer ask “who wrote this?” and receive a single answer.

Implication for the registry: Maschinenschrift is the institutional form that fills the social gap the substrate leaves. It does not restore the old categories of authorship. It creates a new one: the declared, catalogued mediation.

Subject matter (Sachdimension)

The substrate knows linguistic patterns, not matters. It fills gaps by linguistic probability, not factual accuracy. This is why hallucination is not a bug but the structural consequence of pure language-modeling. The older form of this finding — that form can be delegated to the substrate and substance cannot — is ordered under §1 and now reads more precisely: what is not delegable is not a class of content but the commitment to it. Wording and content are both available from the substrate, in any degree; what never comes with them is anyone who answers for them.

Implication for the registry: the seal’s three axes (W / G / B) are the direct structural answer to this asymmetry. They are not arbitrary categories. They follow from what the substrate is and is not — and their number follows from §1: what someone stands behind is not a further axis, because it cannot be imported.

5. The iterative writing process

Writing with the substrate displaces the iterative process from inside the author to a materialized exchange. Each iteration is now visible as text: prompt out, response in, judgment, modification, repeat. Three consequences follow.

The prompt becomes a writing act. The author must now articulate intent precisely enough to elicit a useful actualization. Vague prompts return vague text. Precision in instruction precedes precision in output.

Authorial position clarifies against the average. The substrate returns the statistically likely phrasing. The author’s own voice becomes legible by difference from this average — by what is rejected, modified, sharpened against the returned text. Selection becomes a constitutive act of authorship. Selection is not only rejection of the average; it is also recognition of what the author already held, articulated through the iteration but not contributed by the substrate.

The final text is a cut through a much larger field. Before the substrate, a final text was one of a few drafts. Now it is one of potentially thousands of actualizations that the substrate could have produced and the author rejected, modified, or accepted. Authorship shifts from production toward curation.

The catalog does not document each iteration. It condenses the cumulative mediation into the two mediation axes, W and G.

6. Pain points addressed and not addressed

The substrate produces a landscape of concerns. Maschinenschrift addresses some and explicitly does not address others. This limitation is intentional and gives the institution its credibility.

Addressed:

• Authenticity (Echtheit) — through self-declaration
• Mediation (Vermittlung) — through the seal’s two mediation axes, which make mediation legible
• Voice (Stimme) — indirectly, by making mediation visible and thereby making the unmediated also identifiable

Not addressed (and not claimed to be):

• Copyright of training data
• Concentration of linguistic infrastructure in few corporations
• Ecological footprint of model operation
• Cognitive atrophy from habitual delegation of formulation
• Truth and hallucination at the level of individual factual claims

The frontend does not promise more. Honesty about scope is part of the institution.

7. Seal axis semantics

The seal carries three axes. Their typographic form (square brackets, five slots, three rows) is unchanged. Their semantic content has passed through four states and thus three refinements: from the early triple (K/T/R) to (F/S/V), to (D/G/K) on 2026-05-24, and, as of 2026-08-02, to the present (W/G/B) — Wortlaut, Geist, Beleg (glossed in English contexts as Wording, Geist, Record; the letters follow the German canon, as Geist always has — see §10 on Geist as untranslated loanword). The numeric values carry across unchanged: a D 4 declaration is a W 4 declaration.

The axes measure substrate mediation — what the machine contributed. The human is the unmarked default position: an empty seal [ ] means the work was produced by a human alone, without substrate intervention. Dots mark where the Geist aus der Maschine enters: the more dots, the more the spirit of this work was imported from outside the writing person. This polarity preserves the original Maschinenschrift visual convention and is consistent with §1’s anti-anthropomorphism: the substrate is not someone whose presence we must positively detect; it is a medium whose traces we mark.

The earlier names Form and Substanz/Substance were retired because they described what the axis covered abstractly while the values measured something more specific — the linguistic surface (a matter of diction) and the contribution to the work’s spirit (German: Geist). The earlier name Verifikation/Verification was retired because, once Maschinenschrift was established as register-keeper rather than assay office, verification falsely implied a checking authority. Since Spec v8 the B-axis records whether the work’s factual statements were checked against sources outside the substrate — a statement of the author’s, not of the registry, so the objection to Verifikation does not reach it.

The numerical scale (0–5) and the conversion from earlier declarations remain unchanged across every rename: an F=4 declaration became a D=4 declaration and is now a W=4 declaration; an S=2 declaration became and remains G=2. Only the axis labels change.

Axis W — Wortlaut (Wording)

Scale: 0–5 Measures: degree of substrate contribution to the linguistic surface — idiomatic phrasing, stylistic register, syntactic flow, sentence-level formulation.

Axis G — Geist (Spirit, in the sense of esprit / Hegel’s Geist — intellectual presence, not religious essence)

Scale: 0–5 Measures: degree of substrate contribution to the work’s spirit — ideas, arguments, structural decisions, factual material, content. This is the axis on which the contemporary AI debate centers: Geist aus der Maschine — how much of what appears as the work’s mind was imported from outside the writing person?

Canonical reading of the axis name. Geist sounds mentalistic but measures a question of provenance and vouching: how much of what the work holds out as its own came from a source that stands behind nothing. The axis is subject-indexed — for a private author the question reads does this Geist mean it?, for an institution does this body bind itself to this? — and what it counts is import from a source to which neither question can be put. Read this way the axis requires no access to anyone’s inner states, and the standing objection against arguments from authorial intent does not reach it: the axis does not record what was meant, it records who vouches. The name is retained; this reading is canonical.

What the scale cannot hold. The five values count how much was imported, not in what manner the author stood in the exchange. Two works can carry the same G value and mean different things by it: one where the substance arrived as proposals and the author chose among them, one where the substance arrived the same way but was set in motion by the author’s own positings and governed by rules the author had laid down. The axis has no slot for that difference. The narration field does, and this is one of the reasons it exists. Read the number as the measure of import and the narration as the account of what the author held against it. Declarants should not move a value up or down to compensate for something the narration is there to say.

Genre conditionality. The interpretation of Geist (and Wortlaut) is genre-conditional, and the condition is one of commitment, not of content class. The truth register changes the object of what the author stands behind, not the fact that they stand behind it: in fiction the author does not vouch for factual accuracy but for the text as a work; in non-fiction for both. The same value therefore reads differently in the two registers — where no claim on the world is made, imported substance touches nothing the author has undertaken to answer for; where one is made, it touches all of it. Readers should interpret W/G values in light of the declared genre field.

Axis B — Beleg (Record)

Type: binary Measures: whether the work’s factual statements were checked against sources outside the substrate.

What the axis is for. The substrate operates coherentially — text-with-text, no access to a referent. Confabulation is therefore not a defect of that mode but its normal condition wherever nothing outside holds it. B is the axis on which the author states whether anything outside held it. This is the point at which Maschinenschrift restores a controlled, narrow correspondence claim: not that what the work says is true, but that someone put it against the world.

ausstehend is neither an accusation nor a confession of error. It says that the factual layer has not been set against anything outside the machine, and that a reader who needs it to be must do that work themselves. It is the more common state, and declaring it costs nothing.

Independent means outside the substrate, not outside the author. A check is independent when it runs against something the model did not produce — a source, a document, a measurement, a person who knows. It does not require a third party to certify it. An author who verified her own citations has checked independently in the sense of this axis; an author who asked the model whether the model was right has not.

B is a self-declaration, like W and G. All three axes are statements of the declaring author, and their truth is the author’s responsibility. The catalog confirms nothing on this axis. Maschinenschrift is a register, not an assay office: nothing here makes the institution a checking authority, and nothing here makes it liable for the accuracy of anyone’s facts. This is a change from the previous semantics, under which B was the one value the registry could confirm. It is also a simplification — the seal is now a single speech act throughout.

Admission is not attestation. The plausibility threshold the catalog applies at the point of admission is untouched by this: before an entry is taken up, the declaration is checked for whether it coheres with the work it describes — whether it holds together in itself. That is a curatorial threshold for entry into a visibility layer, not an attestation of the values. Admission says the declaration does not fall apart; it does not say the values are right, and it never makes them the institution’s statement. The procedure itself is published on the submission page and is the governing form of it.

B and G are orthogonal, and both are needed. G counts how much of the factual material came from the substrate. B says whether it was checked against anything else. A work can be high on G and belegt — the model supplied the material, the author verified it. A work can be low on G and ausstehend — the author gathered the material herself and checked none of it. The pair is what makes the declaration informative: the number says how much of the work’s factual layer arrived by a coherential route, the binary says whether that route was ever corrected.

Genre. The axis reads across registers, not only in non-fiction. A novel has a factual layer — places, procedures, chronology, terminology — and it is exactly the layer a substrate fills most confidently and least reliably. Where a work makes no claim on the world at all, nothing stands unchecked and the axis reads belegt; the marker is a warning, and no warning is warranted where there is nothing to warn about.

Bibliographic findability is an admission criterion, not an axis value. Until Spec v8 the B-axis notated whether the declared work stood in the bibliographic world — findable, datable, identifiable. That requirement has not been dropped; it has moved to where it was enforced anyway. The submission procedure requires at least one platform on which the work is publicly findable, and an entry that does not meet it is not taken up. The seal no longer spends a slot on a condition every catalog entry satisfies by construction.

On the name. The axis has carried three names. As Verifikation (until 2026-05-24) it implied a checking authority, which Maschinenschrift is not — and the present semantics does not reinstate one, because the statement is the author’s. As Korrespondenz (until 2026-08-02) it stood in a known tension with the pragmatic ordering of §1, because accounts of truth as correspondence are the classical foil for pragmatist positions; that tension does not return either, because B records that a check was made, not that a proposition matches the world. Beleg fits the present semantics better than it fitted the previous one: a belegte Behauptung is a claim with something behind it. The letters W/G/B are unchanged and remain the terminal set.

The dagger is the pending marker, realized on two layers: on the seal glyph as a right-aligned half-stroke per the canonical seal specification (Spec v8; the geometry has been unchanged since v5 — the axis names were renamed twice, and the B-axis semantics once), and on the textual identification as the literal † character. Both layers carry the marker independently — readers who see only the glyph identify ausstehend by the half-stroke; readers who see only the identifier identify it by the dagger.

8. Frontend copy directives

The following replacements apply across all pages, both German and English.

Exception, as of 2026-08-02: the machine-readable meta layer — <title>, meta description, Open Graph, Twitter Cards — is exempt from this section and carries the common terminology (KI, AI, KI-Kennzeichnung, AI disclosure, KI-Anteil). That layer addresses the searcher in the words they type; the page they arrive on is canonical. Everything rendered on the page — headlines, body copy, navigation, seal labels, interface strings — follows this section unchanged. The phrase “literature written with large language models” / “Literatur, die mit Large Language Models geschrieben wurde” is admissible everywhere, body copy included.

Page-level headlines

Headlines only; for the meta layer see the exception above.

Old (DE): “Selbstdeklaration KI-gestützter Literatur” New (DE): “Selbstdeklaration schrifttechnisch vermittelter Literatur”

Old (EN): “Self-declaration of AI-assisted literature” New (EN): “Self-declaration of machine-mediated literature”

Section titles

Axis names

Numerical values from earlier declarations transfer unchanged through both renames: F=4 → D=4 → W=4, S=2 → G=2. Only the label and its semantic framing change — the scale direction (high = more substrate mediation, low = more human) is preserved.

Body copy patterns

Wherever the body copy currently describes “AI” or “the AI”, replace with das Sprachsubstrat / the language substrate. Wherever it describes “AI-assisted writing”, replace with maschinell vermittelte Schrift / machine-mediated writing. Wherever it describes “collaboration with AI”, replace with Schreiben durch das Sprachsubstrat / writing through the language substrate.

Catalog pages

Entry detail pages show the three axes as W, G, B (Wortlaut / Geist / Beleg). The German term Maschinenschrift remains untranslated in both language versions of the site — see §10. As of Spec v7 the axis names follow it: they stay German in both languages and carry an English gloss on first use (wording, record); Geist is glossed but never replaced. The YAML field names in the data layer are the same German words (wortlaut, geist, beleg) — the field names are identifiers, not display text.

9. Data-layer directives

The schema below is the current, intentionally minimal canonical form. No timestamp register, no backend, no submission endpoints.

• YAML primary store: entries use wortlaut:, geist:, beleg: keys. The earlier diktion: / korrespondenz: keys, the form: / substance: / verification: keys, and the pre-Foundation k_value / t_value / r_status columns are all retired.
• Validation constraints: wortlaut and geist each accept integers 0–5; beleg accepts the strings belegt (documented) or ausstehend (pending).
• Migration of existing values: any entry declared under an earlier spec keeps its numerical values (F=4 → D=4 → W=4, S=2 → G=2). Field names migrate form → diktion → wortlaut and verification → korrespondenz → beleg; the verification value verified maps to beleg: belegt, unverified to beleg: ausstehend. Scale direction is preserved — only labels change. Entries still carrying an earlier key are normalized to the v7 schema at the rendering layer; no display path reads an axis key directly.
• Field name language: keys remain German across language versions of the site. They are identifiers, like Maschinenschrift and Geist themselves (see §10). The display strings are not translated either: the axis names stay German in both languages and carry an English gloss on first use (wording, record), and Geist is never replaced — §7 and the Spec are the governing form of this rule.
• Semantic migration of the B-axis (Spec v8, 2026-08-24): the field name, its two values and the glyph are unchanged, but what they assert is not. A belegt declared under v5–v7 asserted bibliographic findability; a belegt under v8 asserts that the work’s factual statements were checked against sources outside the substrate. The two do not carry across. Entries declared belegt before 2026-08-24 must be re-declared on the new question; entries declared ausstehend keep the marker but change its ground, and should likewise be reviewed. Numeric W and G values are untouched. The pre-v6 value verification: verified addressed the same question v8 asks and is the one historical value that maps directly.
• Dagger notation: the † glyph is applied to the textual axis line at the rendering layer when beleg: ausstehend. It is not stored in data.
• No genesis schema. Fields for filing dates, publication dates, lead times, or manuscript states are not introduced. Authors who want cryptographic timestamping of manuscript states should use OpenTimestamps independently of Maschinenschrift; the resulting .ots proofs remain with the author.
• declared_at remains the single temporal field — it records the day a work entered the catalog, derived from the Git commit history. It is not a priority anchor; the project does not make priority claims.

10. German terms in the English context

Two German words are not translated in English material — they enter the project’s English vocabulary as proper terms, the way Zeitgeist, Schadenfreude, or Bildung have entered English.

Maschinenschrift

The word Maschinenschrift is not translated. It remains German on the English site, in English documentation, and in English press materials. The German compound holds the entire relation in a single noun: the human writes, the machine mediates, the trace of both is in the script. English has no equivalent. Typescript names the artifact, not the practice. Machine writing is literal and misleading. The German word also takes its place in a tradition of compound nouns for institutional forms of writing — Schrifttum, Schriftverkehr, Hochschrift — that English approaches only through periphrasis.

Treated as a proper noun in English contexts, capitalized but not italicized after first use. First use in any English document should be set in italics with a brief gloss: Maschinenschrift (German: machine-script).

Geist

The name of the G-axis is not translated either. Geist carries weights that English spirit loses (religious overload) and English mind loses (Hegel, Geisteswissenschaften, esprit-tradition). The project’s diagnostic formula — Geist aus der Maschine (see §1) — is most precise in German, and the axis label should match the formula.

Treated like Maschinenschrift: first use in italics with brief gloss, Geist (German: spirit / intellectual presence; cf. Hegel’s Geist, Geisteswissenschaften). After first use, plain noun. An occasional English gloss Ghost from the Machine may accompany the formula for readers familiar with the Ryle–Koestler–Shirow lineage.

11. Inaugural entry

The inaugural entry of the Maschinenschrift catalog is La Scuola dei Silenzi by I. O. Vargas (slug: la-scuola-dei-silenzi). This entry establishes the methodology in practice. Its declared values are W=5 / G=3 / B=ausstehend — comprehensive substrate contribution to the wording, substantive contribution to the spirit, and a factual layer that was never set against anything outside the substrate. The narration says so in as many words: research material was structured through the substrate, not verified.

The entry has been declared twice. The first declaration (2026-06-01) read D=4 / G=2 / K=ausstehend and was identical, in numbers, with the pre-Spec-v6 form F=4 / S=2 / unverified: through both renames the values carried across untouched. The values changed for the first time on 2026-08-02, when the finished work was re-assessed against its own process — W to 5, G to 3. A rename never moves a value; only a re-assessment does. On 2026-08-24 a third event occurred that is neither: the B-axis changed what it asks, and the entry’s value stayed ausstehend — under the old question because no listing existed at the time of declaration, under the new one because the research was never checked. All three events belong in the record, and the catalog entry names them.

Earlier drafts of this document anticipated a numbered registry with identifiers of the form MS-YYYY-NNNN. The current architecture — an open standard with a curated catalog rather than a centralized backend registry — addresses works by slug. The textual dagger notation defined in §7 and §9 is not dormant under this scheme: the catalog listing sets the † after the textual axis line of every entry declared beleg: ausstehend. It stands beside the glyph, not inside it, and is therefore untouched by the negative list of the seal specification, which forbids additional marks on the seal.

Companion documents

This Foundation is the linguistic canon. Two companion documents complete the constitutional layer of Maschinenschrift: the legal canon (the CC BY 4.0 architecture, the trademark posture, the anti-capture rationale) and the speech-act canon, which establishes the declaration as a Transparenz-Bekenntnis — assertive with a commissive overtone, not third-party attestation — and the institution as register-keeper rather than assay office. Both are working documents and are not published.

The three are disjoint truth-layers; they remain consistent with each other. Precedence, where they touch (notably the B-axis in §7 above): each layer governs its own subject matter — this Foundation the language and the axis semantics, the licensing strategy the legal construction, the self-declaration paper the speech-act classification and the role of the institution. Outside its own subject a layer describes; it does not decide. Where two layers decide the same question, the most recent constitutional clarification governs the reading. A derivation that a later clarification has withdrawn does not survive in the layer that first carried it.

Stand: 2026-08-11. This document is the canonical source for all language and seal-semantics decisions of Maschinenschrift. The seal geometry is specified separately in Spec v7; the shorter account of the same position is under Idea.