Tried by the Flow — Architecture Volume

Note on the translation — This translation re-fixes the source text without displacing its substance. A few terms of the corpus have no obvious equivalent in English; the choices made, and how they are to be read, are set out in the section “Translation choices” at the end of the volume. This translation is not frozen: it will evolve with readers’ feedback and with the improvement of the means of translation. The French version is authoritative.

§0. Threshold — The Status of This Volume

The tomes speak the language of the canal; this volume speaks the language of categories. Both say the same substance.

This volume is neither a summary nor a fourth tome. The synthesis of the first edition condensed the narrative; the Architecture Volume assembles the systematic register: the theses, the diagram, the families, the layers, the minimal ontology, the formal machinery, the map of neighbours, the two formulas.

Its rule of circulation is that of the whole edition: one contribution, one locus. Each contribution has its single place of demonstration in the tomes; here it is stated, connected and, when necessary, formalized. The cross-references indicate where the demonstration lives.

The formal machinery lives here and nowhere else: the state vector, the operational times, the persistence package, the formalization of over-throughput. In the tomes: zero occurrences for the state vector and the persistence package; a single mention, in prose, for the operational times. The long synthesis formula closes this volume; it never appears in the tomes.

§1. The Twelve Theses

The backbone of the edition. Each thesis is stated here; its demonstration lives at the locus indicated. Three movements follow the dramaturgy of the trilogy — the spring, the torrent, the reader.

First Movement — Constitution

Thesis 1. Nothing is by nature a datum under every description. The status of datum is constituted within a regulated practice. Cross-reference: A1 — T1 “Preface to the Edition” + “Part I — The Two Ontologies of the Datum”; support A5. Against the datum by nature, without falling back into arbitrary decree — constitution is regulated and imputable, carried by the three intentions rather than by an act of will at each inscription. The entity is processually constituted, historically individualized, relationally maintained.

Thesis 2. Every empirical datum is constrained by a material that resists it, but this material never alone determines its description. Cross-reference: B9 — T1 “The Flow and the Annotation”; support A3. A double safeguard. Against arbitrariness: the material resists; not every description can be sustained. Against naive realism: to annotate is to choose — the real relevant to an observation always exceeds the cuts it produces, and the undrawn is part of the gesture.

Thesis 3. The trace is neither transparent nor necessarily univocal. It itself results from a process of formation and preservation. Cross-reference: A4 — T1 “The Trace and the Reader”, pillar I. Traceability is differential — causal marking, preserved trace, accessible trace, interpretable trace — and equivocation remains possible: different processes leave similar marks. “No knowledge has any route other than the trace” stands; “everything is legible, everywhere” falls.

Thesis 4. Inscription does not capture a bare value. It stabilizes a result produced by an operation, with a regime of uncertainty or an evidentiary regime. Cross-reference: A3 — T1 “Part IV — The Act of Annotation” + “The Two Times of the Datum”. The result carries its operation — object aimed at, rule, instrument, model, uncertainty. Hence the divide between the times: the documentary fact (this apparatus produced this result) is irreversible; the assertion about the phenomenon remains revisable.

Second Movement — Identity and Persistence

Thesis 5. Every occurrence is material, but the identity of a datum is not that of a particular medium. Cross-reference: A6 — T2 “The Properties of Movement”; echo T1, duplication. Material dependence without identity of medium. The datum of origin is identified by its act — which drawing, on which fragment of the real, under which criteria — not by the number of its carriers. Occurrences are counted; the datum remains one: copying without division, seen from the medium.

Thesis 6. Identity is historical. It rests on origin and provenance; the copy multiplies occurrences, transformation engenders a filiation. Cross-reference: A6 — T2 “The Properties of Movement”; support A3 (inscribing with its origin); PROV model in the formal machinery. The individuation bundle — operation of constitution, object aimed at, relevant time, result, provenance. Two recordings of the same value may be two data; two different files may realize the same datum. Provenance is constitutive, not decorative; for a derived datum, it includes the model and the parent data.

Thesis 7. To persist requires more than to survive materially. Depending on the objective, integrity, authenticity, representation, context and possibility of access must be maintained. Cross-reference: B2 — T1 “The Two Times of the Datum” for the images; formal demonstration here, §4 (the persistence package) and §6 (the state vector). A durable datum is not a value-medium pair but a stratified package intended for a community that will know how to read it; the loss of a layer does not always destroy the whole, it reduces what the datum can sustain. The two deaths remain, in the tomes, the images of the typical collapses; universal death is a limit-ideal.

Thesis 8. Freshness is not age, nor a verdict of use. It is the recency of the act: the proximity of the act of inscription to the present of the world, which can never be recovered. Cross-reference: B3 — T1 “The Two Times of the Datum”. Lapse, once decomposed, becomes validity for a use — a function of the object, the question, the instant, the tolerance. The same inscription may be stale for the alert and accurate for history; the statement concerning the past does not age like the one that claims to tell the present. Freshness, for its part, is observed: the recency of the act, the same for all uses — what varies from one use to another is the verdict of validity, a function of the object, the question, the instant and the tolerance.

Third Movement — Circulation, Reading, the Torrent, the Loop

Thesis 9. Circulation can preserve, alter or engender. It must be assessed by the invariants it claims to maintain. Cross-reference: A6, B4, B8 — T2 “The Properties of Movement” + “The Act of Exchange”. The five identities — origin, content, encoding, binary, filiation — give the grid: the faithful copy maintains, conversion displaces, derivation engenders. Exchange extends availability and the possibility of use; it accomplishes nothing more. And resumption at the other end has no medium — it has conditions.

Thesis 10. Reading actualizes without constituting the datum anew. It is situated, selective and costly; it can produce new derived data. Cross-reference: A2 — T3, document 1 (+ document 4); supports C2 “The Situated Reader”, B5. The status of constituted datum does not depend on rereading; actualization does — it is this, and this alone, that reading adds. The scale replaces falling back into trace everywhere except in the limit case. The reader is finite and situated — the window is multi-constrained — and the machine that processes occupies one more situated position. When reading re-inscribes, it opens a filiation: the meaning produced can become a datum again.

Thesis 11. The torrent is a regime of mismatch between production, selection, processing and deadline, liable to be amplified by sociotechnical loops. Cross-reference: C1, B6, B7 — T2 “The Torrent”; T3 “The Over-Plus”; formalization here, §6. An incompatibility of cadence, not the goodness of the arrivals — distinguish raw, relevant, priority, processable, read throughput. The seven outcomes of the queue are not ontologically equivalent: a datum in waiting does not become a trace, a rejected one remains a datum, one expired for a use may serve another. No exhaustive classification at finite cost — but windows, summaries, priorities, partial archives: what framing and delegating organize. The rule that turns the raw into the presented is a decision — political in the proper sense.

Thesis 12. The process is reflexive. Data contribute to transforming the real, the categories and the infrastructures that will produce the next data. Cross-reference: A8 — T3, closing; supports T1 Part VI, T2 “Self-Amplification”. The chain closes — action, modified real, new traces, new regime of observation — and the loop can stabilize or amplify; delays can make a decision correct at calculation and harmful at application. The passage from reading to action requires the bridge observation / inference / intervention: a prediction is not a fresh datum about the future. Ontology becomes responsibility of the chain — each link answers for its gesture and passes on to the next what it needs in order to answer; provenance is its documentary form, decomposed probity gives its seven places.

The word processual thus qualifies first of all the mode of constitution and of relational existence of the datum, not the obligation to classify the datum itself as an event. Explanatory priority goes to the process — over the identity, persistence, circulation, actualization and value of the datum.

§2. The Consolidated Looped Diagram

The guiding movement of the trilogy, with its mediations, and its closure: from the trace to action, and from action to new traces.

flowchart TD
    R["Process or event of the real"] --> M["Causal marking or interaction"]
    M --> T["Preserved and accessible trace"]
    T --> O["Situated observation: object, rule, instrument, model"]
    O --> D["Datum of origin: result, time, uncertainty, provenance"]
    D --> P["Persistence package: occurrence, code, context, fixity"]
    P --> X["Copy, transport or transformation"]
    X --> A["Occurrence accessible to a reader"]
    A --> L["Situated reading and interpretation"]
    L --> I["Inference, decision or new inscription"]
    I --> U["Action on the real"]
    U -- "modified real, new traces" --> R

The diagram indicates logical dependencies, not a compulsory chronology. An institutional datum is constituted by an act that institutes rather than by a measurement; a derived datum starts from other data rather than from a primary trace; a copy can circulate without ever being read; a reading may lead to no action (§3).

The loop can be stabilizing or amplifying. Delays between observation and action can produce oscillations, or make a decision correct at the moment of calculation and harmful at the moment of its application; a policy based on data sometimes modifies the very behaviours that produce those data. The passage from reading to action further requires the bridge of the three relations to the real — observation, inference, intervention (thesis 12; locus: T3, closing).

The torrent itself can arise from an endogenous loop: more data make more automation possible; automation multiplies capture points and decisions; these decisions modify the real and produce new traces; these traces become new data; the increase in the flow justifies still more automation (locus: T2, self-amplification).

§3. The Three Families of Constitution

Three ways for a datum to be born; the measurement drawing ceases to be the implicit prototype of every datum. Locus: A7 — T1, constitution; echoes T2 and T3.

Observational. Event or state → interaction → trace → measurement or description → inscribed result. Regimes: physical measurement, biological observation, testimony, photograph, survey. An interaction leaves a trace; the act constitutes it as a result under criteria.

Institutional. Social or legal act → category and constitutive rule → recording. The inscribed marriage, the counted vote, the registered title, the closed balance: under the constitutive rule, the inscription does not describe an independent phenomenon — it takes part in bringing into existence the fact it registers. Its truth is not checked against the world; it is contested before the rule and the authority that instituted it. Its provenance names the act, the rule, the institution.

Derived. Parent data + model or operation → transformed result → new provenance. A genus of which the corpus holds three species — the consolidated, the verisimilar, recounting in the technical sense — and which the family extends: correction, translation, aggregation, annotation, classification, estimation, prediction, synthetic datum. Neither direct drawings nor faithful copies: every derived datum opens a filiation, and its provenance includes the model and the parent data; translation belongs to it only if a declared invariant is altered — if the invariants are preserved, it is a faithful conversion, a new occurrence of the same datum —, the operation and the source text being inscribed in the provenance in both cases.

Real chains braid the three families together: an observational entry goes into a derived consolidation that grounds an institutional act, whose recording will feed other derived data.

§4. The Layers of the Datum

A durable datum depends on a stratified set:

1. content — value, symbol, proposition or structure;
2. occurrence — determinate material realization;
3. encoding — rules that map material states to symbols;
4. schema — organization of fields and relations;
5. provenance — origin, agents, activities and transformations;
6. context — object aimed at, units, population, method, purpose;
7. representation information — what a community must know in order to interpret;
8. access conditions — location, rights, interfaces, technical means;
9. validity regime — question, instant, tolerance and use to which the result can respond.

It is here that the persistence package takes shape: what a durable datum is — not a value-medium pair, but a package of these nine layers, intended for a community that will know how to read it. The archival science of digital preservation established its principle; the corpus adopts it as its own.

The loss of a layer does not always destroy the whole; it reduces what the datum can sustain — and makes it go down the scale, from accessible to intelligible, down to the mute datum and, at the limit-ideal, the dead one. Cross-references: conditions of interpretive resumption, locus T2 “The Act of Exchange”; scale, locus T3, document 1.

§5. The Minimal Ontology

To make explicit the categories and relations without which statements remain ambiguous. A sober register — prose and tables; the aim is not to produce a computing ontology, but to fix what the tomes demonstrate.

Primitives

Necessary Relations

• causedBy(T, R) — the trace is causally linked to the process;
• uses(O, T) — the observation mobilizes a trace;
• generates(O, D) — the operation constitutes a datum of origin;
• realizes(C, D) — a material occurrence realizes this datum;
• encodedBy(C, E) — the occurrence uses an encoding;
• derivedFrom(D₂, D₁) — a datum descends from another;
• copyOf(C₂, C₁) — an occurrence was produced as a copy of another;
• accessibleTo(C, r, t) — an occurrence is accessible to a given reader at a given moment;
• interpretableBy(D, r, t) — the reader has the resources to interpret it;
• readAt(D, r, t) — a reading has actually taken place;
• about(D, R) — the datum claims to concern a process or a state;
• validFor(D, Q, t, ϵ) — the datum is acceptable for a question, at a moment, according to a tolerance;
• affects(U, R) — an action modifies the observed process or its environment.

Criteria of Identity

A datum of origin is individualized by a minimal bundle: operation of constitution + object aimed at + relevant time + result + provenance. Two occurrences may realize the same datum. A transformation produces a new datum if it significantly changes the content, the epistemic status, the object aimed at or the validity regime; the boundary between faithful conversion and derivation is not natural — it is defined by the invariants that the practice intends to preserve.

Five identities settle the cases: identity of origin (same documented event of constitution); identity of content (same logical content according to a given criterion); identity of encoding (same structured representation); binary identity (same sequence of bits); continuity of filiation (documented relation of transformation). “The two copies are the same datum” becomes defensible if the sentence means: two occurrences realize the same datum of origin; it becomes false if it claims that there exists only one material object.

The Five States

Relative to a reader and to an instant, cumulative. Locus: A2 — T3, document 1.

• constituted — a result has been produced and stabilized according to a rule; its existence owes nothing to reading;
• accessible — an occurrence can be reached by this particular reader;
• intelligible — the reader or their community possesses the resources of interpretation; unintelligible for this particular reader, the datum is that reader’s mute datum; unintelligible for all, at the limit-ideal, the dead datum;
• actualized — an actual reading has produced a meaning from it in a situation;
• operative — this actualization has contributed to an inference, a coordination or an action.

A datum can be constituted without being accessible, accessible without being intelligible, intelligible without being read, read without nourishing anything — and used without being valid.

§6. The Formal Machinery

The formalisms live here; in the tomes, the prose of the canal.

The State Vector

The state of a datum, for a reader r, an instant t, a use u, decomposes into seven independent components:

S(D, r, t, u) = (E, F, I, A, X, G, V)

• E — existence of at least one occurrence;
• F — fixity of the occurrence examined;
• I — integrity of the content;
• A — authenticity and continuity of provenance;
• X — accessibility for the reader r at the time t;
• G — intelligibility for this reader or their community;
• V — validity for the use u.

Each component evolves without dragging the others along. The two deaths remain, in the tomes, the images of its typical collapses; universal death is a limit-ideal — rigorously, one speaks of current unintelligibility for a defined community, of documented loss of representation information, of practical undecidability of reconstruction. Locus of the images: B2 — T1 “The Two Times”; T3, document 1.

The Operational Times

The operational deployment of the two regimes of time. Locus of the regimes: B1 — T1 “The Two Times”.

Time of the event — when what is represented occurred; of observation — when an interaction with the phenomenon took place; of inscription — when the result was fixed; of processing — when it was transformed or aggregated; of transmission — when it left a node; of reception — when an occurrence became available elsewhere; of reading — when a reader interpreted it; of action — when it contributed to modifying the real.

Coincident in the simple apparatus, divergent in real infrastructures: causal order is not to be confused with the order of clocks, and the time of the process is documented — it is not assumed to be single. The dating of an operational time refers to a reference source or version, to an instant and a clock, to an expected level of consistency; freshness is the recency of the act thus dated (thesis 8), and validity for a use remains a function of the object, the question, the instant and the tolerance.

The Formalization of Over-Throughput

Over-throughput is formalized on the basis of: λ, arrival rate of the items liable to be processed; μ, effective capacity of processing or reading; B, stock in waiting; W, mean waiting time; d, deadline beyond which an item loses its value for use; p, policy of selection, priority, aggregation or rejection. When arrivals durably exceed service — λ > μ — the queue grows if no mechanism of rejection, synthesis or reduction intervenes; in the field of queues, Little’s law, L = λW, relates the mean stock, the arrival rate and the sojourn time, averages taken over a long period, in a stationary regime and under conservation of units; its application to the flow of data is a structural analogy, the phase of growing over-throughput, non-stationary, being outside its domain, and no quantitative value is derived from it.

Seven outcomes of the queue, ontologically distinct: accumulated delay; expiry in the queue; explicit rejection; silent loss; aggregation; prioritization; delegation. A datum in waiting does not become a trace; a rejected datum remains a datum; a datum expired for a use may remain useful for another.

Five throughputs not to be confused: raw, relevant, priority, processable, actually read. The rule that transforms raw throughput into the throughput presented to the reader is a decision — political in the proper sense. Loci: C1, B6, B7 — T2 “The Torrent”; T3 “The Over-Plus”.

§7. The Map of Neighbours

Absence found is not proof of absence: a search can miss a vocabulary, a language, a neighbouring discipline, an unindexed chapter. Five registers of originality are therefore distinguished: terminological (is the expression new?), propositional (has this precise thesis already been formulated?), architectural (have these propositions already been articulated in this diagram?), deductive (are the consequences derived from the architecture new?), of application (does the framework shed a different light on objects?).

Six neighbouring families, with their overlap and their difference; the detail of the references lives in the annotated bibliographies of the three tomes — the whole in that of Tome 1, the families proper to movement and to resumption in Tomes 2 and 3.

The datum as constructed, not raw — Gitelman (ed.), Markham. Overlap: non-pre-existence and constructed character. Difference: the trilogy derives from it an ontological architecture covering persistence, circulation and reading.

The relational datum — Leonelli. Overlap: dependence on context, use and practice. Fruitful tension: an object can acquire the status of datum through its later evidential role, which contests a status entirely fixed at inscription; the edition answers by distinguishing constitution and actualization, and through the intention of use that recruits.

The datum as process — Jones. Overlap: rejection of the datum as a stable objective resource. Difference: this work belongs to organization theory and contingent performance; it does not establish the overall architecture — but it forbids asserting that no existing framework thinks the datum in a processual regime.

Trajectories and materiality — the data journeys: Bates, Lin and Goodale; Leonelli and Tempini. Overlap: material movement, transformation, reuse, change of context. Fruitful tension: circulation does not always preserve an invariant datum, it can manufacture the object that circulates — which thesis 9 integrates: preserve, alter, engender.

From traces to data — Rheinberger. Overlap: the experimental mediation from the trace to the datum. Difference: the trilogy generalizes this chain beyond the experimental sciences and links it to circulation, reading and the torrent.

Process philosophy — Whitehead, Bergson, Simondon. The metaphysical foundation: priority of becoming, individuation, duration, relations. It prevents presenting the substance/process reversal as unprecedented in itself.

Where originality can reasonably be located: in the combination. Eight features compose it — the complete chain, from the real to the trace, from constitution to circulation, from reading to action and to the modified real; the flow of observation as a middle term between the real and consolidated data; the two temporal regimes, operationally deployed; the articulation of datum of origin, copies, filiation; the junction between the micro-ontology of the datum and the macro-dynamics of the torrent; the integration of the situated reader and of their responsibility into the same diagram; the loop by which the production of data transforms the real from which it will draw the next data; the attempt to deduce a minimal ethics from the very structure of the process.

The scoping review is deferred, scheduled after publication; until it is completed, the cautious formula holds.

§8. The Cautious Formula of Originality

To our knowledge, no identified work brings together exactly this architecture; several traditions have independently established major components of it.

The Preface to the Edition carries this sentence; the annotated bibliographies carry the detail. The claim is refocused on originality of composition and deduction: the trilogy proposes an architectural synthesis of the datum as a process of constitution, persistence, circulation, resumption and feedback, and deduces from it a theory of the torrent and of the reader’s responsibility.

A new thesis becomes durable not when it claims to owe nothing, but when it knows exactly what it takes up, what it transforms and what it adds.

§9. The Synthesis Formula

The condensed version of the substance; it lives here, never in the tomes.

A datum is a historically individualized result, produced under a description by a regulated and situated operation, realized on one or more media, accompanied by the conditions that make it possible to establish its origin and to interpret it, liable to be copied or transformed, and whose accessibility, intelligibility, validity and effects depend on the relations in which it is taken up.

This definition does not replace the tomes; it condenses their substance in the language of categories.

Translation choices

These notes say how to read the English words that render the terms of the corpus; they are fixed for the whole English translation by the Translation Charter of the Corpus.

• datum / data. The volume defines the datum as a historically individualized result (§9): an individual. The translation therefore says datum in the singular — unusual but exact, the Latin for “what is given” — and data as its plural, construed as a plural. The French text itself uses datum where it speaks of the datum of origin; English cannot mark the difference, and the datum of origin is named as such each time.
• drawing / to draw (prélèvement, prélever). The gesture of drawing from the flow, as one draws water: situated, carrying an intention and criteria. Sampling is avoided. The undrawn (non-prélevé) is part of the gesture (thesis 2).
• lapse; stale (péremption; périmée). A relation, not a decay: the gap that grows between a datum that stays still and a world that carries on. Expiry is kept for regulatory contexts and for the queue (“expiry in the queue”, “a datum expired for a use”), where an item outlives its deadline.
• actualized (actualisée). The act of reading, by which a meaning is produced; never the updating of a file.
• occurrence: a singular inscription on a medium, what is counted when one copies. Provenance is not origin.
• verisimilar datum: guaranteed by a model, not by the real. Recounting (récit, technical sense): the re-inscription that follows the rereading of a lapsed datum. The récit of the first edition’s synthesis, in §0, is the ordinary narrative.
• over-throughput (sur-débit): arrivals that exceed the capacity to read (λ > μ); neither an overload nor an overflow. Throughput (débit) is the volume of data per unit of time.
• persistence package: the echo of the archival information package is intended; representation information is the term of the OAIS model.
• canal; spring (source): the images of the tomes, which “speak the language of the canal”. The spring where the datum is born, and the title of Tome 2, From Spring to Torrent; never a data source.
• substance renders fond, the substance of the corpus as opposed to its form (“Both say the same substance”, §0); in §7, “the substance/process reversal” is the substance of the philosophers, against which Whitehead, Bergson and Simondon think becoming.
• formal machinery renders apparat; apparatus is kept for dispositif, as in “the intention of the apparatus”. Locus (plural loci) renders foyer: the single place where a contribution is demonstrated.
• medium (support, plural media) carries the inscription; carrier (porteur) holds an occurrence.
• resumption (reprise) is the taking up of a transmitted datum at the other end, which requires the conditions of interpretive resumption; the situated position of handling or processing uses handling for the same French word.
• the divide between the times (le partage des temps, thesis 4): partage as division, not sharing.
• Titles. Tried by the Flow renders À l’épreuve du flux: the datum put to the test of the flow. The tomes and their documents are cited by their English titles (The Two Times of the Datum, The Properties of Movement, The Over-Plus…).