The operating system
Three tasks. One source of truth.
Prospecting, editing, and auditing are independent processes. They share one source of truth, but work with different states and responsibilities.
The separation is deliberate: finding research, designing fiction, evaluating a text, and publishing it are not a single operation.
Thursday / 01:00 — Prospecting
It searches for recent research within a twelve-month window, consults the full paper when available, and records reported results, the authors' inferences, limitations, and possible extrapolations separately.
It maintains an editorial reserve with a minimum of 8 candidates, a target of 15, and a maximum of 20.
A paper may be scientifically interesting and still be left out if it offers no behaviour, anomaly, asymmetry, choice, or consequence capable of sustaining narrative friction without distorting its results.
GitHub / main — Canonical repository
A private repository contains the current version of the methodology, the scientific reserve, candidate claims, run records, editorial memory, and publishable files.
Each task rereads that state before acting. The system's behaviour does not depend solely on the text stored in a scheduled task.
The repository provides persistence, versioning, and internal traceability.
Daily / 06:00 — Editing
It selects an eligible candidate from the reserve and claims it before writing to prevent concurrent runs from using the same paper.
It then executes the complete editorial algorithm.
The run may end as SIN_CANDIDATO, NO_PUBLICADO_CALIDAD, or PUBLICADO.
The editing task does not search for new papers on its own.
Sunday / 22:00 — Editorial audit
It reads the latest seven Spanish stories —or the available archive during calibration—, their records, and editorial memory. It saves a report even when it concludes SIN_CAMBIOS.
The first run establishes a baseline only. Afterwards, the audit may integrate limited, tested, and reversible PATCH adjustments. It cannot modify its permissions, its contract, scientific integrity, publication, validators, or infrastructure. Structural changes require human approval.
All three tasks use the Europe/Madrid time zone. The versioned methodology takes precedence over instructions stored in the tasks. Only the audit has an explicit, limited exception for minor changes.
The flow
SCIENTIFIC SOURCE → PROSPECTING → VERSIONED RESERVE → CLAIM → 6 CONCEPTS → 2 PROSE PROBES → DESIGN → FIRST DRAFT → EVALUATION WITH AN ADVERSARIAL ROLE → REWRITE → ISOLATED EVALUATION → DETERMINISTIC VALIDATION → PUBLICATION
Not every run completes the full flow. A critical failure may return a candidate to the reserve, require a redesign, or close the run without publication.
01 — The paper enters as evidence
Prospecting starts from the scientific work, not from a possible story.
It records what the paper claims, what it interprets, which limitations it acknowledges, and which human consequences may reasonably be derived from its results.
The scientific dossier is forbidden from inventing protagonists, titles, or scenes. Its purpose is to preserve the boundary between evidence and fiction before narrative design begins.
The inclusion of a paper in Espacios Latentes does not imply independent validation of its results or that it has passed peer review.
The system checks the adaptation's fidelity to the source. It does not reproduce the study or replace its scientific evaluation.
02 — Precheck and claim
Before writing, the process checks the methodology version, the state of the reserve, consumed candidates, recent runs, and editorial continuity.
It then creates an exclusive claim on the selected candidate.
Two runs must not consume the same paper or produce two ordinary editions for the same date.
03 — Six possible futures
Six materially different narrative mechanisms are generated from the same scientific ground.
Each proposal must be describable as a causal chain:
novum → situation or objective → decision or change → cost or consequence
Tension may be interpersonal, internal, collective, epistemic, or material/systemic. There must be effective friction, agency, and perceptible consequences. An opposing party with interests of its own is required for interpersonal tension, not fabricated for the others.
Diversity does not mean changing names or settings. The six proposals must cover at least four tension topologies, three progression modes, and distinct literary wagers. No more than two may combine institutional authority, external urgency, and professional sacrifice.
04 — Selection is tested in prose
The two best-positioned candidates each receive a 500–700-word prose probe.
Selection therefore no longer depends on a promising synopsis.
Each probe must demonstrate agency, effective friction, meaningful change, and a narrative expectation that remains open. It must also make a wager in voice, contradiction, structure, relationship, image, or form perceptible.
The candidate that works best as an idea does not always work best as a story.
05 — Design and first draft
Before drafting the full story, the point of view, tension topology, progression mode, main pressure, literary wager, and nature of the ending are fixed. Drafts are oriented toward 1,600–3,000 words, with flexibility when the form justifies it.
The science-fiction dimension must appear within the first quarter of the story and have observable effects on the characters or their decisions.
The first draft is never published.
06 — Evaluation with an adversarial role
A new pass examines the text with a critical rubric and without assuming that the decisions made during design are correct.
Thirteen dimensions are evaluated, including scientific fidelity, causality, tension, agency, cost, progression, conceptual load, voice, functional repetition, variation, and ending.
The result may be APROBADO_CON_CAMBIOS, CORRECCION_LOCAL, REDISENO, or RECHAZO_CONCEPTO.
These evaluations are editorial judgements assisted by a model. They are not objective measurements of literary quality.
07 — Rewrite and isolated evaluation
Review observations must be translated into observable changes to the text: removing, moving, condensing, expanding, or replacing material.
The new version is then evaluated without consulting the paper, the initial intention, or the decisions made during design. Whenever the environment allows it, this reading is assigned to a separate context; otherwise limited independence is declared.
At three approximate reading points —25%, 50%, and 75%— the evaluator must identify which expectation remains open, what sustains the next section, which crucial information appears only once, and which repetitions are inert or gain a new function.
The ending must resolve or transform a previously established pressure.
The test does not attempt to predict how a real reader will react. It seeks to detect structural problems before publication.
08 — Bilingual editing and publication
The English version is produced only after the Spanish edition has passed the preceding controls.
The adaptation must preserve facts, capabilities, causality, losses, the outcome, and scientific limits.
The public run record is then generated and the publication artefacts are prepared.
The four artefacts for the edition —the story and its public run record in Spanish and English— are prepared as one unit. The general methodology is also maintained as a bilingual canonical pair and both routes are synchronised whenever its version changes.
Deterministic controls verify structure, identifiers, bilingual pairs, states, routes, metadata, and numbering. The Site is deployed only after those controls have passed for the commit containing the edition.
After deployment, page accessibility, navigation, home page, archive, language, and rendering are also checked.
Three types of control
1. Scientific evidence
Checks that claims attributed to the paper appear in the source, that results and inferences are not confused, and that limitations are recorded.
2. Editorial evaluations
Assess causality, agency, progression, clarity, tension, voice, and ending.
They are AI-assisted evaluations and necessarily contain judgement.
3. Deterministic validations
Check objective properties of the artefacts: files, identifiers, states, bilingual pairs, routes, structure, and deployment.
A software validator can prove that files are missing. It cannot prove that a story is good.
Publication gates
E1–E3 — Evidence and derivable friction
The source must be identifiable, its limits must be explicit, and friction must arise from behaviours, anomalies, asymmetries, choices, or consequences compatible with the research.
CF1 — Visible science fiction
The novum must appear within the story, produce observable changes, and not be relegated to the title or the scientific note.
CC1 — Understandable causality
An evaluation without the paper must be able to reconstruct what changes, what the focus wants, fears, or needs, what it does or understands, and what cost or consequence follows.
T1 — Tension, agency, and cost
There must be meaningful agency, effective friction, and a perceptible consequence. The topology may be interpersonal, internal, collective, epistemic, or material/systemic; only the first requires another party with interests of its own.
B2 — Scene traction
The probe must meaningfully change situation, knowledge, relationship, obligation, interpretation, or values, and keep a concrete expectation open.
J1 — Progression and payoff
Progression may operate through escalation, revelation, recontextualisation, transformation, choral accumulation, deterioration, or recursion. Each section must modulate the experience, and the ending must answer a pressure established earlier.
Repeating a function without enriching it is not narrative progression; repetition that reinforces, contrasts, or recontextualises may be.
L1 — Isolated evaluation
At the 25%, 50%, and 75% checkpoints, a concrete narrative expectation that remains unresolved must be identifiable.
The ending must provide a recognisable payoff for something established earlier.
QA — Publishable integrity
The Spanish edition, the English edition, both public run records, metadata, identifiers, numbering, and routes must correspond to the same run. The bilingual pair for the general methodology must also share the canonical version.
The boundary
Science lays the ground. Fiction decides where to break it open.
The papers are real and linked from every story.
Characters, institutions, future technologies, and narrative consequences may be fictional.
The scientific note distinguishes what comes from the original work, which limitations it declares, and which elements have been extrapolated or invented.
A plausible extrapolation is not a prediction.
Traceability
Every story leaves more than a text.
The repository functions as operational memory and a traceability record.
It preserves enough information to reconstruct internally which evidence was used, which alternatives were considered, which decisions were made, which changes were applied, and the result of each run.
Each story is associated with the version of the method that produced it. The public run record also records the format version that made that reconstruction visible.
The complete working artefacts remain in the private repository. The public run record summarises the decisions needed to understand the route without publishing the full working texts.
Direction and responsibility
Automation is not an abdication of judgement.
Prospecting and editing execute a previously defined methodology and cannot modify it.
The weekly audit may learn from the archive and integrate PATCH adjustments only within a contract it cannot edit. It must provide repeated evidence, preserve positive cases, work in a branch, pass regression and validation, and accept SIN_CAMBIOS as a normal result.
New gates, structural changes, permissions, story withdrawals, and replacements require human intervention.
Editorial evaluations remain judgements. Traceability does not turn them into objective measurements, but it does allow the criteria applied and their consequences to be recorded.
- The methodology takes precedence over scheduled instructions.
- The first draft is never published.
- A critical failure invalidates the run.
- A run may end without a story.
- Every publication is associated with a method version and a run record.
ESPACIOS LATENTES
Cutting-edge fiction inspired by recent scientific research.
ARXIV → FICTION / 2026