Transductive Publication Format
A conformance model for scientific work published as three synchronized reading layers: public article, informed scientific explanation and technical record.
Abstract
Purpose and scope.
The Transductive Publication Format (TPF) defines a scientific publication as a linked research object with at least three human-facing representations of increasing detail. The first representation is a full-length public article intended for an intelligent non-specialist. The second is an informed scientific explanation intended for readers who want methods, uncertainty, history and competing interpretations without requiring field-specific fluency. The third is a technical record intended for expert inspection, reproduction, citation and archival preservation.
The format is designed to coexist with existing scholarly infrastructure rather than require its replacement. DOI registration, JATS XML, PDF exports, conventional reference lists, peer review and IMRaD may be used inside the technical layer. TPF adds cross-layer identity, claim alignment, version propagation, public accessibility and challenge routes.
The central design claim is operational: technical precision and public accessibility need not be competing outputs when multiple synchronized representations can be produced from the same research object.
Normative language
What MUST, SHOULD and MAY mean in this draft.
MUST identifies a requirement for TPF conformance. SHOULD identifies a strong recommendation that may be omitted with an explicit reason. MAY identifies an optional feature.
A conforming publication MUST expose all three reading layers under one publication identity. Each layer MUST identify the publication version and MUST link to the other two layers. The three layers MAY be stored as separate documents, routes or views, but their relationship MUST be machine-readable.
The public and informed layers MUST NOT introduce stronger causal, quantitative or generality claims than the technical record supports. The technical record MAY contain claims that are too detailed to reproduce in full above, but their practical implications SHOULD be represented in at least the informed layer.
Publication identity
The object that stays the same while its representations differ.
Persistent identity
The publication MUST have a stable canonical identifier. A DOI is preferred when available. The identifier names the publication as a whole, not only its technical PDF.
Version
Every rendered layer MUST display or expose the current version. Previous published versions MUST remain retrievable or at minimum cryptographically identifiable and described.
Contributors
Contributions SHOULD be expressed by role where possible. Public writing, visualization, software, data curation and technical analysis are legitimate research-publication contributions and SHOULD NOT disappear behind a single undifferentiated author list.
Funding and conflicts
Funding sources and relevant conflicts MUST be available from every layer without requiring navigation to the technical record.
Layer requirements
Minimum content for each human-facing representation.
Layer 1 · Public article
- Assume no field-specific vocabulary at entry.
- Introduce the human, practical or conceptual problem before internal disciplinary framing.
- Explain specialized terms when first required, preferably after explaining the distinction the term names.
- State the main result, major uncertainty, important limitations and what the result does not justify.
- Disclose funding, relevant conflicts, review state and correction state.
- Provide direct routes to deeper explanation and evidence.
Public-language adapters
A Layer 1 publication MAY expose multiple language-complexity renderings such as ELI5, ELI12, ELI18 and General. If multiple renderings exist, they MUST preserve the same substantive claims, quantities, uncertainty, scope and causal strength. They MAY change vocabulary, sentence length, examples and explanatory scaffolding. The interface MUST NOT present language level as a ranking of reader intelligence.
Layer 2 · Informed explanation
- Describe the relevant history and prior work.
- Explain method choices conceptually before relying on formal names.
- Describe realistic alternative interpretations and unresolved disagreements.
- Expose enough quantitative information for an informed reader to judge scale and uncertainty.
- Link major claims to technical anchors, figures, methods or references.
Layer 3 · Technical record
- Contain sufficient methodological detail for expert audit and, where feasible, reproduction.
- Provide complete references and persistent identifiers.
- Link data, code, protocols, preregistration and computational environments where applicable.
- Describe exclusions, transformations, model choices, parameter settings and uncertainty procedures.
- Expose version, review, correction and known-limitation history.
Cross-layer claims
The feature that makes three articles one publication rather than three unrelated texts.
A major scientific assertion SHOULD have a stable claim identifier. The same claim MAY appear in all three layers using different language. The publication SHOULD expose the relationship among those expressions.
For example, a public statement such as “the treatment reduced symptoms by a modest amount” can link to an informed statement explaining the size and uncertainty of the change, which links to the exact model estimate, dataset and analysis in the technical layer.
A correction to the technical estimate MUST trigger review of all linked higher-level expressions. Publication tooling SHOULD surface stale or contradictory claims automatically.
Evidence and research objects
Data and code are part of the publication, not luggage attached after the trip.
TPF adopts the broad lesson of research-object work: papers are not the only outputs required to understand modern research. Datasets, software, workflows, protocols, models, instruments and provenance SHOULD receive persistent identifiers and direct links where feasible.
The technical layer MUST distinguish an unavailable artifact from an artifact that does not exist. “Available on request” SHOULD be treated as a temporary access state, not equivalent to publication.
Interactive figures MAY provide transformations, parameter exploration or executable analysis. A stable non-interactive representation MUST also be available for preservation and accessibility.
The public layer SHOULD explain what important evidence objects are for. It need not expose file formats or computational details unless they affect the interpretation.
Review and challenge record
Publication should make disagreement addressable.
Review status
Every layer MUST state whether the work is unreviewed, under review, peer reviewed, revised after review, corrected, retracted or otherwise formally updated.
Public-layer review
At least one reviewer or editor MUST check whether the public and informed layers accurately represent the technical record. This review is part of scientific quality control, not copy-editing.
Addressable criticism
Published claims, figures and methods SHOULD expose stable anchors so formal reviews, corrections and public criticism can link to the exact object being challenged.
Open issues
A publication SHOULD maintain a visible list of unresolved problems, replication attempts, material criticism and known limitations after publication.
Accessibility and human factors
Accessible scientific communication is a format requirement, not a courtesy.
The primary HTML representation SHOULD conform to current Web Content Accessibility Guidelines at an appropriate level. Keyboard navigation, semantic headings, focus states, readable contrast, alternative text, captions and non-color encodings are baseline requirements.
Mathematical notation SHOULD include accessible text or MathML where feasible. Interactive figures MUST have a non-interactive explanatory alternative. Tables SHOULD remain navigable by assistive technology.
The public layer SHOULD be tested with readers who are intelligent but unfamiliar with the field. Readability scores MAY be used as diagnostics but MUST NOT substitute for human comprehension testing.
Unexplained abbreviations and specialist terms SHOULD be treated as publication lint: software can flag them, but authors decide whether the term is genuinely necessary.
Machine-readable manifest
The relationships that a browser shows should also be available to software.
A conforming publication SHOULD expose a JSON manifest describing identity, versions, layers, contributors, claims, evidence objects, review state and relationships among them. The manifest is not intended to replace established scholarly metadata; it links the reading architecture to it.
Where a DOI or other persistent identifier exists, the manifest SHOULD reference it rather than invent a parallel identity. Existing metadata standards remain valuable precisely because a new reading format should not require a new island.
Interoperability
The format should degrade gracefully into the infrastructure science already has.
TPF HTML is the preferred reader-facing representation. A publication SHOULD also be exportable to stable PDF for offline use and preservation. The technical record MAY preserve a conventional manuscript layout for journal submission or archival systems.
JATS XML, Crossref metadata, ORCID contributor identities, DataCite records, RO-Crate packages and domain repositories are compatible with the model. TPF does not define replacements for those systems.
This is a deliberate response to earlier reform projects. PubPub, eLife Lens and Elsevier’s Article of the Future show that scientific communication often has to bridge web-native presentation and existing indexing/archival systems. FORCE11 and Research Objects show that the surrounding data model matters. The format therefore treats compatibility as a first-class design constraint.
Known failure modes
The format can reproduce the very problems it is meant to solve.
| Failure | What it looks like | Countermeasure |
|---|---|---|
| Public layer becomes marketing | Only positive findings, exaggerated stakes, missing uncertainty. | Require limitations, negative/null results where material, conflicts and reviewer checking of simplification. |
| Three layers drift apart | A corrected result remains wrong in the public story. | Stable claim IDs, dependency checks and version-propagation review. |
| Author burden explodes | Every paper becomes a bespoke interactive design project. | Standard templates, reusable figures, AI-assisted translation and optional interactivity. |
| Accessibility becomes simplification | Technical nuance disappears in the name of readability. | Keep all depth; public language changes representation, not meaning. |
| New format becomes new prestige ritual | Expensive design becomes another status signal. | Judge conformance by clarity, traceability and accessibility, not visual novelty. |
| Web object becomes fragile | Links rot, interactive code stops running. | Stable exports, archived assets, versioned dependencies and preservation packages. |
Minimal conformance checklist
A publication does not need every advanced feature to begin.
REQUIRED
One publication identity; three linked reading layers; visible version and review state; funding/conflicts; public statement of findings and uncertainty; informed explanation of methods and alternatives; technical record with full references and methods; correction propagation across layers.
STRONGLY RECOMMENDED
Stable claim anchors; public evidence links; accessible HTML; archived PDF export; data and code identifiers; open limitation/challenge record; contributor roles; human comprehension testing.
OPTIONAL
Interactive figures, executable analyses, annotation, public review, live data views, simulation, AI-guided reading and automatically generated alternative explanations.
A plain, excellent three-layer publication is more conformant than a spectacular interactive paper that still assumes the public should remain outside.
Prior art informing this draft
The specification is intentionally evolutionary.