Use an Evidence Passport to identify exact evidence, provenance, transformations, automated checks, reviews, signatures, limitations and corrections. Name the property each layer establishes and never collapse them into a universal trust or validation score.
Start with the question the passport can honestly answer
An Evidence Passport should answer: what evidence was supplied, where it came from, which versions were used, what transformed it, which checks ran, who reviewed what, and what remains missing or disputed. It is a structured inspection surface for a scientist—not a badge that makes the underlying study trustworthy by itself.
NIST's Research Data Framework treats authoritative copies, versions, derivatives, workflow inputs and outputs, responsible parties, timestamps and provenance as connected data-management concerns. W3C PROV supplies interoperable concepts for entities, activities, agents, derivation, revision and attribution. Both support a more inspectable record, but neither certifies the scientific truth of the record.
- Name the passport subject and exact version
- State the intended use and explicit non-use boundary
- Enumerate included, missing, restricted and superseded evidence
- Bind every claim and review to exact object versions
- Avoid a single trusted, validated or compliant status
Make identity, integrity and provenance different fields
Identity records what an object is declared to be. Byte integrity records whether current bytes match a digest. Provenance records asserted relationships among inputs, transformations, outputs and agents. These are useful but different facts. A matching hash cannot prove authorship or accuracy, and a provenance relation cannot authenticate the event it describes.
W3C PROV-O makes relationships such as generation, use, derivation, revision, primary-source assertion, attribution and invalidation machine-readable. A SingularCell passport can use those concepts to connect evidence without claiming that vocabulary conformance makes the evidence biologically correct or sufficient. Each relation should retain its source, assertion method and verification state.
- Record declared source identity separately from source verification
- Store hash algorithm, digest, object version and computation time
- Link every transformation to exact inputs and outputs
- Distinguish people, organisations and software agents
- Report provenance gaps instead of filling them with assumptions
Anchor claims to evidence without claiming the source proves them
A useful claim record points to an exact evidence object, version and locator—such as a page, section, table, row or cell—and records the claimed relationship. That relationship is itself an assertion by the mapper or reviewer. A language model may help propose terminology mappings, but its output should remain unreviewed until a person accepts a bounded interpretation.
NIH-hosted preclinical reporting principles support transparent methods, replicate structure, statistics, omissions, underlying datasets, attribution and biological-material identity for interpretation. NIH's Data Management and Sharing policy defines metadata in its own scope to include methodology, provenance, transformations and contextual variables. These practices improve inspectability; they do not replace replication or establish a universal scientific verdict.
- Give every claim a stable ID and exact text
- Cite exact evidence versions and locators
- Record whether the source states, contextualises or conflicts with the claim
- Label machine-generated mappings as unreviewed
- Preserve competing evidence and unresolved terminology
Use FAIR features for reuse, not as a quality badge
The original FAIR principles call for persistent identifiers, rich metadata, explicit links, qualified references, provenance, licenses and domain-relevant standards. Those features can make evidence easier to find, interpret and reuse across organisations. FAIR does not mean that all data must be open, nor does it establish that a dataset or conclusion is accurate or high quality.
A passport should therefore state access conditions, licenses, confidentiality limits and retention status. When evidence is withheld or redacted, a disclosure manifest should say what category is absent and why. The disclosed package should be called complete only against a defined manifest in which every required item is included or explicitly accounted for.
- Use persistent identifiers where available
- Store rich metadata and explicit links between data and metadata
- Record licenses, access conditions and reuse limits
- Enumerate redacted or restricted items in a disclosure manifest
- Never equate FAIR with open, valid, verified or complete
Name exactly what each automated check verified
Schema validation can show that required fields and relationships satisfy a named schema version. A manifest check can show whether referenced objects resolve or are explicitly unavailable. A signature check can show whether a cryptographic or authentication rule succeeded. None of these checks can decide whether an endpoint, threshold, method or conclusion is scientifically suitable.
Every check output should record its name and version, exact inputs, configuration, responsible software, execution time, result and errors. Replace a bare verified badge with a property-specific statement such as hash match, citation locator present, version chain complete or signature cryptographically verified. If the necessary evidence is missing, use an explicit missing or unresolved state.
- Identify the check and version
- Bind the result to exact inputs and configuration
- Preserve errors and incomplete states
- Use property-specific result language
- Prevent technical checks from issuing biological acceptance
Bind reviews and signatures to scope, meaning and version
A review record should identify the reviewer, role, organisation as declared, exact objects reviewed, review question, intended-use scope, outcome, rationale, limitations and time. A signature should identify the exact signed object and state whether it means authored, reviewed for manifest completeness, scientifically assessed for a bounded use, or authorised for release.
In its applicable scope, 21 CFR 11.50 requires signature manifestations to include signer name, date and time, and the meaning associated with the signature. FDA Part 11 guidance also emphasises predicate-rule scope and preserving record content and meaning. SingularCell can borrow the explicit-meaning pattern without claiming that a passport is Part 11 compliant or that any signature validates the science.
- Bind review to exact passport, evidence, analysis and report versions
- State the review question and intended-use boundary
- Record unresolved issues and re-review triggers
- Display signer, time and explicit signature meaning
- Keep a system seal visually and semantically distinct from human review
Preserve corrections and make limitations impossible to miss
A correction should name the target object and version, old and new values, reason, actor, time, effect-on-meaning assessment, affected downstream objects and re-review status. The earlier version remains inspectable where lawful and permitted. If upstream evidence is corrected or withdrawn, affected claims and reports should be marked stale or held until the impact is resolved.
The passport's strongest safe statement is narrow: it identifies supplied evidence, records declared sources and transformations, binds reviews and signatures to exact versions, and preserves limitations and corrections. It does not establish source truth, biological validity, statistical adequacy, laboratory quality, method or site equivalence, regulatory compliance or acceptance of the study result.
- Create linked successor records for corrections
- Retain superseded evidence and review states
- Propagate stale or held status to affected outputs
- Show access restrictions, conflicts and missing evidence
- End with the exact scientific and regulatory non-claims
Primary sources
Material claims were checked against the organisations responsible for the guidance or measurement work.
- NIST — Research Data Framework, Version 2.0 ↗National Institute of Standards and Technology · Authoritative-copy identity, versions, derivatives, timestamps, responsible parties, workflow inputs and outputs, tools and provenance.
- W3C — PROV Data Model ↗World Wide Web Consortium · Interoperable semantics for entities, activities, agents, derivation, revision, attribution, responsibility and provenance bundles.
- W3C — PROV-O: The PROV Ontology ↗World Wide Web Consortium · Machine-readable relationships for generation, use, derivation, revision, primary-source assertions, attribution and invalidation.
- NIH — Principles and Guidelines for Reporting Preclinical Research ↗National Institutes of Health · Transparent reporting of methods, replicate structure, statistics, exclusions, underlying datasets, attribution links and biological-material identity for interpretation.
- NIH — Final Policy for Data Management and Sharing ↗National Institutes of Health · Metadata including methodology, provenance, transformations, contextual variables, limitations and persistent identification in the policy's stated scope.
- FAIR Guiding Principles for scientific data management and stewardship ↗Wilkinson et al. / Scientific Data · The original FAIR principles for identifiers, rich metadata, explicit links, qualified references, provenance, licenses and domain-relevant standards.
- eCFR — 21 CFR 11.50, Signature manifestations ↗United States Government Publishing Office / United States Food and Drug Administration · A regulated-context example in which a signature displays the signer name, date/time and explicit meaning.
- FDA — Part 11 Electronic Records; Electronic Signatures — Scope and Application ↗United States Food and Drug Administration · Predicate-rule scope and bounded examples for audit history, copies, retention and preserving record content and meaning.
Limitations
- This article explains evidence-record architecture and claim boundaries; it does not define a biological protocol, acceptance threshold or scientific-validation process.
- NIST, W3C, NIH and FAIR sources support provenance and reporting concepts but do not certify SingularCell or any passport conclusion.
- FDA Part 11 examples apply only within their stated regulated scope and do not make a research-use passport Part 11 compliant.
- A complete documentation package may still be scientifically inadequate; context-specific interpretation and acceptance remain with qualified people.
See the handoff as a working system.
Explore one synthetic study from research question through capability comparison, returned results and review-required evidence.