Keep the released original plan immutable. Record every later criterion as a linked, time-stamped version with its actual role, the available evidence about data access, and the exact analyses and reports that used it.

Preserve the released plan as the historical starting point

A study record becomes misleading when a later criterion is inserted into the document that is presented as the original plan. The safer pattern is to freeze the released plan, give it a stable version and content hash, and make every later change a separate linked event. A current-plan view may combine the original with accepted amendments, but it should remain a derived view rather than a replacement for history.

FDA Part 58 and OECD GLP provide bounded examples of explicit change records in their regulated scopes: protocol or study-plan changes retain reasons, dates and attribution instead of silently replacing the earlier record. Those requirements do not automatically apply to an ordinary research-use cell study. SingularCell uses the conservative record pattern without claiming GLP or FDA compliance.

  • Give the original released plan a stable identifier and version
  • Record its author, releaser, release time and content hash
  • Prevent later edits from changing the released bytes
  • Show a current view only as a derivation from preserved versions
  • Keep superseded plans available in the audit history

Classify later events before changing the study record

A planned amendment, an unplanned deviation, a new exploratory criterion and a correction describe different events. Treating them as one generic edit hides why the record changed and when the change became relevant. OECD GLP explicitly distinguishes intended study-plan amendments from unintended deviations in its own scope; the distinction is a useful design analogy for research software because it preserves operational meaning.

Ask the user to choose the event type and explain the choice. If the meaning of a criterion, population, endpoint, operator, unit, timepoint or decision role changes, create a new criterion version. A formatting correction should not be allowed to alter any bound scientific field. When classification is uncertain, store an unresolved state and require human review rather than guessing.

  • Use amendment for an intended plan change
  • Use deviation for an unplanned departure during execution
  • Use exploratory criterion for a new analysis question
  • Use correction for an identified record error
  • Keep uncertain classifications visible and review-required

Track document time and knowledge time separately

A creation timestamp answers when a record was made; it does not prove what the author had already seen. A trustworthy chronology therefore needs two clocks: document events such as authoring, approval and effective time, and knowledge events such as raw-data access, processed-data access, interim-summary viewing or result viewing. Bind each event to the exact dataset or result version involved.

If the system has no access event, it must report missing evidence rather than infer that no one saw the data elsewhere. A person's declaration can be retained as an attributed assertion, but it is not the same as an independently generated system log. Conflicting timestamps or attestations should produce a visible timing-conflict state, never an automatic upgrade to prospective status.

  • Record criterion creation, approval and effective times
  • Record data acquisition, import, access and result-view events separately
  • Bind every event to an exact data or result version
  • Label attestations as assertions with author and scope
  • Use timing unknown or timing conflict when evidence is incomplete

Give every criterion a role that follows it into results

The record should say whether a criterion appeared in the original released plan, was added through a later evidenced amendment, arose after relevant data access, followed a result, or has unresolved timing. NIH and NINDS materials emphasise advance analysis planning, distinction between exploratory and confirmatory work, and transparent disclosure of departures within their stated contexts. They do not provide a universal validation rule for outsourced cell studies.

An exploratory label is not a verdict that a result is false or useless. It tells the reader how the question arose and limits what the analysis can claim. The same idea may later appear prospectively in a new study's original plan, while the earlier analysis remains exploratory. The software should preserve both records rather than retroactively changing the earlier role.

  • Store original, amended, exploratory, post-result or timing-unknown role
  • Carry the role into analysis outputs, reports and exports
  • Keep the criterion expression and applicable context versioned
  • Separate hypothesis generation from later prospective testing
  • Never treat timing status as proof of scientific adequacy

Bind each analysis and report to exact versions

A reader should be able to reconstruct which plan, criterion, dataset, analysis code, configuration and exclusions produced a finding. NIST's Research Data Framework supports authoritative copies, version identification, derivatives, responsible parties, timestamps and provenance as research-data-management concerns. W3C PROV supplies relationships for revision, derivation, generation, use, attribution and invalidation.

These relationships document asserted lineage; they do not prove that a source entry is true or that an analysis is scientifically appropriate. SingularCell should therefore present provenance and human review as separate layers. A report can be byte-identical to a recorded artifact and still require scientific assessment of its method, criterion and interpretation.

  • Link every analysis run to exact criterion and data versions
  • Record code, configuration, software and execution identity
  • Link each finding to the activity that generated it
  • Bind reports to the plan and analysis versions they describe
  • Keep lineage evidence separate from scientific approval

Correct the record through a successor, not an overwrite

When an error is discovered, create a correction record that names the target version, old value, corrected value, reason, responsible person and time. Preserve the superseded version and identify which analyses, reports or reviews may be affected. In its applicable scope, FDA Part 11 guidance describes audit trails that do not obscure prior information and preserve record content and meaning.

For covered Part 58 studies, final-report corrections or additions are signed and dated amendments that identify what changed and why. SingularCell can adopt that additive history pattern without claiming Part 11 or GLP compliance. If the correction changes scientific meaning or a conclusion, the corrected output needs a new bounded review; an approval attached to the prior version should not transfer automatically.

  • Create a new correction record with reason and attribution
  • Preserve the superseded object and its prior review state
  • Identify every downstream artifact that used the old version
  • Reissue affected reports as linked successors
  • Require new review when meaning or conclusions may change

Report chronology without making a scientific judgement

The strongest safe output is factual and bounded: criterion C was absent from the released original plan, was added in amendment A at time T, and the record contains specified evidence or gaps about relevant data access. The output can name the plan, data, analysis and report versions involved. It should not accuse a person of hindsight, infer misconduct or decide that the criterion is scientifically valid.

A useful report separates original criteria, later prospective amendments, deviations, exploratory or post-result analyses, unresolved timing and corrections. It also names the reviewer and exact review scope. This lets a scientist inspect the chronology without allowing the software or a language model to turn record completeness into a biological, statistical or regulatory conclusion.

  • State what the records show and what remains unknown
  • Separate original, amended, exploratory and corrected material
  • Name exact versions and evidence supporting each chronology statement
  • Avoid misconduct, validity, compliance and equivalence conclusions
  • Leave scientific suitability and impact with qualified reviewers

Primary sources

Material claims were checked against the organisations responsible for the guidance or measurement work.

  1. eCFR — 21 CFR 58.120, Protocol ↗United States Government Publishing Office / United States Food and Drug Administration · A regulated-context example in which protocol changes, reasons, signatures, dates and retention remain explicit.
  2. OECD — Principles of Good Laboratory Practice and Compliance Monitoring ↗Organisation for Economic Co-operation and Development · GLP-context distinctions between intended amendments, unintended deviations, retained prior entries and linked final-report changes.
  3. NIH/NINDS — Rigor and transparency training notice ↗National Institutes of Health / National Institute of Neurological Disorders and Stroke · Advance analysis planning, exploratory-versus-confirmatory distinction and transparent disclosure of divergence from plans.
  4. NIST — Research Data Framework, Version 2.0 ↗National Institute of Standards and Technology · Authoritative copies, versions, derivatives, responsible parties, timestamps and provenance across the research-data lifecycle.
  5. W3C — PROV-O: The PROV Ontology ↗World Wide Web Consortium · Standard relationships for revision, derivation, generation, use, attribution and invalidation without erasing prior entities.
  6. FDA — Part 11 Electronic Records; Electronic Signatures — Scope and Application ↗United States Food and Drug Administration · An applicable-regulated-context example of time-stamped audit trails that do not obscure prior information and preserve record meaning.
  7. eCFR — 21 CFR 58.185, Reporting of nonclinical laboratory study results ↗United States Government Publishing Office / United States Food and Drug Administration · A regulated-context example of retaining original-protocol and change context and issuing corrections as identified, reasoned amendments.

Limitations

  • This article is recordkeeping and provenance guidance, not a biological protocol, statistical analysis plan or scientific change-impact assessment.
  • FDA and OECD examples are bounded to their regulated contexts and do not make an ordinary research-use study GLP, Part 11 or FDA compliant.
  • Timestamps, hashes and audit trails preserve recorded chronology but do not prove absence of prior knowledge, source truth or scientific validity.
  • Only qualified reviewers with context-specific evidence can assess whether a criterion, deviation, correction, analysis or result is scientifically suitable.

Related SingularCell guidance

See the handoff as a working system.

Explore one synthetic study from research question through capability comparison, returned results and review-required evidence.

Explore the synthetic study →Discuss a design-partner workflow