Never edit a released study plan in place. Create a new immutable artifact version for every released-content change, and create a new effective-plan version when a prospective amendment changes operative instructions or scientific meaning. Let software identify changed fields; let qualified people assess impact.
Use two version rules instead of one vague materiality test
The first rule is mechanical: once an artifact is released, any content change produces a new immutable artifact version. Even a spelling or layout correction should not replace the earlier bytes. The second rule is operational: create a new effective-plan version when an intended prospective change alters what a laboratory, analyst or reviewer is instructed to do or interpret.
This separates traceability from scientific impact. Software can compare canonical content, flag protected fields and generate a linked successor. It cannot decide one universal materiality threshold for every cell model, method, laboratory or intended use. A borderline change should remain human-impact-assessment required rather than being silently classified.
- Create a new artifact version whenever released content changes
- Create a new effective plan for prospective operative changes
- Keep the original release immutable and inspectable
- Record change category separately from version identity
- Use a review-required state when semantic impact is uncertain
A protected-field change should trigger a new effective plan
Protected fields include purpose and decision context; material and cell identity; design, controls and replication; method, measurement, instrument and timepoint; criteria and evidence; analysis and reporting; required data and provenance; and operational site or storage scope. A prospective change to any of these can alter the instructions that govern future work.
The trigger only says that a successor plan is required under the configured product rule. It does not say the change is acceptable, comparable or scientifically valid. A qualified reviewer still decides whether work should continue, repeat, be reinterpreted or remain outside the intended use.
- Compare exact protected fields and canonical values
- Show old and new values in the amendment
- Record the reason and responsible people
- Bind approval and effective time to the exact version
- Keep scientific impact separate from deterministic detection
Do not turn a deviation into a retrospective amendment
An amendment is an intended change to future work; a deviation is an unplanned departure that occurred during execution. OECD GLP principles and Study Director guidance distinguish these categories in their regulated context and describe attributable, dated amendment and deviation records. The distinction is useful outside GLP because it protects chronology.
Record the planned requirement, observed departure, discovery time, reason if known, affected site, runs, samples and data, and the impact-review state. The plan version that governed the work remains unchanged. Logging a deviation does not decide whether its impact is negligible or disqualifying, and it must not make the departure appear prospectively approved.
- Bind the deviation to the plan version in force
- Identify affected execution and returned evidence
- Keep observation time separate from later review time
- Preserve unknown reasons and unresolved impact
- Never rewrite history to make the departure planned
Treat corrections and administrative edits as explicit events
A correction fixes an identified error while preserving the target record, prior value and successor. An administrative edit changes metadata or presentation without intending to change operative meaning. Neither category should be trusted solely because an author selected a label. The system should compare canonical fields and require review when meaning may have changed.
In covered Part 58 studies, protocol changes retain reasons, signatures and dates, while final-report corrections or additions are signed and dated amendments that identify what changed and why. These are bounded regulatory examples, not universal rules for research-use studies. SingularCell borrows the non-overwriting pattern without claiming FDA or GLP compliance.
- Preserve target and successor versions
- Record old value, new value, reason, actor and time
- Check whether protected or scientific fields changed
- Create a report successor for released-report corrections
- Require re-review when meaning or conclusions may change
Separate approval, effective time, distribution and acknowledgement
An approved amendment may not yet be effective. An effective amendment may not yet be recorded as distributed to every laboratory or recipient. A recipient acknowledgement may mean receipt, review or another defined action, but it does not automatically prove comprehension, agreement or implementation. Keep each event separate and state its meaning.
OECD Study Director guidance describes numbered amendments, reasons, dates, signatures and distribution in its GLP context. For a research-use workflow, store recipient-specific sent, received and acknowledged times, then bind each run to the plan version recorded as governing it. If the timeline conflicts, create an exception and require human review instead of automatically invalidating the run.
- Record proposed, approved and effective times separately
- Track distribution for each recipient or site
- Define what acknowledgement means
- Record the first execution governed by the successor plan
- Expose timeline gaps and conflicts without guessing
Propagate possible impact without deciding it
When a plan, criterion, method or report changes, software can traverse declared dependencies: laboratory capability comparisons, materials, samples, runs, acquisition settings, data transformations, analyses, discrepancy reports, Evidence Passports, reviews and signatures. Each affected object can be marked for assessment, update, reissue, supersession or an explicit no-action decision.
NIST RDaF supports versions, derivatives, timestamps, responsible parties and provenance, while W3C PROV provides semantics for plans, revisions, generating activities, attribution and invalidation. These relationships expose lineage. They do not establish that the declared chronology is true or that a reviewer should accept the affected work.
- Traverse only declared versioned dependencies
- Mark possibly affected objects as review-required
- Invalidate stale review-version bindings
- Require explicit carry-forward or re-review decisions
- Keep dependency detection separate from scientific judgement
Record timing changes without upgrading their scientific status
Clinical-trial guidance in ICH E9(R1) asks reports to identify whether analyses were prespecified, introduced while treatment assignment was blinded or developed post hoc, and to discuss estimand changes separately within its scope. NINDS guidance emphasises planned exploratory and confirmatory components and states that no single criteria set applies to all studies. Neither source supplies a universal outsourced cell-study amendment threshold.
A change record should capture whether relevant data had been acquired, accessed or summarised when the amendment was created. If that state is unknown, keep it unknown. A late amendment cannot retroactively make an analysis prespecified, and an early timestamp does not prove no one had seen relevant information. The output should report chronology and leave scientific interpretation to qualified people.
- Record relevant acquisition, access and result-view state
- Distinguish prespecified, blinded-stage, post-access and unknown timing
- Do not infer knowledge from absent system logs
- Preserve exploratory status when applicable
- Avoid universal materiality or acceptance conclusions
Primary sources
Material claims were checked against the organisations responsible for the guidance or measurement work.
- 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.
- 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 preserving original-protocol context and issuing report corrections as identified, reasoned amendments.
- OECD — Principles of Good Laboratory Practice and Compliance Monitoring ↗Organisation for Economic Co-operation and Development · GLP-context distinctions among intended amendments, unintended deviations, prior raw-data entries, distribution and final-report changes.
- OECD — Role and Responsibilities of the Study Director in GLP Studies ↗Organisation for Economic Co-operation and Development · GLP-context definitions and handling of numbered amendments, deviations, reasons, dates, signatures, distribution and acknowledgement.
- FDA/ICH — E9(R1) Addendum on Estimands and Sensitivity Analysis ↗United States Food and Drug Administration / International Council for Harmonisation · Clinical-trial-context reporting of whether analyses were prespecified, introduced while treatment assignment was blinded or developed post hoc, with estimand changes discussed separately.
- NINDS — Rigorous Study Design and Transparent Reporting ↗National Institute of Neurological Disorders and Stroke · Context-specific planned exploratory and confirmatory components, prospective methods where applicable and the absence of one universal criteria set.
- NIST — Research Data Framework, Version 2.0 ↗National Institute of Standards and Technology · Authoritative-copy identity, versions, derivatives, timestamps, responsible parties, persistent identifiers and provenance.
- W3C — PROV Data Model ↗World Wide Web Consortium · Machine-readable semantics for distinct plan entities, revision activities, attribution, invalidation and provenance bundles.
Limitations
- This article defines documentation and product versioning rules; it does not determine the scientific materiality or acceptability of a change.
- FDA, OECD and ICH examples remain limited to their regulated scopes and do not make a research-use SingularCell workflow compliant.
- Field comparison, version chains and dependency traversal cannot prove chronology truth, plan implementation, method equivalence or biological validity.
- Qualified reviewers must assess the context-specific impact of changes on execution, data, analysis, interpretation and intended use.
See the handoff as a working system.
Explore one synthetic study from research question through capability comparison, returned results and review-required evidence.