Why English? | Writing SPIRIT-AI Trial Protocols Through Algorithm Versions, Data Flow, Human Oversight and Prespecified Errors

SPIRIT-AI trial protocol reporting helps researchers describe an AI intervention before recruitment and results. People searching for AI trial protocols, algorithm version control or human oversight need clear future-tense answers: which system will be used, what data it will accept, how people will interact with it, what errors will be studied and what will trigger an amendment.

The primary Nature Medicine SPIRIT-AI statement, published on 9 September 2020 and checked on 5 November 2026, adds 15 AI-specific items to the protocol framework. The open full text describes 12 extensions and three elaborations. The EQUATOR record confirms its scope: protocols for clinical trials evaluating interventions with an AI component.

English matters because will use, may update, primary outcome and prespecified error analysis are commitments with different force. A complete protocol lets later readers compare plan, amendment and report. This guide shows how to read or write that chain without treating registration as a guarantee of good methods or safe care.

Use this guide beside eduKate’s SPIRIT 2025 protocol guide and the companion CONSORT-AI trial-reporting guide. Protocol and final report answer different questions and should remain separate canonical owners.

---

Identify the AI Protocol

Signal the AI component, trial design, registration and protocol version clearly.

Write the promise before recruitment

SPIRIT-AI adds 15 AI-specific items to the protocol framework, with 12 extensions and three elaborations. For identify the ai protocol, use a complete protocol sentence: who will do what, using which version and data, at what point in the workflow, under which rule, and how the result will be checked. Future tense is not decoration; it fixes the plan before outcome knowledge can reshape the story.

Prespecified protocol scenario

For identify the ai protocol, before recruitment, the team defines which image formats are acceptable and what happens when a scan fails quality checks. Translate the scenario into a timeline from data acquisition to action. Place eligibility, preprocessing, user interaction, output, override, error capture and monitoring on that timeline. This makes dependencies visible and helps reviewers find a missing decision rule before the first participant is enrolled.

Amendment and version-control question

For identify the ai protocol, state what would trigger a protocol amendment, who approves it, how the date and rationale will be recorded and which participants or inputs it affects. A software build number alone is not enough when calibration, interface, thresholds or workflow instructions also change the intervention.

Limit and parent–student transfer

For identify the ai protocol, do not treat a registered protocol as a guarantee that every later deviation is inappropriate; some changes are necessary and should be dated and explained. The transferable English habit is to make a plan falsifiable: later readers should be able to see whether the promised method was followed. Students can apply that habit to investigations and projects; parents can ask whether an impressive result followed a plan written before the result was known.

---

Prespecify Intended Use

Name the problem, users, target population and action the system is meant to support.

Write the promise before recruitment

A protocol uses future-tense commitments so readers can compare what investigators planned with what they later did. For prespecify intended use, use a complete protocol sentence: who will do what, using which version and data, at what point in the workflow, under which rule, and how the result will be checked. Future tense is not decoration; it fixes the plan before outcome knowledge can reshape the story.

Prespecified protocol scenario

For prespecify intended use, the planned output is a risk category with a numerical probability, and the protocol states how each category enters the existing care pathway. Translate the scenario into a timeline from data acquisition to action. Place eligibility, preprocessing, user interaction, output, override, error capture and monitoring on that timeline. This makes dependencies visible and helps reviewers find a missing decision rule before the first participant is enrolled.

Amendment and version-control question

For prespecify intended use, state what would trigger a protocol amendment, who approves it, how the date and rationale will be recorded and which participants or inputs it affects. A software build number alone is not enough when calibration, interface, thresholds or workflow instructions also change the intervention.

Limit and parent–student transfer

For prespecify intended use, a protocol cannot foresee every failure, but it should make foreseeable risks and response rules explicit. The transferable English habit is to make a plan falsifiable: later readers should be able to see whether the promised method was followed. Students can apply that habit to investigations and projects; parents can ask whether an impressive result followed a plan written before the result was known.

---

Summarise Existing Evidence

Separate development performance, retrospective validation, live testing and unresolved gaps.

Write the promise before recruitment

The algorithm version, acquisition process and preprocessing steps belong in the plan because they define the intervention before outcomes are known. For summarise existing evidence, use a complete protocol sentence: who will do what, using which version and data, at what point in the workflow, under which rule, and how the result will be checked. Future tense is not decoration; it fixes the plan before outcome knowledge can reshape the story.

Prespecified protocol scenario

For summarise existing evidence, users complete training and a competency check before they can act on the system’s recommendation. Translate the scenario into a timeline from data acquisition to action. Place eligibility, preprocessing, user interaction, output, override, error capture and monitoring on that timeline. This makes dependencies visible and helps reviewers find a missing decision rule before the first participant is enrolled.

Amendment and version-control question

For summarise existing evidence, state what would trigger a protocol amendment, who approves it, how the date and rationale will be recorded and which participants or inputs it affects. A software build number alone is not enough when calibration, interface, thresholds or workflow instructions also change the intervention.

Limit and parent–student transfer

For summarise existing evidence, data access and privacy plans must match applicable governance; reporting guidance does not grant permission to reuse records. The transferable English habit is to make a plan falsifiable: later readers should be able to see whether the promised method was followed. Students can apply that habit to investigations and projects; parents can ask whether an impressive result followed a plan written before the result was known.

---

State Objectives and Hypotheses

Write testable questions that connect the AI intervention to outcomes and risks.

Write the promise before recruitment

The protocol should distinguish participant eligibility from input-data eligibility and plan how poor-quality data will be handled. For state objectives and hypotheses, use a complete protocol sentence: who will do what, using which version and data, at what point in the workflow, under which rule, and how the result will be checked. Future tense is not decoration; it fixes the plan before outcome knowledge can reshape the story.

Prespecified protocol scenario

For state objectives and hypotheses, the protocol records the frozen model version and a contingency plan if a safety-critical defect requires an update. Translate the scenario into a timeline from data acquisition to action. Place eligibility, preprocessing, user interaction, output, override, error capture and monitoring on that timeline. This makes dependencies visible and helps reviewers find a missing decision rule before the first participant is enrolled.

Amendment and version-control question

For state objectives and hypotheses, state what would trigger a protocol amendment, who approves it, how the date and rationale will be recorded and which participants or inputs it affects. A software build number alone is not enough when calibration, interface, thresholds or workflow instructions also change the intervention.

Limit and parent–student transfer

For state objectives and hypotheses, user training can improve consistency without eliminating automation bias, over-reliance or workflow pressure. The transferable English habit is to make a plan falsifiable: later readers should be able to see whether the promised method was followed. Students can apply that habit to investigations and projects; parents can ask whether an impressive result followed a plan written before the result was known.

---

Define Participant and Input Eligibility

Plan who may enter and which data the system can process as two linked decisions.

Write the promise before recruitment

Human–AI interaction must be prespecified at input preparation, output interpretation and clinical action points. For define participant and input eligibility, use a complete protocol sentence: who will do what, using which version and data, at what point in the workflow, under which rule, and how the result will be checked. Future tense is not decoration; it fixes the plan before outcome knowledge can reshape the story.

Prespecified protocol scenario

For define participant and input eligibility, investigators prespecify error categories, subgroup checks and who will review serious failures without seeing treatment allocation unnecessarily. Translate the scenario into a timeline from data acquisition to action. Place eligibility, preprocessing, user interaction, output, override, error capture and monitoring on that timeline. This makes dependencies visible and helps reviewers find a missing decision rule before the first participant is enrolled.

Amendment and version-control question

For define participant and input eligibility, state what would trigger a protocol amendment, who approves it, how the date and rationale will be recorded and which participants or inputs it affects. A software build number alone is not enough when calibration, interface, thresholds or workflow instructions also change the intervention.

Limit and parent–student transfer

For define participant and input eligibility, prespecified subgroup analysis still requires adequate data, appropriate methods and restrained interpretation. The transferable English habit is to make a plan falsifiable: later readers should be able to see whether the promised method was followed. Students can apply that habit to investigations and projects; parents can ask whether an impressive result followed a plan written before the result was known.

---

Lock the Algorithm Version

Record the build, model, thresholds and change-control plan before enrolment.

Write the promise before recruitment

SPIRIT-AI asks investigators to plan analysis of performance errors rather than invent an error narrative after results appear. For lock the algorithm version, use a complete protocol sentence: who will do what, using which version and data, at what point in the workflow, under which rule, and how the result will be checked. Future tense is not decoration; it fixes the plan before outcome knowledge can reshape the story.

Prespecified protocol scenario

For lock the algorithm version, an override is permitted, but the protocol says how the reason and downstream outcome will be recorded. Translate the scenario into a timeline from data acquisition to action. Place eligibility, preprocessing, user interaction, output, override, error capture and monitoring on that timeline. This makes dependencies visible and helps reviewers find a missing decision rule before the first participant is enrolled.

Amendment and version-control question

For lock the algorithm version, state what would trigger a protocol amendment, who approves it, how the date and rationale will be recorded and which participants or inputs it affects. A software build number alone is not enough when calibration, interface, thresholds or workflow instructions also change the intervention.

Limit and parent–student transfer

For lock the algorithm version, an adaptive system needs especially clear versioning because the evaluated intervention can otherwise drift during the trial. The transferable English habit is to make a plan falsifiable: later readers should be able to see whether the promised method was followed. Students can apply that habit to investigations and projects; parents can ask whether an impressive result followed a plan written before the result was known.

---

Plan Data Acquisition

Specify equipment, timing, operators and conditions used to create inputs.

Write the promise before recruitment

A protocol can describe planned updates or adaptation, but it should state decision rules, timing and how versions will be tracked. For plan data acquisition, use a complete protocol sentence: who will do what, using which version and data, at what point in the workflow, under which rule, and how the result will be checked. Future tense is not decoration; it fixes the plan before outcome knowledge can reshape the story.

Prespecified protocol scenario

For plan data acquisition, input acquisition differs across sites, so the protocol names equipment, preprocessing and a harmonisation check rather than assuming equivalence. Translate the scenario into a timeline from data acquisition to action. Place eligibility, preprocessing, user interaction, output, override, error capture and monitoring on that timeline. This makes dependencies visible and helps reviewers find a missing decision rule before the first participant is enrolled.

Amendment and version-control question

For plan data acquisition, state what would trigger a protocol amendment, who approves it, how the date and rationale will be recorded and which participants or inputs it affects. A software build number alone is not enough when calibration, interface, thresholds or workflow instructions also change the intervention.

Limit and parent–student transfer

For plan data acquisition, the protocol should complement the current core SPIRIT statement and relevant AI, ethics and regulatory guidance. The transferable English habit is to make a plan falsifiable: later readers should be able to see whether the promised method was followed. Students can apply that habit to investigations and projects; parents can ask whether an impressive result followed a plan written before the result was known.

---

Plan Selection and Preprocessing

Describe how inputs will be chosen and transformed before the AI sees them.

Write the promise before recruitment

Protocol reporting is not proof of ethical approval, methodological quality or eventual effectiveness. For plan selection and preprocessing, use a complete protocol sentence: who will do what, using which version and data, at what point in the workflow, under which rule, and how the result will be checked. Future tense is not decoration; it fixes the plan before outcome knowledge can reshape the story.

Prespecified protocol scenario

For plan selection and preprocessing, a school research team borrows the same planning habit by defining the tool, user, data, decision and failure rule before collecting results. Translate the scenario into a timeline from data acquisition to action. Place eligibility, preprocessing, user interaction, output, override, error capture and monitoring on that timeline. This makes dependencies visible and helps reviewers find a missing decision rule before the first participant is enrolled.

Amendment and version-control question

For plan selection and preprocessing, state what would trigger a protocol amendment, who approves it, how the date and rationale will be recorded and which participants or inputs it affects. A software build number alone is not enough when calibration, interface, thresholds or workflow instructions also change the intervention.

Limit and parent–student transfer

For plan selection and preprocessing, transfer to another domain means transferring transparent planning, not assuming a clinical checklist fits unchanged. The transferable English habit is to make a plan falsifiable: later readers should be able to see whether the promised method was followed. Students can apply that habit to investigations and projects; parents can ask whether an impressive result followed a plan written before the result was known.

---

Plan Poor-Quality Input Handling

State rejection, repeat, fallback and missing-output rules in advance.

Write the promise before recruitment

SPIRIT-AI adds 15 AI-specific items to the protocol framework, with 12 extensions and three elaborations. For plan poor-quality input handling, use a complete protocol sentence: who will do what, using which version and data, at what point in the workflow, under which rule, and how the result will be checked. Future tense is not decoration; it fixes the plan before outcome knowledge can reshape the story.

Prespecified protocol scenario

For plan poor-quality input handling, before recruitment, the team defines which image formats are acceptable and what happens when a scan fails quality checks. Translate the scenario into a timeline from data acquisition to action. Place eligibility, preprocessing, user interaction, output, override, error capture and monitoring on that timeline. This makes dependencies visible and helps reviewers find a missing decision rule before the first participant is enrolled.

Amendment and version-control question

For plan poor-quality input handling, state what would trigger a protocol amendment, who approves it, how the date and rationale will be recorded and which participants or inputs it affects. A software build number alone is not enough when calibration, interface, thresholds or workflow instructions also change the intervention.

Limit and parent–student transfer

For plan poor-quality input handling, do not treat a registered protocol as a guarantee that every later deviation is inappropriate; some changes are necessary and should be dated and explained. The transferable English habit is to make a plan falsifiable: later readers should be able to see whether the promised method was followed. Students can apply that habit to investigations and projects; parents can ask whether an impressive result followed a plan written before the result was known.

---

Plan Human Interaction With Inputs

Name which users may select, edit or prepare inputs and what expertise is required.

Write the promise before recruitment

A protocol uses future-tense commitments so readers can compare what investigators planned with what they later did. For plan human interaction with inputs, use a complete protocol sentence: who will do what, using which version and data, at what point in the workflow, under which rule, and how the result will be checked. Future tense is not decoration; it fixes the plan before outcome knowledge can reshape the story.

Prespecified protocol scenario

For plan human interaction with inputs, the planned output is a risk category with a numerical probability, and the protocol states how each category enters the existing care pathway. Translate the scenario into a timeline from data acquisition to action. Place eligibility, preprocessing, user interaction, output, override, error capture and monitoring on that timeline. This makes dependencies visible and helps reviewers find a missing decision rule before the first participant is enrolled.

Amendment and version-control question

For plan human interaction with inputs, state what would trigger a protocol amendment, who approves it, how the date and rationale will be recorded and which participants or inputs it affects. A software build number alone is not enough when calibration, interface, thresholds or workflow instructions also change the intervention.

Limit and parent–student transfer

For plan human interaction with inputs, a protocol cannot foresee every failure, but it should make foreseeable risks and response rules explicit. The transferable English habit is to make a plan falsifiable: later readers should be able to see whether the promised method was followed. Students can apply that habit to investigations and projects; parents can ask whether an impressive result followed a plan written before the result was known.

---

Define the Output

Specify the score, category, alert, explanation or recommendation the system will provide.

Write the promise before recruitment

The algorithm version, acquisition process and preprocessing steps belong in the plan because they define the intervention before outcomes are known. For define the output, use a complete protocol sentence: who will do what, using which version and data, at what point in the workflow, under which rule, and how the result will be checked. Future tense is not decoration; it fixes the plan before outcome knowledge can reshape the story.

Prespecified protocol scenario

For define the output, users complete training and a competency check before they can act on the system’s recommendation. Translate the scenario into a timeline from data acquisition to action. Place eligibility, preprocessing, user interaction, output, override, error capture and monitoring on that timeline. This makes dependencies visible and helps reviewers find a missing decision rule before the first participant is enrolled.

Amendment and version-control question

For define the output, state what would trigger a protocol amendment, who approves it, how the date and rationale will be recorded and which participants or inputs it affects. A software build number alone is not enough when calibration, interface, thresholds or workflow instructions also change the intervention.

Limit and parent–student transfer

For define the output, data access and privacy plans must match applicable governance; reporting guidance does not grant permission to reuse records. The transferable English habit is to make a plan falsifiable: later readers should be able to see whether the promised method was followed. Students can apply that habit to investigations and projects; parents can ask whether an impressive result followed a plan written before the result was known.

---

Define the Decision Pathway

State how the output will be displayed, interpreted and acted on or overridden.

Write the promise before recruitment

The protocol should distinguish participant eligibility from input-data eligibility and plan how poor-quality data will be handled. For define the decision pathway, use a complete protocol sentence: who will do what, using which version and data, at what point in the workflow, under which rule, and how the result will be checked. Future tense is not decoration; it fixes the plan before outcome knowledge can reshape the story.

Prespecified protocol scenario

For define the decision pathway, the protocol records the frozen model version and a contingency plan if a safety-critical defect requires an update. Translate the scenario into a timeline from data acquisition to action. Place eligibility, preprocessing, user interaction, output, override, error capture and monitoring on that timeline. This makes dependencies visible and helps reviewers find a missing decision rule before the first participant is enrolled.

Amendment and version-control question

For define the decision pathway, state what would trigger a protocol amendment, who approves it, how the date and rationale will be recorded and which participants or inputs it affects. A software build number alone is not enough when calibration, interface, thresholds or workflow instructions also change the intervention.

Limit and parent–student transfer

For define the decision pathway, user training can improve consistency without eliminating automation bias, over-reliance or workflow pressure. The transferable English habit is to make a plan falsifiable: later readers should be able to see whether the promised method was followed. Students can apply that habit to investigations and projects; parents can ask whether an impressive result followed a plan written before the result was known.

---

Plan User Training

Describe instruction, competency checks, refreshers and support.

Write the promise before recruitment

Human–AI interaction must be prespecified at input preparation, output interpretation and clinical action points. For plan user training, use a complete protocol sentence: who will do what, using which version and data, at what point in the workflow, under which rule, and how the result will be checked. Future tense is not decoration; it fixes the plan before outcome knowledge can reshape the story.

Prespecified protocol scenario

For plan user training, investigators prespecify error categories, subgroup checks and who will review serious failures without seeing treatment allocation unnecessarily. Translate the scenario into a timeline from data acquisition to action. Place eligibility, preprocessing, user interaction, output, override, error capture and monitoring on that timeline. This makes dependencies visible and helps reviewers find a missing decision rule before the first participant is enrolled.

Amendment and version-control question

For plan user training, state what would trigger a protocol amendment, who approves it, how the date and rationale will be recorded and which participants or inputs it affects. A software build number alone is not enough when calibration, interface, thresholds or workflow instructions also change the intervention.

Limit and parent–student transfer

For plan user training, prespecified subgroup analysis still requires adequate data, appropriate methods and restrained interpretation. The transferable English habit is to make a plan falsifiable: later readers should be able to see whether the promised method was followed. Students can apply that habit to investigations and projects; parents can ask whether an impressive result followed a plan written before the result was known.

---

Describe Workflow and Setting

Explain where the system enters routine work and who remains responsible.

Write the promise before recruitment

SPIRIT-AI asks investigators to plan analysis of performance errors rather than invent an error narrative after results appear. For describe workflow and setting, use a complete protocol sentence: who will do what, using which version and data, at what point in the workflow, under which rule, and how the result will be checked. Future tense is not decoration; it fixes the plan before outcome knowledge can reshape the story.

Prespecified protocol scenario

For describe workflow and setting, an override is permitted, but the protocol says how the reason and downstream outcome will be recorded. Translate the scenario into a timeline from data acquisition to action. Place eligibility, preprocessing, user interaction, output, override, error capture and monitoring on that timeline. This makes dependencies visible and helps reviewers find a missing decision rule before the first participant is enrolled.

Amendment and version-control question

For describe workflow and setting, state what would trigger a protocol amendment, who approves it, how the date and rationale will be recorded and which participants or inputs it affects. A software build number alone is not enough when calibration, interface, thresholds or workflow instructions also change the intervention.

Limit and parent–student transfer

For describe workflow and setting, an adaptive system needs especially clear versioning because the evaluated intervention can otherwise drift during the trial. The transferable English habit is to make a plan falsifiable: later readers should be able to see whether the promised method was followed. Students can apply that habit to investigations and projects; parents can ask whether an impressive result followed a plan written before the result was known.

---

Define the Comparator

Prespecify what usual care or alternative workflow contains rather than calling it standard.

Write the promise before recruitment

A protocol can describe planned updates or adaptation, but it should state decision rules, timing and how versions will be tracked. For define the comparator, use a complete protocol sentence: who will do what, using which version and data, at what point in the workflow, under which rule, and how the result will be checked. Future tense is not decoration; it fixes the plan before outcome knowledge can reshape the story.

Prespecified protocol scenario

For define the comparator, input acquisition differs across sites, so the protocol names equipment, preprocessing and a harmonisation check rather than assuming equivalence. Translate the scenario into a timeline from data acquisition to action. Place eligibility, preprocessing, user interaction, output, override, error capture and monitoring on that timeline. This makes dependencies visible and helps reviewers find a missing decision rule before the first participant is enrolled.

Amendment and version-control question

For define the comparator, state what would trigger a protocol amendment, who approves it, how the date and rationale will be recorded and which participants or inputs it affects. A software build number alone is not enough when calibration, interface, thresholds or workflow instructions also change the intervention.

Limit and parent–student transfer

For define the comparator, the protocol should complement the current core SPIRIT statement and relevant AI, ethics and regulatory guidance. The transferable English habit is to make a plan falsifiable: later readers should be able to see whether the promised method was followed. Students can apply that habit to investigations and projects; parents can ask whether an impressive result followed a plan written before the result was known.

---

Prespecify Outcomes

Separate technical, process, user, clinical and safety outcomes with time points.

Write the promise before recruitment

Protocol reporting is not proof of ethical approval, methodological quality or eventual effectiveness. For prespecify outcomes, use a complete protocol sentence: who will do what, using which version and data, at what point in the workflow, under which rule, and how the result will be checked. Future tense is not decoration; it fixes the plan before outcome knowledge can reshape the story.

Prespecified protocol scenario

For prespecify outcomes, a school research team borrows the same planning habit by defining the tool, user, data, decision and failure rule before collecting results. Translate the scenario into a timeline from data acquisition to action. Place eligibility, preprocessing, user interaction, output, override, error capture and monitoring on that timeline. This makes dependencies visible and helps reviewers find a missing decision rule before the first participant is enrolled.

Amendment and version-control question

For prespecify outcomes, state what would trigger a protocol amendment, who approves it, how the date and rationale will be recorded and which participants or inputs it affects. A software build number alone is not enough when calibration, interface, thresholds or workflow instructions also change the intervention.

Limit and parent–student transfer

For prespecify outcomes, transfer to another domain means transferring transparent planning, not assuming a clinical checklist fits unchanged. The transferable English habit is to make a plan falsifiable: later readers should be able to see whether the promised method was followed. Students can apply that habit to investigations and projects; parents can ask whether an impressive result followed a plan written before the result was known.

---

Prespecify Error Analysis

Plan how failures will be classified, reviewed and connected to consequences.

Write the promise before recruitment

SPIRIT-AI adds 15 AI-specific items to the protocol framework, with 12 extensions and three elaborations. For prespecify error analysis, use a complete protocol sentence: who will do what, using which version and data, at what point in the workflow, under which rule, and how the result will be checked. Future tense is not decoration; it fixes the plan before outcome knowledge can reshape the story.

Prespecified protocol scenario

For prespecify error analysis, before recruitment, the team defines which image formats are acceptable and what happens when a scan fails quality checks. Translate the scenario into a timeline from data acquisition to action. Place eligibility, preprocessing, user interaction, output, override, error capture and monitoring on that timeline. This makes dependencies visible and helps reviewers find a missing decision rule before the first participant is enrolled.

Amendment and version-control question

For prespecify error analysis, state what would trigger a protocol amendment, who approves it, how the date and rationale will be recorded and which participants or inputs it affects. A software build number alone is not enough when calibration, interface, thresholds or workflow instructions also change the intervention.

Limit and parent–student transfer

For prespecify error analysis, do not treat a registered protocol as a guarantee that every later deviation is inappropriate; some changes are necessary and should be dated and explained. The transferable English habit is to make a plan falsifiable: later readers should be able to see whether the promised method was followed. Students can apply that habit to investigations and projects; parents can ask whether an impressive result followed a plan written before the result was known.

---

Plan Harms Monitoring

Name expected and unexpected harms, reporting routes and stopping decisions.

Write the promise before recruitment

A protocol uses future-tense commitments so readers can compare what investigators planned with what they later did. For plan harms monitoring, use a complete protocol sentence: who will do what, using which version and data, at what point in the workflow, under which rule, and how the result will be checked. Future tense is not decoration; it fixes the plan before outcome knowledge can reshape the story.

Prespecified protocol scenario

For plan harms monitoring, the planned output is a risk category with a numerical probability, and the protocol states how each category enters the existing care pathway. Translate the scenario into a timeline from data acquisition to action. Place eligibility, preprocessing, user interaction, output, override, error capture and monitoring on that timeline. This makes dependencies visible and helps reviewers find a missing decision rule before the first participant is enrolled.

Amendment and version-control question

For plan harms monitoring, state what would trigger a protocol amendment, who approves it, how the date and rationale will be recorded and which participants or inputs it affects. A software build number alone is not enough when calibration, interface, thresholds or workflow instructions also change the intervention.

Limit and parent–student transfer

For plan harms monitoring, a protocol cannot foresee every failure, but it should make foreseeable risks and response rules explicit. The transferable English habit is to make a plan falsifiable: later readers should be able to see whether the promised method was followed. Students can apply that habit to investigations and projects; parents can ask whether an impressive result followed a plan written before the result was known.

---

Plan Data Management and Version Control

Keep software, data, thresholds, amendments and analysis files traceable.

Write the promise before recruitment

The algorithm version, acquisition process and preprocessing steps belong in the plan because they define the intervention before outcomes are known. For plan data management and version control, use a complete protocol sentence: who will do what, using which version and data, at what point in the workflow, under which rule, and how the result will be checked. Future tense is not decoration; it fixes the plan before outcome knowledge can reshape the story.

Prespecified protocol scenario

For plan data management and version control, users complete training and a competency check before they can act on the system’s recommendation. Translate the scenario into a timeline from data acquisition to action. Place eligibility, preprocessing, user interaction, output, override, error capture and monitoring on that timeline. This makes dependencies visible and helps reviewers find a missing decision rule before the first participant is enrolled.

Amendment and version-control question

For plan data management and version control, state what would trigger a protocol amendment, who approves it, how the date and rationale will be recorded and which participants or inputs it affects. A software build number alone is not enough when calibration, interface, thresholds or workflow instructions also change the intervention.

Limit and parent–student transfer

For plan data management and version control, data access and privacy plans must match applicable governance; reporting guidance does not grant permission to reuse records. The transferable English habit is to make a plan falsifiable: later readers should be able to see whether the promised method was followed. Students can apply that habit to investigations and projects; parents can ask whether an impressive result followed a plan written before the result was known.

---

Plan Updates and Adaptation

State whether learning or recalibration is allowed and how evaluated versions remain comparable.

Write the promise before recruitment

The protocol should distinguish participant eligibility from input-data eligibility and plan how poor-quality data will be handled. For plan updates and adaptation, use a complete protocol sentence: who will do what, using which version and data, at what point in the workflow, under which rule, and how the result will be checked. Future tense is not decoration; it fixes the plan before outcome knowledge can reshape the story.

Prespecified protocol scenario

For plan updates and adaptation, the protocol records the frozen model version and a contingency plan if a safety-critical defect requires an update. Translate the scenario into a timeline from data acquisition to action. Place eligibility, preprocessing, user interaction, output, override, error capture and monitoring on that timeline. This makes dependencies visible and helps reviewers find a missing decision rule before the first participant is enrolled.

Amendment and version-control question

For plan updates and adaptation, state what would trigger a protocol amendment, who approves it, how the date and rationale will be recorded and which participants or inputs it affects. A software build number alone is not enough when calibration, interface, thresholds or workflow instructions also change the intervention.

Limit and parent–student transfer

For plan updates and adaptation, user training can improve consistency without eliminating automation bias, over-reliance or workflow pressure. The transferable English habit is to make a plan falsifiable: later readers should be able to see whether the promised method was followed. Students can apply that habit to investigations and projects; parents can ask whether an impressive result followed a plan written before the result was known.

---

Protect Privacy and Access

Describe governance, access, retention and accountable data use without overstating anonymity.

Write the promise before recruitment

Human–AI interaction must be prespecified at input preparation, output interpretation and clinical action points. For protect privacy and access, use a complete protocol sentence: who will do what, using which version and data, at what point in the workflow, under which rule, and how the result will be checked. Future tense is not decoration; it fixes the plan before outcome knowledge can reshape the story.

Prespecified protocol scenario

For protect privacy and access, investigators prespecify error categories, subgroup checks and who will review serious failures without seeing treatment allocation unnecessarily. Translate the scenario into a timeline from data acquisition to action. Place eligibility, preprocessing, user interaction, output, override, error capture and monitoring on that timeline. This makes dependencies visible and helps reviewers find a missing decision rule before the first participant is enrolled.

Amendment and version-control question

For protect privacy and access, state what would trigger a protocol amendment, who approves it, how the date and rationale will be recorded and which participants or inputs it affects. A software build number alone is not enough when calibration, interface, thresholds or workflow instructions also change the intervention.

Limit and parent–student transfer

For protect privacy and access, prespecified subgroup analysis still requires adequate data, appropriate methods and restrained interpretation. The transferable English habit is to make a plan falsifiable: later readers should be able to see whether the promised method was followed. Students can apply that habit to investigations and projects; parents can ask whether an impressive result followed a plan written before the result was known.

---

Preserve Human Oversight

Clarify roles, override authority, escalation and responsibility.

Write the promise before recruitment

SPIRIT-AI asks investigators to plan analysis of performance errors rather than invent an error narrative after results appear. For preserve human oversight, use a complete protocol sentence: who will do what, using which version and data, at what point in the workflow, under which rule, and how the result will be checked. Future tense is not decoration; it fixes the plan before outcome knowledge can reshape the story.

Prespecified protocol scenario

For preserve human oversight, an override is permitted, but the protocol says how the reason and downstream outcome will be recorded. Translate the scenario into a timeline from data acquisition to action. Place eligibility, preprocessing, user interaction, output, override, error capture and monitoring on that timeline. This makes dependencies visible and helps reviewers find a missing decision rule before the first participant is enrolled.

Amendment and version-control question

For preserve human oversight, state what would trigger a protocol amendment, who approves it, how the date and rationale will be recorded and which participants or inputs it affects. A software build number alone is not enough when calibration, interface, thresholds or workflow instructions also change the intervention.

Limit and parent–student transfer

For preserve human oversight, an adaptive system needs especially clear versioning because the evaluated intervention can otherwise drift during the trial. The transferable English habit is to make a plan falsifiable: later readers should be able to see whether the promised method was followed. Students can apply that habit to investigations and projects; parents can ask whether an impressive result followed a plan written before the result was known.

---

Did You Know? A Protocol Protects Against Hindsight

A dated plan reveals when results influenced later choices.

Write the promise before recruitment

A protocol can describe planned updates or adaptation, but it should state decision rules, timing and how versions will be tracked. For did you know? a protocol protects against hindsight, use a complete protocol sentence: who will do what, using which version and data, at what point in the workflow, under which rule, and how the result will be checked. Future tense is not decoration; it fixes the plan before outcome knowledge can reshape the story.

Prespecified protocol scenario

For did you know? a protocol protects against hindsight, input acquisition differs across sites, so the protocol names equipment, preprocessing and a harmonisation check rather than assuming equivalence. Translate the scenario into a timeline from data acquisition to action. Place eligibility, preprocessing, user interaction, output, override, error capture and monitoring on that timeline. This makes dependencies visible and helps reviewers find a missing decision rule before the first participant is enrolled.

Amendment and version-control question

For did you know? a protocol protects against hindsight, state what would trigger a protocol amendment, who approves it, how the date and rationale will be recorded and which participants or inputs it affects. A software build number alone is not enough when calibration, interface, thresholds or workflow instructions also change the intervention.

Limit and parent–student transfer

For did you know? a protocol protects against hindsight, the protocol should complement the current core SPIRIT statement and relevant AI, ethics and regulatory guidance. The transferable English habit is to make a plan falsifiable: later readers should be able to see whether the promised method was followed. Students can apply that habit to investigations and projects; parents can ask whether an impressive result followed a plan written before the result was known.

---

Finish With a SPIRIT-AI Audit

Check version, inputs, users, outputs, decisions, errors, harms, oversight and amendments before recruitment.

Write the promise before recruitment

Protocol reporting is not proof of ethical approval, methodological quality or eventual effectiveness. For finish with a spirit-ai audit, use a complete protocol sentence: who will do what, using which version and data, at what point in the workflow, under which rule, and how the result will be checked. Future tense is not decoration; it fixes the plan before outcome knowledge can reshape the story.

Prespecified protocol scenario

For finish with a spirit-ai audit, a school research team borrows the same planning habit by defining the tool, user, data, decision and failure rule before collecting results. Translate the scenario into a timeline from data acquisition to action. Place eligibility, preprocessing, user interaction, output, override, error capture and monitoring on that timeline. This makes dependencies visible and helps reviewers find a missing decision rule before the first participant is enrolled.

Amendment and version-control question

For finish with a spirit-ai audit, state what would trigger a protocol amendment, who approves it, how the date and rationale will be recorded and which participants or inputs it affects. A software build number alone is not enough when calibration, interface, thresholds or workflow instructions also change the intervention.

Limit and parent–student transfer

For finish with a spirit-ai audit, transfer to another domain means transferring transparent planning, not assuming a clinical checklist fits unchanged. The transferable English habit is to make a plan falsifiable: later readers should be able to see whether the promised method was followed. Students can apply that habit to investigations and projects; parents can ask whether an impressive result followed a plan written before the result was known.

---

Frequently Asked Questions

What is SPIRIT-AI?

It is a reporting extension for protocols of clinical trials evaluating interventions with an artificial-intelligence component.

How many AI-specific items are included?

The extension contains 15 items: 12 extensions and three elaborations.

How is it different from CONSORT-AI?

SPIRIT-AI reports the plan before the trial; CONSORT-AI reports what was done and found.

Why name the algorithm version in a protocol?

A changed model, threshold or interface can change the intervention. Versioning makes the evaluated promise traceable.

Can a protocol be amended?

Yes. Necessary changes should be dated, justified and linked to the affected methods and participants rather than hidden.

Does the checklist replace ethics or regulation?

No. It supports complete reporting and must sit beside applicable ethics, governance, regulatory and methodological requirements.

---

Useful Next Reading

• Read the primary SPIRIT-AI statement.
• Read the open full-text protocol guidance.
• Check the current EQUATOR SPIRIT-AI record.
• Connect AI items to the current core SPIRIT guide.

Closing thought: English protects the before-and-after comparison: the protocol says what will happen, the amendment says what changed, and the report says what actually occurred.