Measuring what learners know and can do
Human Authorship and Learning Verification Standard
Educational achievement should be credited only when evidence shows that the learner possesses and can exercise the relevant knowledge, reasoning, skill or judgment. No single detector, assignment product or immediate score is sufficient.
Project, Research, Analysis, Editorial by Patrick Russell; assisted by ChatGPT (OpenAI).
Measuring what learners know and can do when artificial intelligence can imitate performance
Main conclusion
Educational achievement should be credited only when the evidence shows that the learner possesses and can exercise the relevant knowledge, reasoning, skill, or judgment. A polished submission, detector score, account log, or consistent writing style cannot establish that conclusion by itself. Institutions need a recurring chain of evidence that begins with a defined educational objective, records permitted assistance, preserves protected human performance, and tests retention and transfer after assistance is removed.
This standard converts the findings of Phase Twenty Two into an operating framework. It is designed for institutional review, course design, family oversight, and public investigation. It does not assume misconduct when evidence is incomplete and does not require every task to be performed without artificial intelligence. It requires the institution to identify which human capability an activity is intended to develop and to verify that capability directly.
What the standard adds
This table connects each verification need to the response required to establish that a claimed capability belongs to the learner.
Evidence tableView 5 recordsHide 5 records
Need | Required response |
|---|---|
Synthetic authorship | Separate the quality of a product from the learner’s independently demonstrated capability |
Personalized artificial intelligence | Test new application and explanation rather than relying on familiar style or prior answers |
Detector uncertainty | Use detector output only as a review signal and require corroborating human evidence |
Assisted learning | Compare performance while assistance is available with performance after removal and delay |
Public accountability | Report policy implementation and measured outcomes separately from institutional claims |
Verification principles
Capability rather than product
A product can be genuine, assisted, jointly produced, or substantially generated. The educational question is narrower and more important: can the learner independently perform the intellectual act for which credit is awarded? Verification should therefore examine retrieval, explanation, source use, reasoning, error correction, creation, and transfer under known conditions.
The mental act governs the rule
The same artificial-intelligence use may be appropriate in one activity and destructive in another. A language model may help a student compare arguments after the student has written an analysis, but it may displace the lesson if the objective is to construct the first argument. A coding assistant may model professional practice, but it cannot substitute for a learner’s ability to trace logic or repair an unfamiliar error when those are the assessed capacities.
No single measure is sufficient
- A submitted product shows what was delivered, not necessarily who performed each intellectual step.
- Version history and platform logs show recorded activity, not comprehension or unrecorded assistance.
- A detector estimates textual patterns and may misclassify human or altered work.
- An oral explanation can reveal understanding but may also be affected by anxiety, language, disability, or poor questioning.
- A protected task shows performance under one condition but not automatically long-term retention or broad transfer.
- A defensible judgment combines measures that answer different questions and preserves the learner’s opportunity to explain contrary evidence.
Minimum institutional duty
An institution that authorizes or tolerates artificial-intelligence assistance should also preserve a meaningful way to determine what students can do without it. The California Department of Education model policy now recommends original work, disclosure, alternative assessment, documented evidence of authorship, human review, rebuttal rights, curriculum redesign, and recurring monitoring. That guidance is nonmandatory, but it establishes a useful public benchmark. [1]
Evidence architecture
The verification chain should contain six distinct layers. A weakness in one layer should not be concealed by strength in another.
Evidence tableView 6 recordsHide 6 records
Layer | Evidence | Decision question |
|---|---|---|
Purpose | Course objective assignment objective and human capacity to be formed | What must the learner eventually control? |
Rule | Permitted prohibited and disclosure-required uses with applicable policy version | What assistance was authorized? |
Process | Notes drafts sources prompts feedback revisions and collaboration records when relevant | How was the work produced? |
Protected performance | Observed oral written practical mathematical creative or technical work | Can the learner perform without generation? |
Retention and transfer | A delayed task and a new problem using the same underlying knowledge or method | Did learning persist and generalize? |
Experience and effects | Workload accessibility trust motivation attention relationships wellness and family burden | What benefits or costs accompanied the result? |
Evidence classifications
This table defines the strength and limits of each evidence classification so published rules, actual practice and verified outcomes are not confused.
Evidence tableView 5 recordsHide 5 records
Classification | Meaning |
|---|---|
Published | An official current source states a policy objective practice or tool arrangement |
Implemented | Operating records or representative artifacts show that the stated practice occurred |
Measured | A defined measure and comparison show a result under stated conditions |
Verified capability | Protected performance plus retention or transfer supports independent student command |
Unresolved | The necessary evidence is unavailable incomplete or contradictory |
These classifications are not grades. They prevent a policy, anecdote, usage statistic, or polished example from being mistaken for evidence of durable learning.
Capability ladder
Capability should be evaluated as a progression. Institutions may adjust the content and developmental level, but they should not collapse the upper levels into product completion.
Evidence tableView 6 recordsHide 6 records
Level | Student demonstration | Suitable evidence |
|---|---|---|
Recognition | Identifies a concept fact source method or error | Brief oral or written identification under protected conditions |
Recall | Retrieves central knowledge without generated assistance | Low-stakes retrieval writing recitation calculation or sketch |
Explanation | Explains why a claim method or choice is sound | Oral defense annotated solution source explanation or teach-back |
Application | Uses knowledge in a new but related task | Comparable problem new passage practical task or unfamiliar case |
Judgment | Weighs uncertainty alternatives consequences and ethical limits | Case analysis cross-examination design defense or decision memorandum |
Creation | Produces original work and can account for its choices and revisions | Protected composition design demonstration performance or project defense |
What artificial intelligence may contribute
Artificial intelligence can support questioning, examples, accessibility, critique, translation, research organization, simulation, and feedback. Credit for those benefits should be separated from claims about the learner. A learner may use a tool expertly while still lacking the underlying knowledge; another may possess strong knowledge while declining or needing accommodation in a particular tool. The record should preserve both forms of competence rather than treating them as interchangeable.
The removal condition
At least one assessment of each essential outcome should remove generative assistance. Removal need not prohibit all ordinary supports. It should preserve authorized assistive technology, language access, and disability accommodations while withholding the system that can supply the target reasoning or answer.
Artificial intelligence use conditions
This table distinguishes different conditions of artificial-intelligence use, their educational purpose and the evidence required under each condition.
Evidence tableView 6 recordsHide 6 records
Condition | Purpose | Evidence required |
|---|---|---|
Protected human work | Establish a baseline or verify mastery | Known conditions identity controls authorized accommodations and no generative assistance |
Bounded assistance | Support a defined step without replacing the objective | Permitted functions disclosure and comparison with unaided work |
Guided tutor | Prompt reasoning practice or feedback | Conversation sample teacher mediation error review and later independent test |
Source bounded research | Work within a defined evidence collection | Source opening citation verification and learner explanation of selection |
Accountable collaboration | Model professional human and artificial-intelligence work | Contribution record final human responsibility and protected defense |
Open generation | Explore possibilities or produce a tool-centered artifact | No inference of independent mastery unless followed by direct verification |
Assignment statement
Every consequential assignment should state the learning objective, permitted assistance, required disclosure, records to preserve, protected verification method, applicable privacy limits, and consequences for misunderstanding or violating the rule. A generic course policy is insufficient when permissions change by assignment.
Personalized systems
If a system has access to a learner’s prior essays, notes, feedback, messages, source library, or repeated corrections, it should be treated as a personalized assistant even if it is not marketed as a digital twin. The disclosure should identify the data categories used. The protected verification should employ a fair unfamiliar task so the system’s imitation of prior patterns cannot substitute for new student reasoning.
Core verification protocol
1. Define the human capability. State the knowledge, reasoning, skill, judgment, or habit that the learner is expected to possess.
2. Record the operating rule. Preserve the course, assignment, tool, account, date, policy version, permitted assistance, and required disclosure.
3. Establish a baseline. Collect a brief protected sample before substantial assistance or early in the course under known conditions.
4. Preserve proportionate process evidence. Retain the materials needed to understand authorship and learning without collecting unrelated personal data.
5. Review the submitted product and sources. Check accuracy, evidence use, revisions, and whether the student can identify the contribution of assistance.
6. Conduct a protected understanding check. Ask the learner to explain a central choice, correct an error, verify a source, or solve a comparable problem.
7. Test transfer. Use an unfamiliar but equivalent task that requires the same underlying knowledge or method.
8. Test delayed retention when the outcome matters. Repeat a shorter protected measure after sufficient time for temporary performance effects to fade.
9. Review experience and equity. Record accommodations, anxiety, language, workload, access, trust, motivation, and relevant family or wellness conditions.
10. Classify the finding. Distinguish assisted product quality, independent capability, policy compliance, and unresolved evidence.
Proportionality
Not every worksheet requires a formal defense. The strength of verification should rise with the importance of the claim. Graduation requirements, placement, professional competence, advanced credit, major projects, and allegations of misconduct require stronger and more reviewable evidence than exploratory practice.
Protected understanding checks
This table compares methods for verifying understanding and shows how each should be used without claiming more than it can prove.
Evidence tableView 6 recordsHide 6 records
Method | Best use | Protection against overclaiming |
|---|---|---|
Oral explanation | Reasoning interpretation decisions and source use | Questions should be connected to the objective and scored with stated criteria |
Supervised writing | Voice organization recall and argument construction | Use a comparable prompt rather than demanding reproduction of a prior essay |
Worked problem | Mathematics science coding logic and technical procedure | Require explanation and error correction rather than answer matching alone |
Source defense | Research history civics science and professional work | Ask why a source was selected and what evidence could change the conclusion |
Practical demonstration | Laboratory art performance design and applied competence | Observe the learner making decisions rather than judging only the finished object |
Teach back | Conceptual understanding and synthesis | Use follow-up questions to distinguish memorized language from flexible knowledge |
Fair questioning
- Tell students in advance that understanding checks may occur and explain their educational purpose.
- Do not use trick questions or demand facts beyond the stated learning objective.
- Provide an alternative mode when oral performance is not the capability being assessed and disability or language conditions would distort the result.
- Use a short rubric that separates knowledge accuracy reasoning source command and responsiveness to correction.
- Allow the learner to identify anxiety technical difficulty ambiguous wording or an authorized accommodation before an adverse conclusion.
- Preserve enough evidence for consistent review without recording more personal information than necessary.
Retention and transfer
Why immediate performance is insufficient
Artificial intelligence can improve an answer while it is available without producing durable learning. A verification design should therefore distinguish immediate assisted performance, immediate independent performance, delayed independent retention, and independent transfer to a new problem. These are different outcomes and should be reported separately.
Evidence tableView 5 recordsHide 5 records
Measure | Timing | Interpretation |
|---|---|---|
Assisted product | During the assignment | Shows performance of the learner and available system together |
Immediate removal | At completion or shortly afterward | Tests current independent command and possible near transfer |
Delayed retention | After a stated interval | Tests whether essential knowledge or method remains available |
Novel transfer | On an unfamiliar but comparable task | Tests whether understanding can be applied beyond rehearsed examples |
Longitudinal outcome | Across a term year or later course | Tests growth durability dependency and broader educational effects |
Comparison design
Where feasible, compare protected human work, assistance alone, and assistance with active educator or parent mediation. Record starting capability, actual tool use, prior artificial-intelligence experience, course selection, attrition, accommodations, family support, and COVID-era learning conditions. Satisfaction and speed should not substitute for retention, transfer, or judgment.
The Institute of Education Sciences What Works Clearinghouse provides standards for evaluating causal evidence. This investigation does not claim that a local classroom comparison automatically satisfies those standards, but the comparison should still specify groups, measures, timing, attrition, and limitations. [6]
Authorship review and fair process
An authorship inquiry and a learning verification are related but not identical. A student may misunderstand the material without violating a rule, or may use unauthorized assistance while still understanding the material. Institutions should reach separate findings on policy compliance and capability.
Evidence tableView 8 recordsHide 8 records
Stage | Required practice |
|---|---|
Clarify the rule | Identify the exact assignment policy and whether the questioned use was prohibited or required disclosure |
Preserve evidence | Retain the submission sources drafts version history and technical signal without altering the original record |
Human review | Compare ordinary work and process evidence before relying on a detector or platform alert |
Student response | Ask neutral questions and allow the student to explain process collaboration accessibility and contrary evidence |
Capability check | Use a fair protected task connected to the learning objective |
Separate findings | Distinguish weak understanding incomplete disclosure unauthorized assistance and deliberate misrepresentation |
Proportionate response | Use educational repair when appropriate and disciplinary consequences only under established policy and evidence |
Appeal | Provide review by a person not committed to the initial conclusion |
Detector boundary
The California Department of Education model policy says detector software should not be the sole basis for discipline or a grade penalty. Peer-reviewed research has also found bias against non-native English writers. A detector may begin a review, but it cannot identify the author, establish the applicable rule, determine intent, or prove what the learner understands. [1] [10]
Privacy accessibility and research protections
Data minimization
Authorship comparison can tempt institutions to collect a broad archive of student writing, messages, behavior, and biometric or voice data. The archive itself can become the training material for a personalized proxy. Collect only what is necessary for a defined educational purpose, restrict access, state retention and deletion periods, and prohibit unrelated model training or advertising.
Education records and consent
The Family Educational Rights and Privacy Act governs access to and disclosure of education records at covered institutions. California’s model policy additionally recommends annual guardian consent for autonomous artificial-intelligence student interaction tools and human review of outputs affecting a child. The specific legal obligation depends on the institution, tool, data flow, age, and use; a general consent form should not be treated as universal authorization. [1] [4]
Accessibility
The removal condition must not remove an accommodation merely because it uses software. The institution should distinguish the support that provides access from the function that supplies the target answer or reasoning. Students should receive a comparable verification route when speech, handwriting, language, attention, sensory, motor, or anxiety conditions would otherwise distort the measure.
Research involving children
A school improvement review is not automatically human-subject research, but a systematic investigation designed for generalizable knowledge may require institutional review and additional protections. Federal Subpart D sets additional protections for children involved as research subjects. Parent permission, child assent when appropriate, minimal risk, privacy, and limits on coercion should be determined before collecting participant evidence. [7]
Developmental standard for K through 12
This table sets age-appropriate priorities for protected human performance and responsible artificial-intelligence use across K–12 development.
Evidence tableView 4 recordsHide 4 records
Stage | Protected priority | Permitted evidence of responsible use |
|---|---|---|
Early elementary | Language number memory handwriting narration attention and direct relationship | Primarily educator or parent use; child-facing generation only for a specific supervised purpose |
Later elementary | Foundational knowledge composition calculation reading and source habits | Short guided comparison after independent work with parent or educator visibility |
Middle school | Argument evidence revision self-command and growing independent research | Bounded tutoring critique or source work followed by protected explanation and transfer |
High school | Synthesis judgment authorship disciplinary competence and preparation for adult use | Disclosed accountable collaboration with recurring unaided writing oral defense and novel problems |
Parent and caregiver role
- Know which tools and accounts are authorized, what data they receive, and whether the tool retains or trains on student work.
- Understand the assignment rule before supplying outside artificial-intelligence assistance.
- Observe whether the child can explain and reuse knowledge away from the tool.
- Preserve the child’s opportunity to struggle productively rather than optimizing every product.
- Report accessibility needs, family burden, dependency, secrecy, sleep disruption, anxiety, or withdrawal from human help.
- Request human review when automated feedback, grading, monitoring, or an authorship allegation affects the child.
K through 12 threshold
A K through 12 program should not claim educational benefit from student-facing artificial intelligence merely because students used it, enjoyed it, completed more work, or produced stronger-looking products. The minimum claim requires age-appropriate independent evidence and an account of parent notice, consent, alternatives, teacher mediation, privacy, and wellness.
Higher education standard
Faculty authority and institutional duty
Faculty and departments should decide which mental acts a course must protect. The institution should provide the common infrastructure that individual courses cannot supply alone: approved accounts, privacy and security rules, accessibility support, records, due process, training, audit capacity, and transparent outcome reporting.
Evidence tableView 6 recordsHide 6 records
Educational setting | Minimum protected evidence |
|---|---|
Foundational course | Independent recall explanation and application before broad generation is credited |
Writing and humanities | Protected composition source defense revision explanation and preservation of student voice |
Mathematics science and engineering | Worked reasoning error correction novel problems and practical or laboratory demonstration |
Coding and design | Traceable decisions debugging explanation protected construction and defense of generated components |
Professional education | Judgment under uncertainty ethical limits source verification and accountable responsibility |
Capstone thesis or major project | Contribution record protected oral defense novel follow-up and independent command of central claims |
Degree claim
If a degree or credential represents that the graduate possesses certain capabilities, the institution should preserve evidence that those capabilities were demonstrated by the student rather than only by the student working with an unrestricted proxy. Institutions may also certify artificial-intelligence collaboration as a separate competence, but it should not erase the independent capability claim.
Shared responsibility for educational outcomes
This table assigns verification responsibilities across students, families, educators and institutions so accountability does not disappear between them.
Evidence tableView 8 recordsHide 8 records
Participant | Verification responsibility |
|---|---|
Governing body and leadership | Define educational outcomes approve safeguards require audits publish limitations and stop uses that cannot be justified |
Faculty and teachers | Align rules with learning objectives preserve baseline work conduct understanding checks and interpret evidence |
Students | Follow assignment rules disclose material assistance preserve required process evidence and demonstrate independent command |
Parents and caregivers | Understand authorized use observe transfer communicate conditions and seek human review when needed |
Technology privacy and security staff | Document data flows settings retention access incidents vendors and deletion |
Accessibility and wellness staff | Prevent verification from measuring disability or distress instead of learning and monitor broader effects |
Researchers and evaluators | Use credible comparisons report null and adverse results preserve limitations and disclose conflicts |
Tool providers | Enable transparency export deletion human control and independent evaluation without claiming learning from usage alone |
Engagement is part of the safeguard
The increasing ability of artificial intelligence to simulate a learner makes sustained human engagement more necessary, not less. Educators need contact with students’ reasoning and development. Parents need understandable notice and a realistic way to observe transfer. Students need clear rules and meaningful opportunities to demonstrate capability. Institutions remain responsible for designing and measuring the educational outcome rather than delegating that judgment to a platform.
Institution readiness findings
The following determinations apply the standard to the public record summarized in Phase Twenty Two. They are readiness judgments, not rankings and not claims about unobserved classroom practice.
Evidence tableView 7 recordsHide 7 records
Institution | Most useful existing foundation | Current verification status |
|---|---|---|
UC Berkeley | Restrictive Berkeley Law rule and central teaching resources | Partly established at unit level; campus-wide retention and transfer unresolved |
UCLA | Assessment integrity principles literacy module and faculty initiatives | Framework established; recurring protected outcome evidence unresolved |
Stanford University | Student guidance course authority and proposed understanding checks | Strong conceptual model; recurring implementation and results unresolved |
University of Southern California | Managed account governance and faculty control | Infrastructure established; independent capability evidence unresolved |
California Institute of Technology | Instructor authority disclosure privacy and detector skepticism | Strong course-level foundation; cross-disciplinary transfer evidence unresolved |
Loyola Marymount University | AI-free and integrated options with protected tools | Promising choice model; comparable independent results unresolved |
Thomas Aquinas College | Direct oral written mathematical laboratory and thesis performance | Strong protected architecture; measurement reliability accessibility and later AI competence incomplete |
K through 12 readiness findings
This table compares the selected K–12 institutions by their strongest existing foundation and their present readiness to verify independent capability.
Evidence tableView 3 recordsHide 3 records
Institution | Most useful existing foundation | Current verification status |
|---|---|---|
Mother of Divine Grace School | Parent-mediated developmental sequence narration recitation composition logic and discussion | Strong observation structure; age-specific AI rule and comparable outcomes unresolved |
Harvard-Westlake School | Writing research independent inquiry wellness support and AI ethics study | Capability exists to implement the standard; operative schoolwide evidence unresolved |
High Tech High | Authentic projects critique exhibition student voice and board action | Public performance is strong; individual mastery and full policy evidence incomplete |
Cross institution finding
No institution in the reviewed group has publicly demonstrated the complete chain across its programs: defined formation outcome, assignment-level artificial-intelligence rule, actual-use evidence, protected human performance, delayed independent retention, novel transfer, wellness review, and comparable reporting. Several institutions have strong components and could test the standard immediately. The missing evidence should be classified as unresolved rather than interpreted as institutional failure or success.
No final ranking
A numerical ranking would reward differences in public documentation and institutional type as though they were equivalent learning results. The scorecard should remain categorical until common evidence exists. Formation, governance, implementation, verified capability, and wellness should remain separate dimensions.
Initial verification cases
This table identifies a practical first verification case for each institution, allowing the standard to be tested before broader conclusions are drawn.
Evidence tableView 10 recordsHide 10 records
Institution | Recommended first case |
|---|---|
UC Berkeley | Compare one restrictive Berkeley Law assignment with one bounded-use course using source defense and delayed transfer |
UCLA | Evaluate the student literacy module and a Faculty Fellows course with baseline protected work and later independent application |
Stanford University | Implement and evaluate an understanding check proposed through COLLEGE 102 in representative courses |
University of Southern California | Compare managed-tool access alone with access plus active faculty mediation and protected post-use performance |
California Institute of Technology | Use parallel writing coding mathematics and science tasks with explanation error correction and novel problems |
Loyola Marymount University | Compare AI-free bounded and integrated course models under the same independent capability measures |
Thomas Aquinas College | Preserve composition and senior-thesis process evidence oral defense accommodations and later responsible tool competence |
Mother of Divine Grace School | Develop parent-reviewed stage rules and test protected composition narration and transfer before bounded use |
Harvard-Westlake School | Follow one AI ethics inquiry from source selection through protected writing oral defense and wellness review |
High Tech High | Pair a collaborative public project with individual source defense protected reconstruction and delayed transfer |
Common minimum dataset
Each case should preserve the educational objective, participant characteristics, baseline capability, assignment rule, actual artificial-intelligence use, protected measure, delayed interval, transfer task, scoring method, accessibility conditions, student and educator experience, privacy treatment, attrition, and limitations. Raw student work should not be publicly released merely to make the investigation appear transparent.
Participant evidence and open investigation
Several decisive questions cannot be answered from public sources. The project should remain open to evidence from institutions, educators, students, graduates, parents, caregivers, researchers, and relevant staff. Participation should be voluntary, proportionate, privacy-protective, and clearly separated from accusation or compelled disclosure.
Evidence tableView 6 recordsHide 6 records
Participant | Evidence that would materially improve the finding |
|---|---|
Institutional leaders | Operative policies authority records approved tools audit plans implementation reports and revisions |
Educators | Syllabi assignment rules student-facing instructions anonymized process evidence rubrics workload and verification practice |
Students and graduates | De-identified experience of assistance disclosure learning dependence trust relationships and unaided capability |
Parents and caregivers | Notice consent alternatives developmental observations family burden and whether concerns changed practice |
Technology privacy and accessibility staff | Settings data classes retention deletion incidents accommodations and human review |
Researchers and vendors | Methods comparison groups delayed measures adverse results subgroup effects limitations and conflicts |
Recommendation for contributors
Contributors should identify the institution, grade or program, course, term, tool, account type, policy version, and evidence category. They should distinguish personal experience from an official rule and immediate performance from durable learning. Student identities, education records, private communications, and sensitive health information should be removed unless lawful consent and a defined need justify collection.
Correction and response
Named institutions should have a reasonable opportunity to correct factual errors and supply contrary evidence. Conflicting accounts should be preserved and classified rather than silently reconciled. The investigation should publish the basis and limitation of a changed judgment.
Principal public standards and guidance
1. California Department of Education model artificial intelligence policy June 2026 nonmandatory model covering original work disclosure detector limits alternative assessment human review consent curriculum redesign monitoring and accountability.
2. California guidance for safe and effective use of artificial intelligence in public schools 2025 guidance published in 2026 addressing human-centered learning academic integrity privacy equity and educator support.
3. Empowering Education Leaders United States Department of Education toolkit for safe ethical and equitable artificial-intelligence integration including privacy civil rights transparency and opt-out considerations.
4. Family Educational Rights and Privacy Act United States Department of Education information on rights concerning access to and disclosure of education records.
5. National Institute of Standards and Technology Artificial Intelligence Risk Management Framework Voluntary risk framework and resources for testing evaluation verification validation documentation and continuing risk management.
6. What Works Clearinghouse handbooks and standards Institute of Education Sciences standards for reviewing evidence and interpreting intervention effects.
7. Additional protections for children involved as subjects in research Federal research protections concerning children including permission assent risk and review conditions.
8. Artificial Intelligence and the Future of Teaching and Learning United States Department of Education report emphasizing human-centered design evidence safety privacy and educator involvement.
Authorship evidence and institutional sources
9. Turnitin artificial intelligence writing report guidance Vendor documentation warning that the model can misidentify human-generated and paraphrased text and should not determine adverse action alone.
10. Detectors are biased against non native English writers Peer-reviewed evidence of detector bias and evasion limits.
11. Google Docs version history Official instructions for reviewing earlier versions and recorded changes.
12. Microsoft Word version history Official instructions for reviewing earlier versions of supported Office files.
13. Coalition for Content Provenance and Authenticity frequently asked questions Explanation of cryptographically signed Content Credentials for recorded content origin and editing history.
14. Stanford Artificial Intelligence and Your Learning Student guidance distinguishing learning support from delegation and preserving instructor-specific rules.
15. Thomas Aquinas College syllabus Example of a curriculum built around direct discussion demonstration laboratory work composition and thesis defense.
16. High Tech High design principles Example of authentic project work critique collaboration and public exhibition requiring individual verification within group products.
Source and evidence limit
These sources support the design of the standard and establish selected public policies and practices. They do not prove that any institution implements the complete protocol or that the protocol itself improves long-term outcomes. That question requires representative use, participant evidence, comparison, delayed measurement, and transparent reporting.