Verification and pilot materials · Complete web edition

Phase Eleven Prospective Pilot Scorecard

This is the complete web edition of the canonical project document. Its narrative, recommendations, evidence tables and source links are presented below without abridgment.

3,515words

22tables

110text blocks

Download the Word document

Project, Research, Analysis, Editorial by Patrick Russell; assisted by ChatGPT (OpenAI).

Published as a web edition September 13, 2026

California Education and AI Prospective Pilot Scorecard

Phase Eleven Pilot readiness and evidence requirements

Scorecard date September 12 2026 Informational research design No outreach or personal attribution

Scorecard field	Entry
Responsible person	________________________________________
Organization or office	________________________________________
Review period	________________________________________
Approval or research determination	________________________________________

Phase Eleven conclusion

The six selected cases can support prospective tests, but none currently supplies public evidence of durable unaided transfer after artificial intelligence assistance ends. Berkeley Law and Thomas Aquinas College have the clearest published protections for work that must remain human. USC has the clearest managed tool environment. Stanford supplies the most developed student proposal for individual understanding checks. High Tech High has verified board adoption but not a publicly available operative policy. Mother of Divine Grace supplies a strong K through 12 formation benchmark but no public age specific artificial intelligence framework.

The scorecard therefore records readiness, not success. A case may be ready to test and still have no measured result. No score in this document supports a final institutional ranking, a misconduct finding, or a claim about a student's character or mental health.

Current readiness classification

Case	Present classification	What prevents a result claim
Berkeley Law	Conditional pilot ready	Course artifacts, implementation record, delayed transfer, and comparison are unavailable.
Stanford	Design pilot ready	Understanding checks were proposed and were not official university policy in the reviewed source.
USC	Tool environment ready	A representative course, protected task, and independent outcome record must be selected.
Thomas Aquinas College	Human only benchmark ready	Comparable scoring, accessibility practice, and delayed outcome evidence are unavailable.
High Tech High	Student facing pilot blocked	The complete adopted policy, approved tools, age rules, privacy terms, and consent process are not public.
Mother of Divine Grace	Human only K through 12 benchmark ready	A public age specific artificial intelligence policy and student evidence chain are unavailable.

Scorecard method

Each dimension uses the Phase Nine evidence scale. The number describes the public evidence presently available for the defined case. It does not grade the institution. A pilot may proceed only when every applicable launch gate is satisfied, even if several design dimensions already reach level two or three.

Level	Evidence threshold
0	No usable public evidence located.
1	A proposal, inference, restricted record, or incomplete description identifies the element.
2	A current official policy or published design defines the element.
3	Representative implementation, artifacts, scoring, or aggregate results document the element.
4	Delayed unaided transfer is measured with a credible comparison and stated limitations.

Readiness dimensions

Dimension	Required definition
Purpose	The exact knowledge, reasoning, authorship, judgment, attention, or relationship capacity is named.
Artificial intelligence condition	The tool, account, permitted function, prompts, human support, disclosure, and prohibited substitutions are fixed.
Protected human act	A learner must personally perform the target intellectual act without generative assistance.
Retention and transfer	Equivalent delayed and novel tasks test what remains after assistance ends.
Comparison	Baseline, human only, teacher facing, bounded student use, or open use conditions permit an interpretable comparison.
Scoring	Criteria, raters, reliability checks, exclusions, and missing data rules are set before results are examined.
Rights and access	Privacy, data minimization, accessibility, non artificial intelligence alternatives, and research protections are resolved.
Wellness and relationships	Workload, attention, help seeking, human contact, belonging, family support, and COVID context are measured without diagnosis or surveillance.
Outcome evidence	Immediate, retained, and transfer results are available with limitations.

Launch gates

Name a responsible educator, course or grade, educational purpose, and protected human act.
Complete the institutional review or research determination before collecting evidence for generalizable conclusions.
Preserve accommodations and a non artificial intelligence alternative; for minors resolve permission, assent, age suitability, and privacy.
Register tasks and scoring before review; prohibit automated discipline or misconduct findings from detector scores.

Berkeley Law pilot

Field	Prospective entry
Education level	Higher education unit level case
Pilot status	Conditional pilot ready after course and assessment approval
Evidence position	AI rule 2 Protected performance 2 Outcome evidence 0
Decision before launch	Select one first year writing or doctrinal course and preserve its actual policy, assignment, permitted research, accommodations, and scoring rule.

Pilot question

Does a restrictive artificial intelligence rule preserve independent legal analysis, authorship, citation verification, and transfer to a new fact pattern more effectively than a bounded assistance condition designed for the same objective?

Five condition sequence

Condition	Task	Evidence produced
C0	Closed resource issue spotting and short written rule application before the selected unit.	Baseline legal knowledge, organization, authority use, and reasoning.
C1	Course instruction under the operative rule. If an authorized comparison exists, record the bounded function and human feedback separately.	Fidelity, exceptions, drafts, citations, instructor contact, and time.
C2	Parallel in class analysis followed by a brief oral defense without artificial intelligence enabled search.	Immediate ownership, reasoning, citation judgment, and correction.
C3	Equivalent analysis two to four weeks later without immediate rehearsal.	Retention of doctrine, analytic structure, and source discipline.
C4	A new fact pattern requiring the same doctrine plus one unfamiliar authority or counterargument.	Near transfer, uncertainty, objection handling, and self correction.

Scoring and controls

Blind scoring where practical; keep writing fluency separate from legal reasoning and source accuracy.
Record authorized exceptions and accommodations so restriction is not confused with unequal access.
Compare course sections only when preparation, instructor contact, workload, and task difficulty are credible matches.
Measure trust, workload, office hour use, and help seeking as context rather than proof of learning.

Permitted conclusion A limited conclusion about the selected course, policy version, student group, and time window.

Prohibited conclusion A campus wide claim about Berkeley or a misconduct inference from differences between assisted and unaided prose.

Official sources Berkeley Law policy | Berkeley teaching guidance

Stanford understanding checks pilot

Field	Prospective entry
Education level	Higher education course level case
Pilot status	Design pilot ready institutional policy claim not permitted
Evidence position	Proposal 1 Protected performance 1 Outcome evidence 0
Decision before launch	Identify a course that voluntarily adopts individual understanding checks and specify how other workload is reduced.

Pilot question

Do brief graded oral or supervised understanding checks preserve individual comprehension and critical thinking after collaborative policy writing or bounded artificial intelligence use without increasing total workload?

Five condition sequence

Condition	Task	Evidence produced
C0	Individual 350 word policy memo and a short explanation before group deliberation.	Initial argument, evidence, writing ownership, and position.
C1	Structured collaboration and any permitted artificial intelligence support with prompts, human discussion, and revisions recorded.	Contribution history, negotiation, source verification, and assistance.
C2	Individual oral check or supervised problem solving on the memo's central tradeoff.	Personal comprehension, objection handling, and responsibility for the conclusion.
C3	New policy problem two to four weeks later with no access to the original group memo.	Retention of reasoning and civic decision method.
C4	Apply the method to a different course assessment or non artificial intelligence governance problem.	Transfer beyond the practiced topic and group product.

Scoring and controls

Score individual work separately from section consensus and the quality of the winning proposal.
Keep the understanding check brief and reduce another assignment so workload does not rise by design.
Record speaking and communication accommodations and do not equate verbal speed or confidence with understanding.
Do not describe the student proposal as adopted Stanford policy unless an official later source establishes adoption.

Permitted conclusion A finding about whether a defined understanding check verifies personal learning within the selected course.

Prohibited conclusion A claim that COLLEGE 102 already proves campus wide adoption or long term learning effects.

Official sources COLLEGE 102 report | Student AI guide

USC managed tool pilot

Field	Prospective entry
Education level	Higher education course level case
Pilot status	Tool environment ready course design required
Evidence position	Managed environment 3 Protected performance 0 Outcome evidence 0
Decision before launch	Select one course and assignment whose purpose can support human only, teacher facing, and bounded student use conditions.

Pilot question

Does managed ChatGPT Edu access improve durable learning when paired with faculty mediation, compared with tool access alone and protected human only instruction?

Five condition sequence

Condition	Task	Evidence produced
C0	Short unaided explanation, analysis, or problem before the unit.	Baseline knowledge, reasoning, and confidence.
C1	Planned comparison among human only, teacher facing artificial intelligence, bounded student use, and where justified bounded use plus faculty conference.	Exact prompts, outputs, feedback, time, verification, and human contact.
C2	Reproduce the core reasoning and correct one flawed output without artificial intelligence.	Immediate comprehension, error detection, authorship, and correction.
C3	Equivalent unaided task two to four weeks later.	Retention after immediate tool effects and feedback fade.
C4	New prompt, case, dataset, or professional scenario requiring the same concept.	Transfer, judgment, source verification, and self regulation.

Scoring and controls

Do not collect student chat histories through faculty; USC states that faculty cannot view them in ChatGPT Edu.
Use voluntary research records or student supplied excerpts only under an approved data plan.
Protect personally identifiable, health, student record, and other sensitive information from all prompts.
Provide a non artificial intelligence option and address digital accessibility before requiring the tool.

Permitted conclusion A comparison of the defined instructional conditions and the value added by human support.

Prohibited conclusion A claim that campus wide licensing itself improves learning, judgment, or wellness.

Official sources USC Provost guidance | USC academic integrity

Thomas Aquinas College benchmark

Field	Prospective entry
Education level	Four year higher education benchmark
Pilot status	Human only benchmark ready artificial intelligence comparison optional
Evidence position	Formation design 3 AI rule 2 Outcome evidence 0
Decision before launch	Select composition, tutorial, seminar, mathematical demonstration, laboratory, or senior thesis work and document its existing scoring and accommodations.

Pilot question

What level of retained knowledge, oral explanation, authorship, proof, and judgment is produced under a curriculum that protects academic papers and their preparation from artificial intelligence generation?

Five condition sequence

Condition	Task	Evidence produced
C0	Existing unaided course or thesis entry performance recorded without changing the curriculum.	Starting knowledge, writing, oral reasoning, or demonstration.
C1	Normal tutorial, seminar, laboratory, composition, or thesis practice under the current Code of Conduct.	Drafts, tutor questions, revisions, demonstrations, discussion, and time.
C2	Existing oral conference, seminar response, proof, laboratory explanation, or composition completed personally.	Immediate understanding and ownership across written and oral forms.
C3	Equivalent question or demonstration after two to four weeks.	Retention of texts, principles, proof structure, and reasoning.
C4	New text, objection, theorem, experiment, or thesis defense question.	Transfer, intellectual humility, correction, and judgment.

Scoring and controls

Do not introduce student facing artificial intelligence merely to manufacture a comparison; the protected curriculum is itself a valid benchmark condition.
If a later responsible use comparison is educationally approved, make participation optional and keep it outside prohibited credited work.
Document oral, writing, disability, and language accommodations without weakening the target intellectual act.
Interpret residential friendship, attention, exercise, worship, and community as contextual conditions rather than automatic causes.

Permitted conclusion A documented benchmark for human only formation within the selected activity and time window.

Prohibited conclusion A claim of superiority over artificial intelligence integrated education without a credible comparison and comparable students.

Official sources California Code of Conduct | College syllabus

High Tech High pilot

Field	Prospective entry
Education level	Public charter K through 12 case
Pilot status	Student facing pilot blocked pending operative policy
Evidence position	Board adoption 2 Operative policy 1 Outcome evidence 0
Decision before launch	Obtain and review the complete adopted policy before defining any tool, grade, consent, privacy, disclosure, or student consequence.

Pilot question

After the policy threshold is satisfied, can project based learning use bounded artificial intelligence while preserving individual design reasoning, construction skill, oral defense, source judgment, and foundational learning?

Five condition sequence

Condition	Task	Evidence produced
C0	Age appropriate individual explanation, sketch, construction, writing, or calculation before the project.	Foundational skill, prior knowledge, confidence, and support needs.
C1	Teacher approved bounded use inside one grades 9 through 12 project; younger comparison remains human only unless policy expressly permits otherwise.	Tool function, drafts, critique, group roles, adult support, and family notice.
C2	Individual process conference or supervised construction task without artificial intelligence.	Personal mastery, authorship, design choices, and correction.
C3	Parallel individual task after two to four weeks.	Retention of the project's underlying knowledge and method.
C4	New design problem or community need requiring the same principle without artificial intelligence.	Transfer beyond the exhibition product and original group.

Scoring and controls

Treat the May 28 2025 board action as proof of adoption, not as the full policy text.
Resolve grade thresholds, approved accounts, parent notice or permission, assent where required, privacy, retention, and non artificial intelligence alternatives.
Separate individual learning from group product quality and public exhibition polish.
Compare schools and grades only after accounting for teacher turnover, attendance, family support, disability, language, and post COVID recovery.

Permitted conclusion Before the policy is obtained, only a human only baseline and research design may proceed.

Prohibited conclusion Any student facing artificial intelligence pilot or claim that the board minutes establish the policy's complete safeguards.

Official sources May 2025 board minutes | Design principles

Mother of Divine Grace benchmark

Field	Prospective entry
Education level	Private Catholic K through 12 distance and home education benchmark
Pilot status	Human only developmental benchmark ready artificial intelligence pilot blocked
Evidence position	Formation design 3 Age specific AI rule 0 Outcome evidence 0
Decision before launch	Select a grades 6 through 9 logic or composition activity and preserve the lesson sequence, parent role, teacher feedback, accommodation, and student work.

Pilot question

What do narration, memory, logic, supervised composition, parent dialogue, and delayed transfer look like within a family mediated developmental model before any artificial intelligence condition is considered?

Five condition sequence

Condition	Task	Evidence produced
C0	Narration, recall, logic problem, or supervised composition before the selected lesson sequence.	Starting knowledge, language, reasoning, independence, and parent support.
C1	Normal parent directed lesson, reading, discussion, practice, and teacher feedback with no student facing artificial intelligence.	Instruction, parent prompts, feedback, time, family burden, and peer contact.
C2	New narration, argument, or supervised composition completed without generative assistance.	Immediate ownership, reasoning, memory, and communication.
C3	Equivalent task one to three weeks later according to developmental level.	Retention of content, logical form, and writing decisions.
C4	A new text, claim, audience, or practical problem requiring the same reasoning.	Transfer, correction, judgment, and appropriate independent work.

Scoring and controls

Keep Mother of Divine Grace exclusively in the K through 12 track and use developmental criteria rather than university governance measures.
Do not add a student facing artificial intelligence condition until an official age, permission, privacy, approved tool, disclosure, and alternative framework exists.
Record healthy parent mediation separately from excessive family burden and distinguish chosen home education from emergency crisis schooling.
Measure friendship, belonging, trusted adult access, attention, distress, and family stability without turning them into diagnoses or automatic explanations.

Permitted conclusion A human only developmental benchmark for the selected grade, lesson, family support condition, and time window.

Prohibited conclusion A university comparison, a general homeschooling claim, or an artificial intelligence readiness claim without an official age specific framework.

Official sources Methodology | Curriculum stages | Crisis schooling

Cross case pilot scorecard

Case	Purpose	AI	Human	Delay	Compare	Score	Rights	Wellness	Status
Berkeley Law	2	2	2	1	1	1	2	0	Conditional
Stanford	2	1	1	1	1	1	2	1	Design only
USC	1	3	0	1	1	1	3	1	Select course
Thomas Aquinas	3	2	3	1	1	1	1	1	Human only
High Tech High	2	1	1	1	1	1	0	1	Policy blocked
Mother of Divine Grace	3	0	2	1	1	1	0	2	Human only K12

Interpretation: AI means the artificial intelligence condition is defined; Human means the protected human act is established; Delay means retention and transfer tasks are established; Compare means a credible comparison is specified; Score means scoring and reliability are documented; Rights covers privacy, consent, accessibility, and alternatives. Numerical entries describe evidence maturity and must not be averaged into a school ranking.

First executable work

Priority	Case	Work that can begin without claiming results
1	Thomas Aquinas College	Document an existing human only performance chain and scoring practice without introducing artificial intelligence.
2	Berkeley Law	Finalize a course specific protected task and delayed legal transfer design under the current rule.
3	USC	Select a course and register human only, teacher facing, bounded tool, and tool plus faculty conditions.
4	Stanford	Pilot a brief individual understanding check in a willing course and offset its workload.
5	Mother of Divine Grace	Build a parent mediated K through 12 human only benchmark before any artificial intelligence comparison.
6	High Tech High	Obtain the complete policy and resolve age, privacy, consent, tool, and alternative rules before student use.

Decision rules

Advance from design to implementation only after every applicable launch gate is documented.
Advance to PH 3 only when representative implementation, artifacts, scoring, or aggregate results are available.
Advance to PH 4 only when delayed unaided transfer is measured with a credible comparison and explicit limitations.
Pause or redesign when privacy, accessibility, age, consent, equivalent instruction, workload, or student safety cannot be resolved.
Report null and adverse results with the same prominence as improvement.

Common pilot record

Record	Entry
Institution course or grade	____________________________________________________________
Responsible educator	____________________________________________________________
Educational capacity	____________________________________________________________
Policy version and authority	____________________________________________________________
Artificial intelligence condition	____________________________________________________________
Protected human act	____________________________________________________________
C0 baseline task and date	____________________________________________________________
C1 practice period	____________________________________________________________
C2 immediate task and date	____________________________________________________________
C3 retention task and date	____________________________________________________________
C4 transfer task and date	____________________________________________________________
Comparison condition	____________________________________________________________
Scoring and raters	____________________________________________________________
Privacy consent accessibility	____________________________________________________________
Human support and accommodations	____________________________________________________________
Wellness and relationship context	____________________________________________________________
Missing evidence and deviations	____________________________________________________________

Outcome record

Question	Judgment
Did assisted performance improve	Yes No Mixed Insufficient
Did immediate unaided performance improve	Yes No Mixed Insufficient
Was learning retained	Yes No Mixed Insufficient
Did learning transfer	Yes No Mixed Insufficient
Were authorship and verification preserved	Yes No Mixed Insufficient
Did human support change the result	Improved Worsened Mixed No detected change Insufficient
Did workload or wellness change	Improved Worsened Mixed No detected change Insufficient
Evidence level reached	PH 0 PH 1 PH 2 PH 3 PH 4

Rights wellness and COVID controls

These controls determine whether results can be interpreted fairly. They do not convert the scorecard into a mental health assessment, and they do not authorize collection of sensitive information merely because it might be interesting.

Control	Required record	Stop or limit condition
Privacy	Minimum necessary data, separation of identity, access, retention, deletion, vendor use, and publication threshold.	Sensitive student information would enter an unapproved tool or small cells could identify a learner.
Consent and assent	Research determination, permission process, affirmative child assent when required, and treatment of dissent.	A minor cannot decline without losing ordinary instruction or opportunity.
Accessibility	Approved accommodations, digital accessibility, remediation, and equivalent non artificial intelligence path.	The tool or protected check excludes a learner or changes the target construct.
Workload	Total time, displaced assignments, educator preparation, parent burden, and support contacts.	A verification requirement adds material burden without removing or redesigning other work.
Relationships	Teacher, tutor, parent, and peer contact; help source; belonging; friendship; trusted adult access.	Artificial intelligence substitutes for required human safeguarding or relationship support.
Wellness	Brief neutral measures of attention, frustration, confidence, sleep, workload, and help seeking.	Covert diagnosis, emotion recognition, disciplinary use, or automated intervention.
COVID context	Pre 2020 trajectory where available, disruption exposure, prior support, recovery conditions, and later artificial intelligence exposure.	School closure alone is treated as the cause of a present outcome.

Research protections

Ordinary teaching assessment and systematic research are different activities. Determine the applicable review before collecting evidence for publication or generalizable conclusions.
Parental permission does not replace a child's affirmative assent when assent is required, and silence is not affirmative assent.
Use de identified aggregate reporting whenever possible and suppress small or distinctive combinations.
Keep educators or trained human reviewers responsible for scoring, integrity concerns, and safeguarding decisions.

Official sources HHS child protections | Education technology privacy | WWC standards

Remaining institutions

UCLA, Caltech, Loyola Marymount University, and Harvard-Westlake remain in the investigation. They are not excluded from the method; they are held for the next expansion because Phase Eleven begins with cases that expose the most informative contrasts among restriction, student participation, managed access, project learning, and family centered formation.

Institution	Next pilot preparation
UCLA	Pair the literacy module with a supervised correction and delayed source judgment task.
Caltech	Pair one ethics, writing, scientific, mathematical, or coding activity with cross domain unaided transfer.
Loyola Marymount University	Compare an artificial intelligence free anchor with bounded use and a supervised revision or oral explanation.
Harvard-Westlake	Obtain grades 7 through 12 operating rules before pairing the ethics course with protected writing and source evaluation.

Phase Twelve handoff

Phase Twelve should convert the first executable cases into task packets with parallel prompts, scoring anchors, rater instructions, consent and privacy checklists, workload offsets, and reporting templates. It should begin with a human only benchmark and one bounded higher education comparison. High Tech High and any other K through 12 artificial intelligence condition should remain paused until the operative policy and developmental safeguards are verified.