Verification and pilot materials · Complete web edition

Phase Nine Evidence Ledger

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

3,224words

24tables

66text 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 Evidence Ledger

Phase Nine | Claim status, protected performance, and scorecard consequences

Evidence date: September 12, 2026 • Informational working document • No outreach or personal attribution

Review field	Entry
Responsible person	________________________________________
Organization or role	________________________________________
Review date	________________________________________

Phase Nine conclusion

The evidence ledger confirms that the California institutions divide into three broad patterns: mature AI governance without measured independent outcomes; strong human-formation architecture without complete public AI governance; and promising participatory or project-based practice without sufficient individual verification. No institution reaches PH-4, the level requiring measured durable unaided transfer. This prevents the scorecard from mistaking policies, licenses, courses, or attractive student products for proof that minds are being formed more successfully.

Two status changes are especially important. Thomas Aquinas College's earlier public-policy gap is superseded because its current California Code of Conduct now contains a direct student-facing AI prohibition for academic papers and preparatory work. High Tech High's policy adoption is verified from May 28, 2025 board minutes, while the policy's complete operative text remains only partial evidence.

Ledger codes

Code	Meaning
V — Verified	An official source directly supports the limited claim stated.
P — Partial	An official source supports only part of the claim; implementation, scope, or effect is unresolved.
M — Missing	The required evidence was not located publicly. This does not prove the practice does not exist.
R — Restricted	The evidence likely resides in a course, portal, handbook, student record, or internal system not publicly accessible.
S — Superseded	A dated earlier finding has been replaced by newer or more complete evidence and is retained only for history.

Protected human-performance scale

Level	Evidence standard
PH-0	No protected unaided performance identified.
PH-1	A protected performance is proposed, inferred, or recommended, but not publicly established as implemented.
PH-2	A published policy, curriculum, or program design protects identifiable human work.
PH-3	Implementation is documented with representative assignments, artifacts, scoring, or aggregate results.
PH-4	Durable unaided transfer is measured after assistance ends, with a credible comparison and stated limitations.

Master evidence register

Institution	Level	AI rule	Practice	Human measure	Outcome
UC Berkeley	Higher education	V	P	PH-2	M
UCLA	Higher education	V	P	PH-1	M
Stanford University	Higher education	V	P	PH-1	M
University of Southern California	Higher education	V	P	PH-1	M
California Institute of Technology	Higher education	V	P	PH-2	M
Loyola Marymount University	Higher education	V	P	PH-2	M
Thomas Aquinas College	Higher education	V/S	P	PH-2	M
Mother of Divine Grace School	K through 12	M	P	PH-2	M
Harvard-Westlake School	Grades 7 through 12	M	P	PH-1	M
High Tech High	K through 12	V/P	P	PH-1	M

Interpretation: V/P in a master cell means one part is verified and another remains partial. V/S means current evidence is verified and supersedes an earlier finding. Status codes describe evidence completeness, not institutional quality.

Ledger 1 UC Berkeley

Field	Ledger entry
Education level	Higher education
Current judgment	A strong unit-level restriction now exists at Berkeley Law, but implementation and transfer evidence remain unavailable.
Protected performance	PH-2

Claim-by-claim record

Claim	Status	Official evidence	Required confirmation
Formation purpose	V	Berkeley Law explicitly connects its restriction to independent legal reasoning.	Preserve the policy version and authority record.
AI rule	V	The summer 2026 law-school policy restricts conceptualizing, drafting, revising, translating, and editing credited work unless an exception applies.	Map authorized exceptions by course.
Tools and privacy	P	Campus guidance favors institutionally licensed tools and warns against protected-data disclosure.	Capture approved-tool list, settings, retention, and opt-out rules.
Representative practice	R	A first-year legal-writing or doctrinal assignment is the selected unit, but its prompt and student artifacts are not public.	Obtain syllabus, assignment, drafts, feedback, and permitted research methods.
Independent performance	P	The rule protects unaided work, but no common oral defense or supervised transfer measure is publicly established.	Document one closed-resource analysis or oral defense.
Outcomes and wellness	M	No public before-and-after evidence links the policy to durable reasoning, citation accuracy, workload, trust, or well-being.	Compare cohorts with limitations and alternative explanations.

Scorecard consequence Formation score may rise for Berkeley Law only; campus-wide scoring must remain decentralized.

Official sources Berkeley AI Hub | Teaching guidance | Berkeley Law policy

Ledger 2 UCLA

Field	Ledger entry
Education level	Higher education
Current judgment	UCLA has a coherent institutional framework, but public evidence does not yet show that literacy training changes independent student performance.
Protected performance	PH-1

Claim-by-claim record

Claim	Status	Official evidence	Required confirmation
Formation purpose	P	The six principles include literacy and assessment integrity, but do not define a complete campus-wide account of human formation.	Tie each selected course to specific capacities such as judgment, authorship, or reasoning.
AI rule	V	Instructors communicate whether AI is allowed, limited, or prohibited; the central framework does not impose one uniform permission rule.	Archive the policy used in the selected course.
Tools and privacy	P	Responsible-use guidance addresses risk and equitable access at a general level.	Capture the actual tool, account type, data terms, accessibility, and alternative.
Representative practice	P	The Bruin Learn literacy module and Faculty Fellows work are publicly described, but a complete student evidence chain is not public.	Preserve module completion, assignment, disclosure, rubric, and reflection.
Independent performance	M	No common post-AI supervised writing or oral explanation was located.	Add a short unaided verification after the AI-supported task.
Outcomes and wellness	M	Completion, retention, dependency, workload, belonging, and durable-learning results were not publicly reported.	Measure behavior and unaided transfer, not awareness alone.

Scorecard consequence Credit policy maturity and literacy infrastructure; withhold learning-outcome credit pending unaided measures.

Official sources AI teaching initiative | Responsible-use principles | Course policy guidance

Ledger 3 Stanford University

Field	Ledger entry
Education level	Higher education
Current judgment	Stanford provides strong student participation and human-centered guidance; the most promising verification device remains a proposal rather than an implemented university rule.
Protected performance	PH-1

Claim-by-claim record

Claim	Status	Official evidence	Required confirmation
Formation purpose	V	Official guidance emphasizes human responsibility, critical thinking, and checking AI output.	Connect those aims to graded course outcomes.
AI rule	V	Course permissions remain instructor-specific under the Honor Code and disclosure expectations.	Archive the exact rule for every selected course.
Governance participation	V	More than 1,200 first-year students in more than 70 COLLEGE 102 sections developed AI policy memos in 2026.	Preserve the prompt, section process, voting, and final proposal.
Representative practice	P	The COLLEGE 102 exercise is documented, but individual draft histories and later behavior are not public.	Separate individual reasoning from group deliberation and proposal quality.
Independent performance	P	Individual understanding checks were proposed; Stanford stated the winning proposal was not itself official policy.	Verify whether checks were adopted, where, and with what weight.
Outcomes and wellness	M	No longitudinal results establish effects on writing, dependency, agency, trust, belonging, or workload.	Measure individual outcomes beyond the quality of the group proposal.

Scorecard consequence Credit authentic civic inquiry and student voice; do not score proposed understanding checks as implemented protection.

Official sources COLLEGE 102 report | Student AI guide | Course policy guidance

Ledger 4 University of Southern California

Field	Ledger entry
Education level	Higher education
Current judgment	USC offers a university-scale, managed-tool case; access and data protection are better established than educational effects.
Protected performance	PH-1

Claim-by-claim record

Claim	Status	Official evidence	Required confirmation
Formation purpose	P	Guidance preserves faculty authority and student responsibility, but account availability is not a learning theory.	State the human capacity protected in the selected course.
AI rule	V	Spring 2026 guidance places ChatGPT Edu within course policy and academic-integrity requirements.	Record the selected instructor's allowed, limited, or prohibited uses.
Tools and privacy	V	ChatGPT Edu provides an institutionally managed environment with stronger data governance than consumer accounts.	Capture configuration, retention, accessibility, and alternative-use rules.
Representative practice	R	A course using AI for feedback, comparison, research, or revision is required, but prompts and artifacts are not public.	Archive prompt, outputs, verification, drafts, disclosure, and rubric.
Independent performance	M	No university-wide supervised post-assistance check was found.	Require a brief in-person reproduction, correction, or oral explanation.
Outcomes and wellness	M	No public evidence shows changes in unaided learning, faculty contact, office-hour use, workload, belonging, or confidence.	Compare AI-plus-faculty support with tool access alone.

Scorecard consequence Credit procurement and privacy readiness separately from formation and learning outcomes.

Official sources Provost guidance | ChatGPT Edu at USC | Integrity guidance

Ledger 5 California Institute of Technology

Field	Ledger entry
Education level	Higher education
Current judgment	Caltech has unusually clear links among disclosure, confidential data, accuracy, trust, and instructor authority, but durable transfer is unmeasured.
Protected performance	PH-2

Claim-by-claim record

Claim	Status	Official evidence	Required confirmation
Formation purpose	V	Official teaching materials connect AI use with writing, ethical responsibility, trust, and assessment.	Specify the intended capacity in each reviewed assignment.
AI rule	V	Instructor or division rules govern coursework; students remain subject to the Honor Code and disclosure requirements.	Archive the current division and course versions.
Tools and privacy	V	Consumer tools may not receive restricted information; managed or approved tools are required for protected institutional work.	Capture tool list, contracts, settings, and student alternatives.
Integrity enforcement	V	Central guidance strongly discourages consequential reliance on unreliable AI detectors.	Document actual investigation and appeal practices.
Independent performance	P	An ethics-and-writing model is public, but comparable unaided scientific, mathematical, coding, and prose measures are not.	Pair AI-supported work with in-class analysis, oral conference, or proctored examination.
Outcomes and wellness	M	No public series establishes transfer, trust, workload, stress, belonging, or faculty-contact effects.	Measure outcomes across divisions and student support conditions.

Scorecard consequence Credit governance and trust protections; hold outcome credit until unaided cross-domain transfer is demonstrated.

Official sources Institute AI guidance | AI in teaching | HSS policy | Ethics course

Ledger 6 Loyola Marymount University

Field	Ledger entry
Education level	Higher education
Current judgment	LMU offers practical AI-free and AI-integrated options, protected tools, and reasonable alternatives, but implementation breadth and results remain uncertain.
Protected performance	PH-2

Claim-by-claim record

Claim	Status	Official evidence	Required confirmation
Formation purpose	P	Whole-person and ethical aims are visible, but the reviewed AI materials do not establish common unaided outcomes.	Tie course choices to judgment, authorship, vocation, and relationship aims.
AI rule	V	Faculty and departments determine use within academic-honesty, security, and privacy requirements.	Archive one co-constructed or instructor-set course policy.
Tools and privacy	V	LMU provides protected Microsoft Copilot and tells instructors to consider access, cost, privacy, accessibility, and reasonable alternatives.	Verify configuration, retention, accommodation, and actual alternatives.
Representative practice	P	AI-free, AI-integrated, and co-constructed policy models are published as teaching options.	Preserve the initial rule, student discussion, revisions, assignments, and disclosures.
Independent performance	P	AI-free work is structurally available, but a common oral or supervised verification standard is not public.	Use an AI-free anchor assignment plus oral explanation or supervised revision.
Outcomes and wellness	M	No public results establish effects on independent judgment, writing, workload, help-seeking, belonging, or family and peer support.	Measure choices and results rather than policy availability.

Scorecard consequence Credit agency, alternatives, and protected access; withhold outcome credit until independent performance is documented.

Official sources AI teaching hub | AI tools | Academic honesty

Ledger 7 Thomas Aquinas College

Field	Ledger entry
Education level	Higher education
Current judgment	The current California Code of Conduct resolves and supersedes the earlier public-policy gap; implementation, accessibility, and comparative outcomes still require evidence.
Protected performance	PH-2

Claim-by-claim record

Claim	Status	Official evidence	Required confirmation
Prior ledger claim	S	Phase Seven recorded a complete student-facing AI rule as not publicly found.	Retain the old finding only as dated history.
Current AI rule	V	The Code prohibits AI and AI-enabled search engines in generating academic papers, including finding texts, outlining, arguments, sentences, editing, and tools such as Grammarly.	Preserve the current code and senior-thesis handout.
Formation architecture	V	The published syllabus centers original texts, tutorials, seminars, mathematical demonstrations, laboratories, composition, and conversation.	Document how the AI rule operates inside these practices.
Representative practice	R	Composition and the senior thesis are selected, but drafts, tutor comments, oral defenses, exceptions, and accommodations are not public.	Capture the complete assignment chain and senior guidance.
Independent performance	P	The program contains inherently human oral, written, mathematical, and laboratory performances; public scoring and reliability evidence are incomplete.	Document rubrics, consistency, and links among measures.
Outcomes and wellness	P	The College reports weaker engagement during emergency online instruction, but this is not a causal or comparative wellness study.	Examine learning, friendship, exercise, attention, community, accessibility, and later responsible AI use.

Scorecard consequence Raise verified AI-governance and formation-alignment scores; do not infer measured superiority without comparable outcomes.

Official sources California Code | Syllabus | AI and human flourishing | Online transition

Ledger 8 Mother of Divine Grace School

Field	Ledger entry
Education level	K through 12
Current judgment	Mother of Divine Grace supplies the strongest family-mediated developmental model in the sample; public age-specific AI governance remains the central gap.
Protected performance	PH-2

Claim-by-claim record

Claim	Status	Official evidence	Required confirmation
Institutional level	V	Mother of Divine Grace is a private Catholic K-through-12 distance and home-education program, not a university.	Keep all evaluation developmental and parent-centered.
Formation architecture	V	The published method uses primary, grammatical, dialectical, and rhetorical stages with integral parent responsibility.	Preserve the stage-specific learning aims and parent role.
AI rule	M	No complete public generative-AI rule, approved-tool list, age threshold, or consent framework was located.	Locate official guidance by developmental stage, if it exists.
Representative practice	R	Grades 6 through 9 logic and composition are selected, but lesson plans, drafts, rubrics, and digital-tool records are not public.	Capture the home lesson, parent role, feedback, and student artifact.
Independent performance	P	Published pedagogy supports narration, recitation, handwriting, composition, analysis, and discussion, but comparable scored evidence is unavailable.	Document one supervised composition and oral explanation before any AI comparison.
Outcomes and wellness	P	The school distinguishes prepared homeschooling from crisis schooling and recognizes family stress, fear, isolation, and lost friendships.	Compare learning and wellness across family support and peer-connection conditions.

Scorecard consequence Maintain a strong formation score; hold AI-readiness scoring until age, consent, privacy, and tool rules are verified.

Official sources School overview | Methodology | Curriculum stages | Crisis schooling

Ledger 9 Harvard-Westlake School

Field	Ledger entry
Education level	Grades 7 through 12
Current judgment	The school publishes an AI-ethics learning opportunity, but the operating rules that govern actual student use remain unresolved publicly.
Protected performance	PH-1

Claim-by-claim record

Claim	Status	Official evidence	Required confirmation
Formation purpose	P	Independent thinking, integrity, research, counseling, and wellness are visible institutional aims.	Tie the AI case to grade-specific developmental outcomes.
AI curriculum	V	The 2026–27 curriculum guide includes a course on ethics, technology, and society in the age of artificial intelligence.	Preserve the course version and enrollment context.
AI operating rule	M	No complete public grades 7-through-12 framework for permissions, disclosure, age, consent, privacy, or integrity was located.	Capture schoolwide, division, and course rules without inferring portal content.
Representative practice	R	The ethics course is selected, but syllabus, assigned inquiry, prompts, rubric, and student work are not public.	Archive a complete problem-based inquiry and conventional writing comparison.
Independent performance	M	No public individually authored essay segment, oral defense, or supervised source-evaluation requirement was established.	Verify a human-authored and orally defended component.
Outcomes and wellness	M	No public longitudinal series connects AI use with learning, friendship, belonging, sleep, workload, counseling, or teacher reliance.	Measure by grade and prior support condition.

Scorecard consequence Credit the ethics curriculum; withhold AI-governance and independent-performance credit pending operative rules and artifacts.

Official sources Course descriptions | Independent research | Health and wellness

Ledger 10 High Tech High

Field	Ledger entry
Education level	K through 12
Current judgment	Board authority and student participation are verified, but the missing operative policy text prevents a complete age, privacy, consent, and assessment judgment.
Protected performance	PH-1

Claim-by-claim record

Claim	Status	Official evidence	Required confirmation
Policy adoption	V	The board approved an Artificial Intelligence Use Policy on May 28, 2025 after discussion involving students, staff, and the High Tech Education Collective.	Preserve the signed or final adopted policy and revision history.
Operative rule	P	Minutes describe safety, optional teacher use, handbook consequences, and teacher resources, but do not reproduce the full rule.	Capture approved tools, age rules, consent, privacy, disclosure, alternatives, and annual review.
Student participation	V	A December 2023 student presentation recommended higher-grade use, protection of grammar foundations, and AI as guide rather than answer.	Treat the presentation as deliberation, not final policy text.
Representative practice	R	An exhibition-based middle- or high-school project is selected, but the teacher rule, prompts, drafts, critique, and individual roles are not public.	Archive one full project and an elementary foundational comparison.
Independent performance	M	Public exhibition supports authenticity, but no required individual oral defense or supervised construction check was verified.	Separate individual mastery and authorship from group product quality.
Outcomes and wellness	M	No public evidence connects the policy to mastery, equity, attendance, belonging, family burden, friendship, attention, or adult support.	Compare schools and grades while accounting for post-COVID recovery conditions.

Scorecard consequence Credit governance and student voice; hold substantive policy and outcome scores until the full rule and implementation evidence are captured.

Official sources May 2025 minutes | December 2023 minutes | Design principles | Handbook

Cross-institution findings

Governance is ahead of proof Berkeley, UCLA, Stanford, USC, Caltech, and LMU publish substantial guidance, yet none publicly establishes PH-4 durable unaided transfer.

Restrictions can protect formation Thomas Aquinas College and Berkeley Law make a defensible case that prohibition can preserve the very acts by which liberal or professional judgment develops.

Participation needs individual verification Stanford and High Tech High demonstrate the value of student voice and authentic collaborative work, but group quality cannot substitute for personal mastery.

K through 12 requires developmental governance Mother of Divine Grace, Harvard-Westlake, and High Tech High require age-specific rules, parent authority or notice, consent, privacy, foundational knowledge, and gradual transfer of responsibility.

Wellness remains weakly evidenced Attention, frustration tolerance, trust, belonging, friendship, family stability, workload, and human help-seeking matter, but public policy pages do not measure them.

COVID is context, not a universal explanation Comparisons must distinguish students who entered disruption with stable habits, supportive families, and friendships from those already experiencing dysregulation, isolation, or weak support.

Scorecard rules after Phase Nine

Score policy, practice, protected performance, and outcomes separately; never average missing evidence into apparent certainty.
Award no outcome credit for tool access, licenses, training completion, policy publication, or student satisfaction alone.
Require at least PH-2 before describing an institution as protecting independent thought, and PH-3 before describing protection as implemented across a representative case.
Reserve PH-4 for durable unaided transfer measured after assistance ends with a comparison, time window, and limitations.
For K-through-12 institutions, treat age, consent, privacy, parental authority, and non-AI alternatives as threshold requirements rather than optional bonuses.
Record missing public evidence neutrally; do not turn absence into either praise or accusation.

Phase Ten handoff

Phase Ten should build the protected human-performance protocol: a common set of age-appropriate, discipline-appropriate unaided checks that can be applied to the selected case activities. It should specify writing, oral explanation, source verification, mathematical or scientific reasoning, coding, creative production, memory, attention, authorship, and moral judgment measures; establish when AI is removed; define comparison groups and time windows; and include family, friendship, disability, prior preparation, COVID exposure, and wellness controls. This will convert the evidence ledger from a classification system into a testable evaluation method.