Evidence reviews and scorecards · Complete web edition

Phase Five Integrated Institution 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.

4,415words

17tables

130text 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 Integrated Institution Scorecard

Phase Five Formation, AI readiness, evidence, and institutional priorities

Responsible person	____________________________________________
Organization or role	____________________________________________
Review date	____________________________________________

Main finding

The ten institutions do not fall on one ladder. Thomas Aquinas College and Mother of Divine Grace School present the clearest unified accounts of intellectual formation, but their public records disclose little operational AI governance. UC Berkeley, Stanford, UCLA, and USC publish substantially more AI infrastructure, guidance, and teaching support, but responsibility for formation is distributed across schools, departments, courses, and individual instructors. Caltech and Loyola Marymount University occupy the middle: each connects AI guidance to learning, ethics, or human judgment, but neither publishes institution-wide outcome evidence.

The most important common weakness is measurement. None of the ten institutions publicly demonstrates, with strong comparative evidence, that its current AI practices improve durable learning, independent judgment, attention, relationships, or mental fitness. The scores therefore describe public evidence and institutional readiness. They are not rankings of student intelligence, moral worth, prestige, or overall educational quality.

How the scorecard works

Formation and AI readiness are scored separately because a school can protect attention, memory, discussion, and independent thought without having mature AI rules, while another institution can operate secure AI tools without a coherent formation program. Each of the six dimensions receives 0 to 3 points. A zero means that the reviewed public sources did not establish the practice; it does not prove that the practice is absent.

Score	Evidence threshold	Meaning
3	Clear implemented practice supported by several current public sources	The institution discloses purpose, method, and operational evidence.
2	Partial implementation or strong guidance with limited coverage	The practice is credible but varies by unit, course, grade, or educator.
1	Institutional claim, isolated example, or incomplete public record	The evidence signals activity but does not establish breadth or consistency.
0	Not established in the public sources reviewed	Treat as an evidence gap and investigate before drawing a conclusion.

Formation score Purpose and curricular coherence, protected human work and independent verification, and relationships, attention, and support. Maximum 9.

AI readiness score Public perspective and literacy, AI use in education, and governance, privacy, integrity, and verification. Maximum 9.

Outcome evidence score 0 means no relevant measured result was located; 1 means descriptive or indirect results; 2 means repeated quantitative results with important causal limits; 3 requires strong comparative or causal evidence tied to learning or mental fitness.

Integrated scorecard

Institution	Type	Formation	AI	Outcomes	Current posture
UC Berkeley	Public university	5	9	1	AI-ready; formation remains distributed
UCLA	Public university	5	7	1	AI framework expanding; outcomes incomplete
Stanford University	Private university	6	9	1	AI-ready with explicit human-centered principles
USC	Private university	5	8	1	Broad tool access with course-level discretion
Caltech	Private university	7	6	0	Strong reasoning protections; decentralized AI practice
Loyola Marymount	Private Catholic university	8	6	0	Whole-person mission; AI implementation developing
Thomas Aquinas College	Private Catholic college	9	3	1	Formation-led; public AI governance incomplete
Mother of Divine Grace	Private Catholic K through 12	9	0	1	Parent-led formation; public AI framework absent
Harvard-Westlake	Private grades 7 through 12	6	3	0	AI coursework visible; schoolwide rules unclear
High Tech High	Public charter K through 12	9	3	1	Coherent project model; AI evidence incomplete

Interpretation of the comparison

Strong formation scores identify visible structures that require students to read, remember, reason, discuss, make, revise, or defend their work. They do not prove superior long-term outcomes.
Strong AI-readiness scores identify current guidance, supported tools, training, privacy controls, and course design resources. They do not prove that more AI use produces better education.
The strongest institutional model would combine coherent human formation with bounded, transparent, evidence-tested AI use. No reviewed institution publicly demonstrates that complete model at scale.

Public universities and public schools

UC Berkeley

Public research university Berkeley has the most complete public operational record in the group, including a central AI hub, licensed tools, data rules, teaching guidance, and course examples. Its formation purpose is real but distributed across a large university.

Review dimension	Score	Public basis
Educational purpose and coherence	2	Breadth, inquiry, communication, and critical thinking are public aims, but a common formation sequence is not established.
Protected human work and verification	2	Assignment redesign, disclosure, quizzes, examinations, and selected course practices protect independent work; implementation varies.
Relationships attention and support	1	Student services and collaborative instruction are extensive, but no university-wide attention or relationship design was located.
AI perspective and literacy	3	The AI Hub coordinates guidance, tools, training, policy, research, and teaching resources.
AI use in education	3	Licensed tools, a campus sandbox, workshops, pilots, and multiple course models are publicly documented.
Governance privacy and integrity	3	Risk-based data rules, approved tools, privacy guidance, course-policy expectations, and detector cautions are public.

Primary unresolved gap Campus pilots and course observations have not yet produced public, institution-wide measures of unaided learning, dependence, wellness, or durable judgment.

COVID and wellness context Berkeley documents emergency remote teaching, inequity, and learning disruption. Current public evidence does not isolate which later AI practices were caused by those experiences or whether they improved wellness.

UCLA

Public research university UCLA now has a defined teaching-and-learning AI vision, six public principles, an advisory committee, faculty development, student-use research, and responsible-use guidance. Coverage and measured results remain incomplete.

Review dimension	Score	Public basis
Educational purpose and coherence	2	UCLA promotes broad intellectual and professional development, but formation remains distributed across programs and courses.
Protected human work and verification	2	Assessment integrity is a stated AI principle, and instructors receive guidance on aligning use with learning goals; representative verification evidence is limited.
Relationships attention and support	1	Teaching support, accessibility, and wellness resources exist, but no common AI-related attention or relationship measure was located.
AI perspective and literacy	3	The Teaching and Learning Center states principles for equitable access, student literacy, faculty support, assessment integrity, innovation, and research-led adoption.
AI use in education	2	Faculty fellows, workshops, resources, a 2026 symposium, and student research show active implementation, though not a common student requirement.
Governance privacy and integrity	2	Responsible-use and technology guidance address transparency, ethics, risk, and support; the public record does not show one complete course-to-outcome control system.

Primary unresolved gap UCLA should publish representative course practices and direct measures showing whether its guidance improves unaided mastery, critical evaluation, and student well-being.

COVID and wellness context Instructional-continuity resources preserve lessons from disruption, but no equivalent pre-COVID, closure-period, and current cohort comparison was located.

High Tech High

Public charter network serving K through 12 High Tech High has a coherent project-based model built around authentic work, inquiry, collaboration, revision, public presentation, and adult-world connections. Its public AI record is much thinner than its pedagogical record.

Review dimension	Score	Public basis
Educational purpose and coherence	3	The network publicly organizes learning around common design principles and sustained project work.
Protected human work and verification	3	Student products, exhibitions, critique, revision, internships, and public presentation make thinking and contribution visible.
Relationships attention and support	3	Small schools, advisory structures, teaching teams, and authentic collaboration support sustained human relationships.
AI perspective and literacy	1	A current handbook route references technology and AI-use expectations, but a complete public literacy framework was not located.
AI use in education	1	AI may appear in individual projects or classrooms, but representative network-wide examples were not established in the reviewed record.
Governance privacy and integrity	1	Public governance and handbook structures exist; the operational AI policy, vendor controls, and implementation evidence require verification.

Primary unresolved gap The network should publish the complete AI policy and show how individual mastery is verified inside collaborative or AI-assisted projects.

COVID and wellness context Learning-continuity plans and family communications document disruption response. Later wellness and project-learning outcomes have not been causally connected to closure exposure.

Private research universities

Stanford University

Private research university Stanford combines a human-centered AI literacy framework with approved tools, teaching resources, funded course projects, student guidance, and explicit responsibility for human verification. Formation still varies across programs and courses.

Review dimension	Score	Public basis
Educational purpose and coherence	2	Breadth requirements and undergraduate aims address reasoning, expression, ethics, and citizenship, but do not form one continuous sequence.
Protected human work and verification	2	Course examples use direct reading, original writing, staged drafts, presentations, discussion, and policy deliberation.
Relationships attention and support	2	The AI literacy framework expressly includes well-being, inclusion, stress, and human connection; institution-wide effects are not measured.
AI perspective and literacy	3	Functional, ethical, rhetorical, and pedagogical AI literacy are publicly articulated around human capacities.
AI use in education	3	AI Meets Education, campus tools, funded projects, and multiple course examples show broad educational use.
Governance privacy and integrity	3	Licensed environments, data controls, academic-integrity guidance, disclosure expectations, and agentic-AI safeguards are public.

Primary unresolved gap Stanford should publish course-level and campus-level results for unaided learning, metacognition, dependence, belonging, and long-term retention.

COVID and wellness context The university documents remote and hybrid adaptations and losses in direct interaction. Public evidence does not establish a long-term causal comparison across support conditions.

University of Southern California

Private research university USC provides university-wide ChatGPT Edu access in a protected environment while preserving instructor authority over course use. Its public record is strongest on access, integrity, and faculty guidance, and weaker on a shared formation standard.

Review dimension	Score	Public basis
Educational purpose and coherence	2	USC states broad aims for learning and academic excellence, but no common well-formed-mind sequence was located.
Protected human work and verification	2	Course-specific permission, syllabus clarity, rubrics, authorship rules, and misconduct review protect student responsibility unevenly across courses.
Relationships attention and support	1	The university provides extensive student support, but no public AI-related measure of attention, relationships, or mental fitness was located.
AI perspective and literacy	2	USC presents AI as useful when aligned with course purpose and instructor guidance; a common literacy requirement was not located.
AI use in education	3	ChatGPT Edu is available to active students, faculty, and staff, supported by teaching and technology resources.
Governance privacy and integrity	3	The protected environment, academic-integrity process, course discretion, plagiarism rule, and reporting procedures are public.

Primary unresolved gap USC should distinguish adoption from learning and publish whether broad access changes unaided mastery, authorship, dependence, workload, or faculty-student interaction.

COVID and wellness context USC has a substantial remote-teaching history, but the reviewed sources did not provide an equivalent long-term comparison connecting COVID exposure to current AI design or wellness.

California Institute of Technology

Private research university Caltech explicitly asks how instructors can assess actual ability, deepen learning, preserve effort, and prepare students for an AI-shaped world. Its guidance is thoughtful, but many decisions remain course or division specific.

Review dimension	Score	Public basis
Educational purpose and coherence	2	The core curriculum and scientific culture create common intellectual demands, though no single public formation framework was located.
Protected human work and verification	3	Caltech guidance emphasizes actual ability, process, oral and in-person assessment, self-reflection, verification, and rewards for learning behaviors.
Relationships attention and support	2	Small-scale residential and collaborative structures support human contact, but AI-related attention and wellness results are not public.
AI perspective and literacy	2	CTLO and writing guidance address misinformation, equity, privacy, learning goals, and the difference between support and shortcut.
AI use in education	2	Faculty resources and selected humanities, writing, science, and engineering practices show use, but no common student program was established.
Governance privacy and integrity	2	Institute guidance, security advice, citation expectations, and course discretion exist; implementation and approved-tool coverage remain decentralized.

Primary unresolved gap Caltech should publish representative cross-division course evidence and measured transfer from AI-assisted practice to unaided scientific, mathematical, coding, and writing performance.

COVID and wellness context The reviewed public record did not establish which emergency practices continued after COVID or whether later changes improved connection, workload, or learning.

Mission driven private institutions

Loyola Marymount University

Private Jesuit and Catholic university LMU offers the clearest whole-person mission among the large universities in this sample and connects learning to ethical leadership, mentorship, and human dignity. Its AI resources are expanding, but implementation and outcome evidence remain uneven.

Review dimension	Score	Public basis
Educational purpose and coherence	3	The Catholic and Jesuit mission, core curriculum, and whole-person language provide a coherent educational purpose.
Protected human work and verification	2	Faculty resources support AI-free and AI-integrated assignments, disclosure, academic honesty, and learning-goal alignment; common verification is not shown.
Relationships attention and support	3	Intentional mentorship, intimate classes, mission, and student life create visible relationship structures.
AI perspective and literacy	2	Teaching resources address literacy, ethics, limitations, privacy, equitable access, and intentional adoption.
AI use in education	2	Supported tools, faculty examples, training, and course-design resources demonstrate implementation, but not a universal sequence.
Governance privacy and integrity	2	Academic-honesty guidance, data-security rules, supported tools, alternatives, and privacy cautions are public; detector use warrants corroboration and due process.

Primary unresolved gap LMU should show how its mission changes actual AI assignments and publish measures of judgment, authorship, independent performance, relationships, and well-being.

COVID and wellness context The university should compare students by prior condition, remote-learning exposure, family support, and current AI use before attributing present wellness or learning patterns to COVID.

Thomas Aquinas College

Private Catholic liberal arts college Thomas Aquinas College presents the strongest college-level formation model in the sample: original texts, memorization and demonstration, sustained seminar and laboratory discussion, compositions, tutorials, and a defended senior thesis. Its public AI inquiry has not yet become a disclosed operating framework.

Review dimension	Score	Public basis
Educational purpose and coherence	3	One integrated Great Books curriculum explicitly orders study toward wisdom, judgment, and first principles.
Protected human work and verification	3	Direct engagement with texts, oral discussion, mathematical demonstration, writing, laboratory work, and thesis defense make the student perform the central mental acts.
Relationships attention and support	3	Residential life, small tutorials, discussion, friendship, and technology limits protect attention and direct human encounter.
AI perspective and literacy	2	The Mind and the Machine series and AI and Human Flourishing symposium establish a serious public philosophical and theological perspective.
AI use in education	1	AI is examined publicly, but no required AI literacy course or documented institution-wide classroom use was located.
Governance privacy and integrity	0	No comprehensive public rules for permitted use, disclosure, approved tools, data protection, or AI-assisted assessment were located.

Primary unresolved gap The College should translate its public AI inquiry into explicit guidance and test whether any permitted use strengthens rather than substitutes for reading, reasoning, writing, and discussion.

COVID and wellness context The emergency online transition documented weaker engagement and confirmed the importance of in-person discussion. The return to campus is relevant evidence about institutional fit, but not a causal wellness study.

Mother of Divine Grace School

Private Catholic K through 12 distance learning and home education program Mother of Divine Grace School is evaluated as a K–12 program, not as a university. Its developmental method, parent role, consultant support, memorization, narration, reading, composition, and discussion form a coherent path toward later independent judgment. Public AI guidance was not located.

Review dimension	Score	Public basis
Educational purpose and coherence	3	The K–12 sequence explicitly develops memory, imagination, analysis, argument, communication, and judgment by developmental stage.
Protected human work and verification	3	Parents, teachers, and consultants can observe narration, recitation, writing, discussion, and submitted work directly.
Relationships attention and support	3	Parent supervision, recurring consultant contact, small online classes, homeroom support, and moderated forums provide relationship structures.
AI perspective and literacy	0	No specific institutional statement or student-facing framework for generative AI was located in the public sources reviewed.
AI use in education	0	No published AI literacy sequence, required course, or classroom integration model was located.
Governance privacy and integrity	0	Online security guidance exists, but no public AI rules for age, consent, tools, data, authorship, or assessment were located.

Primary unresolved gap The school should publish age-specific guidance for students, parents, consultants, and online teachers while preserving the parent’s authority and providing non-AI alternatives.

COVID and wellness context The school distinguished prepared home education from emergency crisis schooling and recognized family stress and social isolation. Future comparison should separate students with stable support and friendships from those already experiencing dysregulation, isolation, or weak supervision.

Independent secondary school

Harvard-Westlake School

Private independent school serving grades 7 through 12 Harvard-Westlake publishes a broad college-preparatory curriculum, support systems, and courses that examine artificial intelligence, including its ethical and social implications. The reviewed public record did not establish one schoolwide AI-use and verification framework.

Review dimension	Score	Public basis
Educational purpose and coherence	2	The school states aims of independent thinking, integrity, educational excellence, and purpose, but the curriculum offers many paths rather than one formation sequence.
Protected human work and verification	2	Writing, research, laboratories, arts, discussion, and advanced projects can make student work visible; common AI-era verification rules were not located.
Relationships attention and support	2	Learning centers, counseling, health and wellness, clubs, advising, and faculty contact provide substantial support structures.
AI perspective and literacy	1	AI appears in the curriculum and public course descriptions, but a complete schoolwide literacy framework was not located.
AI use in education	2	The 2026–2027 curriculum includes study of AI and its ethical, technological, and social implications.
Governance privacy and integrity	0	No public schoolwide rules for permitted use, disclosure, approved tools, age and consent, privacy, detection, or appeals were located.

Primary unresolved gap The school should publish age-appropriate middle- and upper-school rules and show how independent writing, research, creativity, and oral explanation are verified.

COVID and wellness context Health and wellness support is visible, but the reviewed public record did not provide a comparable longitudinal account of COVID exposure, family conditions, friendship disruption, and later AI use.

What the comparison reveals

Institutional pattern	Institutions	What should be done
AI-ready with distributed formation	UC Berkeley, UCLA, Stanford, USC	Define the protected human capacities every graduate should retain and publish unaided outcome measures.
Strong reasoning with partial AI system	Caltech and Loyola Marymount	Connect thoughtful guidance to representative course practice, consistent verification, and measured results.
Formation-led with AI governance gap	Thomas Aquinas College and Mother of Divine Grace	Convert educational philosophy into explicit age-appropriate or college-level AI rules without weakening the formation method.
Strong human pedagogy with incomplete AI record	High Tech High and Harvard-Westlake	Publish schoolwide expectations, approved tools, parent and student roles, and evidence from representative grades and subjects.

AI tools most compatible with a well formed mind

Compatibility depends more on the use pattern than on the brand. The Phase Four evidence supports tools that preserve productive difficulty, require explanation, ground claims in sources, keep a human educator responsible, and fade assistance before assessment.

Use pattern	Compatibility	Required safeguard
Socratic or hint-based tutoring	Higher	Ask questions, diagnose errors, refuse assessed answers, and require later unaided work.
Teacher-facing planning or diagnostic support	Higher	The educator reviews every recommendation and remains responsible for instruction and judgment.
Source-grounded research assistance	Conditional	Students read the sources, verify claims, cite them, and explain what the tool contributed.
Feedback on a human first draft	Conditional	Preserve the original draft, limit revision help, and require an oral or written account of changes.
General chatbot before student effort	Lower	Do not use it for the central mental act; require retrieval, first attempts, and independent transfer.
Automated authorship or answer generation	Lowest	Exclude from assessed reading, writing, calculation, interpretation, and judgment unless the task is explicitly to critique AI output.
AI companion replacing human support	Unproven	Do not treat simulated relationship as a substitute for parents, teachers, peers, counselors, or community.
AI detector as sole misconduct evidence	Inappropriate	Require corroborating evidence, human review, notice, and a fair opportunity to respond.

COVID wellness and historical comparison

COVID should remain an explicit comparison factor, not a universal explanation. School closure, emergency remote instruction, loss of services, reduced friendship contact, family stress, and unequal adult supervision changed the conditions under which students learned. Students who already had poor regulation, weak family support, few friendships, disability-related barriers, or mental-health difficulties were less likely to maintain their prior level of learning and wellness than students with stable routines and strong relationships. This is a research hypothesis that should be tested with pre-2020 measures and matched support conditions, not assumed from later distress alone.

Comparison factor	Minimum measure	Reason
Prior condition	Pre-2020 achievement, attendance, disability, attention, distress, and social connection	Separates earlier vulnerability from later exposure.
Closure exposure	Duration, instructional contact, services, assignment completion, and technology access	Distinguishes enrollment from actual instruction.
Family capacity	Adult time, household stress, workspace, language, caregiving, and educational preparation	Tests who could sustain learning at home.
Friendship and support	Peer contact, school connectedness, mentoring, help-seeking, and trusted adults	Tests whether relationships buffered disruption.
Current AI use	Tool, purpose, frequency, timing, guardrails, data exposure, and human supervision	Separates tutoring from substitution or companionship.
Later outcome	Unaided mastery, delayed recall, attention, sleep, distress, belonging, and family relationships	Measures learning and wellness rather than completed tasks.

Parents governance and public influence

The institutions offer very different routes for influence. Mother of Divine Grace places parents inside the instructional structure. Harvard-Westlake can consult families directly but remains a private school. High Tech High operates through public charter governance and public records. Public universities use boards, academic senates, administrative offices, and course-level faculty authority. Private universities and colleges rely more heavily on internal governance. A formal structure does not guarantee that a parent, student, or member of the public can understand who made an AI decision, obtain the evidence behind it, or change the practice.

Every institution should identify who approves AI tools, teaching rules, age limits, privacy terms, assessment uses, and exceptions.
K–12 institutions should explain how parents receive notice, give consent when required, request a non-AI alternative, question a practice, and appeal an academic-integrity decision.
Colleges and universities should explain the division of authority among faculty, departments, academic senates, technology offices, integrity offices, and central administration.
Public reporting should include the evidence reviewed, the decision owner, implementation date, affected population, evaluation result, and next review date.

Influence provenance and security

The public sources reviewed rarely explain the origin and jurisdiction of educational AI systems beyond vendor and privacy information. Institutions should distinguish ordinary academic exchange with allied and partner countries from risks involving entities subject to U.S. foreign-adversary restrictions, undisclosed state influence, intellectual-property transfer, model manipulation, or access to student data. This review found no basis for accusing any selected institution of such conduct. It identifies a governance question that should be answered before adoption.

Record vendor ownership, controlling interests, hosting location, data-processing jurisdictions, subcontractors, model providers, and retention rules.
Review research partnerships, funding disclosures, training-data provenance, content controls, and known government access obligations.
Use current U.S. government definitions and restrictions rather than informal political labels when classifying foreign risk.
Do not convert national-origin risk screening into assumptions about individual students, teachers, or researchers.

Institutional work that should now be completed

Priority	Work that should be completed	Minimum result
1	Publish one current AI governance record covering authority, tools, privacy, age or consent, disclosure, assessment, integrity, alternatives, and review dates.	A reader can identify the rule, owner, scope, and effective date.
2	Select representative grades, subjects, and course types, including no-AI, bounded-AI, guardrailed tutoring, and human-plus-AI conditions.	Practice is observed beyond exceptional examples.
3	Require first attempts, visible process, direct source reading, oral explanation, and delayed unaided checks where AI is permitted.	Assisted output is separated from learning.
4	Measure attention, recall, judgment, integrity, dependence, belonging, distress, family burden, and teacher or parent interaction separately.	Mental fitness and wellness are not reduced to one score.
5	Stratify COVID comparisons by prior condition, closure exposure, family capacity, friendship, disability, and instructional contact.	Historical change is tested without a single-cause story.
6	Audit vendor provenance, foreign jurisdiction, data access, training sources, model updates, and security controls.	Educational adoption includes supply-chain and influence risk.
7	Publish results, limitations, corrective actions, responsible person, and next review date.	The scorecard becomes an accountable, updateable record.

Conclusion and next investigation

The California comparison is ready to move from public-source scoring to verification. The next investigation should test a representative sample of programs and courses against the Phase Four research design. The sample should include a public university, a private research university, a mission-driven college or university, a public K–12 program, an independent school, and Mother of Divine Grace as a parent-directed K–12 model. Thomas Aquinas College should remain a central college-level comparison because its curriculum most clearly protects the human acts that AI can either assist or displace.

The investigation should not begin by asking which institution uses the most AI. It should ask whether students can read, remember, reason, write, calculate, discuss, judge, and relate to other people with increasing independence; then determine whether each AI use strengthens or weakens those capacities.

Public sources reviewed

Official institutional and government sources were accessed September 12, 2026. Public evidence can change. Descriptions are paraphrased unless an official program or tool name is used.

Federal and empirical standards