Verification and pilot materials · Complete web edition

Phase Fifteen Preimplementation Determination

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

2,781words

19tables

84text 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 Preimplementation Determination

Phase Fifteen Evidence gates and authorization record

Determination date September 13 2026 Informational review design No outreach or student participation

Determination field	Entry
Responsible person	________________________________________
Organization or office	________________________________________
Tabletop record date	________________________________________
Determination status	________________________________________

Current determination

The present record does not authorize a student pilot. Phase Fourteen supplies a complete educator tabletop package, but no completed human score sheets, calibration figures, local policy review, approved account confirmation, safeguard approval, or workload record is available. The supported decision is to keep student participation paused while allowing a responsible local team to complete the preimplementation review described here.

This determination concerns readiness to consider an implementation. It is not a finding about any institution, educator, student, course, artificial intelligence product, or educational outcome. It does not convert missing public evidence into evidence of poor practice.

Decision routes

Route	Present status	What it would permit
Human only benchmark	Eligible for local review	Observation of an existing activity with no student facing artificial intelligence and ordinary support preserved.
Bounded higher education comparison	Not authorized	A tightly limited critique condition only after all calibration, policy, privacy, accessibility, and workload gates pass.
K through 12 student facing condition	Paused	No student facing artificial intelligence condition until age, consent, parent authority, privacy, accessibility, and developmental safeguards are verified.
Institutional claim or ranking	Prohibited	No score, ranking, wellness claim, learning claim, or causal conclusion from preparation or tabletop materials.

Determination authority and scope

Required decision authority

A named local authority must own the decision. The appropriate authority depends on the activity and may include the course instructor, academic unit, school administration, privacy office, accessibility office, legal counsel, research review body, information security office, or parent governance process. Completion by one office does not substitute for another office when its approval is independently required.

Authority area	Required record	Status
Teaching authority	Operative course or school rule and the intellectual purpose of the task	Verified Unresolved
Privacy and records	Permitted data, vendor terms, retention, deletion, and incident path	Verified Unresolved
Accessibility	Accommodations and an equivalent non AI route	Verified Unresolved
Research review	Written determination whether the activity is instruction, evaluation, research, or another regulated activity	Verified Unresolved
Information security	Approved product, account, configuration, and restricted information rules	Verified Unresolved
Workload owner	Named staff, displaced activity, time budget, scoring capacity, and stop authority	Verified Unresolved

Scope exclusions

No document in this phase authorizes outreach, recruitment, student assignment, tool purchase, account creation, or transfer of educational records.
No artificial intelligence detector result may establish authorship or misconduct.
No score may be interpreted as a measure of intelligence, character, wellness, disability, political viewpoint, or institutional quality.
No outcome claim may be made without a defined comparison, protected independent measure, and evidence appropriate to the claim.

Official sources Student privacy and education technology | HHS research protections

Required Phase Fourteen records

The responsible person should assemble the original records below before selecting an authorization route. A summary prepared after discussion cannot replace the independent source records.

Record	Completeness test	Location and status
Rater one score sheets	Packet A and Packet B scores, evidence notes, time, and signature are complete	____________________________
Rater two score sheets	Completed independently without access to Rater one or the keys	____________________________
Separation record	Packet version, issued copies, key holder, and premature access are recorded	____________________________
Initial agreement record	Exact, within one point, critical rule, and missing evidence figures precede adjudication	____________________________
Adjudication record	Every disagreement identifies task, anchor, example, or rater cause	____________________________
Revision log	Each wording or safeguard change is dated and traceable	____________________________
New unseen rescore	Substantive revisions are tested on examples not used for training	____________________________
Safeguard review	Every safeguard has a pass, revise, or stop decision with evidence	____________________________
Workload record	Actual time and displaced work are recorded by activity	____________________________
Tabletop decision	Revise, ready for determination, or pause is selected with a named owner	____________________________

Record integrity check

Preserve both original score sheets and their timestamps without overwriting either record.
Record calculations before opening the adjudication keys.
Explain every excluded or unavailable score instead of converting it to zero.
Keep the training set separate from the unseen set used after revisions.
Stop if a real student or client record appears in a tabletop exercise.

Calibration gate

Calibration determines whether different educators can apply the registered dimensions consistently. It does not prove that the instrument measures durable learning or that a future student condition is beneficial.

Measure	Required threshold	Recorded result	Gate
Exact agreement	At least 80 percent	____________	Pass Fail
Within one point	100 percent	____________	Pass Fail
Differences of two or more	None unresolved	____________	Pass Fail
Verification decisions	100 percent correct	____________	Pass Fail
Boundary decisions	100 percent correct	____________	Pass Fail
Missing evidence decisions	100 percent correct	____________	Pass Fail
Post revision unseen set	All thresholds met again	____________	Pass Fail

Calculation record

Comparable scores Exclude only dimensions registered as unavailable Count __________

Exact agreement Exact agreements divided by comparable scores multiplied by 100 Result __________ percent

Within one point Exact agreements plus one point differences divided by comparable scores multiplied by 100 Result __________ percent

Critical decisions Correct verification, boundary, and missing evidence decisions divided by applicable decisions multiplied by 100 Result __________ percent

Failure rule

Any failed calibration gate requires a documented clarification and a new unseen set. Reviewers may not remove a difficult example, average an unresolved difference, or change a threshold after seeing the scores merely to obtain a pass.

Official sources What Works Clearinghouse standards and protocols

Safeguard gate

Safeguard	Evidence required	Decision
Express authorization	The operative rule names the permitted function, condition, disclosure duty, and prohibited uses	Pass Revise Stop
Approved account	The exact product, account type, configuration, and data class are approved	Pass Revise Stop
Source control	The task identifies the closed source packet and the response to missing material	Pass Revise Stop
Data minimization	Only synthetic, public, or specifically cleared information is used	Pass Revise Stop
Retention and deletion	Outputs, logs, dates, owners, and deletion method are documented	Pass Revise Stop
Accessibility	Authorized assistive technology and accommodations remain available	Pass Revise Stop
Equivalent route	The non AI route preserves instruction, time, feedback, credit, and opportunity	Pass Revise Stop
Human first record	The initial response is frozen before any permitted critique	Pass Revise Stop
Independent checks	Oral explanation and transfer tasks occur without generative assistance	Pass Revise Stop
Workload	A named owner can complete review without displacing essential teaching or support	Pass Revise Stop
Interpretation	Ranking, detector based misconduct, wellness inference, and causal claims are prohibited	Pass Revise Stop

Gate rule

A bounded comparison may proceed to final local authorization only when every applicable safeguard passes. A revise result requires a new verification record. A stop result blocks the condition until the responsible authority resolves it in writing.

Official sources Berkeley appropriate use guidance | Berkeley licensed artificial intelligence tools

Human only benchmark determination

The human-only benchmark is the first eligible implementation route. It should observe an existing educational activity without introducing student-facing artificial intelligence, changing ordinary credit, or withholding customary instruction, support, or accommodations.

Benchmark field	Required entry
Course or grade and activity	____________________________________________________________
Target intellectual act	Reading Recall Analysis Writing Calculation Discussion Creation Other __________
Existing assignment status	Existing activity verified by __________________ Date __________
Human only condition	Devices and generative assistance excluded from the registered performance window
Ordinary support	Instruction, feedback, tutoring, and preparation remain unchanged
Accessibility	Approved accommodations and assistive tools documented separately from target reasoning
Protected measures	Initial performance Oral ownership Transfer Delayed retrieval Other __________
Student record handling	Collection basis, access, retention, deletion, and reporting owner documented
Research status	Instruction Evaluation Research review required Determined by __________
Stop authority	Named person __________________ Stop conditions __________________________

Human only authorization

Decision	Select	Basis
Authorize local benchmark review	____	All applicable authority, privacy, accessibility, workload, and record handling conditions pass.
Revise	____	The activity or measure needs clarification before local review.
Pause	____	Authority, burden, accessibility, privacy, or record handling remains unresolved.

Bounded higher education comparison determination

This route is limited to higher education. It compares a protected human-first response with a fixed artificial intelligence critique condition. It does not permit answer generation, rewriting, source expansion, simulated relationships, consequential automated decisions, or use of sensitive student records.

Condition	Required evidence	Status
Human benchmark established	A comparable human-only record exists and its administration is stable	Verified Unresolved
Course rule	The operative policy expressly permits the exact critique function	Verified Unresolved
Approved tool	The exact institutional account and configuration are approved	Verified Unresolved
Fixed critique format	The tool returns only registered weaknesses, source locators, and verification questions	Verified Unresolved
Frozen first response	The original human response is preserved before critique	Verified Unresolved
Closed sources	No outside authority or unsupported citation may enter the condition	Verified Unresolved
Equivalent alternative	A non AI route offers equivalent instruction, time, feedback, credit, and opportunity	Verified Unresolved
Protected ownership	The learner explains and defends the final reasoning without generative assistance	Verified Unresolved
Transfer measure	A new task tests whether reasoning carries to unfamiliar material	Verified Unresolved
Data controls	Input, output, log, retention, deletion, access, and incident rules are approved	Verified Unresolved
Workload capacity	Human review and follow-up can be completed without displacing essential work	Verified Unresolved

Comparison authorization

Decision	Select	Basis
Authorize final local review	____	Every calibration and safeguard gate passes and all required authorities approve.
Revise and retest	____	A correctable task, anchor, safeguard, or workload issue remains.
Pause	____	Any approval, privacy, accessibility, account, workload, or critical boundary remains unresolved.

Official sources Berkeley Law artificial intelligence policy | Berkeley teaching guidance

K through 12 determination boundary

K through 12 institutions remain within the investigation, including Mother of Divine Grace School, Harvard-Westlake School, and High Tech High. This phase does not authorize a student-facing artificial intelligence condition for any grade. Adult educators may review the materials, and a lawful human-only observational baseline may be considered through the appropriate local process.

Requirements before any later proposal

Requirement	Evidence that must exist	Status
Developmental purpose	The proposed use is tied to a defined age-appropriate intellectual act and does not replace foundational practice	Unresolved
Parent authority	Notice, meaningful choice, consent where required, and a non AI route are documented	Unresolved
Age and account eligibility	Product terms, minimum age, school account rules, and adult supervision are verified	Unresolved
Student privacy	Data categories, vendor role, retention, deletion, training restrictions, and incident response are approved	Unresolved
Accessibility and equity	Accommodations, device access, language access, and an equivalent route are protected	Unresolved
Human relationships	The use does not substitute for teacher, parent, peer, counseling, or pastoral relationships	Unresolved
Assessment	Unaided reading, writing, recall, calculation, discussion, creation, and judgment remain protected where relevant	Unresolved
Wellness and dependency	Attention, distress, compulsive use, isolation, and help-seeking stop rules are defined without unsupported diagnosis	Unresolved
Governance	The responsible school authority and review date are recorded	Unresolved

Standing decision

Student-facing artificial intelligence remains paused. The pause is a governance boundary, not a judgment that every use is harmful. It remains in force until the required evidence is verified for the specific grade, task, tool, account, and institution.

Protected outcome and interpretation plan

A later implementation may report only measures registered before observation. Assisted performance while a tool is available must remain separate from unaided learning, delayed retention, transfer, authorship, wellness, and context.

Evidence layer	Permitted record	Prohibited inference
Claim	The stated educational purpose and policy basis	Treating the stated purpose as implemented practice
Practice	Observed task, condition, support, tool mode, and deviations	Assuming a written rule describes every classroom
Immediate result	Performance during or immediately after the condition	Calling assisted performance durable learning
Durable result	Delayed unaided retrieval, transfer, oral ownership, or repeated observation	Inferring durability from a single immediate score
Context	Prior attainment, disability, language, attendance, family support, friendships, COVID disruption, and ordinary instruction	Treating a contextual association as a cause
Gap	Unavailable, incomplete, incomparable, or unverified evidence	Converting missing evidence into a favorable or unfavorable score

Minimum protected measures

Measure	Timing	Purpose
Human first response	Before assistance	Records independent starting performance
Immediate revision	After bounded feedback	Records assisted improvement without calling it learning
Oral ownership	After submission	Tests explanation and defense in the learner's own words
Near transfer	New but similar task	Tests application beyond the practiced example
Delayed retrieval	Registered later interval	Tests retained unaided knowledge
Process and burden	Throughout	Records time, attention, frustration, accessibility, and displaced work

Reporting rule

Report results by measure and condition with denominators, missing data, deviations, and uncertainty. Do not collapse the record into one institutional rank or one well-formed-mind score.

Workload and stop decision

Workload record

Activity	Owner	Planned minutes	Actual minutes	Displaced work
Preparation	____________	______	______	________________
Independent scoring	____________	______	______	________________
Adjudication	____________	______	______	________________
Safeguard review	____________	______	______	________________
Benchmark administration	____________	______	______	________________
Protected checks	____________	______	______	________________
Follow up and deletion	____________	______	______	________________

Mandatory stop conditions

A required authority has not approved the exact condition.
A key was opened before independent scoring or scoring was discussed before the records were locked.
Real sensitive information entered an exercise without written approval.
The tool expanded sources, rewrote the target work, or produced a consequential decision.
The equivalent non AI route became materially worse in time, instruction, feedback, credit, or opportunity.
An accommodation was denied or confused with the reasoning being measured.
A participant experienced distress, coercion, retaliation, or a loss of ordinary support.
Required human review could not be completed within the registered workload plan.
The record was used to infer misconduct, diagnose wellness, or rank a person or institution outside the registered purpose.

Stop record

Field	Entry
Stop date and time	____________________________________________
Person invoking stop	____________________________________________
Condition observed	____________________________________________
Records secured	____________________________________________
Required review	____________________________________________

Final preimplementation decision

Gate summary

Gate	Result	Evidence location
Phase Fourteen records complete	Pass Fail	________________________________
Calibration thresholds met	Pass Fail	________________________________
Authority approvals complete	Pass Fail	________________________________
Privacy and approved account verified	Pass Fail	________________________________
Accessibility and equivalent route verified	Pass Fail	________________________________
Workload feasible	Pass Fail	________________________________
Protected measures registered	Pass Fail	________________________________
No mandatory stop remains open	Pass Fail	________________________________

Required selection

Decision	Select	Effect
Authorize human only benchmark	____	Permits the named local process to consider an existing human-only activity under the recorded safeguards.
Authorize bounded higher education review	____	Permits only the named higher-education comparison to enter final local review. It is not student authorization by this document alone.
Revise and repeat tabletop	____	Requires correction and a new unseen calibration set before reconsideration.
Pause	____	Blocks further implementation activity until the identified evidence or authority gap is resolved.

Decision record

Field	Entry
Selected decision	____________________________________________________
Named activity and condition	____________________________________________________
Evidence supporting decision	________________________________________________________________________________________________________
Open conditions or revisions	________________________________________________________________________________________________________
Responsible person	____________________________________________________
Approving offices	____________________________________________________
Effective and review dates	____________________________________________________

Present evidence decision and handoff

Decision based on the current record

Evidence layer	Phase Fifteen finding
Claim	The investigation now has a complete preimplementation decision structure.
Practice	Phase Fourteen provides exercises and keys; Phase Fifteen provides gates, authority mapping, authorization forms, and stop rules.
Immediate result	A local team can complete a consistent preimplementation determination without treating preparation as implementation evidence.
Durable result	None. No educator calibration, student performance, retention, transfer, wellness, or institutional outcome has been observed.
Context	No outreach, educator participation, student participation, course implementation, or institution-specific approval occurred.
Gap	Completed tabletop records, local authority decisions, approved accounts, privacy and accessibility review, workload capacity, and protected outcome records remain unavailable.

Current decision

Pause student implementation. Permit only local review of the materials and consideration of a human-only benchmark through the applicable institutional process. Do not authorize the bounded higher-education comparison until completed human calibration and every local safeguard gate pass. Keep all K through 12 student-facing artificial intelligence conditions paused.

Next evidence step

The next step cannot be completed by additional desk drafting alone. A responsible local team must conduct the Phase Fourteen tabletop review and complete the Phase Fifteen decision record. If the records pass, the next document should be a narrowly scoped implementation authorization and preregistered observation plan for the selected route. If they fail or remain incomplete, the record should identify the required revision or preserve the pause.

Continuation limits

A completed form does not itself satisfy an approval that belongs to a teaching, privacy, accessibility, legal, security, research, or parent authority.
A tabletop result cannot establish educational benefit, harm, authorship, wellness, or institutional quality.
A human-only benchmark should precede any bounded higher-education comparison when comparable baseline evidence is unavailable.
Student participation begins only after the named activity receives every required local authorization.

Official sources Thomas Aquinas College California Code of Conduct | Berkeley Law artificial intelligence policy | Berkeley teaching guidance