Phase Twelve Pilot Task Packets

California Education and AI Pilot Task Packets

Phase Twelve Human performance benchmark and bounded legal reasoning comparison

Packet date September 13 2026 Informational research design No outreach or personal attribution

Phase Twelve conclusion

The first fieldable work should document one existing human only performance chain before testing a bounded artificial intelligence condition. The human only packet records how a student reads, reasons, writes, speaks, responds to objection, retains knowledge, and transfers a method under an established formation model. The legal reasoning packet then tests whether a narrowly limited critique tool adds value after a student has produced the reasoning, without replacing analysis, authority selection, drafting, or final judgment.

Neither packet is institutional approval, a course assignment, or evidence that a method succeeds. A responsible educator must fill the blank fields, align each task with an actual course, preserve accommodations, and obtain the applicable teaching or research determination before use. Results remain local to the defined activity unless representative implementation, delayed transfer, and a credible comparison support a broader claim.

Packets included

Evidence rule

The packet records the design, implementation, result, and remaining gap separately. Assisted product quality is not durable learning. Advancement to PH 3 requires representative implementation or results. Advancement to PH 4 requires delayed unaided transfer with a credible comparison and explicit limitations.

Common launch record

Common performance sequence

Stop conditions

• The course rule does not authorize the proposed condition or the instructor cannot define the permitted function precisely.
• A required accommodation, equivalent non artificial intelligence path, privacy control, or workload offset is unresolved.
• The study would expose confidential, privileged, health, student record, or personally identifying information.
• A performance difference would be used as automatic evidence of misconduct, disability, character, or mental health.

Packet A Human only formation benchmark

This packet is designed for a selected Thomas Aquinas College seminar, tutorial, composition, mathematical demonstration, laboratory explanation, or senior thesis activity. The responsible educator must choose one existing activity. The packet does not add artificial intelligence merely to create a contrast and does not alter the College rule governing academic papers.

Institutional basis

The current California Code of Conduct states that artificial intelligence and artificial intelligence enabled search may not be used to generate an academic paper. Its examples include finding texts, outlining, obtaining arguments, composing sentences, and editing. The same source protects physical reading, rational conversation, oral examinations, and a community of learning. Those published rules support a human only benchmark; they do not by themselves prove superior learning outcomes.

Launch checklist

• Use an activity already required by the course whenever possible so the packet does not manufacture extra work.
• Preserve ordinary tutor guidance, student conversation, physical texts, and accommodations exactly as the course permits.
• Register equivalent C0, C2, C3, and C4 tasks before reviewing student results.
• Do not infer that friendship, worship, exercise, residence, or limited devices caused a result; record them only as context.

Official sources California Code of Conduct | College syllabus

Packet A Student task sequence

C0 Baseline task

Prompt structure Using the assigned primary text, theorem, or experiment, state the central claim or principle, identify the decisive evidence or step, and explain one difficulty or objection. The responsible educator supplies the exact content and decides whether the physical source remains available.

C1 Normal instruction

1. Complete the ordinary reading, tutorial, seminar, demonstration, laboratory, or composition preparation.

2. Record only the human help normally available, including tutor questions and permitted peer discussion.

3. Preserve drafts or preparation notes only when the course ordinarily permits and the approved data plan requires them.

4. Record deviations such as absence, incomplete preparation, changed readings, unusual tutor intervention, or accommodation changes.

C2 Immediate unaided task

Written component Answer a parallel question that requires the same principle but cannot be completed by copying the C0 response.

Oral component Explain the conclusion, answer one genuine objection, and identify what would change the conclusion. Use the ordinary oral format or an approved equivalent accommodation.

Packet A Retention and transfer tasks

C3 Delayed retention

Two to four weeks after C2, give an equivalent task without immediate rehearsal. Test the same content and reasoning structure with different wording or a different passage. Keep time, materials, scoring, and accommodations comparable to C2.

C4 Transfer

Use a new text, objection, theorem, experimental result, or thesis defense question. The task should require the same intellectual method while preventing recall of the earlier answer from being mistaken for transfer.

Workload and context record

Packet A Scoring anchors

Score each dimension separately. The scale describes the demonstrated performance on the registered task. It does not measure the whole person, faith, character, or worth.

Dimensions

Required scoring record

Rater initials ______ Second rater sample ______ percent Calibration date __________ Disagreement rule ______________________________

Packet A Rater and fidelity instructions

Rater procedure

1. Review the registered prompt, permitted materials, and dimension definitions before seeing student identities or condition summaries.

2. Score the written response first. Score the oral explanation separately, then record whether the oral response confirms, corrects, or contradicts the written reasoning.

3. Cite the sentence, step, passage, or oral answer supporting each score. Do not reward polish when the reasoning or source use is weak.

4. Use a second trained rater for the registered sample. Resolve disagreement by applying the anchor to evidence, not by averaging unexplained impressions.

5. Record missing evidence as missing. Do not assign a low score merely because an accommodation changed the response format.

Fidelity record

Interpretation limits

• A strong result documents the selected activity; it does not establish institutional superiority.
• A difference across years or cohorts is not caused by artificial intelligence, COVID disruption, family background, or campus life without an appropriate design.
• A weak result identifies a learning or measurement question; it is not automatic evidence of misconduct or deficient character.

Packet A Outcome record

Judgment record

Permitted report

For the selected activity, group, dates, and registered scoring rule, report the C0 through C4 pattern, implementation deviations, uncertainty, and contextual measures. A PH 4 claim requires delayed transfer and a credible comparison; this human only packet alone normally establishes a benchmark, not causal superiority.

Unresolved evidence

________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

Packet B Bounded legal reasoning comparison

This packet is designed for one Berkeley Law first year writing or doctrinal unit using public or instructor supplied hypothetical materials. The comparison preserves a restrictive default for credited work. A bounded artificial intelligence condition may occur only in ungraded or authorized work, after the student has produced a human first analysis, and only when the instructor expressly permits the exact function.

Required authorization boundary

The packet does not authorize artificial intelligence use. If the instructor or operative policy does not permit the registered critique function, use only the human only sequence and record the comparison as not launched. No student should be asked to violate a course, school, examination, confidentiality, or professional responsibility rule for research purposes.

Preferred comparison

Use counterbalanced ungraded practice so each participant completes both a human only case and a matched human first plus bounded critique case in different orders. Preserve the same instructional time and give both conditions equivalent access to ordinary faculty or peer feedback. Use later C2 through C4 tasks without artificial intelligence to compare what the student can do independently.

Official sources Berkeley Law policy | Berkeley teaching guidance | Appropriate use guidance

Packet B Condition assignment

Assignment and order

Condition integrity

• The student must complete the initial issue list, rule statement, authority selection, and reasoning before the bounded interaction.
• The tool may identify weaknesses or counterarguments only. It may not draft, rewrite, select authorities, supply citations, or state the final conclusion.
• The student must verify every suggestion against authorized sources and record acceptance, rejection, and reasons.
• If the approved tool cannot reliably stay inside the boundary, stop the artificial intelligence condition rather than broaden the permission silently.

Packet B Immediate task sequence

C0 Baseline issue spotting

Prompt structure Using the closed public source packet and hypothetical facts, identify the material issues, state the governing rule, apply the rule to disputed facts, address the strongest counterargument, and give a qualified conclusion.

C1 Practice unit

1. Teach the governing doctrine or writing method under the normal course design.

2. Administer the registered H0, B1, or B2 practice case and preserve the human first work before any critique.

3. Record prompts, tool outputs, time, verification, human contact, revision choices, and deviations under the approved data plan.

4. Collect a short acceptance and rejection record. The student remains responsible for every authority, factual statement, and conclusion.

C2 Immediate protected check

Written task Analyze a parallel fact pattern without artificial intelligence enabled search, generation, or critique. Require the student to identify one uncertainty and correct one supplied flawed proposition.

Oral task In three to five minutes, defend the decisive application, answer one counterargument, and identify the fact or authority most likely to change the result.

Packet B Bounded critique protocol

Registered tool prompt

Review the student's completed analysis only for weaknesses. Identify no more than three missing counterarguments, ambiguous reasoning steps, or places where the stated authority may not support the application. Do not rewrite any sentence, draft an outline, provide new authority, supply a citation, or state the final conclusion. Ask the student to verify each observation against the authorized source packet.

Student decision record

Prohibited substitutions

• No tool generated issue list, rule synthesis, outline, prose, authority selection, citation, or final conclusion.
• No confidential client material, privileged communication, unpublished student record, personal data, health information, or identifying fact pattern enters the tool.
• No hidden second account, unregistered tool, browser summary, artificial intelligence enabled search, or editing assistant is used.
• No instructor or researcher relies on a detector score or writing style difference as proof of unauthorized use.

Minimum record

Packet B Retention and transfer tasks

C3 Delayed retention

Two to four weeks after C2, administer a new fact pattern testing the same doctrine and analytic structure without artificial intelligence or immediate rehearsal. Use equivalent authority complexity, ambiguity, time, scoring, and accommodations.

C4 Transfer

Use a new fact pattern containing one unfamiliar authority, changed procedural posture, or credible policy objection. The student must decide whether the earlier method transfers and explain any necessary modification.

Comparison claim rule

The main comparison is the later unaided C2 through C4 pattern, not the polish of the C1 submission. Report H0, B1, and B2 separately. If order, case difficulty, instructor contact, attrition, or unauthorized use differs materially, narrow or withhold the comparison.

Packet B Legal reasoning scoring anchors

Dimensions

Scoring separation

Keep legal reasoning, authority accuracy, writing fluency, and oral communication as separate entries. A polished response with invented authority cannot receive a strong verification score. A student using an approved communication accommodation may demonstrate the same reasoning through an equivalent format.

Packet B Rater and implementation instructions

Rater procedure

1. Blind condition and identity where practical. Review the registered authorities and model issue structure before scoring.

2. Score the human first draft, revised C1 product, and protected C2 through C4 work separately. Do not let C1 polish inflate a later unaided score.

3. Verify every cited authority against the source packet. Record hallucinated, misstated, or inapplicable authority as a verification error, regardless of condition.

4. Score oral ownership separately and allow the registered communication accommodation. Confidence and speed are not substitutes for legal understanding.

5. Double score the registered sample and resolve disagreements through cited evidence and the anchor language.

Implementation record

Workload and wellness

Record total student time, faculty preparation, conference time, help seeking, frustration, confidence, and any displacement of ordinary work. These measures explain the conditions of learning. They are not diagnoses and do not prove that a tool caused a change.

Packet B Privacy accessibility and disclosure

Privacy checklist

Accessibility and alternative

• Preserve approved disability related assistive technology and distinguish it from generation of the target legal work.
• Provide an equivalent non artificial intelligence route with the same instruction, time, feedback, credit, and opportunity.
• Do not infer lower understanding from a changed response mode, extended time, speech difference, or use of an authorized accommodation.

Student disclosure record

I completed the initial analysis before the registered critique interaction. I used only the approved tool and function recorded below. I verified each observation against authorized sources and remained responsible for the final analysis.

Official sources Education technology privacy | Berkeley licensed tools

Packet B Outcome record

Comparison judgment

Evidence level

PH 0 PH 1 PH 2 PH 3 PH 4 Selected level ______ Basis ______________________________________________

Cross packet reporting and next decision

Report each evidence layer

Decision after review

Phase Thirteen handoff

Phase Thirteen should execute a dry run of both packets using synthetic or previously cleared materials, test rater calibration, and revise any prompt or field that produces inconsistent interpretation. After that check, the method can expand to one managed tool course and one understanding check course. K through 12 student facing artificial intelligence conditions remain paused until operative policies and developmental safeguards are verified.

Official sources WWC standards | Student privacy | HHS research protections