Who Holds the Pen

On buying a way to measure our children — and giving away the power to say who answers for it

Somewhere in New Mexico this fall, a five-year-old will sit down in front of a laptop, put on a headset, and read a story out loud.

She will read to no one.

No teacher leans in beside her. No grandmother nods at the good parts — just a roomful of other five-year-olds, talking and fidgeting and calling out, and a microphone. Behind it, a program listens for the shape of every sound she makes, trying to tell whether she read the word, or guessed, or went quiet. The company says it does this well — steadier than a tired teacher, catching what a person might miss. Maybe. We mostly have the company's word for that. What we can see is the room, and the child in it: five years old, still sounding out her letters the way most children her age are, reading aloud into a headset while the class carries on around her.

And when she is done, it will keep a trace of her.

Not her face. The sound of her reading. Last year alone, across New Mexico, this program listened to children read for fifty-five million minutes. That is not a typo. Fifty-five million minutes of small voices sounding out words. The system can also keep the recordings, turning a passing moment in a classroom into data that lasts.

I want to start here, with the child and the microphone, because everything else in this story is about her, and the records I reviewed do not show her family in the room where the key decisions were made.

Here is a question that sounds simple and is not:

Who decided she would read to a machine?

Not whether she should. Set that aside for a minute; I promise to come back to it and be fair. Just: who decided? Point to the person, the meeting, the signature. Follow the thread back and see where it lands.

I followed it. I read the record. What I found at the end of the thread is the reason I am writing this.

The pen

There is an old image for the power in this story: the pen. Whoever holds the pen writes the rule everyone else lives inside.

A lot of people in this story hold pens.

The company holds one. It writes the product terms and pays for studies that say the program works. The state holds one. It writes the rule that every New Mexico kindergartner through second-grader will use the program. The school board holds one. It signs the checks. Teachers hold a smaller pen, though many will tell you they do not feel as if they hold one at all.

And the child holds a pen too, though almost no one thinks of it that way. The point of teaching a child to read is to help her become someone who can read the world for herself and decide what she thinks. That is the real work beneath all the tests. Not a score. A person who can hold her own pen.

So the question is not simply is the machine good? The question is older, simpler, and harder:

Who is writing — and who should be?

Keep that question with you.

The honest word

Before I tell you what I found, I owe you the strongest case for the other side. Not the weak version — the strong one, the one a smart, good-hearted person running this program would recognize as fair. If I skip it, you should stop reading, because I have not earned anything yet.

Here it is.

New Mexico has struggled to teach reading for a long time. Near the bottom of the country, year after year. That is real. It is not the fault of the children, and it wears on everyone who works in a school here. When you live near the bottom, "keep doing what we have always done" is not a kind answer. It is its own kind of harm.

For years, the way schools checked which children were falling behind was a mess. Different schools used different tests. One tool flagged nearly half of New Mexico's first-graders as possibly dyslexic — a signal so broad that it became hard to act on. A warning that fires too often stops telling you where to look first. And teachers, doing their honest best, can disagree about whether a child read a word correctly. The research is clear and a little uncomfortable: two trained people can listen to the same child and score the reading differently more often than many parents would expect.

Against all that, the state wanted something reasonable to want: one clear, comparable picture, so a child drowning quietly in a room that might miss her does not stay invisible. Wanting that is not the mistake.

So here is the fair question, and it deserves a straight answer. Would you rather go back to six scattered tests, none of them comparable, and a state that cannot see its own children at all?

The honest answer is neither — because that is a false choice, and it is the company's favorite one. As long as the only options are the old mess or their machine, their machine wins by default. The real answer to a broken system is not to go back, and it is not to hand the classroom to the first product that promises one clean number. It is to go forward, on purpose, to something actually shown to work for these children, in every classroom. Plenty of New Mexico educators have been asking for exactly that redesign. They keep being told to shut up and give the shiny thing a chance.

I will not answer a real problem with "AI is bad." That is lazy, and it loses the people who can fix this. But I will not grant the company its own headline either. Steadier, fairer, better than a teacher — that is the company's claim, and we have little besides the company's word for it. An assessment resting on a vendor's unproven claims is not one clear picture. It is one convenient one.

And there is a strange gap in it. We have a careful assessment of how a five-year-old reads — down to a single sound. We have none of the company that assesses her: no independent test of whether it is accurate, whether it is fair, whether it is worth the room it takes. The child is measured to a single sound. The company that measures her is measured by no one. Who assesses the assessor?

So let me be clear:

Whether the tool is any good is a real question — and the honest answer, today, is that we mostly do not know. That is not the thread I am pulling. The tool is not the point of this story.

The state that made me

I should tell you who is talking, because it matters.

I am from here. New Mexico schools taught me to read. I have spent twenty years working in data — the deep, unglamorous middle of it, where you learn how systems behave when no one is watching; how a number becomes a target, then a story, and sometimes a lie; how the trail of what a machine did either exists or does not when someone finally goes looking.

I do not build or sell reading programs. I am not here to burn anything down. This state made me. You do not take an axe to the thing that made you.

You audit it the way you check on someone you love. Carefully. Assuming the best. Hoping to find nothing. And you say what you find, even when it is not what you hoped, because you love it — because a thing you love should be able to say, out loud, how it makes its decisions.

Here is what I found.

The state bought a measurement — and gave away something else

The tool may be worth keeping, if it earns its keep. That was never the thing.

In buying a way to measure its children, New Mexico gave away something much harder to buy back: the ability to explain how the decision was made and who answers for it. The state bought a measurement and, without meaning to, gave away its governance — a big word for a simple thing. Governance means: how we decide, and who answers for the decision.

Let me show you the receipts. Walk the thread with me.

It starts clearly. In 2016, the state ran a real, open contest: a public solicitation, competing companies, a winner chosen. That is how it is supposed to work. The winning program replaced an older one, and the state paid so schools would not have to. Fair and square. Named people made a real decision, in the open.

Then the clarity begins to fade.

A couple of years later, with a new governor coming in, a letter arrives. In January 2019, the company certifies that it is the only available source for this kind of program, allowing the state to keep buying from it without holding another contest. That can be a lawful and ordinary part of procurement. But sit with it for a second: the proof that no one else could do the job was authored by the vendor that would get the job.

From there, the purchasing path begins to wander. First there is a direct state contract. Then the purchase runs through a buying cooperative — a shared arrangement that keeps a small district from having to run its own contest for every pencil. That is a genuinely useful idea. Then, on one school district's purchase order, the trail points somewhere stranger: a memorandum of understanding between the state and a supply-chain company, dated July 2019 and effective "until February 2025."

I went looking for that memorandum. So did professional researchers helping me. We could not find a public copy. We could confirm that the company named on the paperwork exists — through a tax ID — and little more. The document cited as authority for part of the purchase does not appear to be publicly available.

And the cooperative contract used as a purchasing vehicle? The cooperative's own records say it was extended for six months and expired on August 28, 2025. I found no public explanation of what purchasing authority carries the program now.

Follow the thread and watch what happens. It begins as one clear, open decision in 2016. Then it comes apart into pieces: a vendor-authored sole-source letter, a cooperative, a missing memorandum, an expired contract. Each piece is smaller and dimmer than the last, until you are squinting.

Then came the part that surprised me most.

At the school-district level, where the checks are signed, I looked for the moment an elected board voted to do this. In a democracy, that is the checkpoint: public servants saying yes in public, on the record, with their names attached.

I found the votes. Sort of.

In the records I reviewed, boards approved large contracts in bundles, under a single agenda line reading, more or less, "Contracts over fifty thousand dollars." One vote for the whole stack. In the public-facing materials I reviewed, the company was not named. On the internal paperwork, the box marked "approved by the board" was left unchecked on more than one contract. One approval I traced led to a committee meeting, not the full board.

I am not telling you anyone broke a law. I am telling you something quieter: I could not find a room where the whole decision was made, because the system never required one. No villain skipped the meeting. Every person did an ordinary job. The company wrote its ordinary letter. The state used its ordinary shortcut. The board took its ordinary bundle vote. A buyer selected the ordinary five-year option — and yes, one district bought five years of the program for $139,500, extending to about 2028.

Each move was small. Each move was defensible. Add them together and you get a machine listening to fifty-five million minutes of children's voices under a rule whose author is not visible.

Here is the whole thing in one line:

The mechanism is the sum of exemptions.

Every shortcut in that chain exists for a good reason. The cooperative keeps a small district from drowning in paperwork. The bundle vote keeps a board from spending all night reading contracts. The company's letter saves the state a step. Each shortcut, on its own, is reasonable.

But each shortcut also removes a stop — a place where somebody might otherwise have to ask, Wait. Should we? String enough good shortcuts together and you build a road with no place left to ask the whole question. No one decides the whole thing. The decision falls through the gaps between a dozen people, each doing a job that makes sense on its own.

Then, in spring 2026, came the last crack. A state memo set the rules for the next school year. In a single sentence about the tests every child must take, it said the state does not allow a parent to opt their child out.

I found no public vote attaching a name to that sentence. It simply appears as a rule — a wall with no visible hand on the pen.

The child's agent was left outside the room

Look at the grown-ups standing around this child.

The company has lawyers. The state has a purchasing office and a legal team. The district has a purchasing department. Teachers have unions, contracts, or other employment protections. Every adult actor who helped make or carry out this decision has someone whose job is to protect that actor's interests — someone to hold the pen, answer the phone, and take a side when something goes wrong. There is a word for that person: an agent. Someone who acts for you and answers for you.

A child has agents too: her parents or guardians. But the record I reviewed does not show them being brought into this decision.

The five-year-old reading into the microphone is the only party whose actual voice enters the system. Yet the public materials I reviewed do not show a clear notice telling her family that recordings may be kept, an affirmative request for consent, or a meaningful way to decline. In spring 2026, the state said parents could not opt out.

Every party in this system has an agent at the table — except the children.

I could stop there, and it would be true and sad. But there is a deeper version of this, and it is the part I cannot stop thinking about.

The program does not only measure reading. The people who run it told the state legislature that it can do something else. When a child reads aloud and says something that does not belong in the story — the examples given were "I didn't get any breakfast this morning" and "I don't feel safe at home" — the program can notice and send an alert.

Think about what that means. Somewhere, a child alone at a laptop, reading a story to no one, says one of the truest and hardest things she knows how to say. A light goes on in another room.

I will not guess why the feature exists. Maybe someone who built it pictured that frightened child. Maybe it is a line on a sales slide. It does not matter — and that is the point. A child's safety cannot rest on a company's motives, good or bad. It has to rest on a rule. And the rule is not in the record.

When the light goes on, who receives it? What must that person do? How fast? Who answers if no one does?

The company says the machine can identify reading errors with ninety-five percent accuracy. We built the ear. We spent millions on the ear. But in the statewide public record I reviewed, I did not find the answer: a named, accountable person responsible for what happens next.

General child-protection duties may apply. They are not a substitute for an operating rule tied to this feature. I found no statewide public protocol saying who receives the alert, how quickly that person must act, what record must be made, and who is accountable if the alert is missed.

The system can judge the sound of the word cat. The public record does not tell the child who must answer a cry for help.

That is the whole problem in one image. We have become very good at listening. We have not made the answer visible, reliable, or accountable.

Put the picture back together

You may expect the ending to be stop the machine. It is not. Stopping the machine would not fix the thing that is actually broken. The decision came apart into pieces, and no single piece is where accountability lives.

You do not fix a thing that came apart by getting angry at one piece. You fix it by putting the picture back together — by creating one checkpoint where the full arrangement must be reviewed in public, with names attached.

Here is the lucky part, the reason to feel something other than tired: that checkpoint is coming. The state has said it will shop for the program again for the 2027–28 school year, with a new public solicitation possible as early as fall 2026. That contest is the best available chance to reconnect the pieces and decide them together. It is the closest thing to the room the last decade lacked. For a moment, the pen returns to the public.

The work now is not only to protest. It is to write down, before the contest opens, what the next contract must say. Not angry demands. Plain terms any careful family would want, written so a busy state employee can lift them into the document:

Name the law. The contract should say, in words, that it follows the federal student-privacy law. Across the 126 pages I reviewed, that law is never once named.

No permanent license to a child's voice. The company should be allowed to use a recording only to do the job the school hired it to do — not forever, not for unrelated products, and not to train whatever it builds next without new, explicit permission. The current fine print grants broader rights.

Delete by default. When the school year ends, the recordings should go unless there is a specific, named reason to keep them. Retention should be off by default, not on.

Show the proof for every child. The program performs strongly in some grades and, by its maker's own submitted data, more weakly in others. For children with accents, speech differences, or disabilities, the honest answer is: we do not have enough evidence. That is exactly the child the committee chairman described at the hearing — his son, a strong reader with a speech difference, who "gets absolutely frustrated and really dislikes" the program because it scores him poorly. Before requiring the machine for his son, show that it can hear his son fairly.

Give the child another door. When the machine cannot fairly measure a child, there must be another valid way — on paper, aloud, or with a human. Reaching that door should not require a special-education label first. The current pathway appears to.

Use an outside grader. The studies submitted in support of the program were funded by its maker. That does not make them false. It does make independent evaluation essential. The next contract should require someone the company does not pay to check whether the program works, for whom, and under what conditions. A company grading its own homework is evidence with a conflict attached. Who grades the grader? It is not enough.

Answer in New Mexico. A dispute involving a New Mexico child's data should be heard under New Mexico law, in New Mexico — not in a distant forum chosen in the vendor's fine print.

Put a real hand behind the alarm. If we teach a machine to hear a child say "I don't feel safe," then we owe that child a named, trained, accountable human on the other end of the light. The contract should say who receives the alert, how quickly that person must act, what happens next, and what record is kept. Build the responder, not just the alarm.

None of that stops the reading. None of it is anti-technology. Every line puts a name back on a pen. It takes a decision that fell through the gaps and sets it on a table where people have to sign for it.

There is a hard truth beneath this, and it comes from another part of my working life. People who fight fraud learn that a threshold can be defeated by splitting one large transaction into many small ones. Governance can fail by the same arithmetic even when no one is cheating: split one consequential decision into enough ordinary steps, and no step looks large enough to demand public attention.

You cannot build a threshold that addition cannot defeat.

But you can build a ledger that adds the pieces back together — a plain, open record of who did what, under what authority, for how long, and who answers for the result. Addition cannot hide from that. You cannot out-clever every gap. You can make sure that whatever falls through lands in the light, with a name on it.

Not a smarter wall. A record no shortcut can dodge.

That is not fraud-fighting after the fact. It is building the system so it cannot come apart in the dark in the first place.

Show up

I keep coming back to the image from the beginning: the pen.

Whoever holds it writes the rule the rest of us live inside. Here is where I landed after reading all 126 pages.

The pen does not belong to the company alone. It does not belong only to the state, the board, or the teacher. It belongs in public hands: with the officials we elect to answer for children, with the families entrusted to speak for them now, and with the children we are raising to speak for themselves.

The adults' job is not to author what a child becomes. It is to protect the conditions in which she can become the author of her own life. That is the promise beneath a public school. Not a score. A mind with its own pen.

New Mexico is out ahead on this. You can read that two ways, and both are true. We are on the frontier — among the first to use a listening machine with our youngest children at this scale, before much of the country has noticed the question. And we are on the front line — among the first places where the hard questions are landing in a real hearing, a real debate, in front of real parents. Whether this is a frontier or a front line depends on whether we decide out loud and on purpose, or let the decision keep falling through the gaps.

Every spring, in old villages along the Rio Grande, people who share an acequia walk its length with shovels and clear out the winter. The work is shared. The water will not move unless people show up. So people show up. By noon, you know exactly who your neighbors are.

That is the lesson. Not that a machine is bad. That a shared system works only when the people it serves can see it, enter it, and know who is responsible.

As the state prepares the next contest, show up. Not with a torch. With the plain list of what the contract should say. Put the names back on the pens.

And hand the child hers.

A note on sources: none of this is secret. Every number and quotation here comes from the public record — a legislative hearing, the materials handed to the committee, and 126 pages of contracts any citizen can request. I did not uncover anything. I only added it up and wrote my thoughts down.