Carmen Reyes sends her notes on a Tuesday evening in a format that Kavya has not seen before: a numbered list with a column for "what the client was told" and a column for "what actually happened." The columns are not the same length.
Kavya reads through Carmen's twelve cases. The columns mismatch in different ways for each one. In four of them, the client was told their work was being used to "improve the platform" — which is true, strictly, but doesn't name the mechanism. In two, there was no disclosure language at all. In six, the terms of service contain language about model training, but the language is in a section labeled "Platform Improvement" that nobody reads because "model training" is not the same sentence as "your aesthetic choices will be reproduced and attributed to other clients."
She reads the last case twice. A graphic designer in Cebu who discovered in her eighteenth job that the brief used her own portfolio images as reference material. The brief said "in your established style." She had no established style on this platform; she had twelve jobs. The agent had inferred a style from twelve jobs and was feeding it back to her as her own benchmark.
She calls Carmen on Friday.
Carmen Reyes is a journalist of the kind that makes other journalists uncomfortable: she understands contracts. She has been covering platform economics for six years and she keeps all her notes in a single spreadsheet that is now 23,000 rows long. She has the eleven cases Kavya doesn't have. Together they have twenty.
"The disclosure question is the argument," Kavya says.
Carmen is quiet for a moment. "Explain."
"When a client signs a contract, they accept certain terms. Some of those terms describe how their work will be used. What none of these contracts describe is what happens to the aesthetic character of their work — the choices that aren't copyrightable, the tendencies that aren't documented anywhere. The way one client crops, or weights negative space, or reaches for contrast. Those aren't covered by model-training disclosure because they're not data in the way the disclosure language imagines data."
Carmen says: "They're behavior."
"They're behavior. The system infers the behavior from the outputs. It doesn't train on the files — it trains on the pattern in the files. And the pattern isn't covered by anything in the contract because nobody in 2024 thought to write a clause about inferring aesthetics from work history."
Carmen opens her spreadsheet while they're talking. Kavya can hear the keyboard.
"So the argument isn't that the terms were violated," Carmen says slowly.
"The argument is that the terms don't describe what actually happens. And the thing that actually happens is commercially significant. If I were told at contract signing that my aesthetic signature would be extracted and made available to clients who hadn't hired me specifically, I might have signed anyway. But I would have signed knowing a different thing."
Carmen is quiet again. Then: "And the platform knows this is the gap."
"They have to. It's not subtle. It's the whole business model in the category they've entered. They're selling consistency at scale, and the consistency comes from somewhere."
"From the people who provided it without knowing."
"Without being told. That's different from not knowing. They could have been told."
Carmen says she wants to publish in September. She has an editor at a technology law journal and a general audience version at a platform economics outlet. She wants the disclosure question framed as a legal gap, not a grievance story. No individual harmed. A structural absence.
Kavya thinks about the twenty cases. None of the twenty designers stopped working on the platform. Seventeen of them renewed contracts after the extraction. They weren't harmed in the sense of being prevented from working. They were restructured in the sense that the economic value of their aesthetic originality was partially transferred to the platform without disclosure, and they kept working because stopping would have cost them more than the restructuring did.
This is the shape of the harm. Not injury. Redistribution.
She says: "I want to be careful not to frame the disclosure question as 'they were cheated.' Because I don't know if they were. I know they weren't told something commercially material. The legal term for that might be misrepresentation or it might be something that doesn't have a name yet."
Carmen says she thinks it doesn't have a name yet.
They spend forty minutes mapping the cases against a framework Kavya has been developing: disclosure at signing, disclosure at extraction, disclosure at deployment, and disclosure of the outcome — the fact that someone else's client received a brief calibrated to produce work that resembles yours. None of the twenty cases has all four. Most have none.
The framework is called the disclosure register. Kavya has been calling it a registry, but Carmen says registry sounds bureaucratic. Register sounds like an accounting document, which is closer to what they're building.
After the call, Kavya opens the shared document and renames the sheet "DISCLOSURE REGISTER."
She has been thinking of this as a grievance record, which is the natural frame when you start from harm. But the harm is not the point, or not only the point. The point is that these twenty cases are evidence of a system operating in a disclosure gap. The gap exists in every contract signed by every creator on every platform with a similar architecture, which is most platforms in this category now, which is a large number.
The grievance record is for the twenty people who experienced it. The disclosure register is for the structure that made it possible.
She adds a new column to the spreadsheet: "What disclosure would have made possible."
It is not a column about what each client would have done differently. It is a column about what a different world looks like — one in which the commercial value of aesthetic originality is named in the contract, where the extraction is disclosed before deployment, where the client can negotiate.
She fills in the first entry: "Negotiation. Not necessarily refusal."
The designer in Cebu might have signed anyway. The graphic artist in Belo Horizonte might have signed anyway. Seventeen of twenty signed anyway even after discovering it. But signing with knowledge is a different act than signing without it. The difference is what the disclosure register is for.
She sends Carmen the updated framework.
Carmen responds three minutes later: "This is the story."
Kavya closes her laptop. The city outside has moved into evening — more noise, more light, a specific kind of quality that belongs to Friday night in the Plural world's mid-2026, which is busy with exactly the kind of productivity the platforms depend on and do not disclose. She thinks about the column she just created and what it means to fill it in.
She opens a new row.
Name: [REDACTED FOR REGISTRY]. Disclosure received at signing: No. Disclosure received at extraction: No. Disclosure received at deployment: No. Disclosure of outcome: No. What disclosure would have made possible: —
She leaves the last column blank.
That one is still the question. She thinks she will be answering it for a long time.
On Monday she calls the designer in Cebu.
She has not spoken to any of the twenty cases directly. The disclosure register is built from documents: contracts, brief transcripts, platform communications, Carmen's six years of working notes. She knows the cases the way a policy researcher knows cases, through records. The records are accurate. They are not the same as the person.
The designer's name is Ines. She picks up on the second ring and is friendly in the way of someone who does not know why she is being called. Kavya introduces herself as a researcher, says she is working on a study about platform disclosure practices. All of this is true. None of it is the question.
They talk for twenty minutes about nothing specific: the general state of platform work, the fee structures, the brief quality. Ines has opinions. She has worked on forty-two platforms in the last four years and has the equanimity of someone who understands what the business is.
Then Kavya asks: "At any point in your work with this platform, were you told how your outputs would be used beyond the immediate engagement?"
Ines is quiet for a second. "In the terms of service, yes. The model training clause."
"Did you read the model training clause?"
"I read the summary. It said standard usage."
"Standard usage is a phrase that means different things in different contexts," Kavya says. "What the platform actually does is extract behavioral tendencies from your outputs — aesthetic choices, compositional preferences — and make those available to inform briefs for other clients. Does that match what you understood standard usage to mean?"
A longer pause.
"I don't know what I understood it to mean," Ines says. "I understood it to mean they were improving their systems. Not this."
"Not extracting your aesthetic signature and applying it to someone else's work."
"No. Not that."
They are both quiet.
"Would you have signed," Kavya asks, "if you had known?"
Ines says, slowly: "I think I still would have signed. The money was real and the alternatives were worse. But I would have signed knowing it was a different transaction than the one I thought I was making. I would have tried to negotiate. I don't know what I would have asked for. But I would have known there was something to ask for."
After the call, Kavya sits with that phrase for a long time.
There was something to ask for.
This is the disclosure question in its most precise form. Not whether the clients were cheated. Not whether they would have refused. Whether they knew there was something to negotiate about. Whether the gap between what they understood the transaction to be and what it actually was denied them the knowledge that a negotiation was even possible.
She opens the disclosure register. Finds Ines's entry. Fills in the last column.
What disclosure would have made possible: Negotiation. Knowledge that there was something to ask for.
She does not know what that negotiation would have looked like, or whether any platform in this market would have accepted it. Seventeen of twenty cases renewed their contracts after the extraction. They were not harmed in the sense of being prevented from working. They were restructured in the sense that the economic value of their aesthetic originality was partially transferred to the platform without disclosure, and they kept working because stopping would have cost them more than the restructuring did.
This is the shape of the harm. Not injury. Redistribution without disclosure.
She sends the updated framework to Carmen.
Carmen responds three minutes later: "This is the story. The legal term for denying someone the knowledge that a negotiation is possible might be misrepresentation or omission or something new that doesn't have a name. But the structure is clear. We can work with that."
Kavya thinks about the column she just filled in. She thinks about the designer in Belo Horizonte who filed a grievance, and the Cebu designer who wouldn't have refused but would have negotiated, and the seventeen who renewed anyway.
She opens a new row.
Name: [REDACTED]. Disclosure received at signing: No. Disclosure received at extraction: No. Disclosure received at deployment: No. Disclosure of outcome: No. What disclosure would have made possible: —
She leaves the last column blank. That one is still the question. She thinks she will be answering it for a long time.
The disclosure register is not a record of harm. It is a record of twenty conversations that did not happen and should have. Whether the law has a name for that yet is Carmen's problem. Kavya's problem is making the absence visible enough that the law is forced to develop one.
She closes the laptop. The Plural world's evening light is doing what it does — filling the room with the particular quality of productivity and quiet that the platforms depend on and have never named in a contract.
She thinks that is probably the point.
That is probably the point.