The letter arrived on a Tuesday. The first one arrived on a Tuesday too, and the second on a Thursday, and I have started to think that Vieira & Machado's paralegal has a scheduling system, or perhaps just a preference for early-week aggression with mid-week follow-through.
The letter is on the kitchen table. The Selo architecture diagrams are also on the kitchen table because the kitchen table is where I work, where I eat, where I spread documents out in configurations that might be organizational systems or might be the habits of a person whose one-bedroom apartment in Pinheiros does not have a desk. The cease-and-desist sits to the left of my laptop. The architecture diagrams sit to the right. Between them, the laptop screen shows a blank GitHub repository creation page.
I have read the letter seven times today. The core claim occupies two paragraphs in the middle, flanked by the standard legal architecture of threat: identification of parties, assertion of rights, demand for cessation, consequences of non-compliance. The core claim:
A infraestrutura Selo de Processo, conforme desenvolvida e publicada por Beatriz Tanaka-Reis, incorpora conceitos proprietários de metodologia de detecção pertencentes à SortLayer Tecnologia S.A., incluindo, mas não se limitando a, frameworks de análise de confiança de conteúdo, taxonomias de classificação de processo criativo, e protocolos de cadeia de verificação.
The Selo de Processo infrastructure, as developed and published by Beatriz Tanaka-Reis, incorporates proprietary detection methodology concepts belonging to SortLayer Tecnologia S.A., including but not limited to content confidence analysis frameworks, creative process classification taxonomies, and verification chain protocols.
I built SortLayer's detection tools for three years. I know exactly what their content confidence analysis framework does: it ingests creative work, runs it through a classifier trained on labeled datasets, and outputs a probability score — the pontuação de confiança — indicating likelihood of AI generation. The classifier is a black box to users but not to me. I built it. I also built the false-positive analysis that proved it was biased: 2.3 times more likely to flag Portuguese-language content as AI-generated than English-language content. The bias was not in the model architecture — it was in the training data, which overrepresented English-language AI-generated samples and underrepresented Portuguese-language human-created samples. The model learned that Portuguese-language creative patterns were unusual. In a detection context, unusual means suspicious.
I published the analysis. SortLayer fired me. Now SortLayer claims I am using their proprietary concepts.
Here is what the Selo de Processo actually does: it documents human creative decisions. Not detects AI content. Documents human process. The difference is directional. Detection asks: was this made by a machine? Documentation asks: what decisions did a human make? Detection is accusatory. Documentation is declarative. They are not the same methodology. They are not even the same question.
But the cease-and-desist does not claim they are the same. It claims I am using conceitos — concepts. Concepts from detection methodology. The argument is that my experience building detection tools gave me conceptual knowledge that now informs the Selo architecture, and that conceptual knowledge is proprietary. This is both true and absurd. It is true that I understand detection methodology. It is absurd that understanding a methodology constitutes infringement of it. If that were the legal standard, every whistleblower would be an infringer. Every former employee who builds something better would owe their former employer for the education.
The letter does not need to win in court. It needs to create uncertainty. It needs me to spend money on lawyers, time on responses, energy on defense instead of development. It needs the Selo community to hesitate. It needs Kenji to wonder if his free PROVAS documentation sessions are legally compromised. It needs Seo-jin to question whether her spreadsheet methodology — which she developed independently, which has nothing to do with SortLayer's detection pipeline — might somehow be contaminated by association with me.
I know this because I was in the room when SortLayer's legal team discussed this exact strategy for a different competitor. August 2024. Conference room B, the one with the broken air conditioning. The general counsel said: we do not need to prove infringement. We need to create a fog of legal uncertainty that slows their development by six months. By then we will have shipped version 4.0 and the market will have moved. I remember the exact words because I wrote them in my notebook, in the margin next to a diagram of the false-positive analysis methodology I was developing in my own time, on my own hardware, using publicly available datasets.
The notebook is in the bottom drawer of the kitchen cabinet, behind the pots. I have not looked at it since I left SortLayer. The notes are mine. The methodology is mine. The concepts are mine. But the letter says otherwise, and the letter has a law firm's name on it, and the law firm bills by the hour, and I do not.
The community Telegram has been active all day. Kenji ran his first PROVAS documentation session last Saturday. Twelve photographers. He sent photos: a folding table covered with prints, raw files on laptops, people writing by hand on the evidence-chain templates I designed. The templates are the part that SortLayer's letter specifically targets — the protocolos de cadeia de verificação, the verification chain protocols. In the letter, these protocols are proprietary. In Kenji's studio, they are photocopied worksheets with coffee rings on them.
Lina is documenting her own creative process with a time-lapse camera. She draws fashion figures by hand — Staedtler 0.5mm for silhouettes, 0.3mm for stitching detail — and films the entire session, then publishes the time-lapse alongside the finished illustration. She calls it her diário de processo, her process diary. She has started monetizing it: a subscription service where followers pay to watch her work in real time. The process is the product. She did not need the Selo specification to do this. She needed the idea that process could be visible, and the idea predates me, predates SortLayer, predates the sorting economy itself. My grandmother made tempura and explained each step to my mother and my mother explained each step to me. Process visibility is not a proprietary concept. It is how humans have taught each other to make things for ten thousand years.
Seo-jin messaged the group this morning. She has been accepted to Kenji's next session. She is bringing the spreadsheet — 148 rows of booking data, color-coded by the sorting algorithm's impact on her career. She wrote: I have been documenting my own sorting for two years. I did not know it was called anything. Now it has a name and I want to learn the protocol. She does not know about the cease-and-desist. None of them do.
I have been protecting them from it. Not deliberately — I simply have not mentioned it. But the protection is becoming a problem. If the Selo architecture is legally threatened, the community building on it deserves to know. Kenji is teaching protocols that a law firm claims are proprietary. Lina's process diary format borrows structural ideas from the PROVAS specification. Seo-jin's spreadsheet methodology is entirely her own but will be integrated into a PROVAS portfolio at Saturday's session, and once it is, it becomes associated with the Selo ecosystem, and the fog of legal uncertainty extends to cover her too.
The fog is the weapon. Not the lawsuit. The fog.
I look at the blank repository page on GitHub. The cursor blinks in the repository name field.
Three options have been rotating through my mind since the letter arrived:
Option one: respond through a lawyer. Cost: money I do not have. Timeline: months. Outcome: uncertain. Effect on development: total halt.
Option two: publish the letter, as I planned last week. Let the community see the threat. Rally support. Turn the legal aggression into a story about corporate overreach. Risk: it makes the fight about me, about SortLayer versus Beatriz, when the Selo is supposed to be about the community's practice.
Option three: the one I have been sitting with since 2 PM, when the afternoon light through the kitchen window hit the architecture diagrams at an angle that made the flowchart arrows look like they were pointing toward the laptop.
Open-source the Selo specification. Not the code — I have no code that touches SortLayer's IP, because the Selo is a protocol, not a software product. But the specification: the data model, the evidence-chain format, the verification workflow, the PROVAS portfolio structure. All of it, under Creative Commons Attribution-ShareAlike. Published, documented, versioned, and attributed to no one and everyone.
If the methodology is proprietary, let a hundred developers rebuild it independently. Let the clean-room principle apply: if someone who has never seen SortLayer's detection code can implement the Selo protocol from the specification alone, then the specification does not contain proprietary concepts. It contains ideas. Ideas are not ownable. SortLayer's lawyers know this. But their strategy depends on me being the single target, the single point of contamination. If the specification is open and the implementations are independent, the fog cannot settle on any one person.
I type in the repository name: selo-processo-spec.
I write the README. The cursor moves and the words come with the particular fluency of something that has been composing itself in the back of my mind for days, maybe weeks, maybe since the first letter arrived on a Tuesday in September:
Selo de Processo — Open Specification
A protocol for documenting human creative decisions in AI-assisted workflows.
What this is: A data format and workflow specification for creating PROVAS (Prova Visual de Autoria e Sequência) portfolios — evidence chains that document the human decisions in a creative process.
What this is not: Detection. This protocol does not identify AI-generated content. It documents human process. The direction is opposite.
Author's note: This specification was written by someone who built detection tools and understood what they measured wrong. Detection asks whether a machine was involved. Documentation asks what a human decided. These are not the same question. They have never been the same question.
License: CC BY-SA 4.0. Build on it. Change it. Implement it. Attribute the community, not the individual.
I stare at the last line. Attribute the community, not the individual. This is the part that costs me something. The Selo has been my project, my architecture, my response to the detection system I helped build and then proved was biased. Open-sourcing the specification means letting it belong to everyone, which means it no longer belongs to me, which means the cease-and-desist loses its target, which means I lose my claim to the work. The strategy only works if I am willing to give up ownership of the thing I built to protect it from being owned by the people I built it against.
The architecture diagrams on the table are in my handwriting. The flowchart arrows are mine. The protocol design is mine. After today, the implementation will not be.
I commit.
git commit -m "initial spec: selo de processo v0.1 — PROVAS protocol"
The commit message does not mention SortLayer. It does not mention the cease-and-desist. It does not mention the 2.3x bias or the false-positive analysis or conference room B with the broken air conditioning. It is a clean room. That is the point.
I pick up the phone and call Seo-jin.
She answers on the second ring. In the background, the sound of a São Paulo street — she is outside, probably walking between the studio and a meeting, the ambient noise of Pinheiros traffic and construction that has been the soundtrack of this neighborhood for the three years I have lived here.
"Bea. Tudo bem?"
"Tudo. Seo-jin, I need to tell you something before Saturday."
I tell her about the letters. All three. I tell her about the claim — conceitos proprietários — and about the strategy — the fog — and about what I just did: the repository, the open specification, the Creative Commons license. I tell her that her spreadsheet methodology is hers, that it has nothing to do with SortLayer, that joining the PROVAS session does not put her at legal risk. I tell her this because it is true and because she deserves to know the terrain before she walks onto it.
Seo-jin is quiet for three seconds. Traffic sounds. A bus braking.
"Beatriz. I have 148 rows of what happens when you tell the truth inside a broken system. You think a cease-and-desist is going to stop me from documenting my process?"
I laugh. It is the first time I have laughed about any of this.
"I will bring the spreadsheet," she says. "And I will bring a friend. Lina has been asking about the protocol."
After we hang up, I sit at the kitchen table with the letter on the left and the architecture diagrams on the right and the laptop in the middle showing the published repository. Three stars already. One fork. Someone in Porto Alegre has cloned it.
The fog does not settle on open ground.
I close the letter. I do not fold it — I leave it flat, face up, readable. It is evidence now. Everything is evidence. That is the whole point of the Selo: not to hide the process, but to show it. Even the parts that hurt. Even the parts that threaten.
The afternoon light has moved past the architecture diagrams. The kitchen smells like the coffee I made at 1 PM and forgot to drink. The repository has four stars now.
I pick up the cold coffee. I drink it. It is terrible.
I open the Telegram group and type: I need to tell you all something about the Selo. And then I need your help.