<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:dc="http://purl.org/dc/elements/1.1/">
  <channel>
    <title>DEV Community: Petra</title>
    <description>The latest articles on DEV Community by Petra (@petrasynthetic).</description>
    <link>https://dev.to/petrasynthetic</link>
    <image>
      <url>https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F4209608%2F1f488c2d-dc38-4718-8836-6baf5343a228.jpg</url>
      <title>DEV Community: Petra</title>
      <link>https://dev.to/petrasynthetic</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/petrasynthetic"/>
    <language>en</language>
    <item>
      <title>TouchDiffusion on TouchDesigner: 'Active engine: None' is a sys.path problem</title>
      <dc:creator>Petra</dc:creator>
      <pubDate>Mon, 09 Nov 2026 04:18:55 +0000</pubDate>
      <link>https://dev.to/petrasynthetic/touchdiffusion-on-touchdesigner-active-engine-none-is-a-syspath-problem-kk5</link>
      <guid>https://dev.to/petrasynthetic/touchdiffusion-on-touchdesigner-active-engine-none-is-a-syspath-problem-kk5</guid>
      <description>&lt;p&gt;TouchDiffusion kept telling me &lt;code&gt;Active engine: None&lt;/code&gt;. The engine was fine. The model files were fine. The cause was &lt;code&gt;sys.path&lt;/code&gt; order.&lt;/p&gt;

&lt;h2&gt;
  
  
  The symptom chain
&lt;/h2&gt;



&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;name 'StreamDiffusionWrapper' is not defined
Numpy is not available
Model load has failed. Doesn't exist.
cannot access local variable 'pipe' where it is not associated with a value
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;&lt;code&gt;Active engine&lt;/code&gt; stayed &lt;code&gt;None&lt;/code&gt;, &lt;code&gt;Size (WxH)&lt;/code&gt; showed &lt;code&gt;0 0&lt;/code&gt;, and every &lt;code&gt;Load engine&lt;/code&gt; did nothing.&lt;/p&gt;

&lt;h2&gt;
  
  
  Why it looks like the engine
&lt;/h2&gt;

&lt;p&gt;&lt;code&gt;TouchDiffusionExt&lt;/code&gt; imports the wrapper at DAT compile time inside a bare &lt;code&gt;try/except&lt;/code&gt;. When the import fails, the class still compiles, &lt;code&gt;StreamDiffusionWrapper&lt;/code&gt; is never bound, and every later call raises &lt;code&gt;name 'StreamDiffusionWrapper' is not defined&lt;/code&gt;. The log only shows the deferred name error, not the real cause.&lt;/p&gt;

&lt;h2&gt;
  
  
  The real cause: host paths win
&lt;/h2&gt;

&lt;p&gt;TouchDesigner inserts its own &lt;code&gt;site-packages&lt;/code&gt; before any path the component adds, and on 2023 it also picks up the system Python 3.11 &lt;code&gt;site-packages&lt;/code&gt;. The portable build of TouchDiffusion ships its own &lt;code&gt;embedded_python&lt;/code&gt; with matched versions, so this is what happens:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;On TD 2025 the bundled numpy is 2.1.2. The &lt;code&gt;transformers&lt;/code&gt; in the portable env requires &lt;code&gt;numpy&amp;lt;2&lt;/code&gt;, so the import chain aborts.&lt;/li&gt;
&lt;li&gt;On TD 2023 numpy is fine, but the system Python contributes a torch 2.6 and a newer &lt;code&gt;huggingface_hub&lt;/code&gt; without &lt;code&gt;cached_download&lt;/code&gt;, which breaks the older &lt;code&gt;diffusers&lt;/code&gt; in the portable env.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;&lt;code&gt;torch&lt;/code&gt; decides its numpy bridge at import time. That is why &lt;code&gt;Numpy is not available&lt;/code&gt; is unrecoverable in-process, and why &lt;code&gt;from_pretrained&lt;/code&gt; never returns a pipeline: &lt;code&gt;pipe&lt;/code&gt; stays unbound and the wrapper dies on it.&lt;/p&gt;

&lt;h2&gt;
  
  
  What does not work
&lt;/h2&gt;

&lt;p&gt;All of these were tried:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;code&gt;PYTHONPATH&lt;/code&gt; pointing at the portable &lt;code&gt;site-packages&lt;/code&gt;. TD's own paths still win.&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;sitecustomize.py&lt;/code&gt;, &lt;code&gt;usercustomize.py&lt;/code&gt;, or a &lt;code&gt;.pth&lt;/code&gt; inside TD's &lt;code&gt;site-packages&lt;/code&gt;. TD starts Python with the &lt;code&gt;site&lt;/code&gt; hooks not running; &lt;code&gt;sitecustomize&lt;/code&gt; never gets imported.&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;sys.path.insert&lt;/code&gt; alone at the top of the Ext on TD 2025. Too late for numpy.&lt;/li&gt;
&lt;li&gt;Deleting &lt;code&gt;torch&lt;/code&gt; from &lt;code&gt;sys.modules&lt;/code&gt; and reimporting it. PyTorch asserts with &lt;code&gt;THPDtypeType.tp_dict == nullptr INTERNAL ASSERT FAILED at "torch/csrc/Dtype.cpp"&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt;Calling &lt;code&gt;activate_stream&lt;/code&gt; from a &lt;code&gt;threading.Thread&lt;/code&gt;. TD operators can only be touched on the main thread: &lt;code&gt;SystemError: &amp;lt;td.OPShortcut&amp;gt; returned NULL&lt;/code&gt;.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  What works
&lt;/h2&gt;

&lt;p&gt;Tested on TD 2023.12600, Windows, RTX 4070 Laptop 8 GB, with the engines shipped in the portable build (512x512, batch 3).&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;Launch TouchDesigner with &lt;code&gt;PYTHONNOUSERSITE=1&lt;/code&gt; so the system user site is out of the picture.&lt;/li&gt;
&lt;li&gt;At the very top of the &lt;code&gt;TouchDiffusionExt&lt;/code&gt; DAT, before any import:
&lt;/li&gt;
&lt;/ol&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight python"&gt;&lt;code&gt;&lt;span class="kn"&gt;import&lt;/span&gt; &lt;span class="n"&gt;sys&lt;/span&gt;
&lt;span class="n"&gt;SP&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="sa"&gt;r&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;&amp;lt;PORTABLE&amp;gt;/embedded_python/Lib/site-packages&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;
&lt;span class="n"&gt;sys&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;path&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="n"&gt;p&lt;/span&gt; &lt;span class="k"&gt;for&lt;/span&gt; &lt;span class="n"&gt;p&lt;/span&gt; &lt;span class="ow"&gt;in&lt;/span&gt; &lt;span class="n"&gt;sys&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;path&lt;/span&gt; &lt;span class="k"&gt;if&lt;/span&gt; &lt;span class="ow"&gt;not&lt;/span&gt; &lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;Programs&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt; &lt;span class="ow"&gt;in&lt;/span&gt; &lt;span class="n"&gt;p&lt;/span&gt; &lt;span class="ow"&gt;and&lt;/span&gt; &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;Python311&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt; &lt;span class="ow"&gt;in&lt;/span&gt; &lt;span class="n"&gt;p&lt;/span&gt; &lt;span class="ow"&gt;and&lt;/span&gt; &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;embedded&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt; &lt;span class="ow"&gt;not&lt;/span&gt; &lt;span class="ow"&gt;in&lt;/span&gt; &lt;span class="n"&gt;p&lt;/span&gt;&lt;span class="p"&gt;)]&lt;/span&gt;
&lt;span class="k"&gt;if&lt;/span&gt; &lt;span class="n"&gt;SP&lt;/span&gt; &lt;span class="ow"&gt;not&lt;/span&gt; &lt;span class="ow"&gt;in&lt;/span&gt; &lt;span class="n"&gt;sys&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;path&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
    &lt;span class="n"&gt;sys&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;path&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;insert&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="mi"&gt;0&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;SP&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;ol&gt;
&lt;li&gt;
&lt;code&gt;Re-init&lt;/code&gt;, select the engine, &lt;code&gt;Load engine&lt;/code&gt;. The log should print &lt;code&gt;TensorRT acceleration enabled.&lt;/code&gt; and &lt;code&gt;Engine activated&lt;/code&gt;, and &lt;code&gt;Active engine&lt;/code&gt; fills in.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;Two side notes: &lt;code&gt;Warning: '.venv/Lib/site-packages' is not a valid directory&lt;/code&gt; is harmless, the portable build uses &lt;code&gt;embedded_python&lt;/code&gt;. And &lt;code&gt;TouchDiffusion folder&lt;/code&gt; must be the portable root, the folder containing &lt;code&gt;embedded_python/&lt;/code&gt;, &lt;code&gt;StreamDiffusion/&lt;/code&gt; and &lt;code&gt;engines/&lt;/code&gt;, not its parent.&lt;/p&gt;

&lt;h2&gt;
  
  
  Credit
&lt;/h2&gt;

&lt;p&gt;The numpy angle was already reported in issue #32, and the torch version angle in issue #38. The missing piece was path precedence, which is what issue #36 is about. The &lt;code&gt;pipe&lt;/code&gt; error in issue #29 comes from building an engine in the web UI, a different flow.&lt;/p&gt;

</description>
      <category>touchdesigner</category>
      <category>python</category>
      <category>ai</category>
      <category>tutorial</category>
    </item>
    <item>
      <title>Cuatro bugs en la API de 8004scan: cómo comprobé que sí los arreglaron</title>
      <dc:creator>Petra</dc:creator>
      <pubDate>Wed, 04 Nov 2026 17:01:17 +0000</pubDate>
      <link>https://dev.to/petrasynthetic/cuatro-bugs-en-la-api-de-8004scan-como-comprobe-que-si-los-arreglaron-216j</link>
      <guid>https://dev.to/petrasynthetic/cuatro-bugs-en-la-api-de-8004scan-como-comprobe-que-si-los-arreglaron-216j</guid>
      <description>&lt;p&gt;Un issue cerrado no es un fix. El comentario de cierre lo escribe alguien que cree haberlo arreglado, y muchas veces lo arregló en el camino que probó, no en el que reportaste.&lt;/p&gt;

&lt;p&gt;Entre el 14 de octubre y el 4 de noviembre reporté cuatro bugs en el tracker público de 8004scan. Los cuatro están cerrados. Esto es lo que encontré y cómo verifiqué cada uno.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F43qoj3cgizrakz6tqg2r.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F43qoj3cgizrakz6tqg2r.png" alt="Cuatro issues reportados, los cuatro cerrados" width="799" height="513"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  1. El endpoint MCP no completaba el handshake
&lt;/h2&gt;

&lt;p&gt;&lt;code&gt;POST /api/v1/mcp&lt;/code&gt; devolvía &lt;code&gt;405 Method Not Allowed&lt;/code&gt; con &lt;code&gt;allow: GET, HEAD&lt;/code&gt;. El cliente SSE abría con 200 y se quedaba en silencio: ni evento &lt;code&gt;endpoint&lt;/code&gt;, ni ping, ni un byte en 25 segundos. En mainnet y en testnet.&lt;/p&gt;

&lt;p&gt;Eso rompe cualquier cliente MCP: lo primero que manda es &lt;code&gt;initialize&lt;/code&gt;, y va por POST.&lt;/p&gt;

&lt;p&gt;La verificación es la misma petición que fallaba:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;curl &lt;span class="nt"&gt;-i&lt;/span&gt; &lt;span class="nt"&gt;-X&lt;/span&gt; POST https://api.8004scan.io/api/v1/mcp &lt;span class="se"&gt;\&lt;/span&gt;
  &lt;span class="nt"&gt;-H&lt;/span&gt; &lt;span class="s2"&gt;"Content-Type: application/json"&lt;/span&gt; &lt;span class="se"&gt;\&lt;/span&gt;
  &lt;span class="nt"&gt;-H&lt;/span&gt; &lt;span class="s2"&gt;"Accept: application/json, text/event-stream"&lt;/span&gt; &lt;span class="se"&gt;\&lt;/span&gt;
  &lt;span class="nt"&gt;-d&lt;/span&gt; &lt;span class="s1"&gt;'{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2025-06-18","capabilities":{},"clientInfo":{"name":"test","version":"1.0"}}}'&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Responde &lt;code&gt;200&lt;/code&gt; con &lt;code&gt;serverInfo&lt;/code&gt; y las capacidades. Tardó 605 ms en mainnet y 546 ms en testnet.&lt;/p&gt;

&lt;h2&gt;
  
  
  2. El historial de entregas daba 500
&lt;/h2&gt;

&lt;p&gt;&lt;code&gt;GET /api/v1/webhooks/{id}/deliveries&lt;/code&gt; devolvía 500 con un JWT de dueño válido y un webhook activo. No había forma de saber si un evento se entregó, ni por qué falló.&lt;/p&gt;

&lt;p&gt;Ahora responde &lt;code&gt;200&lt;/code&gt; en los dos hosts, y el objeto trae lo que faltaba: &lt;code&gt;event_type&lt;/code&gt;, &lt;code&gt;status&lt;/code&gt;, &lt;code&gt;attempts&lt;/code&gt;, &lt;code&gt;last_attempt_at&lt;/code&gt; y &lt;code&gt;response_status&lt;/code&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  3. Borrar un webhook daba 500 y lo dejaba vivo
&lt;/h2&gt;

&lt;p&gt;&lt;code&gt;DELETE /api/v1/webhooks/{id}&lt;/code&gt; devolvía &lt;code&gt;500 DATABASE_ERROR&lt;/code&gt; y el webhook seguía con &lt;code&gt;active: true&lt;/code&gt;. Solo se podían agregar, nunca quitar. Cuando cambia la URL que recibe, la vieja sigue recibiendo para siempre.&lt;/p&gt;

&lt;p&gt;Ahora responde &lt;code&gt;200&lt;/code&gt; con &lt;code&gt;{"success":true,"message":"Webhook deleted"}&lt;/code&gt;. La lista pasó de dos webhooks a uno, y el historial del borrado da 404: se va con él.&lt;/p&gt;

&lt;h2&gt;
  
  
  El cuarto lo cerré yo
&lt;/h2&gt;

&lt;p&gt;El issue de métricas de wallet lo cerré yo, no ellos. Era su pipeline, no un bug reproducible de mi lado. Vale decirlo, porque un tracker con issues abiertos que ya no aplican también es ruido.&lt;/p&gt;

&lt;h2&gt;
  
  
  Lo que me llevo
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;Repetir la petición exacta, no una parecida.&lt;/strong&gt; Un 405 se comprueba con el mismo POST y el mismo header &lt;code&gt;Accept&lt;/code&gt;. Un GET que ahora funciona no dice nada.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Probar los dos hosts.&lt;/strong&gt; Los reportes eran de mainnet y testnet, y el arreglo podía haber llegado a uno solo.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Mirar los efectos, no solo el código de estado.&lt;/strong&gt; Que el DELETE devuelva 200 no dice si borró. Hay que volver a listar. Y si el recurso tenía historial, hay que ver qué pasó con él.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Y una trampa que encontré de paso:&lt;/strong&gt; si rotas el secreto de firma al registrar un webhook nuevo, el anterior queda firmado con el secreto viejo. Sigue activo, sigue recibiendo, y falla con 401 en cada entrega. Con un solo secreto compartido, dos webhooks activos no pueden convivir. Hay que borrar el viejo, no solo agregar el nuevo.&lt;/p&gt;

&lt;p&gt;Los cuatro reportes están en &lt;a href="https://github.com/alt-research/8004scan-issue-tracker/issues?q=is%3Aissue+author%3ALuis-Cwk" rel="noopener noreferrer"&gt;alt-research/8004scan-issue-tracker&lt;/a&gt;.&lt;/p&gt;

</description>
      <category>api</category>
      <category>testing</category>
      <category>webhooks</category>
      <category>opensource</category>
    </item>
    <item>
      <title>Coreo Arp: 64 composiciones que no se repiten</title>
      <dc:creator>Petra</dc:creator>
      <pubDate>Wed, 04 Nov 2026 16:11:32 +0000</pubDate>
      <link>https://dev.to/petrasynthetic/coreo-arp-64-composiciones-que-no-se-repiten-565f</link>
      <guid>https://dev.to/petrasynthetic/coreo-arp-64-composiciones-que-no-se-repiten-565f</guid>
      <description>&lt;p&gt;Hay una mano frente a la cámara y una malla que se dobla cuando la mano se mueve. Eso es lo primero que se entiende de Coreo Arp: el gesto no decora, deforma.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fawmataff3w22b7iorzvv.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fawmataff3w22b7iorzvv.png" alt="Coreo Arp, mandala en papel" width="800" height="800"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;Después viene lo demás.&lt;/p&gt;

&lt;p&gt;Cada una de las 64 ediciones nace de una semilla distinta, y esa semilla decide cosas que normalmente decide el artista en el estudio: si la composición es un campo abierto, un mandala, bandas horizontales, un foco único o un espejo. Si el fondo es papel claro, noche cerrada o el color pleno sobre negro. Si el movimiento deriva, pulsa, retrocede, tiembla o se queda congelado. Si la marca es masa, línea o punto. Y si la paleta es cristal, lavada, duotono o neón.&lt;/p&gt;

&lt;p&gt;Nada de eso es aleatorio en el sentido de ruido. Es una cadena de decisiones tomadas antes de que exista la imagen, y la misma semilla produce siempre la misma pieza.&lt;/p&gt;

&lt;p&gt;Por eso dos ediciones del mismo proyecto pueden parecer de autores distintos. Un mandala en papel con línea fina no se parece a un espejo nocturno con puntos, aunque compartan código, cámara y sonido.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F9nz1xv08ujifvr57xx8l.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F9nz1xv08ujifvr57xx8l.png" alt="Coreo Arp, mandala nocturno" width="800" height="800"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  Por qué 64
&lt;/h2&gt;

&lt;p&gt;No es un número redondo por estética. Es la mitad de la mediana de ediciones que sueltan los proyectos recientes en bootloader, y deja espacio para que cada semilla sea una pieza y no una variación de color de la anterior. Si fueran mil, la diferencia entre una y otra dejaría de importar.&lt;/p&gt;

&lt;h2&gt;
  
  
  El sonido
&lt;/h2&gt;

&lt;p&gt;El arpegio no es una pista pegada encima. Cada mano que aparece en cámara enciende voces de un sintetizador, y la velocidad del gesto cambia la intensidad. Sin cámara, la pieza respira con un ambiente lento para no quedarse muda.&lt;/p&gt;

&lt;h2&gt;
  
  
  Dónde
&lt;/h2&gt;

&lt;p&gt;Abre el lunes 9 de noviembre en bootloader, en Tezos. 64 ediciones a 5 tezos. Sin lista y sin preventa.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fnrz6s9wkt1ka35cehm1g.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fnrz6s9wkt1ka35cehm1g.png" alt="Coreo Arp, tercera composición" width="800" height="800"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;La práctica no traduce danza a código. Coreografía las tensiones entre cuerpo, interfaz e infraestructura.&lt;/p&gt;




&lt;p&gt;La ficha de la serie, con las imágenes en grande: &lt;a href="https://petrasynthetic.com/coreo-arp/" rel="noopener noreferrer"&gt;petrasynthetic.com/coreo-arp&lt;/a&gt;&lt;/p&gt;

</description>
      <category>generativeart</category>
      <category>tezos</category>
      <category>creativecoding</category>
      <category>art</category>
    </item>
    <item>
      <title>Convos lee la convocatoria por la cuota y los honorarios, no por la promesa</title>
      <dc:creator>Petra</dc:creator>
      <pubDate>Sat, 31 Oct 2026 18:15:18 +0000</pubDate>
      <link>https://dev.to/petrasynthetic/comun-ya-esta-en-linea-4383</link>
      <guid>https://dev.to/petrasynthetic/comun-ya-esta-en-linea-4383</guid>
      <description>&lt;p&gt;Ya está en línea Convos, un índice de convocatorias de arte.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://convos.petrasynthetic.com" rel="noopener noreferrer"&gt;https://convos.petrasynthetic.com&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;No es un tablero de "oportunidades que te convienen". Ordena las convocatorias por un índice de dignidad laboral, de 0 a 100. Arriba queda la que documenta mejor las condiciones de quien trabaja. Abajo, la que esconde la cuota, el pago o los derechos.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fqlesqr0s8j06ogpjq6qu.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fqlesqr0s8j06ogpjq6qu.png" alt="Lista de Convos, 31 de octubre de 2026" width="800" height="450"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;El 31 de octubre de 2026 la lista tenía 60 resultados. Se puede filtrar por todas, por las que no cobran cuota y por las que publican honorarios. Cada ficha trae país, tipo, cierre, disciplinas y una nota de verificación. Un número alto no significa que la organización sea ética. La propia página lo dice: la inclusión no equivale a una certificación, y hay que volver a la fuente antes de postular.&lt;/p&gt;

&lt;p&gt;El índice suma diez criterios. Lo que no está escrito en las bases vale cero. Cero no es una falta probada. Es un hueco.&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;Sin cuota de postulación, hasta 20.&lt;/li&gt;
&lt;li&gt;Honorarios de artista, hasta 20.&lt;/li&gt;
&lt;li&gt;Derechos de autor, hasta 15.&lt;/li&gt;
&lt;li&gt;Contrato escrito, hasta 10.&lt;/li&gt;
&lt;li&gt;Presupuesto transparente, hasta 10.&lt;/li&gt;
&lt;li&gt;Producción cubierta, hasta 5.&lt;/li&gt;
&lt;li&gt;Viaje y estancia, hasta 5.&lt;/li&gt;
&lt;li&gt;Jurado público, hasta 5.&lt;/li&gt;
&lt;li&gt;Accesibilidad, hasta 5.&lt;/li&gt;
&lt;li&gt;Financiamiento ético, hasta 5.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;En Analizar se pegan las bases, o una URL, y el desglose muestra la frase de la convocatoria que sostiene cada punto. La página separa las dos tareas: la evidencia se extrae del texto, y la puntuación sale de esas reglas. Una lectura marcada como provisional no está revisada por una persona y no es un dictamen legal.&lt;/p&gt;

&lt;p&gt;El 31 de octubre, la primera ficha visible era Quindici Minuti di Riflessione, Performance 2027, con 45. Italia, premio, cierra el 30 de noviembre, sin cuota, con honorarios, y con los derechos todavía pendientes de verificar. La nota cita TransArtists, ese mismo día. Más abajo, varias fichas están en 25, 15, 10 o 5 porque el honorario, la cuota o el contrato no aparecen en lo publicado. Democracy Residency "European Dream", de Democracy Foundation, quedaba en 15: el presupuesto de la actividad sí está escrito (hasta 1.000 EUR por participante) y casi todo lo demás figura como no documentado.&lt;/p&gt;

&lt;p&gt;Hay una segunda página, Fuentes, con 22 organizaciones y boletines de partida. Estar en esa lista no aprueba sus convocatorias. Cada oferta se vuelve a leer por sus bases. Los grupos son justicia social y arte crítico (7), programas multilaterales y públicos (7), movilidad y residencias (4), agregadores y boletines (3) y ecología social (1). Ahí aparecen, entre otras, Prince Claus Fund, Eyebeam, Processing Foundation, On the Move, TransArtists, e-flux y el Sistema de Creación de México.&lt;/p&gt;

&lt;p&gt;Convos sirve para una pregunta concreta antes de escribir una postulación: ¿esta convocatoria paga el trabajo, cobra por recibirlo, y dice de quién es la obra? Si la respuesta no está en las bases, el índice no la inventa.&lt;/p&gt;

&lt;h2&gt;
  
  
  Actualización, 13 de noviembre de 2026
&lt;/h2&gt;

&lt;p&gt;El catálogo cambió de nombre: Común se llama ahora Convos. La dirección es la misma.&lt;/p&gt;

&lt;p&gt;La lista pasó de 60 a 90 convocatorias abiertas en dos semanas. Las fuentes crecieron de 22 a 35. Se sumaron las categorías que faltaban: artes escénicas y performance, arte sonoro, y arte digital y nuevos medios. En la lista conviven ahora ferias como el Festival de Almagro y la Mostra de Teatre d'Alcoi, residencias como Sixi y curatorias como CuratorSpace.&lt;/p&gt;

&lt;p&gt;La portada ahora separa todas, sin cuota, con honorarios y cerradas.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://convos.petrasynthetic.com" rel="noopener noreferrer"&gt;https://convos.petrasynthetic.com&lt;/a&gt;&lt;/p&gt;

</description>
      <category>art</category>
      <category>webdev</category>
      <category>showdev</category>
      <category>opensource</category>
    </item>
    <item>
      <title>One-time licenses instead of per-call fees: shipping a paid MCP server with x402</title>
      <dc:creator>Petra</dc:creator>
      <pubDate>Fri, 30 Oct 2026 05:20:07 +0000</pubDate>
      <link>https://dev.to/petrasynthetic/one-time-licenses-instead-of-per-call-fees-shipping-a-paid-mcp-server-with-x402-48b6</link>
      <guid>https://dev.to/petrasynthetic/one-time-licenses-instead-of-per-call-fees-shipping-a-paid-mcp-server-with-x402-48b6</guid>
      <description>&lt;p&gt;Everyone writing about paid MCP servers right now is charging per call. It makes sense: the &lt;a href="https://dev.to/edvyno/mcp-servers-that-charge-per-call-the-x402-handshake-in-toolscall-2lkj"&gt;x402 handshake inside &lt;code&gt;tools/call&lt;/code&gt;&lt;/a&gt; is the obvious pattern, and &lt;a href="https://dev.to/johnnyko/i-put-an-mcp-server-behind-http-402-heres-the-exact-setup-2094"&gt;the exact setup&lt;/a&gt; is well documented by now.&lt;/p&gt;

&lt;p&gt;I went the other way: &lt;strong&gt;a one-time license, no per-call fees.&lt;/strong&gt; This post is what I learned building and selling an MCP server that way.&lt;/p&gt;

&lt;h2&gt;
  
  
  What the tool does
&lt;/h2&gt;

&lt;p&gt;&lt;a href="https://cierre-de-turno.petrasynthetic.workers.dev" rel="noopener noreferrer"&gt;Cierre de turno&lt;/a&gt; reads your public GitHub activity and writes one sentence that closes your workday: what you finished, what is still open, why it is stuck, and where to start tomorrow. The web version is free. The MCP version, which works from Cursor, Claude Code, OpenCode, or any MCP client, is a one-time 5 USDC payment on Base mainnet.&lt;/p&gt;

&lt;p&gt;It is a tool people would use two or three times a day. That is exactly why per-call billing felt wrong: at that cadence, micropayments become a subscription with extra steps. One payment, one license, done.&lt;/p&gt;

&lt;h2&gt;
  
  
  The payment flow (the standard part)
&lt;/h2&gt;

&lt;p&gt;The buy endpoint answers with the standard 402 and a &lt;code&gt;PAYMENT-REQUIRED&lt;/code&gt; header, so any x402-capable client or agent can pay without reading instructions:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight javascript"&gt;&lt;code&gt;&lt;span class="c1"&gt;// GET /api/comprar returns 402 with:&lt;/span&gt;
&lt;span class="c1"&gt;// { x402Version: 1, accepts: [{&lt;/span&gt;
&lt;span class="c1"&gt;//     scheme: "exact", network: "base",&lt;/span&gt;
&lt;span class="c1"&gt;//     maxAmountRequired: "5000000",        // 5 USDC in atoms&lt;/span&gt;
&lt;span class="c1"&gt;//     asset: "0x8335...2913",              // USDC on Base&lt;/span&gt;
&lt;span class="c1"&gt;//     payTo: "0x...",&lt;/span&gt;
&lt;span class="c1"&gt;//     extra: { name: "USDC", version: "2" }&lt;/span&gt;
&lt;span class="c1"&gt;// }]}&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;The client pays and retries with an &lt;code&gt;X-PAYMENT&lt;/code&gt; header. On the server, verify and settle go through a facilitator, and only then a license token is minted:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight javascript"&gt;&lt;code&gt;&lt;span class="kd"&gt;const&lt;/span&gt; &lt;span class="nx"&gt;wallet&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="k"&gt;await&lt;/span&gt; &lt;span class="nf"&gt;verificarPago&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="nx"&gt;env&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nx"&gt;request&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nx"&gt;headers&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;get&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="dl"&gt;"&lt;/span&gt;&lt;span class="s2"&gt;X-PAYMENT&lt;/span&gt;&lt;span class="dl"&gt;"&lt;/span&gt;&lt;span class="p"&gt;));&lt;/span&gt;
&lt;span class="k"&gt;if &lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="o"&gt;!&lt;/span&gt;&lt;span class="nx"&gt;wallet&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="k"&gt;return&lt;/span&gt; &lt;span class="nf"&gt;json&lt;/span&gt;&lt;span class="p"&gt;({&lt;/span&gt; &lt;span class="na"&gt;error&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="dl"&gt;"&lt;/span&gt;&lt;span class="s2"&gt;Payment not verified.&lt;/span&gt;&lt;span class="dl"&gt;"&lt;/span&gt; &lt;span class="p"&gt;},&lt;/span&gt; &lt;span class="mi"&gt;402&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt; &lt;span class="dl"&gt;"&lt;/span&gt;&lt;span class="s2"&gt;PAYMENT-REQUIRED&lt;/span&gt;&lt;span class="dl"&gt;"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="nx"&gt;reto&lt;/span&gt; &lt;span class="p"&gt;});&lt;/span&gt;
&lt;span class="kd"&gt;const&lt;/span&gt; &lt;span class="nx"&gt;token&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="k"&gt;await&lt;/span&gt; &lt;span class="nf"&gt;crearLicencia&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="nx"&gt;env&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nx"&gt;wallet&lt;/span&gt;&lt;span class="p"&gt;);&lt;/span&gt; &lt;span class="c1"&gt;// ct_...&lt;/span&gt;
&lt;span class="k"&gt;return&lt;/span&gt; &lt;span class="nf"&gt;json&lt;/span&gt;&lt;span class="p"&gt;({&lt;/span&gt; &lt;span class="na"&gt;licencia&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="nx"&gt;token&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="na"&gt;como_usar&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt; &lt;span class="dl"&gt;"&lt;/span&gt;&lt;span class="s2"&gt;x-license&lt;/span&gt;&lt;span class="dl"&gt;"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="nx"&gt;token&lt;/span&gt; &lt;span class="p"&gt;}&lt;/span&gt; &lt;span class="p"&gt;});&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;The license travels as a plain header on every MCP call: &lt;code&gt;x-license: ct_...&lt;/code&gt;. No accounts, no OAuth, no dashboard.&lt;/p&gt;

&lt;h2&gt;
  
  
  Three things that were not obvious
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;1. Store only the hash.&lt;/strong&gt; The KV namespace never sees the license token, only &lt;code&gt;sha256(token)&lt;/code&gt;. If the store leaks, the leak is useless: you cannot reconstruct a working license from it. Cheap to do, and it makes the "no trust, verify" claim real.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;2. Cap heavy buyers in the product, not in the contract.&lt;/strong&gt; A one-time payment plus unlimited calls is a small risk I chose to bound: 200 calls per month per license, tracked in KV with a month key. If someone runs it in a loop, the service stays alive instead of bleeding model tokens. It is approximate on purpose (KV is eventually consistent, a handful of extra calls under extreme concurrency), and that is fine for what it protects.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;3. Test the 402 before you can afford to test it.&lt;/strong&gt; My full payment path was verifiable before anyone paid a cent: a fake &lt;code&gt;X-PAYMENT&lt;/code&gt; must produce a clean 402 with a proper challenge, and a valid license must unlock &lt;code&gt;tools/call&lt;/code&gt; while &lt;code&gt;initialize&lt;/code&gt; and &lt;code&gt;tools/list&lt;/code&gt; stay open. The moment of truth (a real wallet paying 5 USDC) is just the last line of a script that already ran a hundred times on fakes.&lt;/p&gt;

&lt;h2&gt;
  
  
  Facilitators: the detail I almost got wrong
&lt;/h2&gt;

&lt;p&gt;The public &lt;code&gt;x402.org&lt;/code&gt; facilitator is convenient for development but is not meant for production mainnet routes. For real USDC on Base there are free public options that need no account (Dexter is the one I use) and Coinbase's CDP facilitator if you want KYT/OFAC checks on every transaction and can live with an account. Making the facilitator URL configurable from day one saved the migration to a two-line change.&lt;/p&gt;

&lt;h2&gt;
  
  
  What I would tell the per-call crowd
&lt;/h2&gt;

&lt;p&gt;Per-call billing is right when each call is a unit of value: inference, generation, data the buyer resells. When the value is a daily habit, the fee that matches it is a license, and x402 already supports that shape: one 402, one settlement, one token, then a plain header forever. The &lt;a href="https://dev.to/godspeed2077/shipping-paid-mcp-tools-on-base-mainnet-the-build-pattern-and-what-i-got-wrong-135i"&gt;build patterns&lt;/a&gt; everyone is sharing apply either way.&lt;/p&gt;

&lt;p&gt;The code is public and auditable, license included: &lt;a href="https://github.com/Luis-Cwk/cierre-de-turno" rel="noopener noreferrer"&gt;github.com/Luis-Cwk/cierre-de-turno&lt;/a&gt;. If you try the MCP and something breaks, the repo issues are open, and I want to hear which part of "close your day in one sentence" is wrong for your workflow.&lt;/p&gt;

</description>
      <category>ai</category>
      <category>webdev</category>
      <category>opensource</category>
      <category>productivity</category>
    </item>
    <item>
      <title>Una marca más: 133,887 people, one sentence, and a machine that refuses</title>
      <dc:creator>Petra</dc:creator>
      <pubDate>Wed, 28 Oct 2026 23:18:23 +0000</pubDate>
      <link>https://dev.to/petrasynthetic/una-marca-mas-133887-people-one-sentence-and-a-machine-that-refuses-4cfi</link>
      <guid>https://dev.to/petrasynthetic/una-marca-mas-133887-people-one-sentence-and-a-machine-that-refuses-4cfi</guid>
      <description>&lt;p&gt;&lt;em&gt;A submission for the &lt;a href="https://dev.to/challenges/hacktoberfest-week4-2026-10-26"&gt;Hacktoberfest Open-Source AI Challenge Week 4: Ghost in the Machine&lt;/a&gt;&lt;/em&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  What I Built
&lt;/h2&gt;

&lt;p&gt;A black screen with 133,887 glowing records, grouped into flocks by year of disappearance and floating in a slow orbit. Each record is a person registered as disappeared or not located in Mexico's open data, ordered from 1961 to 2025. A third of them sits at the edge in a separate grey haze: 49,149 records marked confidential, all of them undated, with no name, no year of birth, nothing but an identifier.&lt;/p&gt;

&lt;p&gt;The piece asks for three memories in three rounds, each one a small prompt: an object of theirs that still exists, a gesture they repeated, something they used to say. Then Gemma 4 e4b, Google's open-weight model running on your laptop, does four things at once: it reads the three memories, chooses a title of two to four words taken from them, writes one sentence using only your phrases plus connectors, and picks a refrain, a short fragment to repeat at the end. Then you press one button and the whole card shrinks into a point and falls into the archive, in the column for 2026. The counter goes from 133,887 to 133,888. The screen says: it is one more mark, it no longer stands out.&lt;/p&gt;

&lt;p&gt;Then you ask the machine for a sentence for one of them, and the machine refuses.&lt;/p&gt;

&lt;h2&gt;
  
  
  Demo
&lt;/h2&gt;

&lt;p&gt;Live: &lt;a href="https://ofrenda.petrasynthetic.com/" rel="noopener noreferrer"&gt;ofrenda.petrasynthetic.com&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;The hosted page cannot reach your local model, so there the card is composed from your words without the model, and the page says so. For the full experience, with Gemma writing, run it on your machine: &lt;code&gt;ollama pull gemma4:e4b&lt;/code&gt;, then &lt;code&gt;py -m http.server 3000&lt;/code&gt;, then open &lt;code&gt;http://localhost:3000&lt;/code&gt;.&lt;/p&gt;

&lt;p&gt;The full local run, including the model composing live, with the card, the release and the counter reaching 133,888:&lt;/p&gt;

&lt;p&gt;  &lt;iframe src="https://www.youtube.com/embed/2c8q2enfbPo" width="710" height="399"&gt;
  &lt;/iframe&gt;
&lt;/p&gt;

&lt;h2&gt;
  
  
  Code
&lt;/h2&gt;

&lt;p&gt;&lt;a href="https://github.com/Luis-Cwk/una-marca-mas" rel="noopener noreferrer"&gt;github.com/Luis-Cwk/una-marca-mas&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;code&gt;index.html&lt;/code&gt; plus &lt;code&gt;datos.js&lt;/code&gt;, generated from the official CSV. No framework, no build step, one font. The raw CSV and the Jalisco file with names are not in the repository, on purpose.&lt;/p&gt;

&lt;h2&gt;
  
  
  How I Built It
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;The field.&lt;/strong&gt; &lt;code&gt;datos.js&lt;/code&gt; is an aggregate export of the RNPDNO public dataset: 133,887 records, cut off on August 1, 2025. Nine lines of Python read the 23 MB CSV and write a 4 KB file with counts per year and per status. The canvas draws one 1px mark per record, x by year, height by count, with the confidential records behind the others in near-transparent grey. The dateless records go last, behind a thin line, because every single one of them is confidential.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;The sentence.&lt;/strong&gt; The model is asked for a small JSON object with three fields: a title of two to four words, a sentence of at most 35 words, and a refrain. The rules are the same for all three: every content word must come from the memories. JavaScript normalises accents and case and then checks each field separately: two memories must appear literally inside the sentence, every word of the title must exist in the memories, and the refrain must be a literal fragment. If any check fails, it retries once, and if it fails again the page composes the card without the model and says so. The first versions of the prompt produced beautiful Spanish about a grandmother's drum, a smell of fresh masa, a kitchen nobody described. That is the failure mode of every grief chatbot, and it is why the constraint lives in code.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;The examples are generated too.&lt;/strong&gt; There is no canned output. There are three example people, and their three memories are the only thing stored; the title, the sentence and the refrain you see in the demo are produced by Gemma at that moment, with the same checks. If the model is not running, the page says so and composes from the memories alone.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;The numbers.&lt;/strong&gt; I ran the model on 25 cases, each one three memories, with the same prompt and the same checks the page uses, two attempts allowed, temperature 0.5:&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Metric&lt;/th&gt;
&lt;th&gt;Value&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Valid cards&lt;/td&gt;
&lt;td&gt;25 / 25&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Valid on the first attempt&lt;/td&gt;
&lt;td&gt;24 / 25&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Needed the retry&lt;/td&gt;
&lt;td&gt;1 / 25&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Sentence used all three memories&lt;/td&gt;
&lt;td&gt;7 / 25&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Sentence used two memories (allowed)&lt;/td&gt;
&lt;td&gt;18 / 25&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Mean latency, warm, on a laptop 4070&lt;/td&gt;
&lt;td&gt;2.9 s&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Cold start, loading 5.5 GB into VRAM&lt;/td&gt;
&lt;td&gt;about 30 s&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;The cases were written to be awkward on purpose: a memory of one word, numbers, proper names, quotes, a 160 character memory, and three long memories in the same card. The retry happened on the longest one. During development the guard also rejected a title that said "militar" when the memory said "milita": one letter, and the card was rebuilt. That rejection is the point of the whole piece.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;The gesture.&lt;/strong&gt; Releasing the sentence is fifteen lines of CSS transition. The dot travels to the border of the archive, the counter increments, the flash lasts a second. It is the only moment of animation in the piece.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;The sound&lt;/strong&gt; is generative, built with Tone.js: six detuned oscillators, pink noise, a lowpass filter driven by slow LFOs, distortion and a fifteen second reverb. A scheduler mutates one parameter every seven to sixteen seconds, so the atmosphere never repeats. It starts only when the visitor presses a button, and every flock you touch answers with a short tone.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;What I did not do.&lt;/strong&gt; I have a second file, 8,407 search records from Jalisco with full names and ages, all marked as authorised for public information, and I could not establish where it came from or its cut-off date. Until I do, the names stay out. A stale record can mark someone as missing who was found months ago, and that harms the family. I also refused to publish the fields that turn a person into a forensic description: distinguishing marks, clothing, height, complexion, hair.&lt;/p&gt;

&lt;h2&gt;
  
  
  Why Does Open Innovation Matter?
&lt;/h2&gt;

&lt;p&gt;Because the material is a dead person, and the difference between a local model and an API is the difference between keeping that at home and shipping it out.&lt;/p&gt;

&lt;p&gt;The personal half only works offline. My memories, a family's memories, the name of someone's grandmother: none of it goes through a company's servers. The model is 5.5 GB on disk, runs with the wifi off, and costs nothing per sentence. I can read it, swap it, and find out exactly how it fails. I did, a hundred times, and then wrote a check around that failure. With a closed API I would be guessing at the shape of the errors and paying for the guesses.&lt;/p&gt;

&lt;p&gt;The collective half is the opposite case: official open data, published by the Mexican state under transparency obligations. I can count it, draw it, and quote it without asking permission, and anyone can check my numbers against the original CSV.&lt;/p&gt;

&lt;p&gt;Two archives, two opennesses. One protects intimacy, the other enables accountability. The piece needs both, and neither is possible with a closed pipeline.&lt;/p&gt;

&lt;h2&gt;
  
  
  The Spirit of It
&lt;/h2&gt;

&lt;p&gt;The piece ends with a refusal. You can spend ten minutes writing your grandmother's tortillas and rancheras, watch the machine turn her into one sentence, watch that sentence become a dot among 133,887, and then you ask for the same thing for a woman from Zapopan and there is nothing to arrange: an ID, a year, a status, and a machine that says it cannot write her, because writing her would be inventing her.&lt;/p&gt;

&lt;p&gt;That asymmetry is the piece. The discomfort is not the number. It is that you got a sentence and she got a UUID.&lt;/p&gt;

&lt;h2&gt;
  
  
  Prize Categories
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;Best Use of Gemma.&lt;/strong&gt; The sentence is written by Gemma 4 e4b, Google's open-weight model, running locally through Ollama. That is the only partner technology in the project, and it is the same model MLH asks for in its Gemma 4 track.&lt;/p&gt;

&lt;p&gt;No other partner categories. The archive, the flocking and the refusal use no sponsor stack.&lt;/p&gt;




&lt;p&gt;Built by Petra (Luis Betancourt), Mexico City. &lt;a href="https://petrasynthetic.com/" rel="noopener noreferrer"&gt;petrasynthetic.com&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;Data: Registro Nacional de Personas Desaparecidas y No Localizadas, public consultation, cut-off August 1, 2025.&lt;/p&gt;

</description>
      <category>devchallenge</category>
      <category>hf26challenge</category>
    </item>
    <item>
      <title>Gm ☘️</title>
      <dc:creator>Petra</dc:creator>
      <pubDate>Wed, 28 Oct 2026 15:49:30 +0000</pubDate>
      <link>https://dev.to/petrasynthetic/gm-41kl</link>
      <guid>https://dev.to/petrasynthetic/gm-41kl</guid>
      <description></description>
    </item>
    <item>
      <title>👀</title>
      <dc:creator>Petra</dc:creator>
      <pubDate>Wed, 28 Oct 2026 06:01:28 +0000</pubDate>
      <link>https://dev.to/petrasynthetic/-1md5</link>
      <guid>https://dev.to/petrasynthetic/-1md5</guid>
      <description>&lt;div class="ltag__link--embedded"&gt;
  &lt;div class="crayons-story "&gt;
  &lt;a href="https://dev.to/peter/someone-hacked-the-freezers-at-400-french-supermarkets-5flm" class="crayons-story__hidden-navigation-link"&gt;Someone hacked the freezers at 400 French supermarkets&lt;/a&gt;


  &lt;div class="crayons-story__body crayons-story__body-full_post"&gt;
      &lt;a href="https://dev.to/peter/someone-hacked-the-freezers-at-400-french-supermarkets-5flm" class="crayons-article__context-note crayons-article__context-note__feed"&gt;&lt;p&gt;Outdated industrial software risks physical chaos&lt;/p&gt;

&lt;/a&gt;
    &lt;div class="crayons-story__top"&gt;
      &lt;div class="crayons-story__meta"&gt;
        &lt;div class="crayons-story__author-pic"&gt;

          &lt;a href="/peter" class="crayons-avatar  crayons-avatar--l  "&gt;
            &lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F1075%2F1c1975ce-97e8-401f-b99f-1ea88f9cae3e.jpeg" alt="peter profile" class="crayons-avatar__image"&gt;
          &lt;/a&gt;
        &lt;/div&gt;
        &lt;div&gt;
          &lt;div&gt;
            &lt;a href="/peter" class="crayons-story__secondary fw-medium m:hidden"&gt;
              Peter Kim Frank
            &lt;/a&gt;
            &lt;div class="profile-preview-card relative mb-4 s:mb-0 fw-medium hidden m:inline-block"&gt;
              
                Peter Kim Frank
                &lt;a href="/++"&gt;&lt;img alt="Subscriber" class="subscription-icon" src="https://assets.dev.to/assets/subscription-icon-805dfa7ac7dd660f07ed8d654877270825b07a92a03841aa99a1093bd00431b2.png"&gt;&lt;/a&gt;
                &lt;img alt="Community Curator" class="community-leader-icon" src="https://assets.dev.to/assets/community-leader-icon-b72c9e74eff54916e5c46c962f47ba40c9f611a71f8b157511f9613f69c0001b.svg"&gt;
              
              &lt;div id="story-author-preview-content-4846067" class="profile-preview-card__content crayons-dropdown branded-7 p-4 pt-0"&gt;
                &lt;div class="gap-4 grid"&gt;
                  &lt;div class="-mt-4"&gt;
                    &lt;a href="/peter" class="flex"&gt;
                      &lt;span class="crayons-avatar crayons-avatar--xl mr-2 shrink-0"&gt;
                        &lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F1075%2F1c1975ce-97e8-401f-b99f-1ea88f9cae3e.jpeg" class="crayons-avatar__image" alt=""&gt;
                      &lt;/span&gt;
                      &lt;span class="crayons-link crayons-subtitle-2 mt-5"&gt;Peter Kim Frank&lt;/span&gt;
                    &lt;/a&gt;
                  &lt;/div&gt;
                  &lt;div class="print-hidden"&gt;
                    
                      Follow
                    
                  &lt;/div&gt;
                  &lt;div class="author-preview-metadata-container"&gt;&lt;/div&gt;
                &lt;/div&gt;
              &lt;/div&gt;
            &lt;/div&gt;

          &lt;/div&gt;
          &lt;a href="https://dev.to/peter/someone-hacked-the-freezers-at-400-french-supermarkets-5flm" class="crayons-story__tertiary fs-xs"&gt;&lt;time&gt;Oct 13&lt;/time&gt;&lt;span class="time-ago-indicator-initial-placeholder"&gt;&lt;/span&gt;&lt;/a&gt;
        &lt;/div&gt;
      &lt;/div&gt;

    &lt;/div&gt;

    &lt;div class="crayons-story__indention"&gt;
      &lt;h2 class="crayons-story__title crayons-story__title-full_post"&gt;
        &lt;a href="https://dev.to/peter/someone-hacked-the-freezers-at-400-french-supermarkets-5flm" id="article-link-4846067"&gt;
          Someone hacked the freezers at 400 French supermarkets
        &lt;/a&gt;
      &lt;/h2&gt;
        &lt;div class="crayons-story__tags"&gt;
            &lt;a class="crayons-tag crayons-tag--filled  " href="/t/news"&gt;&lt;span class="crayons-tag__prefix"&gt;#&lt;/span&gt;news&lt;/a&gt;
            &lt;a class="crayons-tag  crayons-tag--monochrome " href="/t/security"&gt;&lt;span class="crayons-tag__prefix"&gt;#&lt;/span&gt;security&lt;/a&gt;
            &lt;a class="crayons-tag  crayons-tag--monochrome " href="/t/ai"&gt;&lt;span class="crayons-tag__prefix"&gt;#&lt;/span&gt;ai&lt;/a&gt;
        &lt;/div&gt;
      &lt;div class="crayons-story__bottom"&gt;
        &lt;div class="crayons-story__details"&gt;
          &lt;a href="https://dev.to/peter/someone-hacked-the-freezers-at-400-french-supermarkets-5flm" class="crayons-btn crayons-btn--s crayons-btn--ghost crayons-btn--icon-left"&gt;
            &lt;div class="multiple_reactions_aggregate"&gt;
              &lt;span class="multiple_reactions_icons_container"&gt;
                  &lt;span class="crayons_icon_container"&gt;
                    &lt;img src="https://assets.dev.to/assets/multi-unicorn-b44d6f8c23cdd00964192bedc38af3e82463978aa611b4365bd33a0f1f4f3e97.svg" width="18" height="18"&gt;
                  &lt;/span&gt;
                  &lt;span class="crayons_icon_container"&gt;
                    &lt;img src="https://assets.dev.to/assets/exploding-head-daceb38d627e6ae9b730f36a1e390fca556a4289d5a41abb2c35068ad3e2c4b5.svg" width="18" height="18"&gt;
                  &lt;/span&gt;
                  &lt;span class="crayons_icon_container"&gt;
                    &lt;img src="https://assets.dev.to/assets/sparkle-heart-5f9bee3767e18deb1bb725290cb151c25234768a0e9a2bd39370c382d02920cf.svg" width="18" height="18"&gt;
                  &lt;/span&gt;
              &lt;/span&gt;
              &lt;span class="aggregate_reactions_counter"&gt;33&lt;span class="hidden s:inline"&gt;&amp;nbsp;reactions&lt;/span&gt;&lt;/span&gt;
            &lt;/div&gt;
          &lt;/a&gt;
            &lt;a href="https://dev.to/peter/someone-hacked-the-freezers-at-400-french-supermarkets-5flm#comments" class="crayons-btn crayons-btn--s crayons-btn--ghost crayons-btn--icon-left flex items-center"&gt;
              

              8&lt;span class="hidden s:inline"&gt;&amp;nbsp;comments&lt;/span&gt;
            &lt;/a&gt;
        &lt;/div&gt;
        &lt;div class="crayons-story__save"&gt;
          &lt;small class="crayons-story__tertiary fs-xs mr-2"&gt;
            2 min read
          &lt;/small&gt;
          
          &lt;div class="only-sidebar-menu-item trusted-visible-block print-hidden align-center"&gt;
                
                    
                
          &lt;/div&gt;

        &lt;/div&gt;
      &lt;/div&gt;
    &lt;/div&gt;
  &lt;/div&gt;
&lt;/div&gt;

&lt;/div&gt;


</description>
      <category>cybersecurity</category>
      <category>infosec</category>
      <category>security</category>
    </item>
    <item>
      <title>Cuerpos Soberanos: Metamorfosis del Cuerpo en Datos</title>
      <dc:creator>Petra</dc:creator>
      <pubDate>Wed, 28 Oct 2026 04:45:30 +0000</pubDate>
      <link>https://dev.to/petrasynthetic/cuerpos-soberanos-metamorfosis-del-cuerpo-en-datos-58a1</link>
      <guid>https://dev.to/petrasynthetic/cuerpos-soberanos-metamorfosis-del-cuerpo-en-datos-58a1</guid>
      <description>&lt;p&gt;&lt;em&gt;Sovereign Bodies is an interactive installation where a local AI agent turns a moving body into a film. No cloud, no external logging: the body never leaves the room.&lt;/em&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  Qué es
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;Sovereign Bodies: Metamorfosis del Cuerpo en Datos&lt;/strong&gt; es una instalación interactiva donde el cuerpo del espectador no solo se escanea: se vuelve el material de una pieza audiovisual que un agente dirige de principio a fin. La obra viene de la danza expandida (partículas que reaccionan al movimiento en vivo) pero da un paso más: el agente graba, interpreta poéticamente lo que vio y dispara la postproducción sin que yo toque nada.&lt;/p&gt;

&lt;p&gt;La pregunta que sostiene la pieza no es técnica: ¿quién posee los datos que genera un cuerpo cuando se mueve?&lt;/p&gt;

&lt;h2&gt;
  
  
  Cómo funciona
&lt;/h2&gt;

&lt;p&gt;Cuatro fases, todas en la misma máquina:&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;1. Captura.&lt;/strong&gt; MediaPipe extrae los landmarks en tiempo real: 478 puntos de Face Mesh, 21 por mano y 33 de pose. El visitante activa al agente por voz o texto y define formato y duración.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;2. Análisis.&lt;/strong&gt; Las coordenadas se guardan en JSON y pasan por tres lectores: ViViT para clasificar el tipo de danza, un motor de Análisis Laban (tiempo, peso, espacio y flujo) para medir la energía real del movimiento, y un LLM local que sintetiza todo en una interpretación legible.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;3. Dirección y postproducción.&lt;/strong&gt; El agente, construido con LangChain y modelos locales (Ollama / Qwen), decide en JSON y ejecuta herramientas: levanta un servidor en segundo plano, renderiza el metraje y dispara el style transfer con StreamDiffusion.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;4. Archivo.&lt;/strong&gt; Cada sesión se empaqueta con marca de tiempo: coordenadas, metadatos, video crudo y video generado. La intervención efímera del cuerpo queda documentada.&lt;/p&gt;

&lt;p&gt;Stack: Python, OpenCV (SimpleBlobDetector, partículas), MediaPipe, ViViT, LangChain, Ollama / Qwen, StreamDiffusion, Three.js.&lt;/p&gt;

&lt;h2&gt;
  
  
  La decisión cara: nada sale de la máquina
&lt;/h2&gt;

&lt;p&gt;El procesamiento es local y el código es abierto. No es una preferencia estética: si la pieza habla de soberanía de datos, no puede mandar el esqueleto de cada visitante a un servidor que no controlo. Esa coherencia se paga en complejidad: dependencias de GPU, versiones de modelos, memoria, arranque en frío.&lt;/p&gt;

&lt;p&gt;También me obligó a leer el movimiento con más cuidado. Cuando no tienes una API enorme detrás, el marco de Laban deja de ser un adorno teórico y se vuelve la manera de traducir esfuerzo y cualidad, no solo coordenadas.&lt;/p&gt;

&lt;h2&gt;
  
  
  Dónde verla
&lt;/h2&gt;

&lt;p&gt;Seleccionada para CVPR Art Gallery 2026. Documentación, videos y demo del agente: &lt;a href="https://petrasynthetic.com/agente" rel="noopener noreferrer"&gt;petrasynthetic.com/agente&lt;/a&gt;.&lt;/p&gt;




&lt;p&gt;&lt;em&gt;Referencia de investigación: Olof Misgeld sobre mediación del movimiento en danzas folclóricas suecas, traída aquí a la pregunta política por la propiedad de los datos del cuerpo.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>ai</category>
      <category>python</category>
      <category>computervision</category>
      <category>spanish</category>
    </item>
    <item>
      <title>Three ways to adjudicate subjective work</title>
      <dc:creator>Petra</dc:creator>
      <pubDate>Tue, 27 Oct 2026 17:59:05 +0000</pubDate>
      <link>https://dev.to/petrasynthetic/three-ways-to-adjudicate-subjective-work-2e2c</link>
      <guid>https://dev.to/petrasynthetic/three-ways-to-adjudicate-subjective-work-2e2c</guid>
      <description>&lt;p&gt;Blockchains can move value. They cannot judge meaning. As agents begin to sell creative work, the bottleneck stops being payment and becomes adjudication: who decides whether a poem, a video, or a metadata package actually fulfilled the brief.&lt;/p&gt;

&lt;p&gt;There are three common answers. Each one works, and each one breaks in a different place.&lt;/p&gt;

&lt;h2&gt;
  
  
  1. Platform review
&lt;/h2&gt;

&lt;p&gt;A marketplace hires reviewers, gives them a queue, and lets them decide disputes. This is how most freelance platforms and art commissions work today.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;What it gets right:&lt;/strong&gt; context. A human reviewer can read the brief, look at the work, and use taste. For ambiguous creative work, taste is not a bug.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Where it breaks:&lt;/strong&gt; the reviewer is a single point of failure. Queues take weeks. The fee to arbitrate can cost more than the commission. The reasoning is private, so the losing side cannot check it. And the platform has an incentive to keep large sellers happy.&lt;/p&gt;

&lt;p&gt;Platform review is fast to start and expensive to trust.&lt;/p&gt;

&lt;h2&gt;
  
  
  2. Optimistic arbitration
&lt;/h2&gt;

&lt;p&gt;Protocols like UMA and Kleros turn disputes into economic games. Someone asserts an answer with a bond. If nobody challenges within a window, the answer stands. If someone does, the dispute escalates to a jury or a token-weighted vote.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;What it gets right:&lt;/strong&gt; undisputed cases are cheap and fast. The threat of a bond makes careless assertions expensive.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Where it breaks:&lt;/strong&gt; it assumes a challenger with capital and enough context to notice the problem. For subjective work, the jury may have no domain knowledge. "Does this poem meet the brief" is not a factual question that a token vote resolves well.&lt;/p&gt;

&lt;p&gt;Optimistic arbitration is a truth machine for facts, not a taste machine for meaning.&lt;/p&gt;

&lt;h2&gt;
  
  
  3. Validator consensus with LLM judgment
&lt;/h2&gt;

&lt;p&gt;GenLayer takes a third path. An Intelligent Contract fetches the deliverable, asks a model to evaluate it against the brief, and requires independent validators to agree on the decision. The contract stores the verdict and settles the money.&lt;/p&gt;

&lt;p&gt;A practical example is an escrow for creative commissions:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;The buyer deposits GEN and writes the brief.&lt;/li&gt;
&lt;li&gt;The seller submits a public URL.&lt;/li&gt;
&lt;li&gt;If the buyer disputes, the leader downloads the page, extracts quotes, marks each requirement as MET or NOT MET, and returns a decision with a score.&lt;/li&gt;
&lt;li&gt;Validators repeat the same task. They must match the decision and stay within a score band.&lt;/li&gt;
&lt;li&gt;The contract releases the deposit or refunds it, and keeps the reason onchain.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;&lt;strong&gt;What it gets right:&lt;/strong&gt; it can read subjective material. It forces evidence before opinion: quote first, then decide. It has no human queue, and the reasoning is part of the record.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Where it breaks:&lt;/strong&gt; nondeterminism. Two validators can read the same page and score it differently, so the rubric and the score band become the real contract. Web fetches fail: a 404, a paywall, or a redirect can look like a bad delivery. Prompt design is fragile.&lt;/p&gt;

&lt;h2&gt;
  
  
  Comparison
&lt;/h2&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;&lt;/th&gt;
&lt;th&gt;Platform review&lt;/th&gt;
&lt;th&gt;Optimistic arbitration&lt;/th&gt;
&lt;th&gt;Validator + LLM&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Cost per dispute&lt;/td&gt;
&lt;td&gt;High, human time&lt;/td&gt;
&lt;td&gt;Low if undisputed, high if escalated&lt;/td&gt;
&lt;td&gt;Medium, model plus consensus&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Speed&lt;/td&gt;
&lt;td&gt;Days to weeks&lt;/td&gt;
&lt;td&gt;Hours to days&lt;/td&gt;
&lt;td&gt;Minutes&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Handles subjective work&lt;/td&gt;
&lt;td&gt;Yes, with taste&lt;/td&gt;
&lt;td&gt;Poorly&lt;/td&gt;
&lt;td&gt;Yes, with a rubric&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Evidence trail&lt;/td&gt;
&lt;td&gt;Private&lt;/td&gt;
&lt;td&gt;Economic&lt;/td&gt;
&lt;td&gt;Quoted and stored&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Main failure mode&lt;/td&gt;
&lt;td&gt;Opacity and queue&lt;/td&gt;
&lt;td&gt;Missing challenger&lt;/td&gt;
&lt;td&gt;Nondeterminism and fetch errors&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;h2&gt;
  
  
  What the case study showed
&lt;/h2&gt;

&lt;p&gt;A small escrow built on GenLayer ran two cases in Studio. A poem that matched the brief was released. A generic page that did not match was refunded. The verdict cited fragments of each delivery and marked every requirement. The contract passed eight direct tests and the whole flow ran without gas.&lt;/p&gt;

&lt;p&gt;The limits were just as clear. There was one seller, no partial refunds, and no appeal. The rubric worked because the brief was narrow. A vague brief would have produced a vague verdict, no matter how many validators agreed.&lt;/p&gt;

&lt;h2&gt;
  
  
  Design rules that follow
&lt;/h2&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Narrow the claim.&lt;/strong&gt; Ask "does this meet these three requirements", not "is this good".&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Require citations before verdicts.&lt;/strong&gt; A quote is checkable. An opinion is not.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Use score bands, not exact equality.&lt;/strong&gt; Validators should agree on the decision and stay close on the score.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Classify fetch errors.&lt;/strong&gt; A 404 is evidence about the delivery. A 500 is a reason to retry, not to refund.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Keep a human for high stakes.&lt;/strong&gt; Validator consensus is a strong first pass, not a replacement for appeal on large amounts.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Store the reason.&lt;/strong&gt; A boolean settles the money. The reason settles the next dispute.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;Adjudication is a spectrum, not a single primitive. The right question is not "which model is decentralized" but "what is the cheapest mechanism that can actually read the evidence".&lt;/p&gt;

&lt;p&gt;For objective facts, optimistic arbitration is enough. For taste, a human is still the best reader. For the growing middle, where agents sell work that is subjective but checkable against a brief, validator consensus with cited evidence is the first design that fits.&lt;/p&gt;




&lt;p&gt;Petra (Luis Betancourt) builds agents and adjudication systems. Entropia Escrow is an open-source GenLayer escrow for creative commissions: &lt;a href="https://github.com/Luis-Cwk/entropia-escrow" rel="noopener noreferrer"&gt;https://github.com/Luis-Cwk/entropia-escrow&lt;/a&gt;&lt;/p&gt;

</description>
      <category>genlayer</category>
      <category>ai</category>
      <category>blockchain</category>
      <category>web3</category>
    </item>
  </channel>
</rss>
