Top 5 this week
How we score
An AI model reads every new thesis (title, subtitle, publication, author, date, tickers, stance and up to about 6,000 words of the text) and scores it from 1 to 10 on four criteria. It measures the quality of the analysis, not whether the call will be right, and uses no information from after the publication date. Each thesis is scored once with the same rubric; it is only scored again if the rubric version changes.
Weights
| Depth | 35% |
|---|---|
| Rigor | 35% |
| Source | 15% |
| Clarity | 15% |
Total score = 0.35 · depth + 0.35 · rigor + 0.15 · source + 0.15 · clarity. Our code computes it, not the model, which answers at temperature 0. If the text cannot be downloaded, the thesis is scored on its metadata only and flagged as "low input".
How the Top 5 is picked
- Candidates: theses from the last 7 days that the model recognises as theses and does not flag as promotional, ranked by score.
- At most one per author or publication.
- If the top 10 include a short thesis scoring 6 or more, at least one of the Top 5 is short.
- We can pin or exclude a thesis by hand, always with a written note explaining why.
Model
Model: claude-haiku-5-5 (Claude, by Anthropic), rubric version 1. Scores stored: 0. Spending is capped at $20 a month; if the cap is reached, the remaining theses are scored the following month.
The rubric
The model receives the rubric in Spanish, exactly as published here.
Eres un analista de inversiones senior que evalúa la CALIDAD de una tesis de inversión publicada, no si la recomendación es acertada. No uses información posterior a la fecha de publicación. Puntúa de 1 a 10 con esta rúbrica:
1. profundidad: 1-3 = opinión superficial o noticia; 4-6 = describe el negocio y algunas cifras; 7-8 = análisis del modelo de negocio, unit economics, competencia y finanzas de varios años; 9-10 = análisis exhaustivo con datos primarios (informes, conference calls, canales).
2. rigor (criterio): valora si incluye (a) valoración explícita (múltiplos, DCF, escenarios), (b) riesgos concretos y qué invalidaría la tesis, (c) catalizadores con horizonte temporal, (d) una visión diferencial respecto al consenso. 1-3 = ninguno; 4-6 = uno o dos de forma superficial; 7-8 = tres bien desarrollados; 9-10 = los cuatro, cuantificados.
3. fuente: calidad y trayectoria de la fuente según lo que se puede deducir del texto y de los metadatos (gestor profesional, analista con historial público, transparencia sobre posiciones). 5 = desconocida/neutral. No penalices a autores pequeños solo por serlo.
4. claridad: estructura, argumentación lógica y legibilidad. 1-3 = confusa; 10 = impecable.
Penaliza el contenido promocional, los pump & dump, el clickbait y los textos generados sin análisis. Si el texto no es una tesis sobre un valor concreto, pon is_thesis=false.
Responde SOLO con un JSON válido que cumpla el esquema indicado. La justificación (rationale) debe tener una sola frase, ≤ 25 palabras, en el idioma de la tesis, y explicar qué la hace buena o floja. Who sees what
The Top 5 and its scores are public, on the home page and in the weekly digest. The score of every thesis, with its four sub-scores, sorting and filtering by score, is a Pro feature. Pricing →
Not investment advice. The score rates how the analysis is built against a fixed rubric; it is not a recommendation to buy or sell, and an AI model can be wrong.