Two hundred fifty milliseconds
Severo's spaced-repetition system gets an automatic signal: speed. Instead of asking the user to self-rate easy/hard, it measures retrieval latency. Under 250 milliseconds, the translation is automatic, you've mastered it; 500 to 1000 milliseconds means you recognize but haven't interiorized it; slower means you're guessing. The distinction it operationalizes is recognized versus mastered (Juan can read German but can't produce it cold), and the whole thing needs no Severo judgment call, just a clock. Built and previewed live in Gemini Canvas, with a caveat borrowed from the polyglots: everyone's method is their own, so a copied method that doesn't engage you is still the wrong one.
// trace: where this idea came from
- ↳ video diary @ 16:39 (menos de 250 ms es que ya la sabes)
- ↳ video diary @ 3:13 (palabras reconocidas contra masteradas)
- ↳ Entry 224-2: Score the word, not the sentence (medir la palabra en vez de que Severo la juzgue)
The redesign that scored the word, not the sentence now gets its automatic signal: speed. Classic spaced repetition asks you to grade yourself, easy, hard, again ▸ 1:41, but Severo instead times how long you take to translate. The thresholds Juan sets on the whiteboard: under 250 milliseconds the word is automatic, “eso es que tú ya la sabes” ▸ 16:39; 500 to 1000 milliseconds and you recognize it but haven’t interiorized it ▸ 16:59; slower and it still costs you, or you don’t know it ▸ 17:10. A clock replaces a judgment call, “speed space repetition system.”
The distinction it operationalizes is one Juan feels in his own study: recognized versus mastered ▸ 3:13. He can read most German, but asked to produce it from zero, nothing comes ▸ 3:27, so the review drills exist to drag words across that line, from something you’d know if you saw it to something you can summon. Latency is what tells the two apart without anyone declaring which is which.
la velocidad delata lo que de verdad sabes →
The form gets prototyped live in Gemini Canvas, which generates and runs a little frontend from a one-paragraph description, a review picker with dropdowns for verb tense, person, part of speech, working across any language Stanza supports ▸ 20:49, though the first version ships with only the lemmas ▸ 20:10. The caveat comes from a polyglot’s video Julia sent: everyone’s method is their own, and you can borrow a fourteen-language expert’s seven tricks, but a method that doesn’t engage you is still the wrong method, “igual también es válido” ▸ 5:46. The system judges the learner by the clock, and lets the learner judge the method by whether they come back…
// continued in
no entry has continued this idea yet: the arc is still open