Map what you own
The learner universe records known language, exposure, recall, productive evidence, interests and what is due—not merely completed lessons.
The language method that adapts twice
Learn what expands your voice.
In the life you have today.
Most courses teach a fixed list. Probandolenguas models the language you already own, calculates which next word or phrase will let you express the greatest number of new ideas, then rebuilds the lesson for the time, energy, device and privacy you have right now.
The English 500→503 demo is an inspectable proof of a method designed for every level and language.
YOUR CURRENT EXPRESSIVE UNIVERSElive
01suggest
02difference
03story
The live proof turns the ranking into contexts, a podcast, retrieval, writing, voice and conversation—all inside language the learner can understand.
2decisions computed: what next + how now
5working context-adaptive formats
1continuous learning coordinator
500→503inspectable live English proof
The mathematical engine · E(L)
The curriculum is a function—not a chapter list.
THE OPTIMIZATION
Most curricula ask which words are common. E(L) asks: from everything this learner can almost say, which single addition unlocks the largest weighted set of complete meanings?
next atom= arg maxE(L ∪ atom) − E(L)
WHAT STRICT SCORING HIDES
100Meaning appears before perfect form.
known atoms
48 construction-disjoint held-out conversational nuclei · 555 project-authored CC0 rows · lexical-distance diagnostic, not a proficiency claim
COMMUNICATION NOW · PRECISION LATER
“Yo querer ir playa mañana.”✓
The meaning landed.I want to go to the beach tomorrow.
conversationcontinues
verb agreementremembered
a + larevisited later
The app does not confuse imperfect writing with failure. It protects the attempt, then schedules precision when correction will not kill the conversation.
Count meanings unlocked
The engine ranks words and reusable chunks by their marginal ability to complete conversational intentions this learner cannot yet express.
Make the math lived
The winner becomes comprehensible context, audio, retrieval, writing, speech and conversation. Generated language is mechanically checked before delivery.
One ranking becomes a complete learning loop.Candidate → verified context → personalized audio → retrieval → production → evidence → the next recomputation. The learner changes, so the curriculum changes.
Open the current measurement note
The engine, staged curriculum and English ranker are working. On one strict held-out mechanical comparison, the current Spanish 100-atom curriculum ties a frequency baseline; the preregistered semantic evaluation is the next measurement. That benchmark limits one claim—it does not reduce the product to a research demo.
The second compiler · how now
Your life becomes an input to the method.
Most apps shorten the same lesson. Probandolenguas changes the learning operation around your time, energy, eyes, hands, voice, device and interruption risk.
01
Telegram
A conversation, not a course menu
02
Voice
Listen and answer with your hands free
03
Desktop
Write the longer thought
04
TV · next
Planned: video-first, phone as controller
05
Low energy
Gentle exposure, zero pressure
06
Paper
Print it, write it, photograph it
Interactive context compiler
One target. Five genuinely different sessions.
Try phone, desktop, no-screen, low-energy and print. The coordinator preserves the learning target while changing input, output, pace and evidence rules.
The learning loop
Every interaction changes what comes next.
- 01Model
- 02Rank
- 03Encounter
- 04Produce
- 05Measure
- 06Recompute
Important: listening records exposure, never mastery. Evidence stays skill-specific.
Semantic voice coaching
It hears what you mean—not just what you missed.
The coach acknowledges the learner’s actual idea, gently recasts it, and moves the conversation forward. Communication is rewarded now; recurring errors return later.
NOWPermissive in the momentLATERCorrective over time
P
probandolenguasvoice practice · active
▂▅▇▄▆▃▅▂0:04
“One is yellow and the other is red.”
Exactly—the difference is their color.
One is yellow; the other is red. What story could explain why?
understooddifference ✓color observed
Human meaning · adaptive support
Personalization without sanding away personality.
Human intention, humor, culture and curiosity decide what is worth saying. AI quietly calibrates the support, remembers the evidence and brings the right language back.
- 01 goals shaped by the learner’s real interests
- 02 transparent generated voices in the live proof
- 03 skill-specific evidence and spaced re-exposure
Looping engineering with Codex
The method learned
while we built it.
Design hypotheses became running code, then met real Telegram behavior. Every failure fed the next loop instead of being hidden.
- 01DesignDefine the learner model and constraint
- 02BuildTurn it into a usable conversation
- 03TestMechanically verify words and flows
- 04AuditIndependent model review
- 05UseReal voice notes in Telegram
- 06RepairFix the behavior, rerun the loop
Failures become product knowledge.When an early evaluation revealed order bias and overfitting, the loop corrected the evaluator, the code and the claim. Real learner failures became regression tests too.
A small proof of a much larger method
Learn what moves your voice forward—in the life you actually have.
- 1Inspect the assumed 500-word universe
- 2Choose 3 mathematically ranked candidates
- 3Watch them become context, audio and conversation
English proof · method designed for every level and language · isolated post-submission beta