Don't just recognize words.
Learn to use them.

A multilingual, sense-aware foundation for words and expressions.

• Lexical items can include single words, multiword expressions, fixed phrases, and patterns with variable parts.
• Individual meanings have their own examples, relationships, and learning progress.
• Word forms, parts of speech, pronunciation variants, IPA, and audio can be represented when content is available.
• An internal TMRND usage model organizes tone, mode, register, nuance, and dialect or language variety.
• A semantic knowledge graph connects senses, concepts, related expressions, and contextual examples.
• Content curation considers duplicate detection, sense disambiguation, provenance, and validation.
• The shared architecture supports multiple languages, while a first release may focus on one language.

Build understanding before expecting expression.

• Acquisition introduces a lexical cue, meaning, pronunciation when available, and a useful example.
• A quick placement check can recognize prior familiarity without treating self-report as proven mastery.
• Meaning recognition, listening recognition, sense recognition, and context comprehension provide varied evidence.
• Passive competency tracks understanding of a particular meaning.
• Relearning and remediation support recovery when evidence suggests forgetting.

Choose which familiar meanings deserve deeper productive practice.

• Learners can move eligible meanings into an Active Sense Pool.
• Controlled recall asks for a word or expression from a meaning cue.
• Guided production asks for an expression in a communication situation or sentence frame.
• Active competency and active review scheduling are separate from passive recognition.
• A path toward writing and speaking assessment is being explored for later stages, outside the core MVP.

Study the differences between related expressions, not just their definitions.

• Binary relations connect related senses and concepts.
• Relation guides explain meaningful usage boundaries and distinctions.
• Controlled distinction and contextual discrimination practice choosing the right expression.
• Relation clusters group multiple related senses with contrast guides.
• Relation understanding has its own competency and scheduled review.

Binary relations and relation clusters belong to the planned core MVP.

Estimate what is understood, retrievable, and ready for review.

• FSRS-6 is planned for memory scheduling using difficulty, stability, and retrievability.
• An IRT-like proficiency model is planned to consider learner ability and interaction difficulty.
• Passive, active, and relation competency remain separate modeled estimates.
• Evidence may account for hints, repeated attempts, elapsed time, diversity, and uncertainty.
• Repeated easy answers in one session should not imply lasting mastery.
• Interaction selection and review timing can adapt to session goals and current knowledge.

These models are designed for implementation and testing; no learning outcome is claimed as validated.

Learn from different materials while keeping one knowledge base.

• A learning library can organize books, courses, exam frameworks, topics, and other materials.
• Flexible structures can represent book chapters, course modules, or exam topics.
• Study plans resolve chosen materials into a practical learning scope.
• Preview and intake decisions can include Learn, I Know, and Skip.
• Vocabulary search is planned for exact and prefix lookup, word forms, and basic typo assistance.
• Daily Study, Learn New, Review Only, Active Practice, and New Only are planned study modes.
• Sessions can mix acquisition, passive, active, relation, and remediation work with flexible new/review order.
• Time-aware sessions and progressive disclosure make practice manageable.

A clearer picture than one global completion percentage.

• Coverage shows which parts of a chosen scope have been encountered or assessed.
• Passive readiness, active adoption, active competency, and relation knowledge are separate views.
• Review health reflects the maintenance needs of knowledge over time.
• Learning history keeps the evidence behind progress.
• Study planning can consider available time, review backlog, retention preferences, and target balance.

Account-based progress designed to travel with the learner.

• Learning identity and progress are planned to persist in an account, beyond one installation or textbook.
• Windows is the first planned production client; Android is the next target.
• Previously cached study material can support bounded offline sessions.
• Offline study events can synchronize when a connection returns.
• The canonical vocabulary corpus remains server-controlled, with a protected, bounded cache.

A full offline corpus download is outside the product scope.

AI can assist content work where it helps, with human and system checks.

• AI may help draft examples, exercise questions, plausible distractors, relation explanations, and linguistic metadata.
• Generated content must pass suitable validation before becoming authoritative scored material.
• The core MVP learning loop is not dependent on a runtime language model.
• Future exploration may include contextual explanation, writing feedback, and speaking practice.
• Dedicated grammar learning, additional language corpora, and personal content authoring are later possibilities.

No specific AI provider or external partnership is claimed.