Learn territory / Curio

Learning that changes direction for a reason.

Curio is a set of skills and tools we built to help me learn any topic, at the depth the topic deserves. It turns that mission into bounded, sourced lessons, checks what I can actually explain or apply, and records the evidence honestly.

The map, lessons, sources, evidence, and next step remain durable, so I can leave a topic and return later without starting over. On the back end, Curio links its source references and learning artifacts with my LLM Wiki knowledge layer, which keeps the material retained and easy to retrieve across sessions.

Not a fixed course. Not a folder of generated lessons. A durable learning map that stays accountable to understanding.

Editorial atlas of Curio showing a learner moving from a mission through a provisional map, a bounded lesson, evidence, and a revised next step.
The Curio world: mission, map, lesson, evidence, revision, and a checkpoint another session can resume.

Content completion is not the same as learning.

A lesson can be beautifully written and a poster can be ready while the learner still cannot trace the idea, use it, or recognize a misconception. Most learning tools preserve the output. Curio preserves the reason for the lesson, what the learner demonstrated, and why the map moved or stayed put.

That distinction makes learning resumable. A later session does not have to trust a progress bar or hidden chat history. It can inspect the mission, the latest evidence, the current frontier, and the next safe action.

One coherent step, then evidence.

The map is provisional. Each pass through the loop is narrow enough to teach, check, and record without manufacturing confidence.

  1. OrientConfirm mission, map, prior evidence, and privacy.
  2. SelectExplain why one bounded objective is next.
  3. TeachUse reviewed sources at the learner's current level.
  4. MaterializeCreate an artifact only when it improves learning.
  5. CheckAsk the learner to retrieve, explain, compare, apply, or troubleshoot.
  6. RecordSave the observation without confusing it with readiness.
  7. AdaptRemediate, deepen, branch, or advance because of evidence.
  8. CheckpointLeave the next action and uncertainty durable.

One Request Through the Machine

This public-safe lesson shows one full Curio state transition, from mission to revised learning map. It is evidence of the pattern, not a complete export of the topic.

Mission

Build a durable backend-up mental model.

The learner wants to understand how visible text becomes model input, how inference produces output, how serving constraints shape systems, and how later topics such as retrieval, adaptation, safety, and evaluation fit together.

Why lesson 0001 came first

Establish the request-trace spine before zooming in.

Tokenization, vectors, transformer computation, logits, decoding, and serving pressure would otherwise be disconnected terms. One bounded trace from visible text to streamed output created shared boundaries for later lessons. This was a provisional map decision, not a fixed-curriculum rule.

Sourced lesson

Name the form of information at every major boundary.

Visible text token pieces token IDs learned vectors transformer computation logits sampled output token

After the input is processed in a prefill phase, generation repeats a next-token decode step until a stop condition is reached. Exact toy IDs and scores are illustrative, not model claims.

Reviewed public sources

Teaching artifact

The poster supports boundary tracing and later recall.

It places input context, tokenization, IDs, vector lookup, transformer computation, logits, selection, and streamed output on one surface. The poster is one teaching artifact inside the loop. Its readiness does not prove learner understanding, and the poster is not Curio by itself.

Teaching poster tracing one language-model request from visible text through token pieces, token IDs, vectors, transformer computation, logits, selection, and streamed output.
Open the live tokenization lesson to zoom into the first boundary shown in this request trace. If the image is unavailable, the sourced explanation remains available in the lesson.

Understanding check and evidence

Four boundary questions, one modest conclusion.

The check asked what first turns visible text into model-visible pieces, what an embedding lookup receives, what logits represent, and why output length affects latency. The retained public-safe record contains no raw response. It records four criterion-level observations: visible text was distinguished from tokenizer output, integer token IDs from learned vectors, logits from probabilities or final output, and longer output from additional decode steps. All four criteria were met.

That evidence supports the basic request trace and its major boundaries as shared vocabulary. Confidence is moderate because the result comes from one reviewed check without a repeated transfer task or independent implementation. It does not prove mastery of tokenization, transformer mathematics, inference serving, production architecture, or durable transfer.

Lesson artifact readiness Ready before check Ready after check
Learner understanding Not checked Demonstrated for the basic boundary trace

Artifact readiness and learner understanding are separate. A ready lesson begins with no understanding evidence.

Safe future assumption

Later lessons may use the request-trace vocabulary.

Text, tokens, token IDs, vectors, logits, and repeated decode may now be used as shared terms. No deeper mastery is assumed.

Evidence-causal map revision

Evidence evidence_request_trace_0001 closed the introductory request-trace node. Revision revision_tokenization_next_0001 moved the frontier to tokenization because token pieces and IDs were recognizable, while their model-specific behavior, context pressure, cost, and latency consequences still needed focused study.

If the learner had collapsed the tokenizer, model, and serving boundaries, understanding would remain uncertain. Curio would record the gap, offer a smaller sorting exercise, and keep the map in place.

Seven durable objects keep the learning legible.

These are logical roles, not prescribed filenames, tables, pages, or prompts.

Mission
Defines the capability, scope, evidence standard, and privacy boundary.
Map
Makes the current learning theory and frontier inspectable.
Lesson
Bounds teaching around one objective and source basis.
Teaching artifact
Supports comprehension, recall, or application when useful.
Evidence record
Preserves what the learner demonstrated, including ambiguity.
Map revision
Makes adaptation causal, reviewable, and reversible.
Checkpoint
Lets another session resume without hidden conversation history.

Agency stays with the person. Adaptation is shared.

Human authority

The learner or adopter chooses the mission, topic boundary, evidence standard, privacy policy, acceptable sources, and consequential tradeoffs. Humans approve publication, private-source use, and changes to what counts as evidence.

Agent responsibility

The teaching agent interviews before assuming, selects the smallest useful next lesson, teaches from reviewed sources, checks understanding, records uncertainty, proposes map changes, and stops when evidence or permission is insufficient.

A working topic, not a finished curriculum.

The AI Systems and Backend Mechanics topic is ongoing. Seven lessons have been built so far: five are quiz-passed and two are ready but not yet reviewed. One applied practice lab has also been built. The map continues beyond lesson seven.

7Lessons built so far
5Quiz-passed
2Ready, not yet reviewed
1Applied practice lab

All seven lesson artifacts are ready. Only five have quiz-passed evidence. Readiness reports whether an artifact is usable; understanding reports what learner evidence supports.

Preserve the behavior, adapt the machinery.

A valid reconstruction can use a command line, documents, a database application, or another interface. Storage, model, rendering, source ingestion, visual style, and cadence may change.

Continue to the packet handoff status

The site tells the human story. The packet is the agent authority.

This page explains Curio through a designed narrative and lived proof. The canonical multi-file Pattern Packet gives a reconstruction agent the ordered product, experience, state, protocol, artifact, behavior, golden-example, minimum-build, and acceptance contracts.

The validated packet candidate passed independent reconstruction. This verified immutable GitHub Gist revision is the public packet authority. This page never fetches or renders Gist content at runtime.

Canonical Gist Published packet Open immutable Gist

Specific enough to teach. Narrow enough to protect.

This packet shares behavioral guidance for reconstructing a similar learning experience. Curio is not open sourced, and PersekOS is not a reconstruction dependency.

Shared
The learning loop, logical object roles, state transitions, failure behavior, sanitized golden lesson, and acceptance scenarios.
Remain private
Private topics, family context, private source collections, machine-specific paths, sensitive assets, raw learner responses, and internal PersekOS wiring.
Permission
Readers may use the packet to guide an implementation for their own system. No license to Curio or PersekOS source code, assets, private materials, internal wiring, or brand is implied. The packet makes no broad license grant.

Curio is one coordinate in a larger practice.

I am Dustin Persek. I build useful systems, contribute to open source, and learn deeply with AI. Curio grew inside PersekOS, but the public pattern stands on its own. I am especially interested in conversations with builders working on adaptive learning, agent systems, durable memory, and practical human-AI collaboration.