A connected path through knowledge

From what you know to what you want to build.

Warren maps knowledge as connected prerequisites, finds missing steps, and guides the next concept you can use.

Explore a learning pathWhy this exists
Prerequisite route previewEach connection explains what comes before the goal.
Prerequisite route preview for Deployed Python programSolid arrows lead from prerequisite knowledge to later concepts. Short dashed lines show related concepts. Node borders and verification marks identify current learning state and evidence.
  • Prerequisite
  • Related
  • programming
  • tooling
  • web
  • logic
  • quantitative
  • history

Why Warren exists

The design is over a century old. So is the problem.

The American school day was built to move large groups of children through fixed subjects on a fixed timetable. It coordinates people well. It was never built to track what any one person understands.

Teachers are not the problem. They work inside a structure that hands them thirty different starting points and one shared schedule. No amount of effort inside that structure can fix the structure.

  • Birth year sets the pace.

    A learner is placed by age, and the curriculum moves on a shared clock. When a prerequisite is missing, the clock does not stop for it. The gap does not announce itself either. It surfaces months later as a lesson that will not make sense, by which time the cause is several chapters back.

  • The same clock is too slow.

    A learner who already understands the lesson waits for the calendar. The reward for being ahead is more waiting, and the curiosity that got them there gets no room to run. A child who reads past the syllabus is doing exactly the right thing, and the timetable has no way to say so.

  • The reward becomes the goal.

    Grades, points, and class rank are what the system can see, so they are what learners optimize. The knowledge behind the score is easy to skip. Study for the test, forget it the week after, and the record still shows success. Nothing in that record separates a score from an understanding.

  • The schedule fights the body.

    Children need more sleep than adults: 9 to 12 hours at ages 6 to 12, 8 to 10 through the teens, against 7 or more for an adult. At puberty the body clock shifts later, and pediatricians report the average teenager cannot easily fall asleep before 11:00 p.m. and is best suited to wake at 8:00 a.m. or later. Schools average an 8:03 a.m. start, which puts the alarm nearer 6:30. That is about seven and a half hours a night, for twelve years. Pediatricians advise 8:30 a.m. or later. Only 17.7% comply.

  • Subjects are filed apart. Knowledge is not.

    Course boundaries hide real dependencies. A student meets the same idea twice under two names and never learns that it is one idea. Rates in a mathematics class and rates in a physics class are the same rates; split across two rooms and two years, they get learned twice and connected never. The reverse costs more. A history essay can stall on a data table and a physics problem can stall on algebra, but the timetable files each subject as its own separate failure. A learner is told they are weak at history when the missing step is statistics, and nobody in either room is placed to see it.

  • The path does not meet the work.

    Terms and subjects are organized around the timetable, not around what any trade actually asks for. A learner can finish thirteen years of it and still meet the first real requirement of the job on the first day of the job. The skills that decide the first year of a career often arrive last, or never arrive at all.

  • Reform buys more of the same design.

    US public schools spent about $927 billion in 2020–21, near $18,614 per pupil, up 13% per pupil in a decade. Spending grew. The century-old structure did not. More money moved through the same timetable, the same grade bands, and the same subject walls, and came out the other side as the same school day.

  • The design resists its own repair.

    Every proposed change has to pass through the calendars, budgets, and approvals it is meant to change. The proposals that survive that path are the ones that alter the least. Changing a pattern means breaking it, and no committee sitting inside the pattern is built to do that.

  • The gap shows up after graduation.

    In 2023, 28% of US adults aged 16 to 65 scored at or below the lowest literacy level, and average adult literacy fell 12 points since 2017. That is the result measured years after the last bell rang, and across the same decade it moved down rather than up.

School is still worth attending. Children learn to work with people, to disagree, to share a room with someone unlike them. Those are real, and no digital system replaces them. What school no longer holds a monopoly on is knowledge.

The evidence is already visible in hiring. In software, people who taught themselves from documentation, forums, and their own projects often out-perform graduates who first met data structures in their third year. They were not smarter. They followed a better order and they built things.

After school

A degree sells two things at once.

School ends and a second gate opens. A degree is sold as one thing, but it is two: the knowledge, and the proof that you hold it. They are priced together, and only one of them is still scarce.

Universities do real research and real teaching, and the best of it is worth the room. The point here is narrower. The price, the debt, and the four-year wait attach to the proof, not to the knowledge.

  • The price is for the certificate.

    In 2022–23 the average total price of a year on campus was $58,600 at a private nonprofit four-year institution and $27,100 at a public one. Lectures, notes, and problem sets from institutions of that rank are published openly. What the price buys that an internet connection does not is the credential at the end.

  • The bill outlasts the course.

    US student loan balances stood at $1.65 trillion in June 2026. A four-year course is repaid across decades. The payment schedule does not adjust for whether the knowledge was needed, retained, or ever used in the work the borrower ended up doing. The certificate is bought once and paid for far longer.

  • The certificate places fewer graduates than it used to.

    In the second quarter of 2026, unemployment among recent US college graduates sat near 5.6%, and 42% were underemployed: working a job that does not require a degree. Two of every five people who just paid for the certificate are doing work that never required it.

  • Employers started separating the two.

    Between 2017 and 2019, 46% of middle-skill occupations and 31% of high-skill occupations went through a material reset of their degree requirements. The pandemic sped that up, but the change began before it. More employers now ask what a candidate can do rather than which institution signed for it.

  • Most adults never got one.

    About two-thirds of Americans hold no college degree. That is not two-thirds without skill. Analysts project that the move away from degree screens could open 1.4 million more jobs to workers without one over five years. The ability was already there. The filter sat in front of it.

  • Teaching is not where the money is.

    Princeton states that well over half of its total operating revenue, 55%, comes from net investment income, mainly from its 4,400-plus managed endowments. That is a sound way to fund research. It also means the budget of the wealthiest institutions does not depend on how many people they teach.

Warren does not issue a degree and does not claim to replace one. Hiring runs on signals, and that signal is not ours to hand out. Where a licence or a degree is required, it is required, and no map changes that.

The other half can be separated out. An order to learn things in, a check that they stuck, and something built with them need no enrolment, no campus, and no payment plan. As more employers test for the skill directly, that half is the half that keeps counting.

Why now

The information is already free. The order is not.

More knowledge is reachable from an ordinary phone today than any university held a generation ago. Schools were designed when knowledge was scarce and a building was the only way to reach it. That constraint is gone. The schedule built around it stayed.

The gap is no longer access. It is sequence, retention, and application: what to learn next, whether it is still there a month later, and what to build with it.

  • Solved

    Access to the material.

    Lectures, textbooks, papers, datasets, and simulations are a search away, in most subjects, at no cost.

  • Open

    Knowing what comes first.

    A search returns a page, not an order. A learner who is missing one prerequisite cannot tell which page will make sense and which will waste an afternoon.

  • Open

    Knowing that it stuck.

    Reading feels like learning. Recalling shows whether it happened. Almost nothing online asks the learner to recall before it moves on.

  • Open

    Having somewhere to use it.

    A concept that is never used is a concept that fades. Free material rarely comes with a project attached to it.

Those three gaps are what Warren is being built to close. The map supplies the order. Recall and spaced return test whether the knowledge held. A project supplies the use. None of that requires a building, a bell, or a birth year.

The best self-taught learners already do this by instinct: they find the missing step, they come back to it, and they build something with it. Warren's goal is to make that method explicit, so it does not depend on instinct, a good mentor, or luck.

How Warren works

Fixed goal. Flexible path.

The destination stays visible while the route responds to the prerequisites that are already secure and the next gap that needs attention.

  1. Choose a meaningful goal

    Begin with something you want to make or understand.

    goal · selected

  2. Trace its prerequisites backward

    Follow the concepts that the goal depends on, in reverse.

    route · connected

  3. Check what is already secure

    Mark only the prerequisites that have already been demonstrated.

    concept · secure

  4. Learn and verify the next gap

    Work on the first unmet concept, then check it with another attempt.

    next gap · active

  5. Use it in a real artifact

    Put the newly connected knowledge to work in the thing you chose.

    artifact · applied

Planned pathways

Different goals need different paths.

The public demo shows one computing path. Warren's broader map is being built to connect pathways through mathematics, computing, physical science, and history. Each subject keeps its own structure, while real prerequisites can cross subject boundaries.

  • Mathematics

    Number, algebra, geometry, probability, and statistics.

  • Computing

    Programming, systems, data, networks, and deployment.

  • Physical science

    Measurement, mathematical models, experiments, and explanation.

  • History and humanities

    Chronology, sources, context, comparison, and causal argument.

A programming project can depend on algebra. A physics problem can depend on graphs and proportional reasoning. A historical argument can depend on chronology and source evaluation. Warren keeps those dependencies visible instead of forcing every goal into one fixed course sequence.

Across subjects

One skill, three subjects.

School keeps history, mathematics, and computing in separate rooms. Real work does not. This map traces one goal: explain a historical change and show it in a chart the reader can check.

This slice is illustrative. It shows how cross-subject prerequisites are recorded, not a finished map of any subject.

Cross-subject prerequisite map for Model of a historical changeSolid arrows lead from prerequisite knowledge to later concepts. Short dashed lines show related concepts. Node borders and verification marks identify current learning state and evidence.
  • Prerequisite
  • Related
  • programming
  • tooling
  • web
  • logic
  • quantitative
  • history

The links that cross

  • Reading a data tableSource criticism

    A census table is a historical source. Reading it wrong changes the history.

  • Rates and graphsPlot a series

    The mathematics of a rate is the same mathematics the plotting code draws.

  • Reading a data tableSeries data

    One quantitative skill feeds both the argument and the program.

  • History nodes

    A history node is checked by its sources. It asks which records exist, who made them, and what they leave out.

  • Mathematics nodes

    A mathematics node is checked by computation. A worked answer is either right or wrong, and the same skill is reused everywhere.

  • Programming nodes

    A programming node is checked by running it. The program produces the chart, or it does not.

Reading a data table is one skill. It carries a history student through a census return and a programmer through a data file. A course catalogue teaches it twice, or once and never again. A graph records it once and points both ways.

Retention and application

Remember it. Choose it. Use it.

Learning runs in three stages: take the idea in, keep it, then use it. Most study stops at the first stage, because reading feels like progress. Retention and application are where knowledge becomes usable.

None of these mechanisms are new. Researchers have measured them for decades, and the strongest self-taught learners already use them without naming them. Warren is designed to combine them deliberately instead of treating exposure as learning.

  1. Map.

    Start from a goal, trace prerequisites, and focus practice on the next missing concept.

  2. Retrieve.

    Recall or explain an idea before rereading it. Retrieval practice supports learning and retention across many studied settings.

  3. Revisit.

    Return after a delay. Spacing generally improves long-term retention, but useful intervals depend on the learning goal and retention period.

  4. Distinguish.

    Mix related problem types so the learner must choose a method rather than repeat a visible pattern. Interleaving benefits depend on the task and the similarity of the material.

  5. Correct.

    Use specific feedback and another attempt to correct an error while the concept is active. Feedback effects depend strongly on the information provided and how the learner can act on it.

  6. Apply.

    Use the concept in problems, explanations, and projects. Transfer depends on prior knowledge, task similarity, guidance, and practice; it is not automatic.

The graph sets the order. Retrieval and spacing strengthen access to knowledge. Interleaving and feedback sharpen choice and correction. Application checks whether the learner can use the concept beyond the prompt. A schedule cannot run this loop for thirty people at once; software can run it for one person at a time. These mechanisms are well studied; Warren's integrated implementation is still being built and evaluated.

Guided graph demo

Build and deploy a small Python program

This illustrative, hand-reviewed map traces a bounded route through programming, command-line use, version control, HTTP, deployment, logic, and basic quantitative reasoning.

No learner data is stored. The route returns to its starting state when the page reloads.

  1. Start from a goal

    Highlight the goal node and its immediate dependencies.

  2. Find the next gap

    Mark a small set as secure and focus the first unmet prerequisite.

  3. Inspect trust

    Open the concept detail and explain its evidence and verification label.

Guided prerequisite map for Deployed Python programSolid arrows lead from prerequisite knowledge to later concepts. Short dashed lines show related concepts. Node borders and verification marks identify current learning state and evidence.
  • Prerequisite
  • Related
  • programming
  • tooling
  • web
  • logic
  • quantitative
  • history

Text equivalent

Ordered prerequisite route

The concepts are listed from foundations toward the selected goal.

  1. Basic arithmeticquantitative
  2. Boolean logiclogic
  3. Text valuesprogramming
  4. Command linetooling
  5. Files and pathstooling
  6. HTTP requests and responsesweb
  7. Variablesprogramming
  8. Python runtimeprogramming
  9. Version controltooling
  10. Comparisonslogic
  11. Environment variablestooling
  12. Conditionalsprogramming
  13. Processes and portsweb
  14. Loopsprogramming
  15. Functionsprogramming
  16. Deployment servicetooling
  17. Collectionsprogramming
  18. Modulesprogramming
  19. Errors and exceptionsprogramming
  20. Package managementtooling
  21. Automated testingtooling
  22. JSON dataweb
  23. Network requestsweb
  24. Deployed Python programprogramming

Trust and AI

AI can help construct material. Evidence must earn trust.

Warren shows how each node was checked. AI supports learning; it does not take the learner's turn.

  • execution-checkedRequires the relevant example or behavior to run successfully against an expected result.
  • computation-checkedRequires a reproducible calculation to match its expected result.
  • multi-source citedRequires support from more than one named source.
  • human-signedRequires an identified person to review and approve the material.
  • not yet verifiedLabeled openly and withheld from the public demo material.

These public-demo labels are illustrative: they explain Warren's evidence categories. They are not confidence scores, production-corpus records, or measures of learner mastery.

Build status

Warren is a work in progress.

Its learning material is built and checked on locally operated GPUs. Public serving stays separate, so the site remains available while new material is being produced.