A practical operating system for becoming a high-leverage generalist without losing depth.

Mission

Reduce the time required to move from unfamiliar domain to independent, useful judgment, while preserving deep anchor expertise in software architecture and platform systems.

The unit of progress is not content consumed. It is demonstrated capability:

Question -> Model -> Retrieval -> Application -> Feedback -> Transfer -> Compression

Default operating constraints

The system has a strict work-in-progress limit. Breadth is built sequentially, then maintained in parallel.

Architecture

1. Direction layer

Defines the kind of generalist you are building:

• Anchor depth: software architecture, distributed/platform systems.
• Adjacent leverage: AI systems, cloud economics, product and business strategy.
• Economic leverage: finance, capital allocation, negotiation, sales, entrepreneurship.
• Human systems: leadership, organizational design, psychology, communication.
• Exploration: deliberately distant fields that may produce novel transfer.

2. Portfolio layer

Use four learning modes:

3. Campaign layer

Each 12-week campaign must define:

1. A real capability, not a broad topic.
2. A baseline test.
3. A proof-of-competence artifact.
4. Weekly retrieval and application.
5. At least one source of expert or market feedback.
6. A final independent assessment.
7. A maintenance plan or explicit archive decision.

4. Session layer

Every focused session uses the L.E.A.R.N. loop:

A session that only consumes material is incomplete.

5. Knowledge layer

Store knowledge by its future use:

• Domain maps: the structure of a field.
• Mental models: reusable explanatory and decision models.
• Projects: proof that knowledge works under constraints.
• Retrieval prompts: durable memory and diagnostic tests.
• Transfer notes: cross-domain connections.
• Reviews: operating feedback on the learning system itself.

Standard weekly cadence

This is a default, not a streak. Sessions may be moved, but the weekly outputs should remain.

Minimum viable week

When work or family pressure is high, preserve the loop with:

• One 45-minute acquisition/retrieval session.
• One 90-minute application session.
• One 20-minute weekly review.

Do not compensate later with binge learning. Resume the normal cadence.

Proof-of-competence levels

A campaign normally targets a move of one or two levels, not mastery.

Metrics that matter

Track only evidence of learning:

Hours are capacity data, not success data.

Decision rules

• Retrieval above 90% twice: increase complexity, spacing, or realism.
• Retrieval between 70% and 90%: continue and add varied examples.
• Retrieval below 70%: repair prerequisites or reduce scope; do not simply reread.
• Two weeks without artifact progress: shrink the deliverable or change the campaign.
• No strategic relevance or curiosity after week 3: stop deliberately and record why.
• A note never retrieved, used, linked, or cited in 90 days: archive it unless it is reference material.
• New interesting topic during a campaign: place it in the backlog; do not start it immediately.

Physical notebook integration

Use paper for low-friction capture and thinking. The digital repository remains canonical.

Mark entries with five symbols:

Q  Question worth pursuing
M  Reusable model or principle
D  Decision or changed belief
A  Action or experiment
X  Cross-domain transfer

At the weekly review, migrate only M, D, and X entries that are likely to be retrieved or reused. Do not transcribe the notebook.

Folder map

00-system/       Rules, portfolio, levels, metrics, research basis
01-inbox/        Temporary capture; emptied weekly
02-domains/      Maps of fields and subfields
03-models/       Atomic reusable mental models
04-projects/     Proof-of-competence work
05-campaigns/    Active and completed campaigns
06-retrieval/    Closed-book prompts and results
07-synthesis/    Transfer notes, essays, teach-backs
08-reviews/      Daily, weekly, monthly, campaign reviews
09-prompts/      AI tutor and examiner prompts
backlog/         Prioritized future campaigns
scorecards/      Lightweight measurements

Start here

Use these as launch actions. Each link opens the matching handbook section or campaign workspace.

1. Understand the operating foundationRead the setup path and the research behind the system before configuring a campaign.

   Open start guideSTART-HERE.mdOpen research basis00-system/research-basis.md
2. Map your capability portfolioIdentify anchor expertise, acquisition targets, maintenance domains, and the next proof required in each area.

   Open skill map00-system/generalist-skill-map.md
3. Compare campaign candidatesUse strategic relevance, leverage, application, transfer, scarcity, and sustained curiosity to select one campaign.

   Open campaign portfoliobacklog/learning-backlog.csv
4. Create the campaign contractDefine the capability, proof artifact, constraints, quality bar, and completion criteria through guided setup.

   Open campaign setup05-campaigns/_campaign-charter.md
5. Record the baseline before studyingCapture your closed-book explanation, representative task, score, confidence, and visible gaps.

   Open baseline fields
6. Schedule the evidence pathCreate the standard or minimum viable week and define what each week must produce.

   Open campaign plan
7. Begin with a governing questionWrite your initial answer from memory before selecting sources or asking for assistance.

   Open question coach

The goal of this week is to install the system and begin a real learning campaign. Do not spend the week perfecting tooling.

Day 1 — Define your generalist thesis

Complete the Generalist Skill Map00-system/generalist-skill-map.md.

Write one sentence:

I am building breadth in ______ so I can make better ______ decisions and create ______ leverage, while retaining depth in ______.

Day 2 — Select a campaign

Score no more than five candidates using the campaign portfoliobacklog/learning-backlog.csv.

Use this weighted score:

Priority =
  25% strategic relevance
+ 20% compounding leverage
+ 20% near-term application
+ 15% cross-domain transfer
+ 10% differentiation/scarcity
+ 10% sustained curiosity

Select the highest-scoring candidate that can produce a credible artifact within 12 weeks. When two candidates are within 0.25 points, prefer the one that broadens the portfolio beyond your existing anchor expertise.

Day 3 — Run the baseline

Without studying:

1. Explain the field's purpose and primitives.
2. Answer 10 representative questions.
3. Solve one realistic problem.
4. List likely failure modes and tradeoffs.
5. Rate confidence separately from correctness.

Save the results. The gap between confidence and correctness is useful diagnostic data.

Day 4 — Build the domain map

Build the domain map from memory first in the Questions workspace02-domains/_domain-map.md. Then use a small number of authoritative sources to correct it.

Day 5 — Create the first models

Create three model notes through the Question and model workspace03-models/_mental-model.md. Each note needs a mechanism, assumptions, failure conditions, and at least one retrieval prompt.

Day 6 — Charter the proof project

Define the proof artifact in the Campaign Charter04-projects/_proof-project.md. The project must force decisions under constraints, not merely repeat a tutorial.

Day 7 — Review and schedule

Complete the Weekly Review08-reviews/_weekly-review.md. Put the next week's sessions on the calendar. Define the exact question that starts Monday's session.

Bootstrap completion criteria

The OS is installed when all of these are true:

• One active campaign exists.
• A baseline is saved.
• A domain map exists.
• A proof project is chartered.
• At least five retrieval prompts exist.
• The next week's sessions are scheduled.

Anything beyond this is optional refinement.

The OS is a practical synthesis of several well-supported learning mechanisms. It does not treat any fixed schedule or ratio as a universal law; those are operating defaults to be adjusted from measured results.

Important boundary

Memory techniques alone do not create expert judgment. The OS therefore combines retention mechanisms with realistic projects, ambiguity, feedback, adversarial review, and cross-domain transfer.

Heuristics versus findings

The following are system heuristics, not claims of scientifically optimal universal values:

• Six hours per week.
• Twelve-week campaigns.
• A three-source initial budget.
• The 1, 7, 21, and 60-day review sequence.
• A 60% active-output floor.
• The 70–90% retrieval target band.

Keep them until operating evidence suggests a better value for a specific domain and schedule.

1. Questions before sources

Start with a concrete question, prediction, design problem, or decision. Sources are selected to close a known gap.

2. Attempt before assistance

Before consulting notes, search, or AI, produce an answer. Even a weak attempt makes gaps visible and creates material for feedback.

3. Retrieval over familiarity

Being able to recognize an explanation is not equivalent to being able to reconstruct or use it. Closed-book recall and problem solving are mandatory.

4. Output over accumulation

Every week must change a project, decision, explanation, model, or behavior. Saved links and highlighted pages do not count as output.

5. Feedback over repetition

Practice should expose errors and produce correction. Repeating comfortable tasks is maintenance, not acquisition.

6. Transfer is a deliverable

A model is not considered integrated until it has been compared or applied across contexts.

7. Breadth is sequential

Maintain multiple domains, but acquire one major capability at a time. The campaign work-in-progress limit is one.

8. Compression follows understanding

Summaries are written after retrieval and application, not copied during initial reading.

9. Tooling must stay subordinate

Do not redesign the repository, tags, or dashboards unless an observed failure requires it.

10. Stop deliberately

Ending a low-value campaign is a valid outcome. Record the evidence, extracted models, and future trigger for resuming.

Review quarterly. Use capability evidence rather than self-image.

Generalist thesis

I am building breadth in ____________________ so I can make better ____________________ decisions and create ____________________ leverage, while retaining depth in ____________________.

Portfolio

Portfolio health questions

1. Is anchor depth still producing career or business differentiation?
2. Which missing domain currently limits important decisions?
3. Which domain would connect at least three existing strengths?
4. Am I studying adjacent topics because they are comfortable?
5. Which distant field could challenge my current models?
6. What can be dropped or maintained at lower intensity?

Skill-level rubric

Use observable evidence. A level is not awarded because material was consumed.

Assessment protocol

For each assessed capability:

1. Define a representative task.
2. Restrict references appropriately.
3. Set a time limit.
4. Define objective or expert-reviewed quality criteria.
5. Record confidence before receiving feedback.
6. Record correctness and error classes afterward.
7. Reassess after a meaningful delay, not immediately.

A balanced year normally contains four campaigns:

Recommended sequence for the next four campaigns

1. Corporate finance and capital allocation for technical builders — expand beyond the engineering anchor and strengthen business, startup, and investment judgment.
2. AI systems economics — connect model architecture, infrastructure, product constraints, and unit economics.
3. Negotiation and enterprise value communication — convert expertise into compensation, partnerships, sales, and influence.
4. Organizational design for high-leverage technical leadership — improve execution through systems of people, incentives, ownership, and decision rights.

This sequence is a strategic default. Re-score it when a concrete opportunity changes the expected value.

Maintenance protocol

For each completed campaign, choose exactly one:

• Active maintenance: one retrieval or application session every two weeks.
• Passive maintenance: use only through normal work or projects; assess quarterly.
• Archive: preserve artifacts and models, but spend no scheduled time.

Do not maintain everything actively.

Metrics

Weekly measures

Monthly measures

Health classifications

Green: artifact progressing, retrieval 70–90%, feedback current
Yellow: high consumption, weak application, or feedback older than 14 days
Red: no artifact progress for two weeks or campaign purpose no longer matters

Anti-metrics

Do not optimize for:

• Books completed.
• Hours of video watched.
• Number of notes.
• Number of highlights.
• Saved links.
• Daily streak length.

These may support learning but do not prove it.

type: learning-campaign
campaign:
status: proposed
start:
end:
primary_domain:
starting_level:
target_level:
weekly_budget_hours: 6

Campaign: [capability, not broad topic]

Strategic thesis

Why is this worth learning now? What decisions, opportunities, or leverage should improve?

Capability statement

By the end of this campaign, I can independently ____________________ under ____________________ constraints, and defend the result against ____________________ criteria.

Baseline

• Current level:
• Closed-book map score:
• Representative task result:
• Confidence before feedback:
• Actual correctness:
• Primary gaps:

Driving questions

Proof-of-competence artifact

• Artifact:
• User/audience:
• Quality bar:
• External or objective feedback:

Source budget

Weekly plan

Kill or pivot criteria

Stop, shrink, or replace the campaign when:

• The strategic thesis is no longer valid.
• The proof artifact cannot be made realistic.
• Two consecutive weeks produce no application progress.
• Prerequisites make the original scope unrealistic.
• A higher-value opportunity materially changes priorities.

Definition of done

• Baseline and final assessment are comparable.
• Proof artifact meets its quality bar.
• Core models can be reconstructed without notes.
• Major failure modes and limitations can be explained.
• At least three cross-domain transfers were tested.
• A teach-back, memo, article, or presentation exists.
• Maintenance or archive decision is recorded.

Scorecard

Evidence of progress

• What can I now do that I could not do last week?
• Which artifact changed?
• Which decision improved?

Error analysis

• Most important misconception:
• Error class:
• Why it survived:
• Corrective exercise:

Retrieval health

• Models above 90% twice; increase difficulty:
• Models at 70–90%; continue:
• Models below 70%; repair prerequisites:

Transfer

• Strongest cross-domain connection:
• How it was or will be tested:

System health

• Where did consumption replace output?
• Where did tooling create friction?
• What should be deleted, simplified, or automated?

Next week

• Single most important milestone:
• Exact starting question for Monday:
• Feedback event to schedule:
• Items moved from inbox:
• Topics deferred to backlog:

Use these after making your own attempt. The AI should increase cognitive effort, not remove it.

1. Domain cartographer

I am learning [domain] to reach [observable capability]. Build a domain map containing primitives, core models, constraints, tradeoffs, standard workflows, failure modes, competing schools, representative problems, and prerequisite dependencies. Separate foundational knowledge from optional detail. Then give me a 20-question baseline assessment. Do not provide the answers until I submit my attempt.

2. Socratic examiner

Act as a demanding oral examiner in [domain]. Ask one question at a time. Start with mechanisms, then move to prediction, diagnosis, tradeoffs, edge cases, and transfer. Do not rescue me quickly. After each answer, score correctness, completeness, calibration, and reasoning quality; identify the exact misconception; then ask a harder follow-up.

3. Misconception auditor

Here is my closed-book explanation: [paste]. Compare it against authoritative understanding. Return: materially wrong claims, missing assumptions, hidden leaps, confused adjacent concepts, failure conditions I omitted, and the smallest corrective exercise that would expose whether I fixed each issue. Do not rewrite the full explanation for me.

4. Scenario generator

Generate five realistic scenarios that require choosing among [models/approaches]. Make surface details different enough that keyword matching will fail. Include incomplete information, constraints, and misleading signals. Do not identify the applicable model until after I answer.

5. Adversarial reviewer

Review this artifact as a skeptical expert who must reject weak reasoning: [artifact]. Attack assumptions, evidence, boundary conditions, alternatives, second-order effects, and implementation realism. Rank issues by decision impact. Do not focus on wording unless it obscures reasoning.

6. Transfer coach

I learned [model] in [source domain]. Help me identify structurally similar situations in [target domains]. For each candidate, state the shared structure, the differences that could invalidate the analogy, a testable prediction, and a low-cost experiment. Reject superficial analogies.

7. Compression examiner

Ask me to teach [topic] at three levels: 30 minutes, 5 minutes, and 30 seconds. Check whether each version preserves mechanism, constraints, and failure modes rather than becoming slogans. Question anything I cannot defend.

8. Project designer

Design a proof-of-competence project for [capability] that cannot be completed by following a tutorial. It must require ambiguous judgment, tradeoffs, multiple models, realistic constraints, objective or expert feedback, and a clear quality bar. Offer a minimum, standard, and stretch version.

9. Calibration trainer

Give me ten questions in [domain]. Before answering each, require a 0–100 confidence estimate. Afterward, score correctness and calculate calibration gaps. Group my errors into missing knowledge, faulty mechanism, transfer failure, and overconfidence.

10. Source triage

I need to answer [question] to achieve [capability]. Propose the smallest source set: one overview, one authoritative or primary source, and one practitioner source. Explain what question each source answers and when I should stop reading it. Flag claims that require current verification.

Prompt discipline

Always include:

• Target capability.
• Current attempt.
• Constraints.
• Desired feedback criteria.
• Instruction not to reveal answers prematurely.

Avoid prompts that merely ask for a comprehensive explanation. They maximize information volume, not learning.

Use the notebook as a fast thinking surface. Do not try to make it the permanent knowledge base.

Setup

Reserve the first four pages for an index. Number every page. Each day begins with:

Date
Campaign
Today's question
Expected output

Use these marks in the margin:

Q  Question
M  Mental model
D  Decision or belief update
A  Action/experiment
X  Cross-domain transfer
E  Error or misconception

Project spread

For a project or campaign, reserve two facing pages:

• Left: questions, map, constraints, and hypotheses.
• Right: decisions, tests, results, and next actions.

Continue on any later page and link page numbers rather than reserving large empty sections.

Daily notes

Write freely. At shutdown, box only:

• The strongest model.
• The most important error.
• The next question.

Weekly migration

Move only reusable M, consequential D, tested X, and unresolved high-value Q entries into the digital system. Actions move to the task system. Everything else remains in the notebook as a historical trace.

This avoids duplicate note maintenance and preserves the speed of paper.