The Craft of Teaching — Media Track (Public Edition) #

Companion track to the Craft of Teaching Mastery Curriculum (Modules 0–13).

For: working software engineers — including those who write docs and READMEs, review code, give talks, make videos, or are weighing a paid course or newsletter.

A media track is a rest track. You are not supposed to complete it. Reach for it when the spine's deliverables are too much and what you need is somebody else's careful engagement with how learning actually works, what teaching well costs, and what honest knowledge products look like. Watching twenty hours of conference talks does not make you a better speaker; giving one talk does. Use this as input between outputs.

Two tag systems, both preserved:

QUICK PICKS BY MOOD #

When you know the mood but not the title, start here.

MODULE 0 — The Teaching You Already Do (the audit) #

MODULE 1 — How Humans Learn #

MODULE 2 — The Curse of Knowledge #

MODULE 3 — Backwards Design & Assessment #

MODULE 4 — Retention Engineering #

MODULE 5 — One-on-One Teaching (code review, pairing, mentoring) #

MODULE 6 — Writing for Learners #

MODULE 7 — Speaking #

MODULE 8 — Video #

MODULE 9 — Across the Technical / Non-Technical Divide #

MODULE 10 — Workshops & Curriculum Architecture #

MODULE 11 — Knowledge Products as a Business #

MODULE 12 — Ethics of Public Teaching #

MODULE 13 — The Long Arc #

CHANNELS WORTH SUBSCRIBING TO #

HOW TO USE THIS TRACK #

1. Less than this list suggests. Two to four hours a week during the curriculum, one after. Production over consumption — the deliverables are the curriculum, not the watching.
2. Match mood to evening. After a hard pricing session, watch Patrick Campbell or read Charge More — not an Ali Abdaal revenue video, which feeds the cargo-cult flinch.
3. The Cautionary entries require the lens, not avoidance. Watch one Forte/Abdaal/Welsh piece a month with the critical lens on; don't chain three in a weekend.
4. Re-watch the high-leverage entries. Metz, Winston, Procida, one Bjork lecture, Inventing on Principle — these deepen on second viewing, especially when you bring a real artifact (a README, a talk draft) to it.
5. Honor the grades. Practitioner experience is not Empirical evidence; Empirical research is not the whole territory. Both are real; they are different; the literacy to know which is which is the curriculum's gift.
6. Don't pair two heavy films in one day. Whiplash deserves The Holdovers afterward, or a long walk.
7. If you're in India: nearly everything here is free YouTube/podcast content that works without a VPN; the films rotate across Prime Video, Netflix, and Mubi India — check justwatch.com/in, since rights move quarterly.

Watching is calibration. The work is somewhere else.

The Craft of Teaching — Community Guide (Public Edition) #

Companion track to the Craft of Teaching Mastery Curriculum (Modules 0–13).

For: working software engineers — including those who write docs, give talks, mentor, or are building a course, newsletter, or video channel on the side. Pairs with: the Mastery Curriculum + CRAFT_OF_TEACHING_MEDIA_TRACK.md

This is the ambient layer: the writers, voices, communities, and rooms that keep the teaching discipline alive between sessions. Tiered ruthlessly, because the course-creator economy is enormous, the research-backed-or-honestly-marked canon is small, and confusing the two is the most expensive failure mode this guide exists to prevent. The right number of subscriptions is the number you actually open. Five well-read newsletters beat fifty unread ones.

Grades: Empirical (research-backed) / Practitioner (lived operator experience) / Mixed (both, or popular synthesis needing a critical lens). The grade is truth-in-labeling, not a quality ranking.

1. NEWSLETTERS & BLOGS #

Tier 1: Must-Follow (read every issue) #

Load-bearing for the curriculum's sensibility. Deliberately capped at eight.

Tier 2: Subscribe; Read Selectively #

Tier 3: Sample When Relevant #

How to Engage #

• Reply to issues. Authors read replies; a year of thoughtful reading-and-replying makes you someone they know.
• Cite their work in yours, liberally, with links. Citation builds your credibility with their audience.
• Don't pitch. "I'd love to guest post" cold emails fail; "your last post made me think of X — here's mine" sometimes lands.

2. PODCASTS #

Pick three; listen to them; let the rest go.

Tier 1 #

Tier 2: Sample and Decide #

• The Pragmatic Engineer Podcast — companion to the newsletter, same quality.
• The Bootstrapped Founder (Arvid Kahl) — "sell to your audience first" applied to courses.
• Build Your SaaS (Justin Jackson) — indie-founder-with-day-job reality, applied to knowledge products.
• Smart Passive Income — 2010–2017 episodes only; the original is grounded, the modern genre isn't.
• Andy Matuschak interview appearances (Lex Fridman, Dwarkesh) — search by guest, not feed.
• Oxide and Friends; Signals & Threads — systems-engineering depth; teaching-by-conversation models.
• The Knowledge Project — selected learning-science episodes; search rather than subscribe.

How to Engage #

• Take notes on episodes that move you — podcast notes are where blog drafts come from.
• Don't listen at 2x. The compression destroys the reflection time the medium provides. If you lack the time, read the transcript.

3. COMMUNITIES #

The hardest tier — communities drift into echo chambers, dramaland, or graveyards. The good ones are gold.

How to Engage #

• Lurk first — two weeks before posting. The reputational cost of misfiring on a first post is real.
• Help before asking. Answer three questions before you ask one. Answering small questions in public is itself low-stakes teaching practice.
• Leave gracefully when you outgrow. Periodic cull; nothing personal.

4. COGNITIVE-SCIENCE RESEARCH RESOURCES #

The empirical substrate for M1 and M4, in research-community voice.

5. KNOWLEDGE-PRODUCT PRICING & BUSINESS VOICES #

The M11 canon. Read before you price anything.

6. DIÁTAXIS & TECHNICAL-WRITING RESOURCES #

7. TOOLS (with cost honesty) #

Don't over-tool early: ship the MVP teaching artifact on an existing platform first. Over-engineering the toolchain is itself an avoidance pattern.

8. CONFERENCES #

The fastest way to test your speaking practice — and the most expensive way to learn what the recordings teach for free.

Tier 1: Worth Travel If Submitting #

Tier 2: Watch the Archives #

Strange Loop (ended 2023; archives are gold) · Papers We Love chapters (teaching-by-explaining-research practice) · ML conference archives (NeurIPS/ICML) if your teaching touches ML.

If you're in India: RubyConf India and the regional meetup circuits (Mumbai, Bangalore, Pune) are the highest-leverage speaking venues by far — ₹5–25K plus minimal travel, versus $4,000+ all-in for a US conference whose talks are free on YouTube three weeks later. Submit locally first; travel only when a specific talk is accepted.

How to Engage #

• Meetup before conference. Skipping the rehearsal usually fails.
• CFP rejection is data, not verdict. Most accepted speakers were rejected first; submit to two or three.
• The hallway track is the conference. Talks are recorded; conversations aren't.

9. IF YOU'RE IN INDIA — additional context #

Honest note: there is no India-specific equivalent of the Matuschak/Moore/Kao canon yet. The canon above is US/UK-centric; the translation into your context is your work — and a publishing opportunity.

10. PAID COURSES — honest tradeoffs #

Most high-signal teaching-craft education is free or in books under $50. Paid courses earn a place only when (a) the free layer is exhausted, (b) you specifically need the structured peer cohort, or (c) an employer learning budget pays.

11. WHAT NOT TO SUBSCRIBE TO #

Refusing these inputs is itself an M0 deliverable. If you catch three of them on autoplay the weekend a real teaching artifact is due, close the tab and open a Matuschak essay, a Cathy Moore post, or a Campbell pricing study instead.

How to Use This Guide #

1. Tier 1 is the spine — eight newsletters. Most weeks they produce more signal than you can absorb. The discipline is the small list.
2. Tier 2 is ambient; Tier 3 is a search target. Dip in; don't keep up.
3. Multi-home your own teaching. Engineering discourse has fragmented across X, Mastodon, Bluesky, Discords, and newsletters — there is no town square. Don't invest exclusively in any single platform.
4. The guide does not replace the deliverables. Reading more pedagogy without running the M0 audit, more pricing content without doing the M11 math, more writing advice without shipping the rewrite — that's the avoidance pattern this whole document is warning you about.

The list is not the work. The list is so you can find the work, and the audience that wants the work you'll do.

The Engineer Who Teaches Deliberately #

Reading placement: read after Module 0. Re-read before Module 11, when the pricing-and-format decisions arrive and the stakes of operating your teaching surfaces deliberately become concrete.

What this is: A meta-essay on why most engineers underuse teaching as career and revenue infrastructure despite already operating multiple teaching surfaces by accident — and what it costs them across a 20-year career-and-studio lifecycle.

The Pattern #

Watch any technically competent engineer operate over a decade and one of four things tends to happen with respect to teaching.

The first: never-teach. The engineer who treats teaching as someone else's job. Documentation goes to whoever volunteers; conference talks go to whoever wants the visibility; mentoring is for managers; blog posts are for influencers. The engineer ships code, ships well, and reaches year ten with no teaching surface area beyond the occasional internal Slack message answering a junior's question. The compounding curve never starts because the surface never exists. This is the most common pattern in engineering, and it produces a perfectly respectable career trajectory that ends, predictably, at senior-engineer or staff-engineer-at-one-company, with no transferable authority. Asked at year fifteen why teaching never happened, the engineer says I'd rather just code, which is true and was the correct sentence at year three and was a strategic error by year seven.

The second: accidental-teach. The engineer with a library at 20,000 downloads whose docs explain how to instrument production errors. The engineer with a couple of blog posts that did better than expected. The engineer who occasionally answers Stack Overflow and discovers, three years later, that the answers have 50,000 cumulative views. Teaching surface area exists. It compounds, somewhat. Recruiters reach out occasionally; the library produces consulting referrals occasionally; the blog post that did well produces inbound for two years. None of this was directed; it accumulated as a byproduct of doing real engineering work in public. The compounding is real and undirected. The engineer at year ten has more career capital than the never-teach engineer and dramatically less than they would have had if any of the surface had been operated deliberately. This is, statistically, the second-most-common pattern.

The third: performative-teach. The LinkedIn cargo cult. The hashtag-team-building content. The "I'm hiring" posts that generate engagement and no hires. The course-creator pattern that copies Justin Welsh's hooks without Welsh's underlying competence. The substack that posts twice a week and produces no actual reader retention. Teaching-as-personal-branding without learning outcomes. The engineer accumulates followers and produces approximately zero engineers who are better at engineering after consuming the content. This pattern compounds negatively over a decade because the audience eventually notices the substance gap. The engineers who follow at year one stop following at year four; the recruiters who reached out at year two stop reaching out at year five; the consulting rate that should have risen to authority-level instead stagnates because the authority signal turns out to be hollow. Many engineer-influencers in 2026 are inside this pattern and do not yet know they are.

The fourth pattern is the rarest, and is what this curriculum is for: the engineer who teaches deliberately. Watches each teaching surface as a system with inputs, outputs, and a compounding curve. Operates the library docs as a teaching artifact, not as a record of what changed. Operates the conference talk as career infrastructure, not as a vanity slot. Operates the blog as a discovery layer that produces recruiter inbound for two years per quality post. Operates the course or curriculum as a knowledge product priced at the value the customer derives. Watches career and revenue compound across a decade. Reaches year fifteen with transferable authority, multiple revenue streams, hiring access from former mentees who are now hiring managers, and consulting rates two-to-five times the never-teach engineer's. The work is learnable. The work is engineering work, applied to learner cognition and audience economics. The work is what the rest of this curriculum trains.

This essay is about why so few engineers reach the fourth pattern, and what it costs the ones who don't.

What's Actually Going On #

The technical capacity is not the problem. An engineer who has architected a multi-tenant web application, debugged a production incident at 3 AM, and reasoned about cache invalidation across a distributed system has, demonstrably, the cognitive equipment to model how learners learn and how audiences pay. The reasoning is genuinely simpler than the work the same engineer does on a Tuesday morning. The CAP theorem is harder than designing a course module.

The barrier is cultural, and it shows up in specific patterns inside the engineering subculture.

"Teaching is for academics." A persistent piece of cultural mythology in engineering communities is that pedagogy is what credentialed teachers do at universities. The engineer reads about Backwards Design or retrieval practice, recognizes that the language is academic, and concludes that the discipline therefore does not apply to a working engineer who writes the README. This is empirically wrong. Backwards Design is API-first design. Retrieval practice is the test pattern. Spaced repetition is scheduled cache refresh. The academic register is a register, not the substance. The substance transfers cleanly to engineering work the moment the register is translated. The engineer who refuses to translate is refusing the most accessible cognitive-science evidence base available, on the grounds that it sounds like school.

"I'd rather just code." The engineer's default. Coding is comfortable; teaching feels slow. A function that compiles is observable in seconds; a course module that teaches is observable over weeks. The build cycle is fast; the teaching cycle is long. The "I'd rather just code" instinct routes around the slowness by skipping the teaching, which is the same as skipping the only step that would have converted the engineering work into compounding career and revenue value beyond the day-job paycheck. The code ships. The compensation arrives. The compensation is what it is, and stays what it is, because no teaching surface exists to leverage the engineering work into the next step. The never-teach engineer at year ten is paid the market rate for senior-engineer-with-no-public-surface, which is significantly less than the market rate for senior-engineer-known-for-X.

"The docs explain it." The engineer's specific failure mode in writing. Documentation as record-keeping rather than as deliberate skill transfer. The README explains what the library does; it does not explain what to do, why, and how to know it worked. The release note explains what changed; it does not explain why the change matters for the engineer using the library in production. The engineer believes the writing is adequate because it covers the facts. Adequate writing is not teaching. Teaching writing is engineered for the reader's cognitive load, sequenced for working-memory capacity, designed to produce action — and most engineering writing does none of this because most engineers have never been told that any of this is engineering work too.

"I shouldn't charge for what I learned for free." The pricing flinch on knowledge products. The engineer who learned their stack from free blog posts feels morally allergic to charging 9 a month. Same shape, same cost.

"The course-creator industry is gross." Partly true. Justin Welsh's "$5M solo creator" content is partly substantive and partly cargo-cult. Tiago Forte's "Building a Second Brain" is partly useful and partly productivity-cult. Ali Abdaal's YouTube empire is partly about content quality and partly about the YouTube algorithm. The substance and the cult are interwoven; the engineer who refuses the entire space because the cult is real is also refusing the substance. The discipline is to extract the substance and refuse the cult, not to refuse the entire space. The space contains real economic mechanics: knowledge products at the right scale and the right price are a credible revenue channel. The cult content insists this is easy and obvious; it is neither. The substantive content shows the actual mechanics, and the actual mechanics are learnable.

The engineer's specific failure mode: treating teaching as record-keeping. When a feature ships, the engineer's instinct is to update the README to reflect the new state. When a release goes out, the engineer's instinct is to write a changelog entry that lists what changed. When a learner asks how to do X, the engineer's instinct is to point at the docs. Each of these is a record-keeping move. The actual teaching moves — engineer for retention, sequence for cognitive load, design for the reader's working memory, build feedback loops that tell you whether the teaching worked — are the moves engineers default away from because they are not record-shaped. The teaching situations that route through these moves are, in practice, the situations where engineers underdeliver and never know they are underdelivering, because the metric they look at (the docs are accurate) is not the metric that matters (the reader can do the thing).

The Indian-engineer additions. For engineers teaching from and into India (treated at length in CRAFT_OF_TEACHING_FOR_INDIA.md), the cultural barriers stack with additional layers. The "free YouTube content saturates the market" assumption — the belief that because Indian YouTube has thousands of free tutorials on every stack, there is no demand for paid teaching. This is empirically wrong. The free content is mostly low-quality; the paid content at the right quality and right segment has stronger demand than the free content at scale, because quality teaching is rarer than the noise suggests. Indian engineer-founders consistently underestimate this and consequently underprice into the same market the American course creators charge 1,499 in. The INR-vs-USD pricing flinch on knowledge products — I'll charge ₹999 for the Indian audience and <span class="katex-inline" data-formula="99 for the international audience, where the Indian audience is correct for some segments and the">99 is the same flinch that produces $9-a-month SaaS pricing. The "course-creator industry is gross" reaction stacked with the additional cultural layer that public visibility for teaching feels like self-promotion, which is among the most powerful flinches in the cultural stack alongside the salary anchoring. The reframe is the same reframe in both cases: visibility for teaching that produces real learning outcomes is not self-promotion; it is doing the work in public.

These patterns persist not because engineers can't do the teaching-economics math, but because the cultural inputs to the math are uncomfortable to confront. I do not actually know if my docs teach. I do not actually know what my course is worth. I do not actually know whether my conference talk would land. I do not actually know whether I am an authority or whether I just feel like one because I have downloads. The discomfort is the obstacle. Once it is named, the math is the easy part.

Why Engineers Are Uniquely Positioned to Teach #

Here is the move that makes everything afterward easier.

Teaching skill is engineering reliability applied to learner cognition and audience economics.

This is not a metaphor. The substrate is identical.

Backwards Design is API-first design for teaching. You specify what the learner should be able to do (the contract); you specify the evidence that would confirm they can do it (the test); you specify the assessment that produces the evidence (the integration test); you build the minimum content scaffolding that prepares them to pass (the implementation). Forward-designed teaching — the engineer's default of here is what I know, let me write it down — is the equivalent of building a database first and hoping the API emerges. It usually doesn't. Engineers who have architected from contract backward to data model already have the cognitive shape of Backwards Design; the work is to apply the shape to teaching.

Cognitive science of learning is production observability for the learner's brain. Retrieval practice is did the lesson actually persist in working memory? — a read query against the learner's mental cache. Spaced repetition is is the working-memory write being consolidated to long-term storage? — scheduled refreshes against the Ebbinghaus forgetting curve, which has the same shape as cache TTL. Interleaving is the discipline of varied input — the equivalent of fuzzing the learner's emerging mental model so it generalizes rather than memorizes. The cognitive-science evidence base from Brown, Roediger, and McDaniel (Make It Stick, 2014) is solidly empirical and mostly ignored by practitioners, the same way distributed-systems literature is solidly empirical and mostly ignored by application developers. Engineers who can read distributed-systems papers can read cognitive-science papers; the substrate is similar; the discipline is to read with intent.

Retention engineering is cache invalidation. The Ebbinghaus forgetting curve specifies the cache TTL: the learner forgets roughly 50–80 percent of new content within 24 hours unless retrieval practice intervenes. Spaced repetition is scheduled cache refresh at the optimal intervals (Wozniak's SuperMemo / Anki implementations). Retrieval practice is the read that confirms the cache is warm. Engineered teaching builds these into the design; un-engineered teaching offloads retention to the learner ("review this on your own time") and consequently produces low retention. The engineer who designs cache strategies for production systems can absolutely design retention strategies for learner brains; the move is to apply the same engineering rigor to the cognition layer.

Assessment is integration testing for the learning. If the assessment is a quiz, you've shipped a unit-test-only project to production — you know the individual functions work, you do not know the system works. If the assessment is a deliverable that demonstrates the learner can do the thing in real conditions, you've shipped end-to-end tests that confirm the system works. The empirical evidence on this is brutal: quizzes are weak signals of real-world capability; deliverables are stronger signals. Most courses default to quizzes because they are easy to grade; the engineer with integrity refuses the easy grading and designs for actual capability.

Curriculum architecture is system design for learning paths. Modules are services. Dependencies are call graphs. Checkpoints are SLOs. Time estimates are capacity planning. Most knowledge products fail at this level even when individual lessons are good — the path from lesson 1 to lesson N does not actually produce a competent practitioner because the dependency graph is wrong, the prerequisite documentation is missing, the time estimates are sandbagged. Engineers who have designed distributed systems with real constraints have already done this shape of work; the discipline is to apply it to the curriculum layer.

Pricing knowledge products is admission control for the audience-relationship. Below a price floor, the learner does not value the content and does not ship the deliverables — same shape as a rate limiter that lets through too much traffic and consequently degrades service quality for everyone. Above a ceiling, conversion drops below sustainable. The skill is finding the band, not the single price. Engineers who have implemented rate limiters know exactly this shape of decision.

Format selection is capacity planning under real constraints. Live cohorts are high-touch, high-cost, high-pricing-power. Self-paced is low-touch, low-cost, low-pricing-power. Hybrid is the realistic middle. The decision is constraint-driven: your time, your timezone, your audience, your subject matter. The decision is the same shape as choosing among synchronous and asynchronous architectures for a distributed system; you make the choice based on the actual constraints, not on preference.

The lens is what makes the curriculum unusually accessible to engineers. Most teaching content reads as fluffy because it does not speak the language of someone who has architected real systems. You can. The teaching discipline is engineering discipline applied to learner cognition and audience economics. The math is not harder; the squeamishness is.

The lens has a known limit: the assessment-design work is the place where engineering thinking is partly the obstacle. Engineers default to "build a quiz" because quizzes are deterministic and easy to grade. The empirical reframe is uncomfortable: course-graduate-shipped-real-thing is the success metric, not course-completed-quizzes-passed. The reframe makes the course harder, the completion rate lower, the satisfaction scores worse. It also makes the reputation of the course dramatically better over time, because graduates ship things and recommend the course based on outcomes rather than vibes. The discomfort is the diagnostic that you are doing the right work.

The Two Failure Modes #

Most engineers who do attempt teaching collapse into one of two failure modes. Both are partly true, mostly wrong, and the curriculum is built to refuse both.

The academic-pedagogy register. Reads Wiggins & McTighe at full depth, internalizes the K-12 instructional-design canon, attempts to apply Bloom's taxonomy and rubrics-derived-from-state-standards to a web-framework course. Concludes that to do teaching correctly, you need an instructional designer, formal validation, peer review, university-level rigor — none of which is operationally available to a working engineer writing a course in evenings between the day job and family time. The frameworks are not wrong inside their target configuration. The frameworks are wrong for you because you are not inside that configuration. The academic-pedagogy register, applied to working-engineer reality, produces theater — instructional-design documents written for an audience that does not exist, learning-objective taxonomies that map to nothing the learner will actually do, assessment rubrics written for credentialing rather than for capability. The work looks like teaching; the output is bureaucracy. Most engineers who try this give up at week three because the overhead is unsustainable.

The academic-pedagogy register is partly right about something real: structured pedagogy beats unstructured teaching, and the academic canon contains the structure. The academic-pedagogy register is wrong about almost everything else, including which of the structures actually apply at your scale.

The course-creator-cult register. Reads no academic content, dismisses the canon as bureaucracy, models off Justin Welsh / Tiago Forte / Ali Abdaal, builds the course as a personal-branding exercise, ships, observes the result, calls early traction "social proof," doubles down on the personal brand. Sometimes this works at the level of revenue — when the engineer happens to have a real audience and a real product underneath, the cult-flavored marketing produces real income for a year or two. The cult register is partly right that excessive process kills speed.

The cult register is wrong about almost everything else. It produces engineers who ship courses that get bought and not finished. It produces the audience-quality decay where the followers who arrived for substance leave when the substance turns out to be thin and only the hype-followers stay. It produces the learning-outcome failure mode, because no pedagogy was ever read and the course consequently teaches engagement-with-the-creator rather than capability-in-the-subject. The cult register is the failure mode that produces the "$5M solo creator" claims that turn out, on inspection, to be revenue-mostly-from-cohort-driven-FOMO and not sustained value delivery.

The honest middle is what the curriculum trains. Brown / Roediger / McDaniel evidence base + Wiggins & McTighe Backwards Design discipline + Wes Kao cohort-economics math + April Dunford positioning + patio11 essay-as-business framing + the pricing math from this curriculum's M11. The substantive parts of the cognitive-science canon, the substantive parts of the cohort-economics canon, the substantive parts of the writing-online-compounds canon, refused at the cult-content boundary in each. Not academic theater; pedagogy with engineer-economics. Not creator-cult engagement-farming; teaching that produces real capability priced at real value. Not pure-academic register; not pure-cult register. The honest middle, which is harder than either pure position because it requires sustained calibration across years.

The honest middle is the discipline of choosing which frameworks apply at your scale, applying them with intent, and refusing both the bureaucratic-overhead failure mode and the cargo-cult failure mode.

This is the work.

What This Costs You, Specifically #

This section is for you — the engineer with teaching surfaces already operating accidentally. The worked configuration below is common among senior engineers who build in public: an open-source library with real adoption, a course platform or content site, a personal curriculum collection, a side teaching tool, a blog. Substitute your own inventory; the audit transfers.

The library with real adoption. Suppose you maintain an open-source library at 20,000+ downloads — sixty releases, a few thousand tests. By surface area alone, it is plausibly the highest-leverage teaching surface in your portfolio. The README and release notes reach more engineers per quarter than any conference talk you could plausibly give. The teaching question for this surface is not do the docs cover the facts (they probably do); the question is do the docs teach the engineer reading them how to think about the problem domain, when to use what, why the design choices were made, what the tradeoffs are. The honest answer for most maintainers is partly. The release notes record what changed. They mostly do not teach. The hours that would convert the release notes from records into teaching artifacts — four to six hours per release, applied across the next twenty releases — are hours that would compound into a quality of authority signal that the current state of the docs does not produce. The career and revenue cost of not doing this work over the next five years is not a number anyone can quote precisely; it is a credible fraction of the difference between known engineer with a popular library and recognized authority whose library teaches a generation of engineers how to think about the domain.

The course platform. Suppose your platform holds ~50 courses, ~100 projects, ~400 lessons — the single largest teaching surface in your portfolio by content volume, built organically over years. The pedagogy embedded in it has likely never been audited against the cognitive-science evidence base. The Backwards Design retrofit has not been run on any course. The retention engineering has not been built. The assessment design has not been re-engineered for deliverable-over-quiz. The pricing audit has not produced an anchoring-honest answer. The platform is the surface where a 12-month deliberate-teaching investment would produce the largest single change in your revenue trajectory — and that investment has not been made because the surface is treated as content-platform-that-grew rather than as the largest knowledge product you own.

The personal curriculum collection. Suppose you have built a large personal curriculum collection — structured guides, reading paths, module maps you wrote for your own mastery. The collection is itself a teaching product, whether or not it is ever published externally. The honest audit, treated unsentimentally in TEACHING_FOR_THE_CURRICULUM_AUTHOR.md: most personal curricula state outcomes weakly, almost none have validated assessments, the retention layer does not exist, the format-decision question has not been answered. The collection teaches if you read it; for any other learner, it is unproven. The decision tree on whether and how to publish externally is in that applied-vertical file; the cost of not having that decision tree run is one of the larger open questions in the next 24 months of your teaching strategy.

The side teaching tool. The learning tool, starter kit, or interactive tutorial you built once and never positioned. Same shape as the others: surface area exists, deliberate operation does not, and its strategic position relative to the platform and the collection has never been specified.

The blog at a few hundred followers. The smallest teaching surface and the one with the strongest writing-online-compounds research underneath it. Two-year half-life on a quality engineering post. Five-to-fifteen recruiter inbounds across two years per substantive post. The blog at a few hundred followers and irregular cadence is producing approximately zero of the compounding the same blog at 1-2 substantive posts per quarter would produce. The cost of the irregular cadence over the next five years, in recruiter-inbound terms, is in the dozens; in consulting-rate-increase terms, plausibly in the tens of thousands; in career-optionality terms, larger again, because the irregular cadence is what kept the blog at a few hundred followers instead of at 5,000 across the same time window.

The anchoring math, applied to the knowledge-product layer. The flinch that anchors salaries low anchors knowledge-product prices low — this curriculum's M11 attacks the same root cause that the Negotiation, Money & Wealth, and PM & Discovery curricula attack from their own angles. The math transfers directly: 499 across 100 customers across 5 years per knowledge product is $200,000 of foregone revenue per product. Across a portfolio of 3-5 knowledge products over the next decade, this compounds to seven figures. The math in M11 is run with specifics; the seven-figure number is not hyperbole. It is the standard lifetime-compound-cost shape applied at the knowledge-product price point.

The mission-funding implication. If you have goals bigger than a career — the kind that need funding across decades rather than quarters — knowledge products are a credible second leg of revenue alongside whatever you build and sell. The history of self-funded missions includes speaking-and-teaching as a recurring funding mechanism. Teaching is not orthogonal to goals that size; it is potentially load-bearing for them. The fourth pattern (the engineer who teaches deliberately) is the configuration where teaching surface area becomes mission-funding surface area in the second decade. The cost of staying in pattern two (accidental-teach) is the cost of those goals being underfunded for the lifetime of the work.

The cost compounds invisibly per teaching surface. Catastrophically across a 30-year career.

This is not a moral failing. This is the standard engineer default, applied to a teaching layer that no prior curriculum or certificate in most engineers' trajectories has operationalized. The cure is the same cure for any underinstrumented system: name what is happening; build the model; run the diagnostics; calibrate the responses; iterate.

The cure is the rest of this curriculum.

Closing: The Fourth Pattern #

What it looks like to be the engineer who teaches deliberately.

Not the LinkedIn influencer. Not the academic. Not the never-teach engineer at year ten with no transferable authority. Not the accidental-teach engineer at year ten with a library and a blog and no integrated strategy. Not the performative-teach engineer at year ten with 50,000 followers and no real authority underneath. Not the cert-completer who has the frameworks in memory and has never re-engineered a real teaching surface under real time pressure.

The integrator. The engineer who has built a working model of every teaching surface they operate, who knows the topology of the library docs and the curriculum collection and the blog and the course platform and the side tools, who has classified each by career-compounding signal, revenue-compounding signal, learner-outcome signal, and the gap between current state and deliberate state. Who has decided which surfaces deserve the next year of focused investment and which ones do not. Who has applied the cognitive-science evidence to the highest-leverage one. Who has run the Backwards Design retrofit on the highest-priority course. Who has built the retention layer on the most-used artifact. Who has run the pricing audit and made the anchoring-honest decision. Who has refused the academic-pedagogy register and the course-creator-cult register and lives in the honest middle, which is harder than either pure position because it requires sustained calibration across years.

Goals bigger than a career require this. Goals that size are not solo missions. They happen inside organizations and across decades, funded by capital that has to come from somewhere — and the somewhere, for the engineer-founder, is the studio. The studio is funded by products that sell and knowledge products that sell. Knowledge products that sell are downstream of teaching skill. Everything downstream of revenue is downstream of products and knowledge products that sell. The chain runs through this curriculum.

The next twenty years require this. The career compounding from the conference talks that land, the blog posts that produce inbound, the library docs that establish authority, the internal sessions that make you the team's de facto staff engineer. The revenue compounding from the course platform re-engineered, from the curriculum collection eventually published, from the cohort offering layered on top, from the corporate-training format for B2B. The compounding from the second decade where the revenue mix produces the discretionary capital that goals bigger than a career require to be funded. Every step in that trajectory is a series of teaching decisions: what to write, what to charge, what to retire, when to expand, when to compress. The decisions cumulate. The cumulative is whether the work reaches the revenue that funds the goals or stays in pre-revenue indefinitely.

The cost of not having this curriculum is the cost of the next twenty years of teaching decisions made on intuition rather than discipline. The cost compounds invisibly per surface, per quarter, per year. The cure is the curriculum.

The work is learnable.

The work is engineering work — observability, capacity planning, adversarial input testing, admission control, graceful degradation, applied to learner cognition and audience economics. You have done this work, in different shape, for years already. The shift is to do the same work on the teaching layer of the career you are building.

The naming is the first move. Then the audit. Then the cognitive-science layer. Then the Backwards Design retrofit. Then the retention engineering. Then the assessment redesign. Then the curriculum architecture. Then the pricing audit with the anchoring-honest math. Then the format decision. Then the next quarter's repetition of the cadence.

Teach deliberately.

Run it.

Companion essay to the full Craft of Teaching curriculum. The integrator-archetype-applied-to-teaching is the curriculum's central frame. The work was always engineering work. The work was always there.