Learning plan · v1

Howard learns Claude by building Markster

Six weeks, 15–20 hours a week, two tracks. On the technical track (weekdays) you build the real Markster product with Claude doing the heavy lifting, learning to give good instructions and challenge the answers. On the business track (Saturday mornings) you learn to analyze a business idea the way it's done at BONO — need-first, scored, with a verdict. By week 6: a working Markster app on a real URL, and a full business verdict on it — plus one on an idea of your own.

6weeks

15–20hper week

76tasks on two tracks

2verdicts: Markster + your idea

The one principle

You are not learning to code. You are learning to direct — the way a producer directs a film crew. Claude writes every line of code, sets up every account step by step, and explains every concept as many times as you want. Your job is to know what you want, say it completely, and never accept an answer you don't understand. That skill is the whole curriculum.

Two tracks, one plan

⚙️ Technical — how Claude works

Monday to Friday. Artifacts, Projects, Claude in Chrome, Claude Code, git, databases, deployment — learned by building Markster for real. Every task trains the same two muscles: complete prompts in, challenged answers out.

64 tasks · ~87h · the app ships in week 6.

📈 Business — how Markster wins

Saturday mornings, ~2.5h. The business-idea analysis method Bogdan built at BONO: reframe from product to need (Jobs to Be Done), talk to customers without polite lies (the Mom Test), score the problem and the market (Y Combinator, Sequoia, MIT), audit the unfair advantage, run the profit math and the Revenue Reality Check — and end with a scored verdict: GO / REFINE / PIVOT / PARK. You learn it on Markster; you graduate by running it on an idea of your own.

12 tasks · ~15h · the full method is in the Business tab.

👉 Start here

Read the Markster tab — what we're building, for whom, and your role in it.
Read the Question Playbook tab — the prompt recipe and the 12 questions you'll use every single day. You'll come back to this page constantly.
Open The 6 Weeks to see the shape of the journey, then go to All Tasks and start with T01. It takes 30 minutes and ends with something on your screen that you made.
One more thing: anywhere in this plan, a dotted word is clickable — tap it for a plain-English definition and an example. The full list lives in the Vocabulary tab.

🔌 How this plan runs in practice

Where the plan lives: right now at hlb.bgn.ro, hosted by Bogdan. That page also holds your two downloads: the Markster PRD in English (you attach it in T02 and upload it to your Project in T03) and the plan as a text file (the version Claude reads best). Click each link once, and save both into a folder called Markster on your computer — everything you will ever attach comes from that folder. In week 6 you buy a domain of your own, move the plan there (T55), and publish your learning journal at a subdomain (T59).
Your Claude can read all of this: in T03 you upload this plan and the PRD to your Markster Project's knowledge base — from then on, ask your Claude anything about any task, term, or prompt in here: "what does T28 want from me?", "explain this week's challenge again". You can also just give any Claude the hlb.bgn.ro link and ask it to read the plan.
Ticking tasks: in weeks 1–3 you click the checkboxes in your tracker yourself. From week 4 (T41), Claude Code can tick them for you through the tracker's own API — but only after you explicitly say so. That rule gets written into your project's rulebook, in your own hand.
What you need: a computer with Chrome and a Claude subscription (Pro is fine to start; weeks 4–6 lean hard on Claude Code — if you hit usage limits, upgrading to Max fixes it). Every other account — Supabase, Vercel, Loom, your domain — gets created along the way, free or cheap, with Claude guiding every click and you typing every password.

☀️ The daily ritual: one brave question

Every day, before your first task, ask Claude one question you'd be slightly embarrassed to ask a person. "What actually is a file?" "What's the difference between the internet and a browser?" "What does 'deploy' mean?" Claude has no eyebrows to raise. Ten minutes a day of brave questions, for 30 days, will do more for you than any course — because the questions you're embarrassed about are exactly the ones blocking you.

🅿️ The word parking lot

New words will arrive faster than they can be explained — that's by design, not an accident. The rule: an unknown word never blocks a task. If it's dotted, click it for the one-line idea. If it still feels foggy, drop it into your parking lot — a running note called "Parking lot" in your Markster Project. You empty it twice a week: Wednesday's brave-question hour and Friday's review both start with "Claude, here's my parking lot — explain each word using something I built this week as the example." By then you'll have used each word for days, and the explanation lands ten times better than it would have on day one.

How the training wheels come off

The plan has three levels. The point of the whole plan is to move you from level 1 to level 3.

Level 1 · Week 1

Prompts provided — and explained

Every task gives you the exact prompt to paste, plus a "Why this prompt works" note that takes the recipe apart. You're not just copying — you're studying the anatomy.

Level 2 · Weeks 2–3

Fill in the blanks

You get the skeleton of the prompt; the [HIGHLIGHTED PARTS] are yours to write. The blanks are always the parts only a human can supply: intent, taste, priorities.

Level 3 · Weeks 4–6

You write it yourself

Tasks give you a mission and a checklist of what your prompt must cover. A reference prompt is hidden behind a click — write yours first, then compare. The comparison is where the learning happens.

Every level · Every task

Challenge the answer

Every task ends with a 🥊 Challenge — a specific question that pokes at Claude's answer. Claude is brilliant and confidently wrong sometimes. Spotting the difference is a skill, and you'll practice it daily until it's a reflex.

What you'll be able to do at the end

💬 Write complete prompts

Context, goal, format, constraints — from scratch, without templates, for any task: writing, research, building, debugging.

🥊 Interrogate any answer

Ask for sources, assumptions, and confidence. Catch Claude when it's wrong, correct it without drama, and know when to verify at the source.

🎨 Build interfaces in Claude.ai

Artifacts, multi-screen prototypes, iterating without breaking what works. Your own tracker app with a real cloud backend.

🌐 Research with Claude in Chrome

Competitor analysis and API documentation extracted by Claude driving your browser — with you as the skeptical supervisor.

⚡ Command Claude Code

Real files, git save-points, plan mode, project rules in CLAUDE.md. The difference between chatting with AI and directing a builder.

🚀 Ship a real product

The Markster app — landing page, AI chat flow, pricing, dashboards, Supabase database — deployed to a public URL, plus your first custom skill.

A typical day (~3 hours)

10 min — the brave question ritual.
2 tasks — 1–2 hours each. Each task: do it, read the why, run the challenge.
5 min — tick the tasks off in your tracker app (you build it in week 1 and use it every day after).

When you're stuck: tell Claude you're stuck and paste exactly what you see — that's not cheating, that's the skill. If you're still stuck after 3 honest attempts, call Bogdan. Every Friday: a review call with Bogdan. From week 5: a short Loom demo to the team.

Context — read before T01

Markster: what we're building, and your part in it

📥 Get the PRD now: this tab is the summary; the full document is markster-prd-en.md — click to download it, and save it in your Markster folder. You will attach it to a chat in T02 and upload it to your Project in T03. (If the link does not work where you are reading this, ask Bogdan to send you the file — it lives next to this page on hlb.bgn.ro.)

One-liner: Markster is an AI platform that registers an EU trademark for a small business in 15 minutes instead of 2–4 weeks with a lawyer — for €939 all-in (€89 service + €850 EUIPO fee), versus €1,500–3,000 the traditional way.

Who it's for

A founder or owner of a small EU business — SaaS, e-commerce, an agency — who just settled on a brand name and wants it protected fast, without lawyer drama.

Job to be done: "Register my trademark across the EU without paying €2,000 and without chasing a lawyer who doesn't answer email."

Why it can win

EUIPO (the EU trademark office) publishes free, complete APIs for searching, classifying, and filing trademarks — and almost nobody in the market uses them. Markster combines those APIs with a conversational AI. Cheaper, faster, and 10× friendlier than the alternatives.

~197,000 EU trademark applications per year, and growing.

How it works — the customer's journey

Chat with the AI (3–5 min): name of the brand, what products/services it covers, logo or text-only.
Availability check (automatic): EUIPO's search API returns similar existing marks; the AI scores the risk 🟢 clear / 🟡 moderate / 🔴 high.
Classification (automatic): the AI suggests the right "Nice classes" (the official categories of goods and services); the customer confirms.
Checkout & filing: €89 + €850 EUIPO fee → the application is filed through the EUIPO API, with Iulia as legal representative.
Dashboard: the application moves through 12 statuses (Draft → Filed → Published → Registered…), with an email at every step.

Person	Role	What they mean for you
Isi	CEO / operations	Gets your Loom demos from week 5. The operator dashboard you build in week 5 is for her.
Iulia	Trademark expert, legal rep	The "expert review queue" screen you build is her workbench.
Bogdan	Product & strategy	Your reviewer and unblocker. Friday calls. Call after 3 stuck attempts.
Prodi	Professional developer	Takes what you build and hardens it for production. Your week-6 README is written for him.
Howard	Builder of the foundation	You build the working app: every screen, real data in a real database, with the EUIPO and payment parts mocked (realistic fakes). Real enough to demo, learn from, and hand over.

What "mock" means — your most important vocabulary word

Some parts of your app will be mocks: realistic fakes. The EUIPO search returns invented results; the payment button pretends to charge a card. Mocks are not a lesser version of the work — they're how professionals build: prove the experience first, connect the real services after. Whenever you ask Claude to build something, saying "mock" vs "real" is one of the highest-leverage words in your prompt.

The stack (the technologies — Claude handles all of them)

Claude.ai for chat, artifacts and Projects · Claude in Chrome for research · Claude Code for building the real app · Supabase as the database (free tier) · Vercel to put the app on the web (free tier) · Stripe / EUIPO APIs — mocked. You will create the free accounts yourself (you own them); Claude guides every click.

Keep this page open

The Question Playbook

Everything in this plan practices two skills: writing complete prompts and challenging answers. Here is the whole theory, on one page. In week 6 you'll write your own version of this playbook — that's the graduation exam.

The prompt recipe

A complete prompt has four parts. Not every prompt needs all four written out — but when a result disappoints you, the fix is almost always one of these four being missing.

Part	Answers the question	Example
Context	What does Claude need to know?	"I'm building Markster, a trademark registration site. Here's the PRD. I'm not technical — explain terms."
Goal	What exactly do I want, and why?	"A landing page whose only job is to get a visitor to click Start."
Format	What shape should the answer take?	"One HTML file. / A 10-row table. / Three options with trade-offs, then your recommendation."
Constraints	What must be true — or must NOT happen?	"Pink #E8386F, Inter font. Keep everything that already works. Don't use jargon. Under 200 words."

The golden rule: every detail you leave out is a decision Claude makes for you. Sometimes that's fine — Claude decides well. But when the result isn't what you pictured, don't say "wrong" — ask yourself which of the four parts did I leave out?

The 12 questions that always work

Learn these the way you'd learn phrases in a foreign country. Each one is attached to a task in the plan, so you'll practice all of them.

"What did you assume that I didn't tell you?" — reveals the invisible decisions.
"What's the weakest part of your answer?" — Claude critiques itself surprisingly well, but only when asked.
"Explain it simpler. Again." — repeat until it's actually clear. You are allowed to say "still not clear" five times.
"What should I have asked you that I didn't?" — the question that finds the questions.
"Give me 3 options with trade-offs, then recommend one." — turns Claude from an oracle into an advisor.
"Before you do anything — what's your plan?" — for any big request. Approve the plan, then let it build.
"What did you actually test, and what did you assume works?" — for every "done ✅". Done is a claim, not a fact.
"How confident are you, and what should I verify at the source?" — makes uncertainty visible. Numbers, dates, prices, laws: verify.
"Explain what was wrong before you fix it." — if the explanation doesn't make sense, the fix probably doesn't either.
"What did you change that I didn't ask you to change?" — after every edit to something that was working.
"What could go wrong if I share / publish / delete this?" — before anything irreversible or public.
"What would a skeptic say about this?" — for anything you're about to believe because you want it to be true.

The courage rules

No stupid questions exist here

Claude doesn't judge, doesn't sigh, doesn't remember you asked the same thing yesterday. The embarrassing question is always the most profitable one — it's the one that's been blocking you.

You can always start over

A conversation gone sideways costs nothing to abandon. Open a fresh chat, paste the last good version, ask for one thing. Starting over is a professional move, not a failure.

Being corrected is the product

When Claude corrects your understanding, you just learned something. When you correct Claude, you just practiced the most important skill in this plan. Both are wins; neither costs anything.

You are the supervisor

You never have to accept the first answer, approve a plan you don't understand, or keep up with a pace that's too fast. "Stop — explain that step first" is always allowed, forever.

Signs Claude might be wrong

Specific numbers with no source — prices, dates, statistics, legal fees. Claude states them fluently even when unsure. Ask question #8.
Too smooth — a perfectly confident answer to a genuinely hard question deserves question #2 or #12.
Agrees with your pushback instantly — Claude sometimes folds under pressure even when it was right. If it concedes too easily, ask: "were you actually wrong, or just agreeing with me?"
"Done ✅" on anything complex — ask question #7, then check yourself.

The journey

Six weeks, from first prompt to shipped app

Level 1 — prompts provided

Foundations & first wins

First artifact in 30 minutes. Read the PRD with Claude as your guide. Set up your Markster Project (Claude with memory). Build the tracker app v1 — an app with memory of its own. Learn the prompt recipe, practice being questioned, and catch Claude being wrong for the first time — on purpose.

📈 Business Saturday: the product-to-need reframe (Jobs to Be Done) and the Mom Test rules. You write Markster's job statement.

Deliverables: tracker app v1 · Markster Project with knowledge base · Markster's job statement · 11+2 tasks, ~16h

Level 2 — fill in the blanks

Prompt craft: landing page & research

Build the Markster landing page and make Claude its harshest critic. Set up Claude in Chrome and run real competitor research and an EUIPO API reconnaissance — then interrogate every source. Tracker v2 gets richer data and export. You start writing the key parts of every prompt.

📈 Business Saturday: the beachhead customer and the next-10-customers test, then a live Mom Test interview against Claude-as-customer.

Deliverables: landing page v2 · verified competitor table · EUIPO API map · tracker v2 · beachhead + interview practice · 10+2 tasks, ~16.5h

Level 2 — fill in the blanks

The clickable prototype

A three-screen Markster prototype: landing → AI chat → summary, with the traffic-light risk check. Debug Day teaches you what to do when things break (they will, on schedule). Tracker v3 moves to the cloud with Supabase — your first real backend. Cowork desktop gets installed.

📈 Business Saturday: scoring the problem with YC's six traits, picking Markster's Sequoia archetype, and answering "why now?" against a devil's advocate.

Deliverables: clickable 3-screen prototype · tracker v3 in the cloud + Bogdan View · problem scores + why-now · 10+2 tasks, ~17h

Level 3 — you write the prompts

Claude Code: commanding a builder

The big shift: from chatting in a browser to directing an agent that creates real files on your computer. Install Claude Code, learn git save-points (so you can always undo — and therefore always dare), write your project's rulebook (CLAUDE.md), use plan mode, and scaffold the real Markster app with the landing page and chat screen inside. On Friday, you teach Claude to tick your own tracker through its API — with your confirmation as the only law.

📈 Business Saturday: the unfair advantage audit ("a bit better" scores zero) and the pre-mortem: how Markster dies, ranked.

Deliverables: Markster app running locally · git save-points · CLAUDE.md · Claude ticking the tracker on command · advantage audit + pre-mortem · 11+2 tasks, ~17h

Level 3 — you write the prompts

Markster end-to-end

The app grows every screen from the PRD: summary & pricing, the 12-status state machine, client dashboard, mock EUIPO search, and the team's dashboards for Isi and Iulia. Supabase stores real application data. Friday is bug-hunt day, and your first Loom demo goes to Isi.

📈 Business Saturday: unit economics of one customer (the €89 question) and the Revenue Reality Check against the team's real targets.

Deliverables: full app flow with database · operator + expert screens · Loom demo #1 · unit economics + reality check · 10+2 tasks, ~17h

Level 3 — you write the prompts

Ship it

Polish pass, deploy to a real URL, and put down roots: buy your own domain, move this plan from Bogdan's site to yours, and publish your six weekly notes as a learning journal at learn.your-domain. Build your first custom skill (an email drafter for Markster's status notifications), write the handover README for Prodi, deliver the final demo, and write your own Question Playbook — the proof that the training wheels are off.

📈 Business Saturday: the verdict — Markster on the full scorecard, as a memo to Isi and Bogdan. Then the capstone: your own idea through the same machine.

Deliverables: Markster live on your own domain · learning journal · custom skill · README for Prodi · final Loom · your playbook · two business verdicts · 12+2 tasks, ~19h

Week	Focus	New Claude capability	Markster deliverable	Hours
W1	Foundations	Artifacts, Projects, prompt anatomy	Tracker v1 + knowledge base	~16
W2	Landing page + research	Claude in Chrome, critique loops	LP v2 + verified research	~16.5
W3	Prototype	Multi-screen artifacts, Supabase, debugging	3-screen prototype + tracker cloud	~17
W4	Claude Code	Files, git, CLAUDE.md, plan mode, APIs	Real app scaffolded	~17
W5	Full build	Database, verification habits	All screens + Supabase	~17
W6	Shipping	Deploy, own domain, skills, handover	Live URL + journal + skill + README	~19

The work itself

All 76 tasks

Two tracks in one list: weekday tasks are the ⚙️ technical track; 📈 Business tasks are the Saturday-morning business track (filter with the chips below). Click any task to open it. W1 gives you the prompt and explains it. W2–3 gives skeletons with [BLANKS] you fill in. W4–6 gives a mission and a checklist — write your prompt first, then peek at the reference. Every task ends with a 🥊 challenge: run it, don't skip it.

The method — read before B01, return every Saturday

How to analyze a business idea

This is the method Bogdan built at BONO for taking any idea from vague to verdict. The twelve Saturday tasks walk you through it on Markster; this page is the reference you come back to.

The philosophy: challenger, not validator

Most idea-analysis tools accept the idea as given and hand back a flattering report. This method does the opposite: it challenges the idea — starting with the one move that matters most: never start from the product; start from the need. "Who has a problem, and what is it?" If there's no need, the analysis finds that out fast, and that's a success, not a failure. Nine out of ten founders start from the product. The method exists to make you the tenth.

The six Saturdays

Week	Question	Frameworks	Tasks
W1	What's the need — and how do we talk to the people who have it?	Jobs to Be Done (Christensen) · Mom Test (Fitzpatrick)	B01–B02
W2	Who exactly is the customer — can we name 10?	Beachhead + DMU (Aulet/MIT) · Mom Test live	B03–B04
W3	Is the problem big, urgent, and well-timed?	6 problem traits (YC) · PMF archetypes (Sequoia) · why-now	B05–B06
W4	Why do WE win — and how could we die?	5 unfair advantages (YC) · pre-mortem · assumptions (Torres)	B07–B08
W5	Does the money work?	Unit economics, LTV ≥ 3×CAC (Aulet) · Revenue Reality Check (BGN)	B09–B10
W6	The verdict — and can you do it alone?	The full scorecard · GO/REFINE/PIVOT/PARK · your own idea	B11–B12

The scorecard

Five dimensions, 1–5 each. The score measures the quality of the thinking, not the size of the dream.

Dimension	What it asks
A · Need	Is it clearly defined, with evidence? How painful? How frequent?
B · Customer	How specific? ("Everyone" scores 1.) Can you reach them — can you name 10?
C · Market	Big enough for the ambition? Growing? Timed right?
D · Differentiation	What do you know that others don't? Why are today's alternatives not enough?
E · Profit formula	Who pays, how much, how do you reach them? LTV ≥ 3× CAC?

Deal-breakers — regardless of the total score, any of these means PARK: no identifiable need · the customer is "everyone" · no plan for how you reach customers.

The Revenue Reality Check

Accessible customers × conversion rate × price = year-1 revenue

Always bottom-up ("I can reach 2,000 founders through these channels"), never top-down ("just 1% of a huge market"). Then compare the estimate with the ambition: 🏠 Lifestyle (€50–300K/yr) · 📈 Growth (€300K–3M) · 🚀 VC-scale (€3M+). Every category is valid — the check exists so ambition and math meet before launch, not after.

Reference conversion rates: cold outreach 1–3% · content/SEO 2–5% · warm referrals 10–25% · partnerships 15–30%.

The verdicts

Verdict	Meaning	The action it demands
🟢 GO	Strong opportunity	Validate for real: customer interviews, Mom Test questions ready.
🟡 REFINE	Potential, pieces missing	Fix the weakest dimension first, then re-score.
🟠 PIVOT	The core is weak	Go back to the need and rebuild from there.
🔴 PARK	The opportunity isn't there	Say it out loud, keep the learning, spend the energy elsewhere.

An analysis that doesn't end in an action — an experiment, a fix, a turn, or a stop — is decoration. Every verdict comes with its next step attached.

Plain English, always

Vocabulary

Every technical term in this plan, in plain English, with an example. You don't need to study this page: everywhere in the plan, the first time a term appears in a section or task it's dotted like this — click it and the definition pops up right there. This page is the full list, for browsing and revisiting.

The rule that beats any glossary: if a word isn't here, or a definition doesn't land — ask your Claude. "What does X mean? Explain it simpler. Give me an example from Markster." That loop is better than any dictionary, and it's the whole point of this plan.

How we know it worked

Success criteria & honest boundaries

✅ At the end of week 6, all of these are true

Markster is live — a public URL anyone on the team can open, with the full customer journey working (EUIPO and payments mocked).
Howard writes complete prompts unaided — context, goal, format, constraints — verified by the week 4–6 tasks done without the reference prompts.
Challenging answers is a reflex — in the final demo, Howard can show three concrete moments where he caught or corrected Claude.
The tracker app tells the story — used daily for six weeks, synced to the cloud, visible to Bogdan — and from week 4, ticked by Claude only on Howard's explicit confirmation.
The plan and the journal live at Howard's own address — a domain he bought himself, the plan moved over from hlb.bgn.ro, and the six weekly notes published at learn.his-domain.
One custom skill exists and works — the Markster email drafter, built and tested by Howard in Claude Code.
Prodi can take over — the README honestly documents what's real, what's mock, and what he'd criticize.
The personal Question Playbook is written — Howard's own top questions, each with a story from these six weeks attached.
Howard can take a business idea from vague to verdict, alone — reframe to the need, customer, market, advantage, profit math, scorecard — proven by the B12 capstone on his own idea.
Two verdict memos exist — Markster on the full scorecard, delivered to Isi and Bogdan in plain language, and Howard's own idea, verdict accepted whatever it says.

Weekly checkpoints

Every Friday: review call with Bogdan — demo, best prompt of the week, one thing Claude got wrong.
Weeks 5–6: Loom demo to Isi & the team.
Stuck rule: 3 honest attempts with Claude, then call Bogdan. Never lose a day to being stuck.

If the pace is wrong

The plan bends, it doesn't break. Any task can take two sessions instead of one. Supabase too much in week 3? Stay on localStorage and do it in week 5. A 6-week plan finishing in 8 weeks is a success; a plan abandoned in week 2 because it felt like a race is not. Adjusting is not failing — say it out loud on the Friday call and re-cut the plan.

🚫 Honest boundaries — what this plan is not

Howard doesn't learn to write code by hand. He learns to read a little, and to direct a lot. That's the point, not a limitation.
Nothing real gets filed at EUIPO and no real money moves. All legal and payment operations are mocked. Iulia and Prodi own the real versions.
The app is a working foundation, not production software. Security hardening, real API contracts, and payment integration are Prodi's job, on top of Howard's foundation.
Accounts and access stay in Howard's hands. He creates every account (Claude, Supabase, Vercel, GitHub) himself, with Claude guiding — passwords are typed by Howard, never by Claude.