Hood Dev · a JAK Labs school
Every course tells you what to learn. None of them look under your hood.
A developer school that assesses how you actually work — then builds the track around what it found. Dev as a trade, not an academy.
Who it is for
The stuck mid-level developer. Not beginners.
You can build things. You have shipped. But you have been roughly the same developer for two years and nobody can tell you specifically why — only that you should “learn system design” or grind more LeetCode.
Hood Dev diagnoses the developer, not the technology. “Your React is Level 2” is what a course library sells. This answers “you cannot operate under ambiguity, and here is the exact moment you proved it.”
If you have never written code, this is not the place yet — it diagnoses developers, and with nothing to measure there is nothing to diagnose. I would rather say that now than take your time.
The Teardown · assessment structure
Seven phases, numbered from zero
You are graded on the path, not the artifact — in 2026 the artifact proves nothing, because a model could have made it.
Intake5 min · ungated
Where you are, what you have shipped, what you are trying to become.
The Written25 min · adaptive
Scenarios, not trivia. The questions get harder or easier based on what you have already shown.
The Bench60–90 min · read and repair
An unfamiliar codebase with real defects. Measures time-to-first-edit and how much you read before you touched anything.
The Build2–4 hours · ambiguity under load
A deliberately underspecified brief. What you put in QUESTIONS.md matters more than what you put in the code.
The Verification Bay30 min · the AI-native phase
AI use is allowed, expected, and itself measured. Nobody is proctoring abstinence. The question is whether you can direct, verify and correct a model — the actual 2026 competency.
The Readdelivered, not taken
Your type, your leaks, and every claim cited to something you actually did.
The Contract10 min
You accept the Work Order, or you do not. The track is generated from the findings, not from a catalogue.
Leaks · the weakness taxonomy
Type is orthogonal to competence. Leaks are what’s missing.
Nine of them, each scored 0–100 with evidence citations.
| # | Leak | Symptom |
|---|---|---|
| L1 | Fundamentals | Shaky on data structures, complexity, concurrency. Solutions work by luck. |
| L2 | Debugging | Guesses instead of observing. Changes several things at once. No hypothesis. |
| L3 | Delivery | Starts, doesn’t finish. Scope creeps until nothing ships. |
| L4 | Reading | Can only write greenfield. Freezes in an unfamiliar codebase. |
| L5 | Systems design | Can’t structure beyond a file. Coupling everywhere. |
| L6 | Production | Never operated anything. No deploys, monitoring, rollback, incidents. |
| L7 | Ambiguity & scoping | Needs a perfect ticket. Won’t interrogate a spec. Builds the wrong thing correctly. |
| L8 | Communication | Can’t explain a tradeoff to a non-engineer. Can’t demo. Can’t say no. |
| L9 | Verification | Trusts output that looks right. No skepticism of AI, or of self. |
L7, L8 and L9 are the ones no competitor measures at all — and they are exactly what separates someone who can be dropped into a real business from someone who needs a perfect ticket.
The blocking order
Hard rule: if fundamentals or debugging score above 60, your track opens with them regardless of what you wanted to learn. You cannot teach systems design to someone who cannot debug — they fail and blame the curriculum. This is why the engine sometimes tells an ambitious learner something they do not want to hear. Saying it plainly is the brand.
How it refuses to be astrology
Four rules, and none of them are negotiable
evidence or nothing
No statement in a Read exists without a citation to something you actually did. That is the difference between a mechanic’s estimate and a horoscope.
deterministic scoring
The model writes the prose, never the verdict. Given identical telemetry the engine returns an identical type and identical leak scores, forever. A test asserts the scoring path makes zero network calls.
unknown never flatters
Insufficient evidence returns INSUFFICIENT — never a generous guess. A confident wrong Read is worse than an incomplete one.
type is stable, leaks move
You are never re-typed — that would destroy the thing you identify with. You are always re-scored on leaks, every eight weeks or at each gate.
Ranks · evidence-based, never time-based
You cannot sit your way up this ladder
| Rank | Requirement | Unlocks |
|---|---|---|
| Rookie | Teardown complete, Work Order accepted | Discord, curriculum, your track |
| Apprentice | Fundamentals and debugging both below 40 | Advanced modules, peer review |
| Journeyman | Your project shipped to production with real users · verified leak deltas on re-assessment · a written scope doc · a 10-minute live defence of your own work | Bay Time — 1:1 with me on your project |
| Craftsman | You operated it: a real incident handled, iteration on user feedback, an uptime record | Recurring Bay Time, teaching privileges |
| Hood Dev | You did it for someone else — client or employer — end to end | Case study, referral pipeline, alumni tier |
Bay Time is earned, not scheduled. You cannot buy an hour of one-to-one. You reach it by shipping something real, and the session is about your project — never a curriculum walkthrough. Earned access gets chased; purchased access does not.
Where this actually stands
Read this before you sit it. The Teardown is live and the curriculum is 23 modules covering every leak — but a browser-only sitting measures 2 of the 9 leaks. Measured, not estimated. The Bench and the Build carry the rest and they are graded by hand, which takes your Read from roughly 2 of 9 to about 6 of 9.
So a Read from the browser alone is a sketch. It is a real, evidence-cited sketch rather than a personality quiz — but you should know which one you are getting. The instrument is early and being calibrated in the open, and the whole thing is free for 90 days from launch precisely because of that.
The person building it runs a live production business on software he wrote himself — $216K invoiced through it, one engineer. The curriculum is the thing I would have wanted at the point I was stuck.
More on the engineering side of what I do: jaklabs.io/engineering.
Start
Find out what you’re actually missing.
Forty minutes, in the browser, free. Or start with the zero-friction version: Aura Rank grades your repositories entirely offline — nothing uploaded, nothing transmitted — and hands you to the Teardown when you want the developer assessed rather than the code.
The school lives at hood.jaklabs.io. Signup, login and the assessment all happen there.