TECHNICAL FLUENCY CURRICULUM
DARK 30 VENTURES // FOUNDER CURRICULUM

TECHNICAL FLUENCY

6 weeks · 18 modules · From framework user to framework builder
THE PROBLEM YOU ARE SOLVING

FROM READING THE SHEET MUSIC TO PLAYING BY EAR

You built the system. You ran it in front of a CTO. It worked. But you felt the gap between having the answer on a page and owning it in your mouth.

This curriculum closes that gap. Not by memorizing more — by understanding the primitives so deeply that you can explain them from any angle, under any constraint, without looking at anything.

The goal is not "I built this." The goal is "I understand why every piece exists and what breaks if you remove it."

// THE TEST

When someone changes one assumption mid-sentence, you don't freeze, you don't hedge, you restructure the explanation in real time because you understand the system at the component level.

CURRICULUM MAP

6 WEEKS, 3 TRACKS

TRACK A
SYSTEM THINKING
TRACK B
VERBAL DELIVERY
TRACK C
LIVE BUILD
Wk 1-2
Primitives
Wk 1-2
Record + Review
Wk 1-2
Rebuild from zero
Wk 3-4
Composition
Wk 3-4
Constraint drills
Wk 3-4
New vertical
Wk 5-6
Architecture
Wk 5-6
Live sparring
Wk 5-6
Ship to prospect
WEEKS 1-2

PRIMITIVES

Understand every building block in isolation. No frameworks yet.
01
What Is a Signal?
The atomic unit of the entire system
45 min CONCEPT
▶
Core concept
A signal is a public event that predicts a future need. Not a lead. Not an intent score. A structural change in the world that creates demand for something specific. A permit filing predicts crane demand. A PE acquisition predicts onboarding standardization. A training coordinator posting predicts that informal systems broke.
The difference between a signal and a data point: a data point tells you what happened. A signal tells you what will happen next.
EXERCISE 1 — IDENTIFY 5 SIGNALS
Pick any industry you know. Write down 5 public events that predict someone will need to buy something. For each one, write: what is the event, what does it predict, and how far in advance does it predict it.
I can explain what a signal is without using the word "lead"
I can explain why a PE acquisition is a signal for onboarding software
I can name the CraneGenius signal that predicts crane demand 12 weeks out
TEST YOURSELF
Someone says "We use Bombora for intent data." What is the difference between intent data and signal data? Say it out loud in under 30 seconds.
Reveal answer
Intent data tells you someone visited a competitor's website. Signal data tells you someone filed a building permit, got acquired by PE, or posted a training coordinator role. Intent is behavioral inference. Signal is structural demand detection. Intent says "they are interested." Signal says "they will need this in 12 weeks whether they know it yet or not."
02
What Is a Scoring Model?
Turning raw signals into ranked priorities
60 min CONCEPT + CODE
▶
Core concept
A scoring model takes a signal and asks: how likely is this to become a real opportunity? The CraneGenius model scores 0-12 based on keyword matches against crane-demand patterns. The Trainual model scores 0-10 based on how strongly the signal predicts someone needs onboarding documentation.
Key insight: the scoring model is wrong on day one. That is fine. It self-corrects with outcome data. Every closed deal teaches the model which scores predicted real buyers. The model you ship is the hypothesis. The outcomes are the experiment.
EXERCISE 2 — BUILD A SCORING MODEL ON PAPER
Pick one of the Trainual verticals (HVAC). Write down 5 criteria that predict a company needs Trainual. Assign point values to each. Now score 3 real companies against your model. Does the ranking feel right? If not, adjust the weights. You just did what the outcome tracker does automatically.
I can explain why a sales rep posting scores higher than an operations manager posting
I can explain what "cold start problem" means for a scoring model
I can explain how outcome data fixes scoring accuracy over time
EXERCISE 2B — READ THE ACTUAL CODE
Open src/score_filter.py. Read the score_and_filter function. Trace one record through the scoring loop. Which keywords match? How does the recency boost work? What does the ICP vertical modifier do? Write a 3-sentence summary of what the function does.
I can open score_filter.py and point to where keywords are matched
I can explain what the 12-point cap does and why it exists
03
What Is a Pipeline Stage?
Input, transform, output. The universal pattern.
45 min CONCEPT + CODE
▶
Core concept
Every stage in the pipeline follows the same pattern: take input from the previous stage, apply a transformation, write output for the next stage, log what happened. Stage 1 ingests. Stage 2 normalizes. Stage 3 scores. Each one is independently re-runnable. If Stage 4 fails, you fix Stage 4 and re-run it — you don't re-run Stages 1-3.
This is the TLAMI pattern in miniature. Every stage has a Trigger (previous stage output exists), Logic (the transformation), Action (write output), Monitor (log counts and errors), and Iterate (improve when outcomes show the logic is wrong).
EXERCISE 3 — TRACE A RECORD
Pick one row from data/scored_records.csv. Trace it backward: what did it look like after Stage 2? After Stage 1? Now trace it forward: what will Stage 3b add? What will Stage 4 add? Draw the transformation on paper. One record, nine stages.
I can name all 9 stages in order without looking
I can explain what changes between Stage 3 and Stage 3b
I can explain why Stage 3b is "the moat"
04
What Is TLAMI?
The universal framework for any automation
30 min FRAMEWORK
▶
Core concept
Trigger → Logic → Action → Monitor → Iterate. Every system you build follows this pattern. Every answer you give in a technical interview can be structured this way. If someone asks "how does X work?" you answer: what triggers it, what logic runs, what action it takes, how you monitor it, and how you improve it.
EXERCISE 4 — TLAMI ANYTHING
Apply TLAMI to three things that are not software: (1) how a restaurant handles a reservation, (2) how a thermostat works, (3) how your morning routine works. Write T/L/A/M/I for each one. The point is: TLAMI is not a tech framework. It is a thinking framework.
I can TLAMI any workflow in under 60 seconds without writing anything down
I can TLAMI each CraneGenius pipeline stage from memory
05
What Is an Outcome Loop?
How systems learn from their own results
45 min CONCEPT + CODE
▶
Core concept
A system without an outcome loop is a static tool. A system with an outcome loop is a learning machine. The outcome tracker logs what happened after each signal: meeting booked, deal opened, closed won, no response. The correlation analysis finds which signal attributes predicted which outcomes. The scoring weights update. The next run is more accurate than the last.
This is the compound advantage. Every customer you close makes the system more accurate for the next customer. A competitor starting today has zero outcome data.
EXERCISE 5 — LOG A FAKE OUTCOME
Run: python3 -m src.outcome_tracker --log --signal_id TEST-001 --outcome meeting_booked
Then run: python3 -m src.outcome_tracker --report
Read what it outputs. Understand what each line means.
I can explain the outcome loop without saying "machine learning" or "AI"
I can explain why the moat compounds over time
06
Verbal Drill: Explain Each Primitive
Record yourself. Listen. Improve.
90 min DELIVERY HARD
▶
The method
Open Voice Memos on your phone. For each primitive (signal, scoring model, pipeline stage, TLAMI, outcome loop), record yourself explaining it in exactly 30 seconds. Play it back. Listen for: filler words (um, like, basically), hedging (kind of, sort of, I think), and vagueness (stuff, things, the system does a lot). Re-record until the explanation is clean.
EXERCISE 6A — THE 30-SECOND DRILL
Record 5 explanations, 30 seconds each:
1. What is a signal? (not a lead, not intent data)
2. How does scoring work? (keywords, weights, outcome correction)
3. What are the 9 pipeline stages? (name them, don't explain them)
4. What is TLAMI? (apply it to one CraneGenius stage)
5. What is the outcome loop? (why the moat compounds)
All 5 recorded with zero filler words
I can deliver each one without looking at any notes
EXERCISE 6B — THE COLD OPEN
Record yourself delivering the 60-second Trainual spiel (the 4-beat version from the walkthrough) from memory. Do not read it. Do not look at the page. Just talk. Play it back. Time it. It should be 50-70 seconds. If it is over 90 seconds, you are explaining too much. Cut a sentence from each beat.
Delivered from memory in under 70 seconds
Zero hedging language ("I think," "kind of")
WEEKS 3-4

COMPOSITION

Combine primitives into systems. Handle constraints.
07
System Design from First Principles
Given a problem, design the signal system from scratch
2 hrs BUILD HARD
▶
Pick a company you have never built for. Example: Gusto (payroll for SMBs). Design a signal system for them on paper. What public events predict a company needs Gusto? (State new-business filings, job postings for "office manager" or "bookkeeper", SBA loan approvals.) What is the scoring model? What is the 9-stage pipeline?
EXERCISE 7 — DESIGN FOR GUSTO
On paper, write:
1. Three signal classes for Gusto
2. Scoring criteria (0-10) for each
3. The 5-step system (signal → score → tier → outreach → brief)
4. One outreach line per signal class
Do not look at CraneGenius or Trainual while doing this. The pattern should come from memory.
Designed without referencing any existing system
Scoring criteria are specific, not vague
Outreach lines reference the specific signal event
08
Constraint Drills
Practice restructuring when assumptions break
60 min DELIVERY HARD
▶
This is the Arash test. You are mid-explanation and someone changes a constraint. The drill: start explaining the CraneGenius pipeline. After 30 seconds, read one of the constraint cards below. Pause for 2 seconds maximum. Then restructure your explanation to account for the new constraint. Record yourself. Review.
CONSTRAINT CARDS — Pull one at random
1. "The data is wrong half the time."
2. "You have no CRM."
3. "You have 48 hours to ship."
4. "The sales team refuses to use the report."
5. "Budget is $0 — no APIs, no paid tools."
6. "You can only use one signal source, not seventy."
7. "The scoring model has never been validated."
8. "A competitor launched the same product yesterday."
I can handle any constraint card with less than 3 seconds of silence
My restructured answer still has clear logic, not just hand-waving
09
PSR Story Delivery
Tell each story at 1-min, 3-min, and whiteboard depth
90 min DELIVERY
▶
You have three stories: Tractian Stage Gates, Tractian Two Mountain, CraneGenius. Each one needs to be deliverable at three depths: 1-minute version (elevator), 3-minute version (interview answer), and whiteboard version (walk them through the actual system). The skill is not knowing all three — it is switching between them fluidly based on how much time you have.
EXERCISE 9 — DEPTH SWITCHING
Set a timer. Record yourself telling the Tractian Stage Gates story in exactly 60 seconds. Then record the 3-minute version. Then draw the whiteboard version on paper while talking through it. Do the same for Two Mountain and CraneGenius. Nine recordings total.
1-min versions are all under 70 seconds
3-min versions add depth without repeating the 1-min version verbatim
Whiteboard versions can be drawn while talking simultaneously
10
Build a New Vertical from Scratch
Prove the pattern is portable by building one
3 hrs BUILD HARD
▶
Pick a vertical not in the Trainual system: restaurants, fitness studios, or auto dealerships. Build the full signal system: identify 3 signal classes, define the source ingestion logic (even pseudocode), define scoring criteria, write outreach lines, and generate a mock report. If you can do this for a vertical you have never thought about, the pattern is yours, not just a thing you memorized.
EXERCISE 10 — NEW VERTICAL BUILD
Vertical: Restaurants (for a POS/ordering system like Toast)
1. What 3 public events predict a restaurant needs new POS?
2. What are the scoring criteria?
3. Write 3 outreach lines specific to each signal
4. Generate a mock 5-signal report by hand
Bonus: actually build it as a new directory in src/
Completed without referencing the Trainual or CraneGenius code
WEEKS 5-6

ARCHITECTURE

Think at the system level. Teach it to someone else.
11
Failure Mode Analysis
What breaks, why, and what you do about it
60 min ANALYSIS
▶
EXERCISE 11 — BREAK THE SYSTEM
For each pipeline stage, answer: (1) What is the most likely failure? (2) How would you detect it? (3) What is the fallback? Write it as a table: STAGE | FAILURE MODE | DETECTION | FALLBACK. This is what a CTO-level operator knows that a user does not.
I can name the failure mode for every stage without looking
I can explain the Indeed RSS deprecation and what I did about it
12
Teach It to Someone
The ultimate test of understanding
60 min DELIVERY FINAL
▶
Find someone who knows nothing about CraneGenius. A friend, a family member, anyone. Explain the entire system in 5 minutes. Then ask them to explain it back to you. If they can explain signals, scoring, and the timing advantage in their own words, you have achieved technical fluency. If they can't, the gap is in your explanation, not their understanding.
EXERCISE 12 — THE TEACH-BACK TEST
1. Explain the CraneGenius system to a non-technical person (5 min)
2. Ask them to explain it back (2 min)
3. Note which parts they got right and which they couldn't explain
4. The parts they couldn't explain are the parts you don't actually understand yet
5. Re-study those parts and repeat
The listener can explain the timing advantage back to me
The listener can explain why outcome data makes the system better
I did not use jargon they had to ask about
EVERY DAY

THE DAILY PRACTICE

15 minutes. Non-negotiable.

5 minutes — Read one module from the codebase. Open a file in src/. Read one function. Understand what it does. Can you explain it in one sentence?

5 minutes — Record one explanation. Pick any concept. Record yourself explaining it in 30 seconds. Play it back. Was it clear? Was it confident? Were there filler words?

5 minutes — One constraint drill. Start explaining something. Pull a random constraint card. Restructure on the fly. Record it. This is the muscle that makes you dangerous in conversations.

// THE STANDARD

You are done when you can explain any part of the system, from any angle, under any constraint, without looking at anything.

CRANEGENIUS // CAPEXLAYER // DARK 30 VENTURES // CONFIDENTIAL