Investigation · Vault-9

The black box is waiting

First lesson: The second signal: Kotomarket's black box

0/83
lessons done
0· +10/lesson

Interactive course · Data Analysis

Investigate the data.
Become an analyst.

Python, pandas, product metrics, cohorts and statistics — on the Kotomarket archive. Code and charts run right in the browser: nothing to install.

Start for freePrologue and chapter 1 are free — 11 lessons
  • Python and statistics from zero
  • 11 investigations
  • an investigator's licence at the end
python · pyodide Run
# where exactly did revenue sag?
orders.groupby("platform")["revenue"].sum()
android    1 412 300
ios        3 980 100
web        2 106 400
checkout conversion30.7% → 15.2%
Case #06−15.5 pp

The broken checkout

Find where exactly the funnel tore — and prove it with numbers.

13
chapters
83
lessons
11
investigations
Python
right in the browser

Outcome

What you will be able to do

Not a list of technologies — the working moves of an analyst.

Find the cause of a metric drop

You will break the change down by segment, localise the moment of failure and cut off coincidences.

Test hypotheses with statistics

Confidence intervals, p-value, power and MDE — through simulation, not textbook formulas.

Compute the product economics

CAC, LTV, margin, payback and cohort retention — up to "this is where we lose money".

Deliver the conclusion

Charts in code, an executive dashboard and a final report a decision can be made on.

How it works

A course you work through with your hands

Code — right on the page

Python, pandas and matplotlib execute in the browser (Pyodide). No installs, no cloud accounts.

11 cases instead of exercises

Every case has the same shape: a metric fell — evidence, hypotheses, verdict. You run the analysis.

Python and stats from zero

The course continues the SQL one, but assumes no prior Python or statistics.

Kotomarket black box · Vault-9 · 2184

Eleven cases from one store

Vault-9 has opened the black box of a dead store. Your mentor L. reached the last entry of her journal and vanished. You are the investigating analyst: eleven of her cases await, and in each one a metric fell for its own reason.

  • 04the lying average order value
  • 05one-rouble subscribers
  • 08review manipulation
  • 09delayed cancellations
  • 10growth that is too smooth

The route

Four acts and a finale

The program is built like an investigation: first read the data, then obtain it, then find the cause, then prove it.

I2 chapters

The language of metrics

What an analyst does and how a product is measured

II3 chapters

The tools

Python, pandas and the first look at data

III3 chapters

The craft

Cleaning, joins, the method of inquiry

IV4 chapters

The proof

Segments, cohorts, statistics, dashboards

I

The language of metrics

2 chapters · 11 lessons · 2 h 46 min

What an analyst does and how a product is measured

Prologue

"The Second Signal"

You will see what an analyst actually does and run your first Python cell right in the browser.

eventobservation → conclusionPython
4 lessons · 51 min
Chapter 1Case №0

Anatomy of Metrics

You will learn to compute DAU/WAU/MAU, funnels and retention — and read human behaviour behind them.

DAU/WAU/MAUfunnelretentionrevenue
7 lessons · 1 h 55 minartefact: a conversion funnel

Checkpoint · end of act I

You already read metrics

  • you compute DAU, WAU and MAU and know the difference
  • you build a funnel and see which step leaks
  • you tell retention from repeat purchases
  • you run Python in the browser with zero installs

What comes next

The first 11 lessons are open with no payment. Beyond them the investigation itself begins: pandas, cleaning, joins, cohorts, statistics, 11 cases and the capstone.

II

The tools

3 chapters · 18 lessons · 5 h 16 min

Python, pandas and the first look at data

Chapter 2 PRO

Python for Analysts: A Bear in the Archive I

You will take exactly as much Python as you need to tame pandas: lists, dicts, loops, functions.

lists & dictsloopsfunctions
7 lessons · 2 h 3 min
Chapter 3 PRO

pandas: A Bear in the Archive II

You will master the DataFrame: boolean masks, sorting and groupby on slices of the Kotomarket archive.

DataFramefiltersgroupby
6 lessons · 1 h 44 minartefact: a groupby summary
Chapter 4 PRO

First Look at the Data: inspection, statistics, charts

You will survey an unfamiliar dataset: descriptive stats, distributions and your first matplotlib charts.

EDAdescribe()matplotlib
5 lessons · 1 h 29 minartefact: a distribution histogram
Case01The night the cats vanishedL.'s journal · chapter 7
DAU 42,100 → 29,500−30%

Evidence

  • Web: 5,600 → 5,656, +1%
  • iOS: 10,000 → 9,800, −2%
  • Android: 26,500 → 14,044, −47%

Your task. The overall metric fell 30%, yet nowhere is there a 30% drop. Find where the hole sits and prove it with a breakdown.

III

The craft

3 chapters · 19 lessons · 5 h 38 min

Cleaning, joins, the method of inquiry

Chapter 5 PRO

Data Cleaning: Archive Hygiene

You will bring a dirty archive into working shape: gaps, duplicates, types, dates and a 9,999,999 ₽ order.

missing valuesduplicatestypes & datesoutliers
6 lessons · 1 h 51 minartefact: a clean dataset
Chapter 6 PRO

Joins and Pivots: Stitching the Archive

You will stitch scattered tables with merge, catch fan-out and build a pivot report in five lines.

mergejoin typespivot_table
6 lessons · 1 h 40 minartefact: a pivot report
Chapter 7 PROCase №1

The Method of Inquiry: hypotheses and checks

You will get the method: a hypothesis funnel, the "check the tracking first" reflex, effect size vs noise.

hypothesestrackingeffect size
7 lessons · 2 h 7 minartefact: a hypothesis tree
Case09Delayed cancellationsL.'s journal · chapter 11
Cancellation rate 5.2% → 32.5%×6

Evidence

  • It must be counted by order-date cohort
  • Orders from May 1–14 cancelled at 5.2%
  • The promo hypothesis is not supported

Your task. A cancellation arrives weeks after the order. The calendar metric shows nothing — the cohort metric shows it at once.

IV

The proof

4 chapters · 30 lessons · 8 h 56 min

Segments, cohorts, statistics, dashboards

Chapter 8 PROCase №2 · №3

Segments and Seasonality

You will find the segment dragging the metric down and tell seasonality from a real incident.

segmentationSimpson's paradoxseasonality
6 lessons · 1 h 50 minartefact: a segment breakdown
Chapter 9 PROCase №4 · №5

Cohorts and Product Economics

You will build cohort matrices and compute the economics: CAC, LTV, margin and payback.

cohortsLTV / CACmarginpayback
8 lessons · 2 h 18 minartefact: a cohort matrix
Chapter 10 PROCase №6 · №7

Statistics and A/B Tests

You will tell signal from noise: confidence intervals, p-value, power and MDE — by simulation, not formulas.

CLTbootstrapp-valuepower & MDE
9 lessons · 2 h 43 minartefact: a confidence interval
Chapter 11 PROCase №8 · №9 · №10

Charts and Dashboards: The Console

You will draw metrics in code, assemble a four-chart executive dashboard and crack three fraud cases.

matplotlibdashboardfraud
7 lessons · 2 h 5 minartefact: a four-chart dashboard

Portfolio

What stays with you after the course

Every artefact is one you build yourself, in code, on real archive data.

A funnel and its leak

Steps, conversions and the step where the product loses people.

A cohort matrix

Retention by cohort lifetime month and an LTV calculation on top.

An A/B-test decision

An effect interval, a p-value and a clear verdict instead of "looks up".

An investigation dashboard

Four charts on one matplotlib canvas — plus the post-mortem report.

Finale

The post-mortem

All of L.'s cases fold into a single chronicle — and you write the verdict on it.

Chapter 12 PRO

The Post-Mortem Report: capstone

You will write the post-mortem on the black box: the GMV break, the retention collapse and a cash-gap forecast.

capstonereportinterview prep
5 lessons · 1 h 35 minartefact: a post-mortem report

The first case is waiting in the prologue

The prologue and the first chapter are open with no payment — 11 lessons, your first Python cell and your first investigation.

Start for free