Data Lab

Data Lab: how analytics works in practice

Interactive demos on synthetic data, learning games about data and real problems I have worked on. New demos are added step by step.

Demos on synthetic data

Demos you can explore yourself

Financial and management analytics

Executive dashboard demo

What is happening, why and what to do: measures with a forecast, an EBITDA root cause tree, cash flow, early warning, a decision simulator and a Be the CEO challenge.

Community passport · twin

Community digital twin

A community passport that updates from registers: 50 indicators with data status, every facility on the map, the real resident count and school, blackout and flood scenarios.

Tool · maturity assessment

Data maturity assessment

32 questions in eight dimensions: a maturity level from 1 to 5, your profile against a target, priorities, a 30, 90 and 180 day roadmap and a PDF report. Your answers stay in your browser.

Learning · forecasts

Forecasting models

22 models from the naive forecast to Prophet, SARIMAX and boosting with SHAP: how each one works and where it fails, scenarios, reconciliation by level and a forecast on your own file in the browser.

Belyakov Consult data quality dashboard demo
Demo · analytics work sample

A data quality dashboard you can try

This is the kind of working tool I build for data teams: processing, quality and source availability indicators, a region × domain matrix and the exceptions worth reviewing first. The data is synthetic; the indicator logic is real.

What is inside

Six pages: overview, analytics, data quality, sources, reports and appearance. The quality target moves with a slider, filters drive every chart, and a management PDF report is built from the current filters in one click. You can upload your own CSV: everything is calculated in the browser and the file goes nowhere.

Belyakov Consult graph analytics demo
Demo · graph analytics

A network of ties and a need index

The graph shows what separate tables hide: who lives without a support network, who holds care together in a community, where social workers are overloaded and which records may be duplicates. The data is synthetic; the indicator logic is real.

What is inside

Six pages: a network of people, households, social workers and clinics; a map of co-occurring demographic, social and medical characteristics; a composite index with weights you can change; a comparison of communities; signals for social work; the method and the rules for protecting sensitive data. Nodes and ties can be exported to CSV for Gephi or Neo4j.

Belyakov Consult transport accessibility demo
Demo · transport accessibility

Travel time zones and access to services

The map shows how long it really takes people to reach a clinic, a hospital or a public service centre: by car, by bus on its timetable or on foot. You can see villages that look close on a map but are in practice cut off from services, and places where a new point or an extra bus would make the biggest difference. The data is synthetic; the calculation logic is real.

What is inside

Six pages: an isochrone map with buses moving on their timetable and the trip to the facility; population coverage curves with people aged 65+ and children shown separately; catchment areas and facility workload; scenarios with a reopened crossing, a demand responsive bus, more frequent services and a search for the best site for a new point; signals; the method. Travel time is calculated right in the browser and the results can be exported to CSV.

Belyakov Consult executive dashboard demo
Demo · financial and management analytics

An executive dashboard: what is happening, why, and what to do

The dashboard is built around three questions. What is happening: five headline measures with a year-end forecast, risk level and causes, and the system picks the three problems with the largest financial impact. Why: a root cause tree breaks the EBITDA gap down to unit, product, line item and factor, and “Explain the change” builds an explanation from calculated factors without invented causes. What to do: a decision simulator shows EBITDA, cash against the threshold and workforce risk, with lags and demand elasticity. The data is synthetic; the logic of the measures is real.

What is inside

Cash flow from clients to spending and a cash gap forecast with payment terms, early warning with unusual changes, confidence levels and forecast breaches, a risk heatmap, staff and client retention cohorts, a dependency graph that separates causes from correlations, and an effect × risk decision matrix with saved scenarios. Plus the “Be the CEO” challenge: three problems, three minutes and your decisions compared with the base case. Two fictional companies, search across the dashboard, full screen and state in the link.

Belyakov Consult fraud analytics demo
Demo · fraud analytics and scoring

Social payment fraud analytics: rules, a model and fairness

12,000 fictional recipients of four assistance programmes and five built-in schemes: networks on shared accounts, payments after death, under-reported income, double programmes, absence despite a residence rule. You can see how transparent rules, logistic regression and a mix of both build a review queue and what it is worth in money. The data is synthetic; the method is real.

What is inside

Six pages: overview, model and threshold (confusion matrix, ROC curves, benefit by review capacity, coefficients), a review queue explaining each score, networks on shared accounts with a link graph and operators, model fairness by group, method. The model trains right in the browser, and scores can be exported to CSV.

Belyakov Consult community terrain 3D demo
Demo · geospatial analysis in 3D

Community terrain in 3D: water, climate, soils and where to build

The same fictional community as in the transport accessibility demo, now on 3D terrain. Raise the river level and see which villages flood and who loses the road to a hospital. You can also see where sirens cannot be heard, where to put telecom towers and which sites suit a new shelter or clinic. Plus climate, soils, erosion and groundwater: point at any spot on the model to see its values. The data is synthetic; the method is real.

What is inside

Seven pages: terrain and a profile between settlements; natural conditions with 13 layers on the model (slope, aspect, solar radiation, temperature, precipitation, frost, soils, fertility, erosion, land cover, groundwater), climate by month, a wind rose and flood recurrence; flooding with an animated water rise; warning and coverage; site suitability; signals; method. The model can be rotated and zoomed with adjustable vertical exaggeration, and there is a top-down map mode.

Belyakov Consult anomaly detection demo: a 3D cloud of payments with signals
Demo · anomaly detection in 3D

Anomalies a dashboard cannot show

Almost 6,000 payments made by a fictional organisation in a year. The monthly report is green and the deviation from plan is under one percent. Now the same data as a cloud of records: each point is a payment, unusual ones glow ruby. Rotate the cloud, zoom in and fly through it. Click a point to see why the record was flagged, how strong the deviation is and why it may be normal rather than an error or fraud. The data is synthetic; the method is real.

What is inside

Five pages: the record cloud, which unfolds from a regular dashboard; a list of signals with reasons; threshold and review (how many errors and risks are found, how many rare but normal cases, a comparison of three methods: feature deviation, isolation forest, nearest neighbours; a group’s own baseline or a shared one); anomaly search in your own CSV or Excel file right in the browser with a download of suspicious records; the method.

Belyakov Consult community digital twin: model layers from terrain and roads to facilities, shelters and risk zones
For communities · digital twin and passport

A community digital twin: a passport that updates itself

The fictional community of Velyka Ruta in one model: 32 settlements, 139 register objects, roads, terrain and risks. A community passport of 50 indicators where every number has a source, an owner, a date and a status: up to date, stale, sources disagree, no data. You see not only the value but whether you can trust it.

What is inside

Twin layers in 3D that you can rotate and take apart right on the page, and a map of every register with a card for each object; a passport for every settlement; the resident count reconciled from the register, GP registrations and meters, with a projection to 2035; the budget per actual resident, farmland cultivated without documents and buildings without ownership records; school network, long blackout and flood scenarios with their effect on the passport; 14 signals, an indicator catalogue, Excel exports and a check of your own passport file.

Belyakov Consult route dispatch room: a 3D map of the region with routes from two depots, the road graph and the fleet console
For business · logistics and geoanalytics

Route planning: a graph, a map and a dispatch room on one screen

A regional distributor: two depots on opposite banks of a river, 12 vehicles and 66 orders a day to supermarkets, shops, hospitals, schools and Invincibility points, each with its own delivery window. The optimiser builds routes in tens of milliseconds and shows what they cost, how many kilometres they take and who will arrive late.

What is inside

A 3D map you can tilt and rotate: drag a depot and the routes rebuild as it snaps to settlements; slice a road with the knife; press and turn a settlement to change its order; knobs for the fleet, fuel and shift; a vehicle schedule you can scrub through the day. The road graph with pallet flow, centrality, critical links, dead ends and weight limits, and a slider that turns the map from kilometres into minutes. Scenarios for a bridge under repair, a flood, black ice and a demand peak, a search for the best third depot, signals, a comparison with the manual plan and Excel exports.

Programme monitoring and evaluation demo: the overview of a fictional programme with reach, indicators on track, budget spent, data quality, effect on income and a reach map of six oblasts
For programmes and donors · monitoring and evaluation (M&E)

Programme monitoring and evaluation: from the results framework to an honest effect

A fictional donor programme restoring services in six oblasts: five partners, 36 communities, €12m and 22 results framework indicators. One system instead of a dozen spreadsheets: plan and actual for any quarter, reach without double counting, money against work done and checks of partner data. Move the reporting quarter back and see when each problem first became visible.

What is inside

An overview with the programme manager’s attention: a repair cost overrun, a micro-grant pause, a blind spot in Kirovohrad Oblast and data that does not check out. A results framework with assumptions and the flow of money from outcomes to oblasts. An indicator table (IPTT) with a reference sheet for every indicator, a forecast to the end of the programme, the chance of reaching the target and the gap by oblast. Earned value (EV, CPI, SPI, estimate at completion) and cost per result. Data quality on the five DQA standards: checks against documents, double counting, the last digit test, late reports. Impact evaluation on a panel of 3,000 households: before and after gives +13 pp, an after-only comparison gives −1, difference in differences +7, and you can see whom the programme did not help. OECD DAC criteria, recipients’ voice, a draft donor report, the indicator table and a reference sheet template in Excel, and a check of your own table right in the browser.

Data maturity assessment: the level on steps from 1 to 5, a radar chart of eight dimensions against the target, three priorities and a 30, 90 and 180 day roadmap
Tool · data maturity assessment

Data maturity assessment: where you are and what to do next

32 questions across eight dimensions: strategy, governance, quality, architecture, registers and reference data, analytics, people and security. Every answer describes a specific practice rather than an abstract “good or bad”, so the assessment takes about 10 minutes and needs no preparation. Your answers stay in your browser and nothing is sent.

What is inside

A level from initial to optimised, a maturity index and a profile by dimension against the target for your type of organisation: a community, a public body, a business, an NGO or a donor programme. Three priorities that account for the foundation (governance, quality, reference data, architecture) weighing more, because everything else rests on it. Caveats such as “analytics is ahead of the foundation” or “analytics rests on one person”, strengths and “what not to do now”: do not buy a new BI platform while metrics are counted differently, and do not automate dirty data. A 30, 90 and 180 day roadmap with ticks for what is done, a PDF report for your team or leadership, a link to the result and three examples: a community, a company and a donor programme.

The Forecasting models section: a revenue series continues as a blue forecast with 80% and 95% intervals, next to a table of model accuracy on held out data
Learning · forecasting models

Forecasting models: how they work and what they show on your data

22 models from the naive forecast to boosting in one section: benchmarks, exponential smoothing, Holt-Winters and TBATS for several seasonal cycles, regression with drivers and Prophet with holidays and trend breaks, ARIMA and SARIMAX, growth curves, decomposition, Croston’s method with the SBA and TSB variants, Monte Carlo, analog search, gradient boosting explained with SHAP and an ensemble. Each model is shown on data where it is easiest to see: steps with formulas, parameters you can move and what the model sees in the series: trend, seasonal cycles, holidays, driver contributions or paths.

What is inside

A fair comparison of every model on one series with repeated tests over four time windows; a scenario forecast with “base”, “optimistic” and “stress” switches and sliders for price, budget and market; a forecast reconciled across levels “region → communities → services” where the totals add up; 12 datasets to download as CSV or Excel; Excel workbooks where the model is calculated with formulas and updates when you change a parameter; Python and R code. On the “Your data” page you can upload your own file: every model is tested on it, the most accurate one builds a forecast with an interval, and the result downloads to Excel. The file never leaves the browser, the examples are synthetic.

Learning games

Games that teach you to read and clean data

Game · charts and analysis

Scooter Cat: Ride the Chart

A Gravity Defied-style game: three riders with laser eyes ride charts. Twelve tracks from easy to hard, a debrief after each, a leaderboard and four Hell tracks with 90s game rules.

Game · probability and statistics

Jackpot Cat: Do the Math

An 80s slot machine that shows its math: cats and capybaras pay, bugs spoil the line, three Data Bosses wipe the balance. Six experiments on probability and risk, each with a prediction and a debrief.

Game · graphs and duplicates

Robo Vac and the Graph

Pac-Man on a relationship graph: remove orphans, merge duplicates, cut false links and do not swallow a bridge node, or the graph falls apart.

Laser Kitty vs the Data Bugs: a pixel kitten with laser eyes and record bugs carrying data errors
Learning · a data quality game

Laser Kitty vs the Data Bugs

An arcade game in the spirit of old cabinets: record bugs fly in formation and dive at the warehouse, and a kitten with laser eyes deals with them. The errors are real: 31 February, month 13, test@test.com, 01.01.1900, a phone in five formats, the same person with an initial instead of a first name. The laser deletes impossible values and placeholders, the fix beam normalises formats, the merge beam fuses duplicates, and correct records must be let through.

What is inside

Five levels: formats, impossible values, duplicates and namesakes, placeholders, and a source that keeps sending errors until you seal it with an entry rule. After each level: warehouse quality, skills and an explanation of every mistake with a tip for doing the same in Excel or Power Query; level records download as CSV. A calm mode without return fire, keyboard and touch controls.

Scooter Cat: a cat on an e-scooter rides a sales chart track and zaps a data bug with eye lasers
Learning · a game about reading charts

Scooter Cat: Ride the Chart

A game in the style of Gravity Defied: guide a rider on wheels along a track that is really a data chart. Throttle, brake and lean, climbs, drops and jumps, landings on both wheels, and eye lasers that clear data bugs sitting on the line and flying above it. There are three riders, each with its own physics: a cat on a scooter, a capybara on a skateboard and the Head of Analytics on an electric unicycle. All to an 8-bit tune.

What is inside

Twelve tracks from easy to hard, each a different kind of chart with a report trap: trend and noise, a histogram (the average receipt versus the median), seasonality, a Pareto chart (group A brings 80% of revenue), outliers, a truncated axis that drops to zero at a gate, a stacked area, a waterfall from revenue to profit, a log scale, a control chart with limits, small multiples on one scale with gaps between the panels and a forecast band where you only see the real road just ahead of your wheels. After each level a debrief with the numbers of that track and a tip on doing the same in Excel; data points, time, bugs, stars and best times. A calm mode, keyboard and touch controls. A leaderboard for each rider, each difficulty group and all 12 main levels: enter a nickname after finishing and share a link to the leaderboard page. For those who want more, four Hell tracks with 90s game rules: candlesticks, a Gantt chart, a temperature corridor and a receivables saw. One life and no checkpoints, a bug kills at first touch, a hard landing wrecks your rider and the timer does not wait; Hell has its own leaderboard and its own hellish tune.

Jackpot Cat: an 80s slot machine with three cats on the line and an experiment debrief with return histograms for 20, 2,000 and 20,000 spins
Learning · a game about probability and risk

Jackpot Cat: Do the Math

An 80s-style slot machine that hides nothing. Cats, capybaras and tangerines pay, data bugs (DUP, NULL, TEST) spoil the line, three matching bugs show a win that is not there, and three Data Bosses on the line mean an audit: the whole balance burns. A lever, ruby LED displays, bulbs and the sound of old machines, and behind the back panel the reel strips, the odds of every combination and the share the machine keeps: 8 credits of every 100. There is no money, no stakes and no prizes.

What is inside

Six short experiments, each with a prediction first, then a few spins and then a debrief: what happened to you and to a thousand other players. The price of a spin (expected value: you win every fourth spin and still lose), small samples (after 20 spins one player in four is ahead, after 20,000 three in a thousand), reel memory (after five losses the odds are the same), the near miss (three matching symbols show up on screen 25 times more often than on the line), two machines with the same return and a different risk (Cat and Capybara), and Data Boss: a once-in-a-thousand event that reaches 63 players in a hundred over a thousand spins. Every experiment comes with a chart, a note on how the same trap looks in reports and an Excel tip. A Math tab with the table of odds, a journal with your result against the expectation, and the machine’s take. Buttons, a lever, the space bar and touch.

Robo Vac and the Graph: a kitten on a robot vacuum rides the corridors of a relationship graph of people, households and companies, sucks up test records, merges duplicates into the golden record and steers around a bridge node
Learning · a game about graph data

Robo Vac and the Graph: clean the network without breaking it

A Pac-Man style arcade where the maze is a relationship graph: people, households, clinics, social services and companies. The kitten from Laser Kitty rides a robot vacuum along the links, collects dust and decides on every record: suck up “TEST TEST”, merge “LLC Soniachnyi Lan” with the same company code into the golden record, cut a link built on “tel. 000 000 00 00”. A namesake with another birth date, a new record without links and a bridge node must stay, or the graph falls apart. Grey bugs crawl the corridors and drop new junk, and the capybara drops an “Entry rule” capsule.

What is inside

Five levels: orphans, duplicates and namesakes, false links, bridges, and a source of dirt, an unchecked import that keeps adding junk until you seal it with an entry rule. A scanner shows bridge nodes and placeholders, and a component counter shows when the graph has split. After each level: the share of correct decisions, skills and an explanation of every mistake with a tip for doing the same in SQL, Power Query, NetworkX or Gephi; the decision log downloads as CSV. Calm mode without a timer, keyboard, swipe and touch controls. All data is synthetic.

Cases

Real problems by domain

Case · 01

Social protection

Registers, payments, services, data quality, analytical marts and management reporting. In these systems, a matching error or a weak indicator definition can quickly become a decision error.

Case · 04

Financial Analytics

Financial control becomes difficult when budgets, payments, donor funding, accounts receivable and reporting sit in separate files. The problem is usually not a lack of numbers. It is the lack of one traceable financial picture.

Case · 05

Management Analytics

Internal business processes are often spread across many spreadsheets, emails and local reports. Management can see individual figures, but not always where a process is slowing down, what is overdue or where intervention is needed.

The data journey 3D scene: from sources through checks and analytics to management decisions
3D scene · the full data cycle

The data journey: 13 stages in 3D

The full data cycle on one 3D stage: from strategy, sources and registers through integration, quality checks, the model and rules to analysis, reports, finance, management, geodata, users and results. At each stage you see what happens to a record and what it gives the organisation.

What is inside

13 stages and 78 system components; a mode that walks you through every stage in turn; the text of each stage below the scene with links to services and Data Lab demos. The scene can be rotated and zoomed.

Social protection
01

Social protection

Registers, payments, services, data quality, analytical marts and management reporting. In these systems, a matching error or a weak indicator definition can quickly become a decision error.

What we work on

The work included building analytics capability, corporate BI, data quality practices and analytical foundations for public information systems.

Health and rehabilitation
02

Health and rehabilitation

Service planning has to combine health data, geography, facility networks, population need and accessibility. One average cannot explain the system.

What we work on

The focus has included geospatial analysis, network assessment, service accessibility, dashboards and data for rehabilitation planning.

International programmes
03

International programmes

Counting activities is not enough. An indicator has to connect to a source, a method, evidence and the decision the programme team makes from it.

What we work on

Work with indicators, data verification, progress analysis, planning and reporting for programme and management teams.

Financial Analytics
04

Financial Analytics

Financial control becomes difficult when budgets, payments, donor funding, accounts receivable and reporting sit in separate files. The problem is usually not a lack of numbers. It is the lack of one traceable financial picture.

What I do

I bring financial flows, plan versus actual, donor funding sources, receivables and obligations into a single model. It can then support recurring reporting, variance analysis and practical control over the use of funds.

Management Analytics
05

Management Analytics

Internal business processes are often spread across many spreadsheets, emails and local reports. Management can see individual figures, but not always where a process is slowing down, what is overdue or where intervention is needed.

What I do

I build a common analytical model for plans and tasks, procurement, deadlines, team workload, operating cycles, KPIs and deviations. Data from different functions is brought into one management view that can be used regularly, not only at reporting time.