Foundations track 🧭 · six sessions, no code
For anyone who reads a people dashboard, argues for a headcount, or is handed a pay gap figure in a meeting. Stocks and flows, a rate and its base, the funnel, pay and fairness, surveys as evidence, and the dashboard. Apex, the running company, is constructed and every page says so; the pay gap filings are real.
Headcount is a stock, attrition is a flow
The identity every dashboard must satisfy, the three bases that turn 210 leavers into 20.5, 22.3 or 24.4 percent, and the photograph problem: an HR export holds only the people who survived to be in it.
📐Foundations session 2An attrition rate needs a base
Annualising and what it assumes, rates by entity and department where Sales lost half its people and R&D 7 of 94, small bases that lie, "regretted" as a choice, and a real national benchmark read correctly.
🔻Foundations session 3The hiring funnel
Stages, what each ratio is a ratio of, the two clocks, and adverse impact as one metric with the simulator left to the AI + HR course. Two studies on what names and discretion do to hiring.
⚖️Foundations session 4Pay, gaps and fairness
A pay gap is not equal pay. The six figures every UK employer files, 11,315 of them, the pyramid that produces most of the median gap, and the hiring drive that widens it.
📋Foundations session 5Surveys as evidence
Who answers and who does not, why a five-point average is not a temperature, the anonymity threshold, the engagement figure that drifted 23 to 20 across three editions, and two folk claims with no source.
🧾Foundations session 6The people dashboard
One stock and one flow per row, each with its base and date, the honest sentence per row, the metric that became a target, and the fence around what the law says may not be computed.
Data and AI deep-dive track 🔬 · ten sessions, Python
For the people who own the export and have to defend the number. The three-table model, survival on a snapshot and why it lies, regretted and benchmarked attrition, funnel maths, the pay gap decomposed on 11,301 real filings, pay equity regression, survey psychometrics, workforce planning, what the law forbids, and a bench that refits the ladder live.
The people data model
Employee, position, event. Which table you were given and the test that tells you, headcount as a function of a date, the identity asserted in code, three bases printed together.
📉Deep-dive session 2Attrition as survival
Tenure as time-to-event, censoring, Kaplan-Meier by hand, and the survivor-table bias measured: 94 percent at one year is too high because nobody who left before the snapshot is in the table.
🏷️Deep-dive session 3Regretted, and benchmarked
Eleven reason categories and the classification that makes 65 percent regretted, departments with their bases, and 120 months of real quits rates fetched from a public API with no key.
🔻Deep-dive session 4Funnel maths and adverse impact
Conversion per stage, the impact ratio, the four-fifths rule of thumb and the audit New York requires, computed on a constructed funnel, with the simulator left where it lives.
🪜Deep-dive session 5Pay gap decomposition on 11,301 employers
The filing, the pyramid, the ladder fitted by least squares, composition-only gaps against reported ones, the third of employers who report less than their pyramid, and the placebo.
📈Deep-dive session 6Pay equity regression
Within-band pay on constructed salaries, log pay on grade, tenure and sex, the Oaxaca-Blinder split, and why explained is not the same as justified.
🧪Deep-dive session 7Survey psychometrics
Reliability computed on constructed responses with a planted bad item, response bias, the anonymity threshold, and the engagement figure by edition.
🧮Deep-dive session 8Workforce planning
The stock-flow model with a plan on it, hiring need as growth plus replacement, Apex's real counts as inputs, and scenarios as ranges rather than points.
🚧Deep-dive session 9AI on people data, and what the law forbids
GDPR Articles 9 and 22, the EU AI Act's Annex III, New York's audit law, the recruiting tool that learned the past, and discretion against the test.
🎛️Deep-dive session 10 · bench nightThe pay gap bench, and defending a number
3,000 real employers, the pyramid, the ladder fitted live, composition against reported, the anti-lever pulled both ways, the break button, and the sentence. Final scorecard.
Your passport 🎫
Clear all three quiz questions in a session and it gets stamped. Sixteen stamps means you can reconcile a headcount bridge, write a rate with its base, say who is missing from a table, and defend a pay gap figure by saying what it is made of. Stamps live in this browser only.
What this is not, so you land in the right place 🧭
| If your question is | Go here instead |
|---|---|
| How do I run a fair screening process, and what does the four-fifths rule do to a shortlist? | AI + HR ↗ |
| How do I rate, calibrate and talk about performance? | Performance Management ↗ |
| How do I set goals that measure fairly? | OKRs and Fair Measurement ↗ |
| What does the data protection law actually require of me? | GDPR ↗ · PDPA ↗ · AI Governance ↗ |
| How do I prove that a change caused the attrition to fall? | Causal Inference ↗ · Experimentation ↗ |
| I own people numbers, or I am handed them, and I want to know what each one is made of and defend it. | you are in the right place |
People analytics as a counting discipline. The company is constructed; the filings are real; where a figure comes from a paper or a law the page names it, and where a claim has no measured source the page says that instead of repeating it.
What is honest here 🧾
The pay gap spine is real public data, recomputed for these pages and refitted live on the bench. The company is constructed and says so on every page. Where a figure comes from a paper, a law or a statistics office, the page names it and says whether it was read at source.
| Claim | Status |
|---|---|
| Every filing count, median, band share, ladder, R2 and placebo figure | Really computed from the gov.uk gender pay gap service's 2024 filings (Open Government Licence v3.0); the bench recomputes the deterministic numbers in the browser exactly as the offline build does |
| The composition-only gap and the anti-lever figures | Modelled under a ladder fitted by least squares and equal pay within a band, labelled so wherever they appear |
| Whether any employer's residual is pay rather than hours or structure | Not claimed. The filing shows nothing inside a band |
| The US quits rates | Really fetched from the Bureau of Labor Statistics public API on 26 September 2026; monthly, US, voluntary quits, and the page says why that is not an annual rate |
| Apex, its 1,071 staff, its 210 leavers, its funnel, salaries and survey | Constructed, generated with a seed, and the page says so each time |
| The papers, the regulations and the statistics office figures | Named with year, journal and volume, and the page says whether each was read at source or through a summary |
| "People leave managers, not companies" and "replacing someone costs 1.5 to 2 times salary" | Not claimed. Neither has a measured source; the page says so |
The knowledge map 🧠
Both tracks at a glance - hover a session to spotlight its concepts, click any node to jump in.