One fictional company. Five business units. Every example in twenty weeks lives inside it, and every module answers a harder version of the same question with a stronger tool.

Source of truth is docs/07_Client_Zero.md, LOCKED at v2.2 on 13 September 2026. These pages add personality, KPIs and case families on top of that file. Where the two disagree, the locked file wins and this page is wrong.

Kalpa is fictional. Any resemblance to a real company is coincidental, and that disclaimer travels with the name wherever it is printed.

Kalpa Retail's revenue grew 4 percent last year against a plan of 15. Meera Raghavan, its CEO, has asked the new data and AI team one question before she approves any marketing spend:

"Where does our growth actually come from, and where is it leaking?"

That question opens Day 1 and is never fully closed. Week 1 answers it with a revenue tree computed in plain Python. Week 2 answers it from the warehouse. Week 5 answers it with a model. Week 12 answers it with an assistant. A learner comes to see analysis, machine learning, deep learning and generative AI as successive answers to one business problem rather than as separate subjects.

Names are fictional, they stay consistent across every artifact, and a trainer says them from memory. Naming a fictional stakeholder is the opposite of naming a trainer, which never happens in a student file.

Two or three are open at any time. Growth is the spine. Trust and Cost are what a skeptic in the room raises against every Growth answer, which is why the Situation Bank's hardest cards live there.

Weeks 1 and 2 are fixed to the day. Weeks 3 to 9 are fixed to the day in the workbook. Weeks 10 to 20 are outline until they are detailed on go-ahead.

flowchart TB
    K["<b>Kalpa Group</b><br/>Singapore HQ<br/>Bengaluru data and AI centre"]
    K --> R["<b>Kalpa Retail</b><br/>consumer commerce<br/><i>the teaching spine</i>"]
    K --> F["<b>Kalpa Financial Services</b><br/>payments and lending<br/><i>Build 2, and transfer from Week 5</i>"]
    K --> H["<b>Kalpa Health</b><br/>labs and clinics<br/><i>Build 1</i>"]
    K --> C["<b>Kalpa Connect</b><br/>telecom and subscriptions<br/><i>Build 3</i>"]
    K --> L["<b>Kalpa Logistics</b><br/>last mile and warehousing<br/><i>a later build</i>"]

Real companies stay welcome in trainer notes and slides as dated, verified references. They never replace Kalpa as the thing the room computes on.

Logistics is the one unit the locked ladder does not schedule inside Weeks 1 to 9. It is named for a later build, so its page is a holding page rather than a scheduled one.

Retail's tables are what the whole programme grows through: a list of dictionaries in Week 1, files on disk, Postgres tables, a pandas frame, a feature table, then a document corpus.

erDiagram
    CUSTOMERS ||--o{ ORDERS : places
    CUSTOMERS ||--o{ EVENTS : generates
    CUSTOMERS ||--o{ SUPPORT_TICKETS : raises
    ORDERS ||--o{ ORDER_ITEMS : contains
    ORDERS ||--o{ PAYMENTS : settled_by
    ORDERS ||--o| REVIEWS : receives
    PRODUCTS ||--o{ ORDER_ITEMS : appears_in
    CAMPAIGNS ||--o{ ORDERS : influences

Orders, customers, order items, payments, products and events carry Weeks 1 to 5. Campaigns enters on Week 1 Thursday for the discount question and returns in Week 2 as the exposure table. Reviews enter on Week 7 Friday and support tickets on Week 8 Monday, when the Growth thread turns to text. Segments are four: Retail-Core, Retail-Plus, Business and Student. A typical order sits between Rs 800 and Rs 3,000.

Every version is generated deterministically from one seed, so all sixty learners hold identical data, and every planted feature is a witness for exactly one teaching point.

Students are never told what is planted.

The witness list is TRAINER material. It appears in the curriculum row's client-zero column, which is labelled TRAINER ONLY, and in the trainer notes, and nowhere a learner can read. The room is meant to find the bulk order by sorting and the duplicated rows by reconciling against Finance. A slide that announces the plant has spent the lesson.

Build weeks do not use this dataset. They use real messy data at scale, relabelled into the unit that hosts the sub-problem.

flowchart LR
    A["<b>1</b><br/>who runs it,<br/>and what they want"] --> B["<b>2</b><br/>the data estate,<br/>what exists"]
    B --> C["<b>3</b><br/>the KPIs, and<br/>how each is gamed"]
    C --> D["<b>4</b><br/>case families,<br/>mapped to modules"]
    D --> E["<b>5</b><br/>the situations<br/>hosted here"]
    E --> F["<b>6</b><br/>what it must<br/>never be used for"]

Section 3 is the one people skip and the one that matters. Every KPI in this world has a definition, a denominator and a way of being gamed, and a learner who can name all three for a metric can hold their own in an interview about a business they have never seen.

The locked file defines five units and no more. Three domains that come up often in interview preparation have no home in Kalpa today:

Nobody should add these to a day pack until the Programme Head rules. The cost of a sixth unit is not the writing, it is that build weeks draw one sub-problem per unit, so a sixth unit changes the shape of every build week. The cheaper move is a second line of business inside an existing unit, which adds the domain without adding a tour stop. Tracked as issue #69.

1. Kalpa's stakeholders are named, and those names are used from memory across every artifact. A trainer's name never appears in a student file; a fictional CEO's always does.
2. A stakeholder has a motive. Meera wants a growth plan she can defend to a board. Anand wants his books to match. The marketing lead wants the acquisition budget approved. A card without a motive is a tutorial.
3. Money is in Rs, written as the two letters, never the currency glyph.
4. Nothing is invented past the lock. If a day pack needs a fact about Kalpa that the locked file does not carry, the build stops and names the gap. It does not guess.