PepperwindTrading Co.

Founded in Mattancherry in 2017 as an exporter of pepper and cardamom, Pepperwind has cleaned, graded, ground and packed its own goods since 2019. When the window opens in October 2023 it has 120 people, all in Kochi. By September 2026 it has 180 across three entities and about ₹80 crore of revenue a year, paid for by a Series B, and it still loses money, less of it each year.

Each system keeps its own IDs in the shape that kind of system uses and records events on its own clock. The CRM holds an old phone number; the helpdesk knows a customer by a second email; analytics knows them by an anonymous ID until they log in. Underneath, one set of people, products and days.

A slice of Pepperwind runs here in DuckDB, on your machine: nothing is sent anywhere. Pick a question or write your own SQL; a result over time draws itself as a chart. 2,870 whole customers and 37 wholesale accounts from Pepperwind's release, every system's own ids kept as it wrote them.

The same shopper is a storefront customer, a billing account, a helpdesk requester, an anonymous ID in analytics and a row in the ledger, each under its own key and on its own clock. Step through what each system wrote.

Direct-to-consumer orders a month, indexed to India in October 2023. Europe opens in February 2024; the US storefront, shipped from Kochi at first, moves to its own entity in January 2025, then loses its gear range to tariffs. Hover the gold marks for what happened.

Daily orders around four festivals in 2025, each against the median day five to ten weeks before. The shapes come from how people buy and how parcels move: a gift has to arrive in time, couriers rest on the festival day, and demand pulled forward leaves a slump behind it.

Orders by hour of the local day. India shops at lunch and again late in the evening, Europe after dinner, the US through the working day. Summer time moves Europe and the US; India has none. Seconds are recorded, and nothing piles up at midnight.

Each market pays the way its shoppers do, and the method changes what happens next. Smaller Indian towns lean on cash on delivery, and a cash parcel is far more likely to come back unopened.

On 1 September 2024 every Indian and European Club price rose 12%, monthly plans at their next renewal and annual plans at theirs. Monthly churn ran higher through the fourth quarter, and the cancellation tickets name the price. Failed renewals go into dunning, with retries on days 1, 3, 5 and 7, and about two thirds are recovered.

Kochi carries the company; Rotterdam has about ten people and the US a handful. Appraisals run every April, packing staff turn over fastest, and festive peaks are worked by fixed-term operators and agency packers. Support tickets nearly double before Diwali, most of them asking where an order is, and nobody in the warehouse or support may take leave in the fortnight before.

Each person and organisation carries traits only the answer key sees: how much they spend, how price-sensitive they are, how likely they are to leave. Those traits cause what the systems record. Every incident works the same way, so each leaves evidence in more than one system and none is found by reading a label. The answer key lists every affected row and amount.

Each line names a shape the data must have, the band it must land in, and where the band comes from: 66 cite a published figure, the rest are stated estimates or release rules. Benford’s law on order totals, expense claims bunching under an approval limit, resignations after appraisals, cash on delivery by town tier.

Beside the 223 tables sit 200 answer-key tables: the hidden traits, the clean version of every messy record, every incident’s rows. Each KPI is a DuckDB query with its value every month, and each task carries its gold answer and a verifier.