Analytics engineering / margin decisions

Margin Control Tower

When weekly contribution margin changes, this browser tool starts from a hash-verified Olist aggregate, traces discounts, returns, cost of goods, and fulfillment, then tests a bounded promotion scenario before recording an action.

Problem: A revenue-only view hides the margin lost to discounts, returns, cost of goods, and fulfillment, and gives a category manager no way to test a response.
Audience: E-commerce category managers and analytics engineers who need to test a margin decision, not just chart it.
What I built: I built the six-table Olist pipeline, source locks, data contracts, metric definitions, diagnosis and scenario engine, holdout check, browser UI, governed fixture, and tests.
Result: The default path verifies the committed Olist artifact before rendering, exposes measured detection and elasticity reports, recomputes a bounded scenario, and records a category action for review.
Links: GitHub repositoryOpens external site

How it works

01
Acquire
Lock six licensed Olist source tables by URL, retrieval date, size, and hash.
02
Reconcile
Join orders, items, customers, products, payments, and reviews at a declared weekly grain.
03
Contract
Stop if source hashes, schema, grain, accounting math, or browser artifact drift.
04
Diagnose
Decompose contribution-margin drivers and load artifact-bound detection and elasticity reports.
05
Test
Apply disclosed scenario assumptions, inspect the later holdout, and record an action.

Try it

The committed Olist artifact is requested first. If its Parquet or hash-bound reports fail validation, the lab keeps the fixed-seed synthetic fallback active and shows the blocked state.

Loading the requested margin source...

Methods / Results / Real-data analysis

Loading the pipeline-backed methods report…

How this was verified

The default source is real: a compact 748-cell preview is generated from the exact Olist Parquet, bound to its SHA-256, and re-verified in the browser without downloading DuckDB or the complete artifact. An explicit complete-data action downloads and verifies olist-margin.parquet, then queries all 15,809 rows through DuckDB-WASM; failure preserves the labeled real preview and blocks full mode.
Real mode loads the committed Olist aggregate produced from 99,441 orders and 112,650 item rows under CC BY-NC-SA 4.0 (retrieved 2026-07-17); source-table hashes, transport commit, transforms, and proxy boundaries are embedded in the Parquet metadata.
Measured fields use observed order items and freight after payment/review reconciliation; discounts, returns, and COGS remain explicitly documented proxies. The first observed week of each category falls back to current item price and therefore has zero proxy discount.
The optional seed-2026071301 fixture contains 9,360 synthetic rows across 52 weeks and remains isolated as a reproducible fallback and test mode.
The inherited Tableau/RFM dashboard is prior work only and is not evidence for this rebuild.

What this does not prove

Synthetic currency and injected anomalies do not show real lift, detection accuracy, or causal impact.
The scenario uses a disclosed elasticity assumption and is not a forecast.
Olist detection and elasticity results appear only from committed detection-report.json and elasticity-report.json files; missing reports remain visibly pending and never render placeholder results.
Detection precision and recall evaluate six deterministic perturbations on observed Mondays after the real totals are reindexed to a complete Monday calendar; 11 weeks with no derived cells are zero-filled. No manually labeled real anomaly is claimed, and neither calendar-completion rows nor perturbations enter the Parquet artifact.
The associational elasticity coefficient is fit on the analysis window; the later eight-week holdout evaluates MAPE only. Reference price, return deductions, and 60% COGS are disclosed proxies; no causal lift, audited company margin, forecast, or production decision is claimed.