cinderhaven-data

The shared dataset behind the Cinderhaven Provisions portfolio. Cinderhaven is a fictional ~$25M specialty food brand with 90 SKUs across three product lines, selling through Walmart, Costco, Whole Foods, regional chains, UNFI/KeHE distribution, and DTC.

This repo is the single source of truth for the database and the scripts that generate it. Other repos consume the built database — they don't maintain their own copies.

The dataset

cinderhaven_product_master.db — a SQLite database with the following tables:

Base tables

Table	Rows	What it contains
product_master	90	SKU-level attributes, GTINs, UPCs, case dimensions, nutritional info, active retailers, 1WorldSync status. Includes deliberate data-quality defects
stores	~902	Store/door list with retailer, chain name, region, state, volume tier
distribution_log	~12,507	SKU × store authorization history. Reflects data-quality-driven delays and deauthorizations
sku_costs	90	COGS, landed cost, retailer-specific wholesale prices, trade spend rates by channel
price_history	~398	Time-based wholesale price and MSRP changes over the 104-week window
promotions	~195	Promotional events with retailer-specific patterns, promo_cost, and funding_mechanism
chargebacks	~391	Defect-driven chargebacks traceable to actual product_master data-quality issues
scan_data	~1.19M	Weekly unit and dollar sales by SKU × store. 104 weeks (2024-05-11 to 2026-05-02)

Deduction tables

Table	Rows	What it contains
retailers	11	Canonical retailer reference (retailer_id, name, channel_type, dispute portal details)
retailer_rules	90	Per-retailer × deduction-type dispute rules (window_days, evidence_required, recovery_rate)
deduction_codes	97	Retailer-specific deduction codes (Walmart Code 22, KeHE UDR, etc.)
edi_requirements	42	Per-retailer compliance specs (label, pallet, ASN, OTIF)
orders	~5,800	Purchase orders from retailers to Cinderhaven
order_lines	~30,000	Line items per order (SKU, units, price)
shipments	~5,800	What was shipped (carrier, BOL, delivery, ASN, short/damage flags)
pack_records	~5,800	What was packed/labeled (label compliance, evidence format/location)
deductions	~3,000	Deduction records: short_ship, label_fine, pallet_fine, damaged, late_delivery, promo_billback, vague, spoilage, slotting
remittances	~500	Payment events bundling deductions per retailer × week
disputes	~1,400	Dispute filings with evidence quality, outcome, recovery amount, labor hours
dispute_evidence	~3,000	Evidence items per dispute (submitted/required/format)
post_audit_claims	~45	Retroactive audit clawbacks (Walmart APL, KeHE pass-throughs)

The thesis

Every downstream problem in the dataset — chargebacks, slow launches, delisted SKUs, missed velocity thresholds — is causally traceable to data-quality defects in product_master. The deduction tables extend this by modeling five compounding failures at scale: no visibility into deduction patterns, process gaps generating avoidable deductions, weak evidence, inaccessible records, and missed dispute windows.

Revenue and deduction targets

• Annual wholesale revenue: $23M–$27M (~$25M target)
• Channel split: Walmart ~50%, UNFI ~18%, Whole Foods ~10%, Regional ~9%, Costco ~9%, DTC ~3%
• Annualized deductions: $750K–$1.2M (3–5% of wholesale revenue)
• Recovery rate: 5–10% (lean team, mostly handwritten evidence)
• Double-dip deductions: 3 flagged cases totaling ~$15K–$20K

Generation scripts

Run via build_db.py. The full pipeline produces both the base dataset and the deduction extension in a single build.

Base pipeline

Script	Purpose
build_db.py	Orchestrates the full generation pipeline
seed_product_master.sql	Seeds 90 SKUs with deliberate data-quality defects
01_generate_stores.py	Store/door list
02_generate_distribution.py	SKU × store authorization history
02b_generate_chargebacks.py	Defect-driven chargebacks
03_generate_costs.py	COGS, wholesale prices, trade spend
04_generate_promos.py	Promotional events with promo_cost and funding_mechanism
04b_generate_price_history.py	Time-based pricing changes
05_generate_scan_data.py	~1.19M rows of weekly scan data
06_validate_dataset.py	Base table validation

Deduction pipeline

Script	Purpose
seed_deduction_schema.sql	DDL for 13 new tables
seed_deduction_static.sql	Static seed data (retailers, rules, codes, EDI requirements)
07_seed_deduction_tables.py	Applies schema + static seeds
08_generate_orders.py	Purchase orders + order_lines
09_generate_pack_records.py	Pack/label compliance + evidence records
10_generate_shipments.py	Shipments + BOL/POD/ASN flags
11_generate_deductions.py	Deduction records + double-dip seeding
12_generate_post_audit_claims.py	Retroactive audit clawbacks
13_generate_remittances.py	Payment events bundling deductions
14_generate_disputes.py	Dispute filings + evidence
15_validate_deductions.py	Deduction-specific validation

The built database is not committed — it's ~164 MB and regenerable. Run:

python scripts/build_db.py          # build if missing
python scripts/build_db.py --force  # rebuild from scratch

Repos that use this dataset

• retailer-deduction-recovery — Interactive React demo making retailer deduction losses visible and actionable. Consumes this database via JSON export.
• retail-velocity-decision-tool — Velocity decision tool for specialty food CEOs. Try it live →
• product-data-audit-demo — SQL diagnostic queries for auditing product master data quality

Setup for consuming repos

1. Clone this repo
2. Run python scripts/build_db.py
3. Place or symlink the database file in the consuming repo's data/ directory

Tools

Python, SQLite