CPG demo kit
A loadable knowledge graph in two layers: the commercial world of a fictional CPG manufacturer, Meridian Foods Group (suppliers, contracts, promotions, and the documents that connect them), and a real product corpus from USDA FoodData Central, both typed with the GS1 Web Vocabulary. It accompanies the guide One Intelligence Layer, From Supplier Contract to Shelf. Fictional commercial facts attach only to fictional products; the real records carry only what USDA publishes, under a CC0 public-domain license.
31 KB zip, seven files. No signup is needed to download. Running it requires a Fluree workspace (free tier, no credit card) or the open-source CLI.
Loading the kit
Two supported paths. Both are free, and both are covered step by step in the README inside the download. Loading takes about two minutes either way.
Create a workspace
Hosted: create a free Fluree AI workspace and add a new ledger. A ledger is a Fluree database, empty until you load it. Local alternative for developers: brew install fluree, then fluree create.
Import the five data files
Load model.jsonld first, then the four instances files in any order. Each file commits in seconds; the model file defines the vocabulary the instance files use.
Run the questions
Ask the AI agent the competency questions in plain English, as written. Compare each response against the expected answer listed with the question. CLI users can run the three included SPARQL queries directly.
The ten questions
Each question is answerable only by combining data from more than one source domain: trade, ERP, procurement, contracts, or documents. The note under each describes what the query touches. Three further questions over the real USDA corpus ship in the kit.
- 01Which completed promotions at Hartland Grocers created profitable incremental margin, and which eroded it?Reads promotion evaluations for one retail account
- 02What baseline was the February TPR's margin-eroding verdict measured against?Reads the baseline note stored on the evaluation record
- 03If we wanted to change the live Back-to-School TPR, what commitments and notice periods apply?Joins the promotion to its JBP lock window, notice period, and terms
- 04Which trade terms govern scan-backs at Hartland, and where did the monthly reconciliation come from?Traces a trade term to the amendment document that introduced it
- 05Which products and live promotions are exposed to cocoa through their bill of materials?Traverses commodity → BOM line → product → live promotion
- 06Cocoa is up roughly 50% since April. Which live promotions still clear their brand margin guardrail?Combines the price series, cost shares, discount depth, and guardrails
- 07Where is our supplier concentration risk, and what promotions sit downstream of it?Reads share-of-supply figures and follows them to live promotions
- 08What early signals preceded the June cocoa force majeure, and how much lead time did they offer?Compares document dates against the declared date (67 and 49 days)
- 09For Hartland in Q2, connect trade spend, account profitability, and cocoa exposure into one picture.Combines account P&L, the live promotion, and upstream cost exposure
- 10Which accounts show margin trending down while trade investment trends up?Compares quarterly account profitability across retailers
What is in the kit
Seven files: the GS1-aligned data model, four instance files, the competency questions, and a README with load instructions and data-source details. The commercial layer is set in August 2026, partway through a cocoa supply disruption.
Data model
23 classes and 92 properties. Product master data uses the GS1 Web Vocabulary (gtin, brand, brand owner, ingredient and allergen statements, net content); the cpg extension covers what GS1 does not: contracts, promotions, trade terms, evaluations, guardrails, and provenance. Each term carries a descriptive comment.
Supply-side data
Eight suppliers and their contracts, five commodities with monthly 2026 price series, one declared force majeure and one quality hold, and nine source documents with dates and excerpts.
Product data
Three brands, ten SKUs with GTIN-14 identifiers, 24 bill-of-materials lines connecting products to commodities, and three brand-level margin guardrails.
Commercial data
Four retailers, four joint business plans with lock windows and notice periods, nine trade terms, ten promotions across completed, live, and planned statuses, six promotion evaluations with baseline notes, 48 weeks of point-of-sale records, six trade deductions, and eight quarters of account profitability.
Real product corpus (USDA)
68 real branded food records (chocolate, cookies, hazelnut spreads) from USDA FoodData Central: manufacturer-supplied via the GS1 Global Data Synchronization Network, CC0 1.0 public domain. Real GTINs, brand owners, and ingredient statements, each linked to a provenance record.
Competency questions
Thirteen plain-English questions: ten on the fictional commercial layer, three on the real corpus. Each includes the expected answer, so responses can be checked against the data. Three are also included as SPARQL for direct querying.
Download
All seven files in one zip, including the README with step-by-step load instructions and the standards and data-source details.
Questions about the kit
For a guided walkthrough of the kit, or to talk about running the same pattern on your own data, contact us. Background reading is on the CPG solutions page.