Implementation reference
Which market price should Brigade use?
This applies before a kitchen supplies its own purchasing data. Brigade should produce a useful professional-kitchen benchmark without treating grocery, wholesale, foodservice, and farm prices as interchangeable.
Zero-state benchmark onlyDefault to professional-kitchen purchasing. Use the closest eligible market level, specification, geography, package basis, and observation date. Never average retail and wholesale observations into one unlabeled number.
Source access
Start with sources we can use now.
Build the public-data baseline and an allowlisted web-search fallback first. Measure what remains uncovered before entering a paid data contract or sales process.
SourceBest useAccess
Implement now · public data · no sales conversation
Search is a fallback collector, not the calculator. Store the URL, observed date, geography, product specification, package, unit, and displayed price. The deterministic engine decides eligibility and conversion. Prefer two compatible independent observations; label a single-source result or keep the line open.
Test without a contract · useful for evaluation, but automated access has a cost or contact step
Evaluate later · paid or partner access · sales conversation required
Selection order
The engine chooses evidence in this order.
1
Classify the ingredientCanonical concept, form, grade, size, cut, brand requirement, and package wording.2
Infer purchasing contextProfessional-kitchen default; quantity, unit, recipe purpose, and R&D context may make retail more appropriate.3
Filter eligible poolsReject incompatible market levels, specifications, units, stale observations, and geography.4
Normalize and rankConvert eligible evidence through deterministic unit and package rules; rank specification, region, freshness, and source quality.5
Return a benchmarkPoint estimate, plausible range, market level, evidence date, coverage, provenance, and any open assumption.6
Fallback honestlyIf only retail evidence exists, use it as a labeled retail proxy. If no safe conversion exists, keep the line open.Trigger cases
What sends a line to each pool?
| Recipe signal | Primary pool | Rule |
|---|---|---|
| Case, carton, sack, bulk weight, foodservice pack | Foodservice → wholesale | Professional pack language excludes grocery evidence unless no compatible alternative exists. |
| Fresh produce by case or weight | Regional wholesale | Use compatible USDA terminal/shipping-point or licensed commodity evidence; preserve grade, size, origin, and region. |
| Raw meat, seafood, dairy, poultry, eggs | Wholesale commodity | Cut, grade, fat level, fresh/frozen state, and market geography must agree before evidence is eligible. |
| Generic pantry staple in professional quantity | Foodservice → wholesale | Prefer restaurant or commodity evidence. Retail is a fallback, not part of the same median. |
| Named retail brand or consumer package | Retail/package | Brand and package are part of the specification. Do not substitute a generic wholesale commodity silently. |
| Small R&D batch with consumer-size wording | Retail/package | Retail can be the realistic purchase context. Label it as retail and keep the professional-kitchen alternative available if supported. |
| Specialty ingredient with no wholesale coverage | Retail proxy | Return the best compatible retail reference with a wider range and explicit fallback wording. |
| Wholesale and retail sources both available | Context-selected pool | Select one market level. Never average the two or infer an unsupported wholesale discount from retail. |
| Unclear form, grade, package, or conversion | Open / clarification | Show the known context or range, but do not let incompatible evidence become a confident point estimate. |
Agent and engine boundary
Interpretation can be flexible. Money selection cannot.
Agent may propose
- Ingredient concept and specification.
- Likely purchasing context.
- Candidate source pools and query variants.
- A plain-language explanation of the chosen evidence.
Engine must decide
- Which observations are eligible.
- Unit and package conversion.
- Market-level separation.
- Point estimate, range, warnings, and provenance.
Example outputs
The chef sees the result and why it was used.
25 lb all-purpose flour
Foodservice benchmarkUse restaurant or wholesale evidence normalized to pounds.
Foodservice market · regional · compatible bulk specification · updated date shown
One retail package of Oreos
Retail package benchmarkThe named consumer product makes retail evidence appropriate.
Retail reference · exact brand/package · local or national geography shown
Specialty saffron, small test batch
Retail proxyNo compatible foodservice evidence was available.
Retail proxy · wider range · improve by confirming grade or package
Source qualification
Test sources before licensing or enabling them.
Send the same 326 audited ingredient queries to Expana, S&P, Buyers Edge, and Foodbuy. Compare specification coverage, freshness, unit compatibility, geography, added coverage beyond USDA, display rights, and error against a held-out restaurant price set.