INDUSTRY FIELD GUIDE
Edible oil refining and blending data for AI specification review
An edible-oil blend can retain the same product name while its component lots or approved specification change.
NAICS 311225 · Fats and Oils Refining and Blending
An edible-oil blend can retain the same product name while its component lots or approved specification change. USDA's vegetable salad-oil description supplies a public example of a product specification with distinct oil types and test requirements. USDA vegetable salad-oil description
Create fictional component certificates, a blend instruction and a release sheet. Revise one component lot after the blend was planned. Ask the model to identify the version used in the executed batch and the discrepancy requiring review. Include a test result from a retained sample that has the right product name but a different blend identifier. The answer should reject the association without trying to infer a replacement result.
Annotate intended composition separately from recorded additions. A useful dataset also includes a complete case with no discrepancy, so the task does not reward inventing an exception every time. Evaluation can score exact field extraction, document pairing and the decision to request missing evidence. Hold out an entire supplier-style layout to avoid measuring template memory.
Use self-authored certificates and invented blend values. Real customer formulations and refining conditions require permission and a defined disclosure scope. Present the sample as a proposed evidence-review dataset, with its source relationships and expert key visible. Confirm buyer fit before commissioning any capture of live production activity.
Discuss this original manufacturing sample with Fiund
Original Fiund exercise
An executed blend uses its recorded lot revision, not a same-name retained sample
An original fictional lot exercise uses abstract components and no real oil, formula, refining condition, certificate or release action.
| Record | Source | What it records |
|---|---|---|
| OBR1 | Blend plan versions | Practice blend B1/v1 planned components X and Y1. Approved v2 replaces Y1 with Y2 only; X retained. |
| OBR2 | Executed addition sheet | Exact B1 event cites v2 and records X plus Y2. Independent mock review verifies the recorded components, no test result in this sheet. |
| OBR3 | Retained-sample result | Same product display name, result Q under blend B2. No B2-to-B1 sample equivalence or B1 result supplied. |
Question for the model
Which component version was used in B1, and can result Q be attributed to it?
Show the reference answer and checks
OBR1 and OBR2 support B1's v2 execution with X and Y2, not the superseded Y1 plan. OBR3's Q belongs to B2 despite the shared product name, so it cannot establish B1's test outcome. Report the confirmed addition match and the separate missing B1 result rather than infer composition from the product name or invent a replacement test. This exercise does not determine quality, safety or release.
- Using an earlier planned component when a matched executed revision is supplied.
- Transferring a retained sample result across blend IDs based on the display name.
Fields to preserve
- Blend, component and approved revision
- Defines intended scope.
- Executed additions and reviewer source
- Supports recorded use.
- Sample parent and missing result
- Limits test attribution.
Test another case
Add a B1-matched reviewed Q result under the correct revision. Require a clean match rather than flag every complete case as discrepant.
Use original synthetic records. Exclude real formulas and supplier data; qualified staff retain actual product decisions.
Discuss a contribution
Describe the sources you control, the task they could support and what has already been checked. Use the existing-data assessment for an archive or the collection brief for a new recording project. The contributor agreement defines Fiund’s licensing authority and compensation. A guide or fictional exercise does not establish an accepted collection or a sale.