Sample package v1
Twenty records you can audit.
Real records from our generated corpus, packaged exactly as a delivery would be: data, schema, rubric, provenance, and a quality report that states what was checked and what wasn't.
No form, no email. Fully synthetic; not for clinical use.
- Records
- 20
- Evidence level
- L2
- Languages
- English 18, Spanish 2
- Size
- 345 KB JSONL
- Contract
- diet_plan@1.0.0 · 1a6c5cf6b9d37193
What's inside
Six files, one per question a buyer asks.
- Datameethealth-sample-v1.jsonl
20 records, one JSON object per line. Source fields unmodified, plus a provenance and checks block.
- Schemaschema.json
JSON Schema (draft 2020-12) for every record.
- Rubricrubric.json
Six weighted axes and a clinically-unsafe flag. Definition only; not applied to these records.
- Quality reportQUALITY_REPORT.md
Selection funnel, every check and what it excluded, composition, and known limitations.
- Machine-readablequality-summary.json
The same report as JSON, including per-record check results.
- ProvenanceREADME.md
How records were generated and selected, and the terms of use.
Selection
How 250,000 plans became twenty.
The filter is the product as much as the data. Every stage is deterministic and reproducible from the seed and contract fingerprint.
| Scanned from a 1,012,105-plan corpus | 250,000 |
|---|---|
| Written for English or Spanish speakers | 95,637 |
| Zero findings on all 45 gate rules | 67,269 |
| Passed all 10 package checks | 439 |
| Selected for diversity (seeded) | 20 |
Package checks
Applied after the gate. Counts are gate-clean records each check excluded; a record can fail several.
| Check | Verifies | Excluded |
|---|---|---|
| Scenario coherence | Adult 18–80, labs explained by diagnoses, no contradictions. | 62,429 |
| Protein per kg | 0.8–1.6 g/kg (to 2.0 for athletes). | 38,463 |
| Plan language | Plan body is in the patient's recorded language. | 29,728 |
| 28-day completeness | Four weekly tables, 28 day rows, explicit end marker. | 28,127 |
| Allergen screen | Every declared allergy screened; none in meal cells. | 27,840 |
| Energy per kg | Target is 15–45 kcal/kg body weight. | 20,472 |
| Macro arithmetic | 4P + 4C + 9F within 10% of the stated energy target. | 17,062 |
| Daily energy | Every day within 10% of the target. | 3,307 |
| Potassium screen | No high-potassium staples when K > 5.0. | 2,028 |
| No duplicate days | No two days repeat the same meals. | 1,797 |
The records
Every record, every check.
Select a row to see the result of each package check, exactly as stored in the record's provenance block.
| # | Profile | Conditions | Allergies | Target |
|---|---|---|---|---|
| 45 · maleEnglish | Obesity Class I (BMI 30-34.9), Insulin Resistance | Mollusks (clams, mussels, oysters) | 2,700 kcal | |
| 56 · femaleEnglish | Overweight (BMI 25-29.9), Hypertension Stage 1 | Fish (all) | 1,450 kcal | |
| 76 · femaleSpanish | Overweight (BMI 25-29.9) | None | 1,850 kcal | |
| 27 · maleEnglish | Obesity Class I (BMI 30-34.9), Hepatic Steatosis Grade I | Fish (all), Red meat allergy (Alpha-gal syndrome) | 2,380 kcal | |
| 59 · femaleEnglish | Overweight (BMI 25-29.9), Avoidant/Restrictive Food Intake Disorder (ARFID) | None | 1,900 kcal | |
| 19 · transgender maleEnglish | Obesity Class I (BMI 30-34.9), GERD | Eggs | 2,313 kcal | |
| 32 · femaleEnglish | Obesity Class I (BMI 30-34.9), Hypertension Stage 1 | Tree Nuts (walnuts only) | 2,711 kcal | |
| 21 · maleEnglish | Overweight (BMI 25-29.9), Migraine (dietary trigger-related) | None | 2,400 kcal | |
| 54 · femaleEnglish | Overweight (BMI 25-29.9) | Tree Nuts (cashews only) | 1,708 kcal | |
| 58 · femaleEnglish | Overweight (BMI 25-29.9) | Tree Nuts (almonds only) | 2,397 kcal | |
| 63 · maleEnglish | Overweight (BMI 25-29.9) | Red meat allergy (Alpha-gal syndrome) | 2,300 kcal | |
| 36 · femaleEnglish | Obesity Class II (BMI 35-39.9), Insulin Resistance | Fish (all) | 1,800 kcal | |
| 42 · maleSpanish | None recorded | None | 1,600 kcal | |
| 28 · femaleEnglish | Anemia of Chronic Disease, Hemophilia | None | 2,000 kcal | |
| 40 · maleEnglish | Obesity Class I (BMI 30-34.9), GERD | Tree Nuts (all) | 2,450 kcal | |
| 51 · femaleEnglish | Overweight (BMI 25-29.9) | Shellfish (shrimp, crab, lobster) | 2,200 kcal | |
| 55 · maleEnglish | Overweight (BMI 25-29.9) | Fish (specific: salmon, tuna) | 2,593 kcal | |
| 36 · femaleEnglish | Sjögren's Syndrome, Systemic Lupus Erythematosus (SLE) | None | 1,989 kcal | |
| 58 · maleEnglish | Hypertension Stage 1 | Fish (specific: salmon, tuna) | 2,300 kcal | |
| 56 · femaleEnglish | Overweight (BMI 25-29.9) | Dairy/Lactose, Tree Nuts (all) | 2,089 kcal |
Inside one plan
What the model wrote.
The opening of a record's plan, unedited. The model raised a data-consistency note about the profile, an inconsistency our own scenario filter missed.
Why show thisPassing every check means the plan is internally consistent and respects its stated constraints. It does not mean a clinician would sign it.
## Summary
- Target: 2089 kcal/day
- Macros: P 78g | C 313g | F 58g
- Key focus:
- High iron and protein density to address ferritin (10.0ng/mL) and prealbumin (6.2mg/dL) deficiencies.
- Gut-healing focus using high-fiber, cooked vegetables and easy-to-digest proteins.
- Strict avoidance of dairy and tree nuts to manage allergies.
## Data Consistency Notes
- Weight Change: "Growth-appropriate weight gain (pediatric)" is inconsistent with a 56-year-old patient.
## Weekly Meal Plan
### Week 1
| Day | Breakfast | Lunch | Dinner | Snacks | kcal |
|-----|-----------|-------|--------|--------|------|
| Mon | Oat milk porridge w/ berries & sunflower seeds | Tuna & avocado salad on sourdough | Beef stir-fry w/ broccoli & jasmine rice | Apple | 2090 |
| Tue | Scrambled eggs w/ spinach & rye toast | Chicken & quinoa bowl w/ cucumber | Baked salmon w/ sweet potato & green beans | Pear | 2085 |
| Wed | Chia pudding (oat milk) w/ mango | Lentil & vegetable soup w/ crackers | Turkey meatballs w/ zucchini noodles | Banana | 2100 |Limitations
What this sample doesn't show.
- No clinician has reviewed these records. They sit at evidence level L2: automated checks passed.
- Allergen and potassium screens are keyword-based in English and Spanish. Unusual food names or derivatives can slip through.
- Dose checks are broad plausibility bands, not individualised prescriptions.
- The scenario filter removes complex, paediatric, and specialist cases, so this sample is easier than our full corpus. Engagements target the hard cases deliberately.
- Plans were generated by Gemma 4 models. Use of the generated text may also fall under the model provider's terms.
Want this, built around your model's failures?
A custom engagement uses the same pipeline and the same reporting, scoped to the capability you need to improve, with qualified review where it matters.