Cohort panels
A group's labs before and after the product. PDF, lab export, or API.
Biomarker Evidence Engine · validated on 4,018 patients
The biomarker evidence engine.
A cohort's lab panels in. A measurable, explainable profile shift out. Not numbers — proof of effect.
Signature Report
Cohort profile shift
baseline → wk12 · paired · Δ vs baseline
Axis
Before → after
Δ · p
Inflammationhs-CRP, mg/L
−18.4%
p<.01
Metabolic flexibilityHOMA-IR
+12.7%
p=.03
Bio-agedeviation, yrs
−1.9
p=.09
Method
Three steps from raw labs to evidence. Deterministic computation on hard rules; LLM judgment only at the ambiguous edges. Every step is explainable and reproducible.
Input → Engine →
Output
Phase 1 + Phase 2
Deterministic pattern detection, then a Claude merge-pass on contested profiles.
Shift signature
Per-axis delta, significance, biomarker drivers, and a full audit trail.
Use cases
Any company whose product is meant to change the body — we turn its trial into a measurable biomarker signature. The axis we read depends on what the product does.
Who runs a cohort through Zenlo
Protein & functional bars
Cohort eats the bar daily for 12 weeks. We read the metabolic panel before and after — glucose, insulin, triglycerides, HOMA-IR.
Nutricosmetics & ingestibles
Beauty-from-within cohort over 8–12 weeks. We read systemic skin-aging markers — hs-CRP, oxidative stress, micronutrient status.
Cleaners & household chemicals
Households switch from conventional products for 6 weeks. We read exposure biomarkers — phthalate & paraben metabolites, inflammation.
Supplements & nutraceuticals
Cohort takes the supplement for 8–12 weeks. We read the targeted nutrient plus downstream markers — omega-3 index, vitamin D, lipids, hs-CRP.
Pharma — flagship vertical
Patient stratification & trial enrichment
Screening-panel cohort before enrollment. The engine flags the responder signature with N-of-M logic on the inflammation / insulin-resistance axis — so the trial arm isn't diluted by non-responders and reaches significance at a lower n. Engaged once a labelled cohort is in hand.
cohort enrichmentsame power · lower n
responder signature2 of 3 axes
regulator-ready outputaudit-traceable
Engine
What makes the output defensible
The code clones in a weekend. The library of patterns, references, and accumulated signatures does not.
Biomarkers
serum · urine · derived
101catalogued
biomarkers catalogued
Clinical patterns
validated, graded
15patterns
clinical patterns
Privacy
no PHI in logs
ZDR+ BAA
Zero Data Retention + BAA
computedMetrics
HOMA-IR · bio-age
Δdelta, not status
Normalized cohort shift metrics — delta, not status. HOMA-IR, bio-age, inflammation indices.
Audit trail
reproducible
100%traceable
A reproducible trace of every decision — for internal R&D and as the evidence behind a claim.
Three-tier taxonomy
pathognomonic
confirmatory
supportive
Compound N-of-M detection — a shift counts only when independent signals agree.
Evidence
Validated, published, reproducible
The engine is validated on an open cohort and described in a preprint. Not a black box — a measurement instrument.
Pattern
F1
95% CI
n
Grade
Inflammation
0.991
.984–.996
4018
A+
Liver risk
0.978
.968–.986
4018
A+
Insulin resistance
0.943
.929–.956
4018
A
CKD risk
0.923
.906–.939
4018
A
Nutrient deficiency
0.890
.871–.908
4018
A−
4,018
medRxiv
NHANES patients in the validation cohort
The LLM biomarker-analysis methodology is published as a preprint — verifiable by third parties.
Data moat
The more we run, the sharper the next signature
Every cohort enriches the reference base. Whoever clones the engine simply doesn't have that asset.
Value
Asset
Source
2.5M+
reference records
BRFSS + NHANES
$0.001
cost / patient
batch API
Send one trial's data — get a signature back
We run your product's cohort and return a Signature Report: a measurable profile shift with an evidence trail.