Catch decline early.Protect people, not data.
Signal Labs turns preventive screening into a single deterministic Signal Score (0–100, Physical + Cognitive) with a Red / Amber / Green readout. HR sees workforce trends — never an individual’s results.
SOC 2 Type II · HIPAA-aligned · data residency on request.
Workforce Signal Index
Aggregate · k≥120
+1.8 vs. last quarter
differentially-private, ε-budgeted
0–0
Deterministic Signal Score
k≥0
Minimum cohort size
0d
Early-decline lead time
0
Individual rows exposed to HR
Deterministic, not a black box
One score. Two sub-scores. A traffic light anyone can read.
Physical
Cardio-metabolic, recovery, sleep, and movement signals ingested from any connected ring, watch, band, or scale — fused into a transparent, reproducible sub-score.
Cognitive
Validated cognitive and stress measures combine into a sub-score that tracks change over time, surfacing drift long before it becomes a problem.
Longitudinal
Kalman smoothing and CUSUM change-detection flag early decline at the cohort level. The math is deterministic — the same inputs always yield the same score.
An LLM only narrates the score in plain language. It never computes, ranks, or decides — the number is deterministic.
Value of Investment
Preventive screening that pays for itself.
We don’t sell wellness theater. Every pilot ships a VOI model tied to your own headcount, claims, and attrition data — so the value case is auditable, not aspirational.
- Early-decline flags surfaced 90+ days ahead
- Reduced preventable claims and absence
- Attrition risk caught while it’s still reversible
Modeled 3-yr net VOI
$0.6M
≈ 6.4× pilot investment · illustrative
Figures are illustrative defaults. Your pilot model is built on your data.
The privacy line we never cross
HR sees the forest. Only the employee sees their tree.
k-anonymity and differential privacy are enforced in the data layer — not as a policy you have to trust, but as a constraint the system cannot violate.
k-Anonymity (k ≥ 12)
Every figure HR ever sees represents at least 12 employees. Cohorts that would fall below the threshold are suppressed automatically — no slice can be narrowed down to a person.
Differential Privacy
Calibrated noise is added to every aggregate under a fixed epsilon budget. The trend is faithful; no single individual measurably moves the number.
No Individual Rows. Ever.
The HR product has no API, export, or screen that returns one person’s score. Individual results are visible only to the employee, in their own app.
What your benefits team actually sees
Aggregate, privacy-protected, and auditable. Never an individual.
Workforce health, never an individual
HR sees one privacy-protected cohort readout — overall band plus differentially-private median sub-scores. There is no screen, API or export for a single person.
- k-anonymous: every figure stands for ≥ 12 employees
- ε-differential privacy on every released number
- Whole cohort withheld automatically when it’s too small
Catch a downward trend early — at cohort level
Department × age-band trends with the same suppression rules, so you can act on a declining group 90+ days before it shows up in claims.
- Released points are k-anon + ε-DP cohort medians
- Small cells are visibly suppressed, never guessed
- Built for the value case, not wellness theatre
Board-ready exports, zero PHI
Export k-anonymous, ε-DP-noised cohort aggregates for your benefits committee. The privacy check re-runs server-side on every cell.
- CSV / PDF aggregates only — never a row per person
- Cohorts below the floor simply can’t be exported
- Every figure is auditable and reproducible
Run screening without touching health data
Import a roster, schedule screening events, and track participation — administration and results stay cleanly separated by role.
- Roster import + enrollment + event scheduling
- Employees own their results; HR sees aggregates
- Consent and participation tracked, not diagnoses
HR dashboard preview
This is the most granular view HR ever gets.
Cohorts below k=12 are suppressed and can never be drilled into.
Field Services has drifted into Red over 90 days. Want the plain-language brief?
Before any health data
A Business Associate Agreement comes first.
Where US HIPAA applies, Signal Labs acts as your business associate. No employee biosignal — no PHI of any kind — is ingested until a signed BAA (plus the DPA) is in place. This is enforced as an order of operations, not a promise.
BAA required before PHI
Amber · gate01
Qualify & scope
Confirm self-insured eligibility, pilot cohort, and data-residency requirements. No health data is touched at this stage.
02
Execute BAA + DPA
Your counsel and ours sign the Business Associate Agreement and Data Processing Agreement. Only now can a device be connected.
03
Employees opt in
Each employee individually consents in their own app before any biosignal is ingested. HR still sees only k-anon + DP aggregates.
- HIPAA BAA
- Data Processing Agreement
- Master Services Agreement
The screening service is governed by the executed MSA / DPA / BAA — see our terms and privacy pages.
AI you can trust: it explains, it never decides
The number that matters is fixed math. The AI’s job is to make it make sense — and to do the quiet work in the background.
Your score is math, not a guess
The Signal Score is a fixed, published formula. Same inputs, same number — reproducible and independently checkable. The AI cannot change it.
Estimates show their uncertainty
Where we infer a trend or a biological age, it comes with a stated confidence band that updates as more data arrives. No false precision.
The AI only explains
It turns the numbers into plain language. It cannot move your score, change a traffic light, or raise an alarm — those are computed, never generated.
On-device meal analysis
Photograph a meal and the analysis runs on your phone — the image never goes to the cloud.
A device library that stays current
A background research job keeps the list of supported wearables and scales up to date on its own.
Run a 90-day pilot. See the signal for yourself.
Fixed scope, fixed price, a VOI model built on your data — and a privacy guarantee enforced in code.