MICROSOFT DRAGON COPILOT/HOST SURFACE/IRIS RENDERS BETWEEN-VISIT SIGNAL/FEATURING CANARY CLINICAL VOICE
Dragon Copilot · Integrated with EHR
Margaret Williams · 72F · VA-enrolledLIVEANNUAL FOLLOW-UP
Transcript
Dragon ambient
Dr. Chen10:30
Hi Margaret, how have you been feeling since we last spoke?
Patient10:30
I've been okay, doctor. A little tired in the mornings. My blood pressure has been higher lately.
Dr. Chen10:30
Have you been taking all your medications regularly?
Patient10:30
I try. I think I've missed the evening one a couple of times. The mornings are okay.
Dr. Chen10:30
Any chest pain, shortness of breath, dizziness?
Patient10:30
No chest pain. No dizziness. Maybe a little winded going up stairs, but that's nothing new.
Dr. Chen10:31
Let me look at your numbers from home. Your BP's been running higher
Note draft
Dragon + IRIS
Chief Complaint
Annual follow-up for hypertension and type 2 diabetes. Patient reports rising home BP and intermittent morning fatigue.
History of Present Illness
Dragon ambient
72yo F with longstanding HTN and T2DM, presenting for annual follow-up. Self-observed rise in home BP with intermittent morning fatigue. Denies chest pain, dyspnea, dizziness. Reports occasional missed evening medication doses.
IRIS · between-visit monitoring
Home BP trending upward, recent readings above goal. Weight modestly improved. Glycemic control suboptimal relative to target. Medication adherence patterns show gap on one antihypertensive (carvedilol). Recent voice check-in consistent with reported adherence pattern. Open quality-measure gaps available to address during this encounter.
Canary · longitudinal voice insights
Elevated cognitive load signal detected across recent voice interactions, coincident with reported adherence pattern. Other biomarker channels (stress, anxiety, mood) at baseline. Not clinically diagnostic; offered as additional context for patient conversation.
Post-encounter: Dragon's documentation updates the CloudCare patient record. Encounter data flows into the next between-visit cycle.
Assessment & Plan
(drafting in progress — Dr. Chen to complete)
CloudCare IRIS
Intelligent Routing & Insights · Encounter
High priority · rising. BP trend elevated, adherence pattern noted; elevated cognitive load signal (Canary); open quality measure. Primary encounter focus.
01
CareSignals Intelligence
Triage · Risk · Insights
High priority↑ risingFOLLOW-UP
Driving factor: Adherence pattern detected, correlated with elevated BP trend.
DTWN · BIOLOGIC PROOF & ACTUARIAL TRANSLATION
·TWIN SYNCED 4 MIN AGO
·92% SIGNAL VERIFIED
·MODEL · ILLUSTRATIVE
1Twin subjectMEMBER
MR
Marcus Reyes
M · 54 · Self-funded employer plan HTN · prediabetes · BMI 30.8
Twin state
Baseline expected claims$11,480 / yr
Composite risk (PICS)68 → 68
Signals in twin8 of 8
Continuity (90d)94%
Program adoption32%
Share of the covered population that enrols a twin and sustains verified signal for a full quarter. Population response is attenuated to 62% of the motivated-individual effect.
Sustained 90 days · creditable at next risk review
Composite risk
68 → 68
PICS Risk · 6-component
Risk reduction
0.0%
vs. baseline expectancy
Time to credit
90d
sustained verified signal
Signal confidence
92%
device + lab attested
◆ IRIS · AI teammate
Move a lever to see where the leverage is.
Place & Exposure
SAME PERSON · SAME GENETICS · SAME BIOLOGY
·DIFFERENT ENVIRONMENT, EXPOSURE AND ACCESS
Switch the ZIP — watch the trajectory move
8Where the twin lives
Life expectancy figures are real. 44103 (St. Clair–Superior) at 70 years and 44124
(Lyndhurst) at 82 are from the VCU Center on Society and Health Cleveland mapping; Pitkin County
(Aspen) at 86.5 is the IHME county estimate. The remaining environmental values are an
illustrative profile, not measured data for these ZIPs.
9Individual risk beyond biometrics
10Environmental exposure & access81611
12Projected trajectory● LIVE
Projected life expectancy
—
—
Healthspan
—
years before first major event
Place cost index
—
population context — not applied to any dollar here
Addressable gap
—
years you or a program can recover
Structural gap
—
needs policy, not a care plan
YouYour programPolicy only
◆ IRIS · AI teammate
Pick a ZIP to see what place does to this twin.
11What can actually be changedTOGGLE AN INTERVENTION · RECOVERY IS CAPPED AT THE ADDRESSABLE GAP
Place is the most dangerous variable on this screen
ZIP code is a strong proxy for race and income. Using it to set premium, eligibility or underwriting
is proxy discrimination, and it is actively regulated — Colorado's SB21-169 governs exactly this
class of external-data and algorithmic input, and one of the ZIPs above is in Colorado. This screen is
built to point the other way: it identifies where to send care, benefits and outreach, never
what to charge. The engine deliberately separates the gap into what a person can change, what a program
can change, and what only policy can change — because the first two are a product and the third
is not something to sell against.
Read this before you show it to an actuary
Every figure on this screen is produced by a transparent, published coefficient model over synthetic member data. It is a
decision-support projection, not a certified actuarial opinion, and it is not an offer, guarantee or determination of premium,
eligibility or benefit. Production use requires: a credentialed actuarial memorandum (ASOP 23 / 41 / 56), state DOI review wherever an
output touches rating, HIPAA/GINA/ADA wellness-incentive review, and member-directed consent for every signal in the ledger. The engine is
built so each of those reviews has something concrete to examine — open coefficients, full provenance, per-lever attribution.
Concept credit: Digital Twin as a biologic-proof and risk-repricing instrument — Dr. Ron Moody, with commercial framing from Dr. Shayan Vyas, CMO. CloudCare® DTWN · Life365, Inc. · Scottsdale, AZ · v1.0 · synthetic data
The genome is fixed. The outcome is not.
Clairvoyant does not edit anyone's DNA. What it changes is penetrance and expression —
the probability that a variant you already carry becomes a disease you actually get, and when.
A pathogenic MLH1 variant is permanent; death from the colon cancer it predisposes to is
largely optional, and the thing standing between the two is a colonoscopy on a calendar.
Digital Twin proves what already changed. Clairvoyant projects what is coming and then goes
and does something about it — under authority the member grants, at a level the member sets,
with hard stops the member cannot be talked out of.
1Subject & sequencingWES · 2026-04
MR
Marcus Reyes
M · 54 · Self-funded employer plan HTN · prediabetes · BMI 30.8 · former smoker
AssayWhole-exome + PRS panel
Consent on fileMember-directed · revocable
Reportable findings5
Confirmed bySanger · CAP/CLIA lab
2Genomic substrateCLICK A VARIANT
Variants are illustrative but real in kind. Penetrance figures are drawn from published
cohort literature — Lynch syndrome colorectal penetrance from the Prospective Lynch Syndrome
Database, heterozygous familial hypercholesterolaemia event rates from the Simon Broome and
Copenhagen cohorts, CYP2C19 and SLCO1B1 actionability from CPIC guidelines. The member is synthetic;
the mechanisms are not.
3Expression levers — what actually moves the outcomeALL CONDITIONS
A
Nothing changes
Variant carried, no surveillance, no therapy. The untreated natural history.
B
Today's care
What Marcus is actually receiving right now, as documented.
C
Guideline-concordant
Everything NCCN / ACC-AHA / CPIC says should be happening.
D
Everything available
Guideline care plus chemoprevention and add-on therapy.
Effects combine multiplicatively, never additively —
1 − Π(1 − eᵢ) within a condition,
capped, and then floored at population baseline risk. Surveillance and statins
do not make a carrier safer than the general population; they bring a carrier back toward it.
Any model that claims otherwise is selling something.
4Readout● LIVE
Lifetime risk — colorectal cancer
54%→54%
± 0 pts · 80% credible interval
Population reference for a 54-year-old male: 4.1%
Risk removed
0 pts
absolute, lifetime
Expected onset
—
median age, if it happens
Healthspan shift
—
years before first major event
Levers engaged
0
of 9 available
◆ IRIS · AI teammate
Pick a variant, then turn a lever on. I will tell you what moved and what did not.
5Family history3 GENERATIONS · SELF-REPORTED
malefemaleaffected⌿ proband
This is other people's genetic information. Under GINA, family medical history
is genetic information — about your relatives, who did not consent to this analysis.
It is used here only to calibrate your own projection. Nothing in this panel is disclosed to
anyone, and outreach to a relative happens only when you initiate it.
6Projected trajectoryCUMULATIVE INCIDENCE · AGE 54–85
Unmodified — variant carried, nothing doneModified — levers above appliedPopulation reference, same age & sex
What a geneticist will say about this screen
Three things, and all three are fair. One: penetrance estimates from referral-based cohorts are
biased upward, because families get sequenced after someone gets sick — the Prospective Lynch Syndrome
Database exists specifically to correct that, and the lower, prospective figures are the ones used here.
Two: a polygenic risk score built mostly on European-ancestry cohorts transfers poorly to everyone
else, and this engine therefore reports PRS with an explicit ancestry-transferability flag rather than a
single confident decile. Three: presenting a lifetime risk number to a member without a genetic
counsellor in the loop is how you cause harm with true information — so above a configured threshold
this screen will not render a projection until a counselling contact is attached to the record.
Tool 1
Power Proxy
The authority instrument. Nothing in Clairvoyant acts until the member says what may be done,
by whom, how far, and for how long. Autonomy is granted per action class — not globally —
and the hard stops are not configurable.
7Autonomy ceilingMEMBER-SET · CLINICAL DIRECTOR COUNTERSIGNS ABOVE L2
L1
Observe & notify
Advocate watches the record and tells you. It takes no action of any kind.
L2
Draft & queue
Advocate prepares the authorisation, referral or appeal. You or your clinician release it.
L3
Act under standing order
Advocate executes against a pre-signed clinician standing order and a named protocol. Bounded, logged, reversible.
L4
Act & reconcile later
Unsupervised clinical action with retrospective review. Hard-stopped in this build. Not offered.
Autonomy levels are the same L1–L4 ladder used by the CloudCare Virtual Clinical Assistant.
L4 exists in the specification so that the boundary is explicit — and it is disabled, for every
member, for every action class, in every environment. There is no toggle for it on this screen
because there is no toggle for it in the product.
8Granted authority8 CLASSES · 2 HARD-STOPPED
Action class
Granted
Scope & cap
Expires
Each grant is a separate, separately revocable instrument. Raising the ceiling above
L2 on any class requires a countersignature from the Clinical Director of record and
a named protocol; the countersignature is scoped to the class, not to the member.
9Hard stopsNOT DELEGABLE AT ANY LEVEL
Attestation of record. Standing orders covering the L3 grants above were reviewed and
signed by the named clinician; Advocate executes only inside those orders and stops at their edge.
Tool 2
Advocate
The action layer. Advocate takes the trajectory from the top of this screen and the authority
from Power Proxy, and then does the work — the scheduling, the prior authorisation, the appeal,
the records chase, the pharmacy flag. This is the AutoPilot: a loop that runs whether or not
anyone is watching, and reports what it did.
10The loopCONTINUOUS · LAST SWEEP 11 MIN AGO
1 · Project
Variants, family history and current biology produce a forward trajectory per condition.
2 · Identify
Compare the trajectory against guideline: what is owed to this member that is not happening.
3 · Authorise
Check Power Proxy. Act, draft, or stop — per action class, per ceiling.
4 · Execute
Book it, file it, appeal it, chase it. Re-attempt on failure. Escalate to a human on the second denial.
5 · Verify
Confirm the thing actually happened, then feed the result back to Digital Twin as verified signal.
11Advocate queue—
Action
Triggered by
Needs
Granted
Status
Counterparty
Queue items are generated from the levers above: turn a lever on and Advocate acquires the work
that lever implies. An item whose required autonomy exceeds the granted ceiling for its class is
not silently downgraded — it is shown as blocked, with the human it was routed to.
Member-hours spent: 0.4. That is the number this screen exists to move. The clinical case for
surveillance in a Lynch carrier has been settled for twenty years; what fails is the scheduling,
the authorisation, the reminder, the re-booking after a cancellation, and the appeal after a denial.
Advocate is not a smarter diagnosis. It is the administrative labour nobody has been paying for.
13Cascade screening — the people this also applies toMEMBER-INITIATED ONLY · L1
A pathogenic variant is the only test result that is simultaneously about four other people.
Each first-degree relative of a heterozygous carrier has a 50% prior probability of carrying
the same variant — and none of them asked to find that out. Advocate will draft the letter and
name the nearest genetic counsellor, and it will not send anything. The member sends it, or nobody does.
⚠ Genetic information is the most regulated input in this product
Title II of the Genetic Information Nondiscrimination Act prohibits an employer from requesting,
requiring, purchasing or using genetic information — which includes family medical history — in any
employment decision. Title I prohibits a group health plan from using it for eligibility, coverage,
underwriting or premium setting. The ADA independently restricts disability-related inquiry, and
state law goes further in several jurisdictions. Accordingly: no output of Clairvoyant reaches pricing,
rating, underwriting, eligibility, employment or any plan-sponsor-identifiable report. There is no
configuration that permits it, no export that contains it, and no aggregate view that re-identifies it.
A self-funded employer sponsoring this benefit receives participation and outcome counts at population
level and nothing else — the same firewall Digital Twin applies to geography, applied here to a
category that carries criminal and civil exposure rather than reputational exposure.
Clairvoyant is fed byCFW FHIR R4 labs, problem list & genomics (HL7 Genomics Reporting)DTWN verified biologic deltas · the proof layerPICS Risk 6-component compositeCCPR consent, proxy & revocation registryCAIG explainability, attestation & audit logORAK identity & delegated-authority tokensBIM prior authorisation & appealsCLVY penetrance projection & autonomous advocacy ← new
Read this before you show it to a clinician
Every figure here is produced by a transparent coefficient model over synthetic member data with a
synthetic variant set. It is decision support, not a clinical diagnosis, a genetic test report, or a
determination of coverage or eligibility. Production deployment requires: CAP/CLIA-validated assay and
confirmatory testing, an ACMG/AMP-classified variant call, genetic counselling in the member pathway,
clinician-signed standing orders for every L3 action class, GINA/ADA/HIPAA and state genetic-privacy
review, and a Clinical Director attestation per protocol. The engine is built so each of those reviews
has something concrete to examine — open coefficients, named guideline sources, per-action authority
provenance, and a complete reversal log.
Concept credit: Clairvoyant as the action successor to Digital Twin — Kent Dicks.
Penetrance-not-genome framing and the autonomy ladder follow the CloudCare VCA specification. CloudCare® CLVY · Life365, Inc. · Scottsdale, AZ · v1.0 · synthetic data
Every running IA workflow on the n8n Bot Factory, in real time. Each card shows one bot's mandate — trigger, model, memory, autonomy gate, action — with animated edges following live event streams.
16:42:11IA-01 nudge sent to Pt-604716:42:08IA-07 HARD STOP — chest pain Pt-582116:42:02IA-02 vitals OK Pt-610216:41:58IA-03 A1c gap closed Pt-591016:41:51IA-04 CHF lesson delivered Pt-618416:41:44IA-09 TCM 14d check Pt-601416:41:39IA-01 nudge sent to Pt-620116:41:33IA-05 visit booked Pt-591016:41:28IA-02 NEWS2 trend +2 Pt-610216:41:21IA-03 outreach sent Pt-604716:42:11IA-01 nudge sent to Pt-604716:42:08IA-07 HARD STOP — chest pain Pt-582116:42:02IA-02 vitals OK Pt-610216:41:58IA-03 A1c gap closed Pt-591016:41:51IA-04 CHF lesson delivered Pt-618416:41:44IA-09 TCM 14d check Pt-601416:41:39IA-01 nudge sent to Pt-620116:41:33IA-05 visit booked Pt-591016:41:28IA-02 NEWS2 trend +2 Pt-610216:41:21IA-03 outreach sent Pt-6047