Ozyra Review · Volume 01 · 2026Ozyra.
Ozyra Review · Volume 01
The Value
of Health
Intelligence.
Why longitudinal, multimodal care data can become Ozyra's most strategic asset.
CONTINUOUS CARE · LONGITUDINAL DATA · GOVERNED AI
HEALTH INTELLIGENCE
AS INFRASTRUCTURE
Ozyra Review · The Value of Health Intelligence01 / 50
Executive summary · Volume 01Ozyra.

Health intelligence begins
with the trajectory.

OZYRA REVIEW
2026

Ozyra's thesis is that health data becomes strategically valuable not because it is abundant, but because it can form a governed learning system that improves care.

01 · Continuous care

Care extends beyond the encounter.

Appointments remain essential, but much of the clinically meaningful trajectory unfolds between them. Continuous observation can make change visible earlier.

02 · Structured learning

Structure creates value.

Time-aligned signals become more useful when they connect physiology, function, behavior, interventions, adherence, and outcomes.

03 · Workflow integration

Predictions must enter care.

A model creates value only when its output supports a decision, improves a workflow, and is evaluated against what happens next.

04 · Governed intelligence

Trust is part of the asset.

Patient rights, lawful use, provenance, privacy, safety, clinical oversight, and measurable performance determine whether learning can endure.

Ozyra is building governed health intelligence infrastructure: a system that can represent change, support decisions, and learn from outcomes.
This review is written for everyone who helps shape, deliver, govern, or evaluate Ozyra's work.
Front matter · Executive summary02 / 50
Executive thesis · Five propositionsOzyra.

The value is not
in the record.

THESIS · 01
CORE PROPOSITION

The “value of one patient” is not a sale price for data. It is the incremental value created when a governed learning system can understand trajectories, predict outcomes, and improve care.

01
Care becomes continuous.
episode → trajectory
02
Data becomes useful when it is time-aligned and outcome-linked.
volume → structure
03
Intervention-response pairs are more valuable than passive archives.
observation → learning
04
Model value compounds only when predictions enter workflow and are evaluated.
model → system
05
Trust, rights, provenance, and clinical governance are part of the asset.
access → legitimacy
Ozyra's strategic asset is not a database of people. It is a governed system that learns from care.
Interpretation note. Patient Intelligence Value is a non-financial, use-case-specific contribution framework—not an accounting fair value, a market quote, or a rationale to sell personal health information.
Front matter · Core thesis03 / 50
Contents · Volume 01Ozyra.

The argument,
at a glance.

OZYRA REVIEW
VOLUME 01
02
Executive summary
The central argument and its implications.
1 page
03
Core thesis
Five propositions for governed health intelligence.
1 page
05–14
Part I · The Shift
From encounters to trajectories and governed learning systems.
10 pp
15–19
Part II · Learning Value
Depth, density, linkage, rights, and governance.
5 pp
20–25
Part III · Contribution Profile
How one governed profile can contribute to a defined question.
6 pp
26–30
Part IV · Intelligence Flywheel
How care, learning, adoption, and trust create a measurable feedback system.
5 pp
31–36
Part V · Intelligence Architecture
Platform layers, temporal representation, lineage, applications, and safety.
6 pp
37–42
Part VI · Value Realization
Scale scenarios, value logic, responsible revenue, and defensibility.
6 pp
43–45
Part VII · Strategic Choices
System risks and the boundary between build, partner, and shared control.
3 pp
46–49
Part VIII · Roadmap + Proof
Sequenced execution, stakeholder evidence, and the closing thesis.
4 pp
50
Sources + method
Declared evidence and analytical discipline.
1 page
Front matter · Contents04 / 50
Part I · The ShiftOzyra.
Part I · The Shift
The
Shift.
Healthcare is moving from a sequence of visits to a continuously observed - and increasingly learnable - system.
2.4bn
people may benefit from rehabilitation
75%
of non-pandemic deaths from noncommunicable diseases in 2021
9.3%
average OECD GDP allocated to health in 2024
Sources · WHO (2024; 2025) · OECD Health at a Glance 2025. Figures provide system context, not Ozyra market-size estimates.
Chapter 01 · Context05 / 50
Part I · Plate 01 · Care modelOzyra.

Healthcare was built
for the encounter.

FIGURE 01
EPISODE / TRAJECTORY

Modern care delivery is organized around appointments, admissions, procedures, and discharges. That structure is necessary: it coordinates people, facilities, reimbursement, and accountability. But it also compresses a complex health trajectory into a small number of timestamped fragments.

For chronic disease and rehabilitation, many of the decisive changes unfold elsewhere: pain fluctuates; blood pressure varies; sleep deteriorates or improves; medication is taken or missed; cognition, mood, mobility, and confidence change; a prescribed program is followed, adapted, or abandoned.

Core implication

The strategic unit of analysis moves from the visit to the trajectory.

The visit remains clinically essential. The shift is that intelligence requires a representation of what happens before, between, and after visits.

Sparse care record · six visible events
Assessment
Home
Review
Home
Adjustment
Home
Outcome
Continuous health trajectory · state changes between events
FUNCTION / PHYSIOLOGY / EXPERIENCEILLUSTRATIVE · NOT CLINICAL DATA
Chapter 01 · From episodes to trajectories06 / 50
Part I · Plate 02 · MeasurementOzyra.

The patient exists
between appointments.

FIGURE 02
SIGNAL LANES

Remote measurement expands the observation surface. It does not automatically create intelligence.

Digital health technologies can acquire physiological and other data remotely. That makes continuous or higher-frequency observation possible, reduces dependence on in-clinic measurement, and can make participation more convenient. But a pile of sensor readings is not a patient model.

Data becomes a useful learning substrate when signals are time-aligned, calibrated, contextualized, and linked to interventions and outcomes. Missingness must be understood. A sudden gap can mean recovery, device failure, disengagement, hospitalization, or something else entirely.

1 in 6
people globally will be 60+ by 2030
2.4bn
may benefit from rehabilitation
Physiology
blood pressure · pulse · EEG · sleep
Function
mobility · cognition
Behavior
adherence · activity
Experience
pain · patient reports · quality of life
Intervention
dose · timing · change
Sources · FDA, “Digital Health Technologies for Remote Data Acquisition in Clinical Investigations” (2023); WHO, “Ageing and health” (2025); WHO, “Rehabilitation” (2024).
Chapter 01 · The observation surface07 / 50
Part I · Plate 03 · Data modelOzyra.

The record is
not the model.

FIGURE 03
SYSTEM COMPARISON

The electronic health record is an indispensable system of record. Ozyra's opportunity is to add the longitudinal representation that a learning system needs.

System of record

What happened?

  • Who documented, ordered, or signed?
  • Which diagnosis, code, result, or procedure?
  • What is the legal and operational record?
  • What must be exchanged or retained?
  • What happened at a particular event?
+
System of intelligence

How is state changing?

  • Which signals move together over time?
  • What preceded improvement or deterioration?
  • How did the patient respond to an intervention?
  • What outcome is plausible - with what uncertainty?
  • What action should be surfaced to whom?
The bridge · shared foundations

Identity + rights

Consent, lawful basis, permissions, access policy, and patient control.

Semantics + time

Terminology, units, timestamps, provenance, calibration, and missingness.

Outcomes + evaluation

Clinically meaningful labels, uncertainty, subgroup performance, and audit.

Do not replace the record. Make it learnable - without weakening its integrity.
The European Health Data Space establishes a framework for access, control, exchange, interoperability, and secure secondary use. Interoperability is necessary; representation quality and governance determine whether data can support reliable learning.
Chapter 01 · Record plus intelligence08 / 50
Part I · Plate 04 · Learning signalOzyra.

From observations to
intervention-response pairs.

FIGURE 04
LEARNING SEQUENCE

Passive observations can reveal patterns. Their value rises when Ozyra also knows what was intended, what was delivered, when it happened, how much was received, and what followed.

An intervention-response pair is a structured episode that links baseline state, a defined action, exposure or adherence, short-term response, meaningful outcome, and follow-up. Repeated across comparable patients, these pairs create a stronger basis for personalization and prospective testing.

Critical caveat

Sequence is not automatically causality.

Confounding, selection effects, measurement error, and changes in care still matter. Observational learning can generate and improve predictions; causal claims require appropriate study designs and validation.

01 · STATE
Baseline
Symptoms, function, physiology, context, prior treatment.
02 · ACTION
Intervention
Protocol, timing, intended dose, clinician rationale.
03 · EXPOSURE
Delivered
Actual dose, adherence, adaptation, interruptions.
04 · CHANGE
Response
Early signal movement, adverse events, engagement.
05 · LABEL
Outcome
Function, quality of life, clinical endpoint, durability.
Why it matters

Personalization

Estimate which patient may respond to which pathway, with uncertainty.

Why it matters

Operations

Detect non-response, friction, or risk earlier and route attention.

Why it matters

Evidence

Generate hypotheses and support prospective, clinically governed validation.

Strategic implication. A dataset that connects actions to outcomes is more difficult to recreate than an archive of observations, because it depends on care integration, workflow discipline, longitudinal follow-up, and trusted patient participation.
Chapter 01 · The learnable episode09 / 50
Part I · Plate 05 · Intelligence layerOzyra.

AI changes the unit of care -
not accountability.

MODEL ROLE
DECISION RIGHTS

AI can make a trajectory easier to interpret and act on. Responsibility remains with the people and organizations that define, deploy, use, and oversee the system.

System support

What AI may support

  • Synthesize time-aligned evidence and known gaps.
  • Estimate risk, recovery, adherence, or response.
  • Prioritize review and surface possible next steps.
  • Communicate uncertainty and relevant limitations.
  • Monitor performance and signal possible drift.
+
Human accountability

What people must retain

  • Define purpose, intended use, and acceptable risk.
  • Interpret personal context, goals, and preferences.
  • Make, communicate, and document consequential decisions.
  • Override, escalate, pause, or stop unsafe use.
  • Investigate harm and approve controlled change.
Product + data

Design

Own intended use, data contracts, measurement quality, and limitations.

Clinical

Decide

Judge relevance, act in context, override, and remain accountable for care.

Safety + governance

Control

Set rights and risk thresholds, monitor use, investigate, and authorize change.

Patients + participants

Participate

Express goals, understand use, ask questions, exercise rights, and seek redress.

Automation may change who sees what and when. It cannot dissolve responsibility.
Sources · WHO, “Artificial Intelligence for Health” (2024); WHO, “Ethics and governance of artificial intelligence for health” (2021).
Chapter 01 · AI as coordination layer10 / 50
Part I · Plate 06 · Compounding assetOzyra.

The asset is the
learning system.

FIGURE 05
WORKING DEFINITIONS

Six working definitions keep the thesis precise. They separate what is stored from what is represented, learned, and changed.

01 · RECORD
Authoritative documentation
A legally and operationally meaningful account of events, orders, results, decisions, and authorship.
02 · OBSERVATION
A signal in context
A time-stamped measurement or report with identity, source, units, quality, provenance, and known uncertainty.
03 · TRAJECTORY
Change over time
An ordered, contextual sequence of observations that makes state, transitions, and gaps visible.
04 · CARE EPISODE
Action linked to result
Baseline, intended action, delivered exposure, response, outcome, and follow-up connected as one evaluable episode.
05 · REPRESENTATION
A computable state estimate
A versioned abstraction of the trajectory that preserves relevant context, missingness, provenance, and uncertainty.
06 · LEARNING SYSTEM
Evidence into controlled change
People, processes, technology, rights, and governance that turn experience into evaluated improvement.
The database stores objects. The learning system links them to decisions, actions, and results.
Strategic implication. Ozyra's position depends on the quality of these connections and the ability to evaluate them - not on record volume alone.
Chapter 01 · From stored objects to a learning system11 / 50
Part I · Plate 07 · European contextOzyra.

A European window
is opening.

FIGURE 06
EHDS RUNWAY

The European Health Data Space Regulation entered into force in March 2025. It establishes a common framework for individuals' access to and control over electronic health data, cross-border exchange for care, and secure secondary use for research, innovation, policy, and regulatory activity.

Application is phased. Key primary-use and secondary-use rules begin in 2029 for major categories, with additional categories following in 2031. The runway matters: Ozyra can design its data model, permissions, provenance, interoperability, and access controls before the operating environment matures.

The important distinction

Reuse is not permissionless exploitation.

EHDS is a governance framework, not a blanket authorization to commercialize personal health data. Ozyra must evaluate each use case against applicable law, purpose, rights, safeguards, and implementation rules.

2025
Regulation in force
Transition begins
2027
Implementing acts
Operational rules mature
2029
Key rules apply
First priority data + most secondary-use categories
2031
Further categories
Images, labs, discharge reports; remaining secondary-use categories
2035
External access
Eligible third countries and international organizations
01

Provenance

Know where every signal came from and how it changed.

02

Permissions

Make access purpose-specific, auditable, and revocable where required.

03

Minimization

Use the least data and identity exposure needed for the purpose.

04

Evaluation

Track performance, drift, and harm across groups and settings.

Source · European Commission, European Health Data Space Regulation (EU) 2025/327 and implementation timeline. Exact obligations depend on use case and implementation; obtain specialist legal advice before operational decisions.
Chapter 01 · Governance runway12 / 50
Part I · Plate 08 · Ozyra implicationsOzyra.

What the shift means
for Ozyra.

SIX MANDATES
DESIGN + OPERATING MODEL

If Ozyra intends to own and operate the intelligence layer, the strategy must shape the care workflow from the first measurement onward.

01
Own the canonical health trajectory model.
Define the temporal schema, terminology, provenance, uncertainty, and interoperability contracts that every product uses.
02
Instrument care around time.
Capture what happens before, during, and after a care episode - not only what appears in a note.
03
Capture intended and delivered intervention.
Store protocol, timing, dose, adaptation, adherence, and clinician rationale as first-class data.
04
Treat outcomes and uncertainty as core labels.
Optimize for meaningful endpoints, durable follow-up, missingness semantics, and calibrated predictions.
05
Design rights and trust at acquisition.
Make purpose, consent or lawful basis, access, provenance, privacy safeguards, and audit operational from day one.
06
Value lift, not raw records.
Measure model improvement, workflow impact, outcomes, retention, and new product option value - not a hypothetical data sale.
Ozyra can own infrastructure, model IP, and lawful operating rights without claiming ownership over people or their identity.
Positioning. “We own the data” is strategically imprecise and ethically weak. “We operate the governed learning infrastructure and own the resulting technology and model IP, subject to patient rights and applicable law” is stronger.
Chapter 01 · Strategic mandates13 / 50
Part I · Plate 09 · Chapter closeOzyra.
“The company that learns from care - safely, continuously, and measurably - can become more valuable with every treated patient.”
OZYRA THESIS · VOLUME 01

The shift from data to intelligence is not a technology upgrade alone. The next question is which data can actually support learning—and which data only adds volume, cost, or risk.

TEST · 01
Depth
Does the data describe meaningful change over sufficient time?
TEST · 02
Actionability
Can it improve a prediction, workflow, intervention, or outcome?
TEST · 03
Trust
Can it be used lawfully, safely, transparently, and repeatedly?
Next · Part II · Conditions for Learning Value14 / 50
Part II · Conditions for Learning ValueOzyra.
Part II · Conditions for Learning Value
The
Useful
Dataset.
Health data becomes strategically useful when it is structured to explain change, evaluate action, and support lawful reuse.
Time
signals share a coherent sequence
Action
interventions connect to outcomes
Trust
rights and safeguards travel with the data
Framework · Ozyra thesis · informed by FAIR, TEHDAS, and the European Health Data Space.
Chapter 02 · Conditions for learning value15 / 50
Part II · Plate 01 · Data hierarchyOzyra.

Not all data has the same
learning value.

FIGURE 07
VALUE LADDER

A record can be accurate, complete, and operationally essential without being sufficient for learning. Its strategic usefulness depends on the question it can answer and the context that travels with it.

Value rises when isolated events become coherent trajectories; when trajectories connect an intervention to a response; and when those patterns produce validated improvement in a real workflow.

Core implication

Volume is the base layer—not the destination.

More rows increase storage and responsibility. Better structure increases the chance that data can improve a representation, prediction, decision, or outcome.

01
Events
A diagnosis, measurement, note, order, or encounter. Useful for documentation and counting.
02
Trajectories
Events organized over time, with state, context, provenance, and missingness.
03
Action + response
A defined intervention, actual exposure, observed change, and meaningful outcome.
04
Validated lift
Evidence that the learned pattern improves performance, workflow, or care in context.
Interpretation. This ladder is a decision heuristic, not a universal price list. A lower layer may be decisive for one use case; a higher layer only matters when its quality, relevance, rights, and evaluation are adequate.
Chapter 02 · From volume to validated lift16 / 50
Part II · Plate 02 · Longitudinal depthOzyra.

Time creates context—
not just more rows.

FIGURE 08
DEPTH PROFILES

The number of measurements says little about whether a dataset describes meaningful change.

Longitudinal depth combines duration, cadence, phase coverage, and interpretable gaps. A dense burst can reveal short-term variability; a long sparse series can reveal drift; a care episode with baseline, intervention, response, and follow-up can support evaluation.

The right profile depends on the decision. Recovery, deterioration, relapse, adherence, and treatment response unfold on different clocks. Ozyra therefore needs measurement plans tied to clinical purpose—not a single definition of “continuous.”

Dense burst
High local resolution; little long-horizon context.
Long + sparse
Long horizon; important change may fall between samples.
Care trajectory
Baseline, action, response, adaptation, and outcome.
01 · Duration

Long enough

Cover the horizon on which the relevant outcome can change.

02 · Cadence

Often enough

Sample at a frequency that can reveal clinically relevant variation.

03 · Phases

Across transitions

Observe before, during, and after consequential interventions.

04 · Missingness

Explained

Distinguish recovery, disengagement, device failure, and care elsewhere.

Sources · FDA, Digital Health Technologies for Remote Data Acquisition (2023); Morin et al., Nature Cancer (2021).
Chapter 02 · Measurement with temporal purpose17 / 50
Part II · Plate 03 · Multimodal densityOzyra.

Modalities gain value
when they share a clock.

FIGURE 09
ALIGNED SIGNALS

Blood pressure, mobility, sleep, cognition, symptoms, notes, adherence, and imaging each describe a different facet of health. Their combined value does not come from simply placing them in the same warehouse.

They become mutually informative when they refer to the same person, time window, care context, intervention, and outcome—with known units, provenance, calibration, and uncertainty.

Physiology
state + variability
Function
capability + recovery
Experience
symptoms + quality of life
Intervention
action + dose + adherence
Outcome
meaningful endpoint
01
Identity
Signals resolve to the correct person and episode.
02
Time
Clocks, windows, ordering, and latency are explicit.
03
Meaning
Units, terminology, context, and calibration are shared.
04
Quality
Noise, missingness, provenance, and uncertainty are visible.
05
Outcome
Signals connect to a decision-relevant endpoint.
Multimodal does not mean “collect everything.” It means combine the signals that clarify the decision.
Sources · Moor et al., Foundation models for generalist medical artificial intelligence, Nature (2023); FAIR Guiding Principles (2016).
Chapter 02 · Synchronization before scale18 / 50
Part II · Plate 04 · Rights + governanceOzyra.

Usability is bounded
by rights.

FIGURE 10
GOVERNANCE GATES

A dataset can be technically rich and still be unusable for a particular purpose. Lawful basis, permits, patient rights, contractual scope, provenance, security, and prohibited uses determine what can actually be done.

Governance principle

Governance makes legitimate use repeatable.

Clear purpose and dependable controls make responsible use easier to explain, audit, and scale.

01 · PURPOSE
Why?
Define the clinical, research, operational, or regulatory purpose before selecting data.
02 · AUTHORITY
May we?
Confirm lawful basis, permissions, permits, contracts, and relevant patient rights.
03 · MINIMIZE
What is needed?
Use the least data and identity exposure required for the stated purpose.
04 · PROTECT
How safely?
Apply access controls, pseudonymisation or anonymisation, secure processing, and monitoring.
05 · ACCOUNT
Can we prove it?
Log access and outputs; evaluate quality, bias, drift, harm, and compliance over time.
Findable + interpretable

Documentation

Metadata, data dictionary, standards, source, provenance, and quality indicators.

Permitted + protected

Controls

Purpose limitation, minimization, approved access, secure environments, prohibited uses.

Repeatable + accountable

Evidence

Audit trails, versioning, performance evaluation, incident handling, and retirement.

Sources · Regulation (EU) 2025/327; European Commission, Reuse of health data; TEHDAS European Health Data Space Data Quality Framework. Exact requirements depend on the use case and applicable law.
Chapter 02 · Rights as an intrinsic quality dimension19 / 50
Part III · The Contribution ProfileOzyra.
Part III · The Contribution Profile
The
Contribution
Profile.
A governed observation profile contributes differently to different learning questions. Its relevance depends on purpose, evidence, context, and permissible use.
Question
begin with a named decision or learning task
Evidence
trace every assessment to observable inputs
Limits
an index is never a price, permission, or clinical grade
OZYRA FRAMEWORK · USE-CASE SPECIFIC · VERSIONED · NON-FINANCIAL
Chapter 03 · A non-financial contribution framework20 / 50
Part III · Plate 01 · PIV frameworkOzyra.

A useful score begins
with a question.

FIGURE 11
PIV FRAMEWORK

Patient Intelligence Value describes how well a governed observation profile can support one named learning question.

PIV is not a patient valuation. It is a versioned decision framework. The primary output is an eligibility status, a seven-driver contribution profile, and the evidence behind each assessment.

A composite index is optional. It only supports comparison inside the same use case, rubric version, scoring anchors, and weights. Change the question or baseline, and the interpretation changes.

Indexu = 25 × Σ (wu,k × sk)
Eligibility first
sk ∈ {0,1,2,3,4}
Σwu,k = 1
01 · ELIGIBILITY
Pass the gates.
Purpose, authority, rights, minimization, safeguards, and accountability must support the defined use.
02 · PROFILE
Show the drivers.
Report all seven scores, supporting evidence, uncertainty, known gaps, weights, and rubric version.
03 · INDEX
Compare carefully.
A rounded 0–100 index is optional and meaningful only within one declared use case.
01 · QUESTION
Name the decision.
Prediction, workflow, research, or care question.
02 · ELIGIBILITY
Confirm the use.
Rights, purpose, controls, and accountability.
03 · EVIDENCE
Define the profile.
Window, signals, provenance, and known gaps.
04 · SCORE
Apply anchors.
Assess each driver from zero to four.
05 · REPORT
Expose assumptions.
Profile, optional index, sensitivity, and version.
Boundary. PIV does not authorize data use, provide clinical advice, establish causal evidence, or measure economic value. A failed eligibility gate returns “not eligible for this use,” not zero.
Chapter 03 · A transparent contribution framework21 / 50
Part III · Plate 02 · Seven driversOzyra.

Seven drivers describe
the profile.

FIGURE 12
DRIVER REGISTER

Each driver answers a different question about learning contribution. None measures the patient.

D1 · DEPTH
Temporal depth
Does the window span the baseline, change, and follow-up relevant to the use case?
Example weight · 15%
D2 · CADENCE
Phase coverage
Are observations frequent enough—and placed at the transitions that matter?
Example weight · 20%
D3 · MODALITIES
Signal density
Do independent, synchronized signals clarify the same patient state?
Example weight · 10%
D4 · INTERVENTION
Fidelity
Are intended care, delivered dose, adherence, adaptations, and interruptions traceable?
Example weight · 20%
D5 · OUTCOME
Linkage
Is change connected to a meaningful endpoint, with timing, uncertainty, and durability?
Example weight · 15%
D6 · COVERAGE
Contribution
Does the profile reduce a documented evidence gap across states, settings, or groups?
Example weight · 5%
D7 · QUALITY
Provenance
Are completeness, validity, calibration, semantics, missingness, and lineage adequate?
Example weight · 15%
GATE · ELIGIBILITY
May it be used?
Purpose, authority, rights, minimization, safeguards, and accountability are evaluated before contribution.
Gate · not a weighted driver
0
Absent
Missing or unusable for this question.
1
Fragmentary
Limited evidence with material gaps.
2
Adequate
Usable, but important limitations remain.
3
Strong
Well evidenced and fit for the stated use.
4
Extensive
Unusually complete and independently verifiable.
Coverage, not rarity. An identity trait is never intrinsically more valuable. A documented evidence gap may matter, but it never overrides rights, fairness, participant benefit, or acquisition burden.
Chapter 03 · Drivers of learning contribution22 / 50
Part III · Plate 03 · Worked profileOzyra.

One profile, scored
for one question.

FIGURE 13
SYNTHETIC EXAMPLE

A transparent score should make its reasoning easier to inspect than its headline number.

DEFINED QUESTION · HOME REHABILITATION
Can this 12-week profile help identify emerging non-response and route timely clinical review?
Synthetic evidence · baseline function and patient-reported outcome; prescribed, delivered, and adapted plan; daily activity and adherence; weekly pain and patient report; function at weeks 6 and 12; 9% missing wearable days, mostly annotated; no post-program durability outcome.
Illustrative
index · v1.0
79/100
Eligible*
gates assumed
WEEK 00
Baseline
Function, pain, patient report, activity, care plan, and starting context.
WEEKS 01–05
Exposure
Delivered exercises, adherence, symptoms, activity, and early response.
WEEKS 06–11
Adaptation
Midpoint function, protocol changes, continued exposure, and non-response signals.
WEEK 12
Outcome
Function and patient report; durability beyond the program remains unknown.
D1 · Depth · 15%
3/4
D5 · Outcome · 15%
3/4
D2 · Cadence · 20%
3/4
D6 · Coverage · 5%
2/4
D3 · Modalities · 10%
3/4
D7 · Quality · 15%
3/4
D4 · Intervention · 20%
4/4
Index
25 × Σ(weight × score)
78.75
The score is useful because its reasons are visible. Change the question, evidence, or weights—and the result should change.
Synthetic example. No patient data were used. Eligibility is assumed only for illustration. This rubric is a strategic framework, not a validated clinical, legal, or psychometric instrument.
Chapter 03 · Worked rehabilitation profile23 / 50
Part III · Plate 04 · Marginal contributionOzyra.

The next observation matters
when it changes what can be learned.

FIGURE 14
NEXT OBSERVATION

Marginal learning contribution asks how much a new observation changes understanding for a defined question. The same measurement can resolve uncertainty inside a gap—or be redundant inside an already dense interval.

Contextual contribution

MLC(x | S,u) = PIV(S ∪ {x},u) - PIV(S,u)

MLC means marginal learning contribution - not monetary value. The result depends on the existing evidence set S and use case u.

documented coverage gap already dense segment ADDITIONAL OBSERVATIONS MARGINAL LEARNING CONTRIBUTION
Redundant evidenceCoverage or linkage gap
Repeated

Little changes

Another stable-day activity reading may add little when the interval is already dense.

Linking

Context appears

Connecting a protocol adaptation to the response that followed can resolve ambiguity.

Coverage-expanding

A gap closes

A durability outcome or documented representation gap may improve transferability.

Burden belongs in the decision. New collection is justified only when expected benefit is proportionate to participant effort, privacy exposure, operational cost, and potential inequity. Source · Ghorbani & Zou, Data Shapley (2019).
Chapter 03 · The value of the next observation24 / 50
Part III · Plate 05 · Sensitivity + boundariesOzyra.

A score is credible only
when its assumptions can move.

FIGURE 15
SENSITIVITY

PIV should expose which evidence and assumptions drive the result—not hide them behind a single number.

Conservative scenario
64/100
Delivered intervention cannot be independently verified. Intervention fidelity falls from 4 to 1; other assumptions remain fixed.
Reference profile
79/100
Seven declared scores and fixed use-case weights. No post-program durability outcome is available.
Extended follow-up
86/100
A four-week durability outcome is added. Temporal depth and outcome linkage rise from 3 to 4.
01
Not a price.
PIV does not express the worth of a person, record, or dataset in money.
02
Not permission.
A score cannot create consent, lawful basis, access rights, or a permit.
03
Not a clinical grade.
It does not assess health, prognosis, adherence quality, or deservingness.
04
Not portable.
Scores compare only within the same use case, rubric, gates, weights, and version.
Publish the question, eligibility, rubric version, weights, driver evidence, and sensitivity range with every reported index.
Robustness

Stress-test missingness, label reliability, subgroup performance, weights, and the evidence baseline. Rounded movements are mechanical scenarios—not measured clinical gains.

Retirement rule

Revise or retire PIV if it fails to anticipate decision-relevant usefulness, creates incentives that conflict with patient benefit, or produces unstable or inequitable assessments.

Sources · FDA, Real-World Data guidance (2024); EMA, Data Quality Framework for Real-World Data (2026). The PIV rubric itself is an Ozyra strategic framework and has not been clinically validated.
Chapter 03 · Sensitivity, interpretation, and limits25 / 50
Part IV · The Intelligence FlywheelOzyra.
PART IV · OPERATING SYSTEM
Learning compounds only when the loop closes.
A learning system connects governed observation to action, outcome, evaluation, and controlled change. Scale matters only when each turn produces evidence that can improve the next.
Observe
capture state, action, context, and outcome
Act
place useful support inside accountable work
Verify
measure benefit, burden, safety, and equity
OZYRA OPERATING THESIS · CONTINUOUS LEARNING REQUIRES CONTINUOUS ACCOUNTABILITY
Chapter 04 · From contribution to compounding26 / 50
Part IV · Plate 01 · Operating loopOzyra.

A flywheel is a sequence
of accountable handoffs.

FIGURE 16
OPERATING LOOP

The loop is not a promise of automatic growth. It is an operating model in which every handoff has an owner, an evidence requirement, and a stopping rule.

Evidence returns to care.governed · traceable · evaluated
01 · QUESTION
Define the decision.
Name the user, action, intended benefit, and acceptable risk.
02 · OBSERVE
Capture context.
State, intervention, exposure, outcome, provenance, and known gaps.
03 · EVIDENCE
Test the signal.
Represent, compare, validate, and quantify uncertainty.
04 · WORKFLOW
Support an action.
Place the output where a trained person can review and use it.
05 · RESULT
Measure what followed.
Use, override, experience, outcome, burden, and safety.
06 · CHANGE
Improve deliberately.
Investigate gaps, approve changes, revalidate, and document.
Closure test. A prediction does not close the loop. The loop closes when a supported action and its consequences are observed, evaluated, and used to inform a controlled next step. Sources · AHRQ, About Learning Health Systems; National Academy of Medicine, Learning Health System.
Chapter 04 · The accountable operating loop27 / 50
Part IV · Plate 02 · Reinforcing loopsOzyra.

One system,
five reinforcing loops.

FIGURE 17
LOOP REGISTER

Compounding is strongest when care, learning, product, adoption, and trust improve together. A gain in one loop cannot permanently compensate for failure in another.

01 · CARE
From signal to a better next step.
signal → review → action → outcome → plan
Evidence: appropriate action occurs sooner, with better outcomes or lower burden.
02 · LEARNING
From experience to more reliable evidence.
trajectory → evaluation → error analysis → version
Evidence: calibration, utility, or robustness improves across relevant groups and settings.
03 · PRODUCT
From friction to a usable workflow.
friction → design change → use → feedback
Evidence: appropriate use rises while time, clicks, interruptions, or avoidable work fall.
04 · ADOPTION
From proven utility to broader participation.
benefit → repeat use → reach → coverage
Evidence: meaningful use and retention expand without coercion or hidden workload.
05 · TRUST
From accountability to durable permission.
clarity → control → confidence → responsible reuse
Evidence: rights are exercisable, incidents are handled, and participation remains informed.
A loop is healthy when it produces evidence of benefit and exposes evidence of harm.
Framework · Ozyra thesis, informed by AHRQ learning health system principles and WHO guidance on ethics and governance of AI for health.
Chapter 04 · Five mutually dependent loops28 / 50
Part IV · Plate 03 · Leading indicatorsOzyra.

Measure whether the loop
is actually turning.

FIGURE 18
INDICATOR REGISTER

The first useful metrics describe the quality and speed of feedback. They precede scale claims and sit alongside clinical, workflow, safety, equity, and participation outcomes.

Indicator
What it reveals
Required evidence
Eligible trajectory coverageobservation
How much of the declared population and care pathway passes the rights, relevance, and quality gates.
Publish numerator, denominator, exclusions, and reasons for missingness.
Outcome-linkage yieldlearning
The share of eligible episodes connected to a defined outcome within the relevant follow-up window.
Declare the outcome, timing, loss to follow-up, and label quality.
Feedback latencyspeed
How long it takes for an action and its consequences to become reviewable evidence.
Show the distribution and slow tail, not only the median.
Appropriate action rateworkflow
Whether surfaced outputs are reviewed, used, or correctly overridden in the intended context.
Record non-use, override reasons, escalation, and alert burden.
Verified liftresult
Change against an agreed baseline or comparator in a clinical, functional, or workflow outcome.
Report effect size, uncertainty, burden, and adverse consequences.
Context stabilitysafety + equity
Whether performance and calibration remain adequate across time, sites, devices, and relevant groups.
Expose gaps, drift, confidence, and predefined response thresholds.
Trust + participationlegitimacy
Whether people understand the use, remain engaged, and can exercise rights without losing appropriate care.
Track retention, comprehension, requests, complaints, incidents, and resolution.
Interpretation. No single indicator proves compounding. The register should be versioned by use case, paired with thresholds and owners, and reviewed with the people affected. Sources · AHRQ; IMDRF Good Machine Learning Practice (2025).
Chapter 04 · Evidence before scale29 / 50
Part IV · Plate 04 · Break pointsOzyra.

Every flywheel has
break points.

FIGURE 19
FAILURE REGISTER

The system should make weak links visible early. A pause is a control mechanism, not a failure of ambition.

01
Learning without a question
More variables, dashboards, and experiments appear, but no decision or intended benefit is named.
Response · define the use case, user, outcome, and stopping rule.
02
Data without context
Signals cannot be linked reliably to state, intervention, exposure, provenance, or follow-up.
Response · repair representation and missingness before model expansion.
03
Output without workflow
Performance looks promising offline, but the right person never sees or can act on the result.
Response · redesign the handoff, ownership, timing, and human oversight.
04
Adoption without outcomes
Usage grows while benefit, burden, overrides, and unintended consequences remain unmeasured.
Response · connect use to comparator-based outcome and safety evaluation.
05
Change without control
Data, workflow, or models change faster than validation, documentation, monitoring, and accountability.
Response · pause affected use, investigate drift or harm, approve a versioned change, revalidate, and communicate.
A trustworthy flywheel can slow down, stop, or reverse when the evidence requires it.
Sources · WHO, Ethics and governance of artificial intelligence for health (2021); IMDRF, Good Machine Learning Practice for Medical Device Development (2025). Regulatory classification and exact obligations depend on the specific use.
Chapter 04 · Detect, pause, repair, and revalidate30 / 50
Part V · Intelligence ArchitectureOzyra.
PART V · PLATFORM CAPABILITY
The model is not the product. The system is.
Ozyra's architecture must connect heterogeneous signals to a stable longitudinal representation, useful applications, and an evidence trail that supports oversight and change.
Connect
exchange data without losing provenance or purpose
Represent
model state, change, context, and uncertainty
Govern
control use, evaluation, monitoring, and change
OZYRA PLATFORM THESIS · REUSABLE CORE · USE-CASE SPECIFIC EVIDENCE
Chapter 05 · Intelligence architecture31 / 50
Part V · Plate 01 · Platform mapOzyra.

A governed path from
signal to decision.

FIGURE 20
PLATFORM MAP

The platform separates exchange, representation, intelligence, and application logic so each layer can be tested, replaced, and governed without breaking the whole.

01 · SOURCES
Care, devices, people, and research
Clinical records, patient reports, sensors, imaging, waveforms, protocols, and outcomes.
02 · EXCHANGE
Ingestion + interoperability
APIs, Fast Healthcare Interoperability Resources (FHIR), files, event streams, terminology mapping, identity resolution, and quality checks.
03 · CORE
Canonical longitudinal representation
State, change, intervention, exposure, context, missingness, uncertainty, provenance, and rights.
04 · INTELLIGENCE
Reusable model + evaluation services
Retrieval, features, embeddings, prediction, calibration, rules, explanation, monitoring, and versioning.
05 · APPLICATIONS
Use-case specific workflows
Monitoring, decision support, protocol matching, operations, research tools, and participant feedback.
Interoperability moves information. Representation makes change learnable.
Sources · HL7 Fast Healthcare Interoperability Resources (FHIR) R5; European Health Data Space Regulation. FHIR supports exchange and structure; local profiles, terminology, validation, and clinical context remain necessary.
Chapter 05 · Layered and replaceable by design32 / 50
Part V · Plate 02 · Temporal representationOzyra.

Health is a state
that changes.

FIGURE 21
STATE SEQUENCE

A temporal model must preserve what was known, what changed, what action occurred, and what remained uncertain at each point in the trajectory.

01 · BASELINE
Starting state
Symptoms, function, physiology, context, prior care.
02 · INTENT
Planned action
Goal, protocol, rationale, timing, intended dose.
03 · EXPOSURE
What occurred
Delivered dose, adherence, adaptation, interruption.
04 · RESPONSE
Early change
Signal movement, experience, side effects, engagement.
05 · OUTCOME
Meaningful result
Function, quality of life, clinical endpoint, burden.
06 · FOLLOW-UP
Durability
Persistence, relapse, new context, next decision.
Time
Event time, recording time, interval, cadence, and latency are distinct.
Missingness
A gap can reflect recovery, disengagement, device failure, or care elsewhere.
Uncertainty
Measurement error, model confidence, and unknown context travel with the estimate.
Context
Setting, goals, comorbidity, environment, and workflow shape interpretation.
Provenance
Source, transformation, device, author, and version remain traceable.
Comparability
Units, definitions, populations, and protocols must align before aggregation.
Boundary. Temporal order can support prediction and hypothesis generation, but sequence alone does not establish causality.
Chapter 05 · State, change, and uncertainty over time33 / 50
Part V · Plate 03 · Evidence lineageOzyra.

Every output needs a
traceable evidence path.

FIGURE 22
LINEAGE CHAIN

Auditability begins before model training. Ozyra must be able to reconstruct how a source signal became a representation, an output, a human action, and an observed result.

01 · CAPTURE
Preserve the source
Record the original payload, producer, device, event time, acquisition context, and applicable rights.
Artifact · raw payload + acquisition event
02 · NORMALIZE
Map without erasing
Resolve identity, terminology, units, and timestamps while retaining source values and mapping decisions.
Artifact · source-to-canonical mapping
03 · QUALIFY
Expose fitness and gaps
Check validity, calibration, completeness, outliers, drift, and missingness; quarantine what fails.
Artifact · quality + quarantine report
04 · TRANSFORM
Version every derivation
Tie features, rules, embeddings, and state estimates to declared code, parameters, and input versions.
Artifact · feature definitions + checksum
05 · SNAPSHOT
Freeze the evidence set
Version cohorts, windows, exclusions, labels, splits, and representation state for training and evaluation.
Artifact · cohort + label manifest
06 · TRACE USE
Connect output to consequence
Link model version and input snapshot to the surfaced output, human response, override, outcome, and incident record.
Artifact · decision + outcome audit trail
If an output cannot be traced back to its evidence, it cannot be governed with confidence.
Principle · Preserve source truth, version every transformation, and connect deployed outputs to what people did and what happened next.
Chapter 05 · Evidence lineage from source to outcome34 / 50
Part V · Plate 04 · Application stackOzyra.

A reusable core,
many evidence thresholds.

FIGURE 23
APPLICATION STACK

The same platform can support several applications, but each application needs its own intended use, workflow owner, validation plan, and governance route.

01 · MONITOR
Make trajectory change visible.
Test · signal validity, timeliness, burden, escalation rules.
02 · PREDICT
Estimate a relevant future state.
Test · calibration, discrimination, uncertainty, subgroup performance.
03 · MATCH
Compare protocols or next steps.
Test · comparability, contraindications, human review, outcome evidence.
04 · ORCHESTRATE
Route attention and work.
Test · usability, override, workload, latency, operational impact.
05 · LEARN
Evaluate and improve the system.
Test · versioning, monitoring, causal design where needed, change control.
Participant-facing

Clarity, accessibility, consent or other lawful basis, burden, support, and routes for questions or redress.

Professional-facing

Intended use, limitations, timing, interpretability, override, training, and accountability.

System-facing

Integration, reliability, audit, cybersecurity, monitoring, incident response, and measurable lift.

Application boundaries determine the evidence burden and may affect regulatory classification. Classification must be assessed for each intended use and jurisdiction.
Chapter 05 · Platform reuse does not mean evidence reuse35 / 50
Part V · Plate 05 · Safety caseOzyra.

Safety is an architecture,
not a review at the end.

FIGURE 24
SAFETY CASE

Every use case needs a living safety case that connects intended use, evidence, human oversight, monitoring, and controlled change.

01
Intended use + classification
Define the user, population, decision, benefit, foreseeable misuse, and applicable obligations.
Artifact · use-case and regulatory assessment
02
Data, security + representativeness
Characterize sources, access protections, threats, coverage, exclusions, missingness, provenance, and population gaps.
Artifact · data, security, and subgroup statement
03
Validation + human factors
Evaluate performance in context, including the human-AI team, workflow timing, comprehension, and overrides.
Artifact · validation and usability report
04
Transparency + oversight
Communicate purpose, inputs, outputs, limitations, uncertainty, responsibility, and safe-use instructions.
Artifact · user and participant information
05
Monitoring + incident response
Track performance, drift, burden, inequity, misuse, complaints, and safety signals over the lifecycle.
Artifact · monitoring and escalation plan
06
Change control + revalidation
Version data, models, workflows, and documentation; define when change requires review, pause, or revalidation.
Artifact · approved change record
The right to override or stop the system must be operational, not ceremonial.
Sources · NIST AI RMF 1.0; Regulation (EU) 2024/1689; IMDRF Good Machine Learning Practice (2025); FDA transparency principles for machine learning-enabled medical devices. Exact obligations depend on the use.
Chapter 05 · A living case for safe and responsible use36 / 50
Part VI · Value RealizationOzyra.
PART VI · FROM CAPABILITY TO OUTCOME
Economic value is a result, not a property of the data.
Value is realized when a governed capability creates repeatable clinical, operational, research, or product benefit - and when that benefit can be sustained at a proportionate cost and risk.
Benefit
an outcome or workflow changes in a meaningful way
Repeat
the result survives people, sites, and time
Sustain
delivery, trust, safety, and economics remain viable
OZYRA VALUE LOGIC · SCENARIO-BASED · EVIDENCE-LED · NON-FINANCIAL AT PATIENT LEVEL
Chapter 06 · Value realization37 / 50
Part VI · Plate 01 · Scale scenariosOzyra.

Scale changes the question
before it changes the value.

FIGURE 25
SCALE SCENARIOS

More participants can increase coverage and statistical power, but they also increase heterogeneity, governance burden, integration work, and the consequences of error.

10k
Illustrative profiles
Instrument and test feasibility.
Establish measurement quality, participation, workflow fit, outcome linkage, and the first prospectively evaluated use cases.
Do not scale if the loop cannot close reliably.
100k
Illustrative profiles
Validate across meaningful variation.
Test calibration, subgroup coverage, operational repeatability, site differences, and monitoring at a larger denominator.
Do not hide weak groups inside an aggregate.
1m
Illustrative profiles
Build multi-site platform capability.
Standardize representations, automate quality controls, manage drift, support multiple workflows, and evaluate external validity.
Do not confuse distribution with clinical utility.
10m
Illustrative profiles
Coordinate network-scale learning.
Operate federated governance, high-reliability infrastructure, public accountability, jurisdictional controls, and ecosystem standards.
Do not scale faster than oversight and benefit.
Scenario note. These profile counts are analytical markers, not forecasts, targets, market-size claims, or valuations. Capability depends on use case, coverage, quality, rights, and verified lift.
Chapter 06 · Scale expands both possibility and responsibility38 / 50
Part VI · Plate 02 · Value bridgeOzyra.

Value must cross
every bridge.

FIGURE 26
VALUE BRIDGE

A contribution profile does not become enterprise value directly. Each step requires evidence, ownership, and a credible alternative or comparator.

01
Governed participation
Legitimate access, clear purpose, proportionate burden, and accountable stewardship.
02
Learnable trajectories
Relevant, longitudinal, outcome-linked, quality-controlled evidence.
03
Verified lift
A model, workflow, or research process improves against an agreed baseline.
04
Repeatable product utility
The improvement persists across users, sites, populations, and time at acceptable burden.
05
Sustainable system value
Benefit, trust, delivery economics, retention, and future capability reinforce one another.
Clinical

Better function, safety, access, experience, prevention, or recovery.

Operational

Less avoidable work, delay, variation, or resource waste.

Knowledge

Faster, stronger, more representative, or more reusable evidence.

Never jump from patient count to enterprise value.
Chapter 06 · From participation to sustainable utility39 / 50
Part VI · Plate 03 · Realization logicOzyra.

A scenario is credible
when its assumptions are visible.

FIGURE 27
REALIZATION LOGIC

The value case should be recomputable from observed reach, verified improvement, actual adoption, durability, and the full burden of delivery and risk.

Realized value = eligible reach × verified lift × adoption × durability - total burden
Use ranges · publish uncertainty · state comparator · version assumptions · separate evidence from ambition
01
Eligible reach
The relevant population for which use is lawful, technically feasible, and clinically or operationally appropriate.
02
Verified lift
Change against a baseline or comparator, with effect size, uncertainty, adverse consequences, and subgroup analysis.
03
Adoption + durability
Appropriate use, retention, workflow fit, and persistence of benefit across sites and time.
04
Total burden
Implementation, integration, monitoring, support, capital, participant effort, privacy exposure, and residual risk.
Downside case
Lower eligible reach, weak adoption, high integration burden, or benefit that does not persist.
Evidence case
Observed inputs, declared uncertainty, and a comparator-based estimate of benefit and burden.
Upside case
Improved coverage or lift only when a named capability and evidence path make the change plausible.
Framework · Ozyra strategic model. It is not an accounting fair-value method, securities valuation, or patient-level price formula.
Chapter 06 · Recomputable assumptions before valuation claims40 / 50
Part VI · Plate 04 · Revenue architectureOzyra.

Commercialize the capability,
not the person.

FIGURE 28
REVENUE ARCHITECTURE

A responsible business model is paid for useful software, services, evidence generation, or licensed intelligence under clear purpose and governance.

01 · WORKFLOW SOFTWARE
Subscription or platform access
Care organizations · programs · networks
Monitoring, coordination, decision support, quality measurement, and operational tools with service levels and use-case controls.
02 · ENABLEMENT
Implementation + managed capability
Providers · research teams · public programs
Integration, configuration, evaluation, monitoring, and change support where accountability remains explicit.
03 · GOVERNED RESEARCH
Evidence partnerships
Academic · clinical · life-science partners
Question-led analyses, prospective studies, cohort discovery, or evaluation services under approvals, contracts, and participant protections.
04 · INTELLIGENCE SERVICES
Licensed models, APIs, or evaluation tools
Qualified product and care partners
Use-case bounded capabilities with documentation, validation, monitoring, support, auditability, and controlled updates.
License intelligence and workflow. Do not position identity-bearing health records as inventory.
Price unit

Capability, service level, evaluated use, workflow, or outcome - not a human identity.

Contract unit

Purpose, roles, access, security, monitoring, change, benefit, and termination.

Proof unit

Utility, safety, equity, burden, retention, and economics for the intended use.

Chapter 06 · Revenue aligned with accountable utility41 / 50
Part VI · Plate 05 · Defensibility stackOzyra.

Defensibility is a system
of reinforcing assets.

FIGURE 29
DEFENSIBILITY STACK

Models can be copied or overtaken. A durable position comes from the difficult combination of evidence, workflow, governance, trust, and repeated execution.

01
Models
Useful intellectual property, but only as strong as the task, data, evaluation, and deployment context.
02
Representations
A coherent temporal and multimodal model of state, change, intervention, and uncertainty.
03
Outcome evidence
Intervention-response pairs, comparable cohorts, prospective validation, and known limits.
04
Workflow
Integration, timing, human oversight, feedback, and measurable operational or clinical lift.
05
Rights + trust
Lawful use, transparency, provenance, privacy, security, participation, and accountable governance.
06
Operating network
Partners, distribution, repeatable implementation, monitoring, shared standards, and a learning culture.

Each layer is insufficient.

A strong model without workflow is unused. Workflow without evidence may scale harm. Access without rights is fragile. Governance without useful outcomes will not sustain participation.

Positioning rule

Own the differentiating system, not the person.

Ozyra can protect technology, representations, evaluation methods, model IP, workflows, and contractual operating rights while respecting patient rights and partner obligations.

Chapter 06 · The stack compounds as a whole42 / 50
Part VII · Strategic ChoicesOzyra.
PART VII · BOUNDARIES + OPERATING CHOICES
Protect the differentiating core. Share the work that requires a system.
Ozyra should build where longitudinal representation, evaluation, and workflow learning create distinctive capability - and partner where standards, access, infrastructure, or specialist accountability are essential.
Build
the canonical model, evaluation logic, and product core
Partner
for access, devices, infrastructure, and specialist delivery
Share
governance, standards, evidence, and accountability
CHOICE PRINCIPLE · CONTROL DIFFERENTIATION · INTEROPERATE AT THE BOUNDARIES
Chapter 07 · Strategic choices43 / 50
Part VII · Plate 01 · System risksOzyra.

The largest risks sit
between the layers.

FIGURE 30
RISK REGISTER

Technical performance is only one part of system risk. Rights, representation, workflow, incentives, and economics can fail independently or reinforce one another.

01
Purpose or rights mismatch
Data is technically accessible but the intended use, authority, transparency, or participant expectations do not align.
Control · use-case gate, rights mapping, access audit, stop rule
02
Coverage and selection bias
Participants, devices, sites, or follow-up systematically exclude people for whom the system is intended.
Control · coverage register, subgroup thresholds, recruitment repair
03
Weak labels or leakage
The model learns documentation artifacts, future information, or outcomes that do not reflect meaningful benefit.
Control · label protocol, temporal audit, blinded evaluation
04
Drift and local variance
Population, devices, workflow, incentives, or care change and performance no longer transfers.
Control · site acceptance, monitoring, recalibration, revalidation
05
Automation or workflow harm
Timing, interface, alert burden, over-reliance, or unclear responsibility makes a technically sound output unsafe.
Control · human factors, override, workload and incident review
06
Benefit cannot support delivery
Integration, monitoring, support, evidence, and capital burden exceed the repeatable utility created.
Control · comparator-based value case, staged commitment, exit criteria
A risk register is useful only when it changes a decision, owner, threshold, or action.
The register is illustrative and must be specialized by intended use, population, jurisdiction, technology, workflow, and accountable organization.
Chapter 07 · Risks across data, people, workflow, and value44 / 50
Part VII · Plate 02 · Sourcing boundariesOzyra.

Build, partner, and share
by design.

FIGURE 31
DECISION MATRIX

The right boundary follows differentiation, accountability, switching risk, specialist capability, standards, and the need for independent oversight.

Capability
Build / control
Partner / procure
Shared responsibility
Longitudinal representation
Canonical state model, temporal semantics, provenance, uncertainty, and feature definitions.
Terminology services and implementation tooling where non-differentiating.
Clinical validation of meaning, mappings, and local fitness.
Models + evaluation
Task logic, evaluation harness, calibration, monitoring, and model orchestration.
Commodity foundation models, specialist algorithms, and external benchmarks.
Independent validation, safety review, and change approval.
Workflow products
Differentiating user experience, decision logic, feedback capture, and analytics.
EHR modules, communications, identity, and local implementation capacity.
Clinical ownership, training, escalation, and outcome evaluation.
Devices + acquisition
Selection criteria, quality controls, event alignment, and device abstraction.
Validated hardware, logistics, connectivity, maintenance, and calibration.
Participant support, acceptance testing, and incident handling.
Infrastructure + exchange
Architecture, security requirements, observability, and data contracts.
Cloud, storage, networking, FHIR connectivity, and commodity security tooling.
Resilience testing, access governance, and breach response.
Evidence + governance
Use-case dossier, audit trail, model documentation, and internal quality system.
Research sites, legal expertise, notified or oversight bodies, and specialist assurance.
Protocol, approvals, participant voice, monitoring, publication, and accountability.
Decision rule. Do not outsource accountability. Contracts and interfaces should make ownership, evidence, access, monitoring, change, exit, and incident response explicit.
Chapter 07 · Control the core and interoperate at the edge45 / 50
Part VIII · Roadmap + ProofOzyra.
PART VIII · EARNED PROGRESSION
Each stage should earn permission for the next.
The roadmap moves from reliable observation to validated use, integrated delivery, broader scale, and platform leverage. Progress is gated by evidence, safety, trust, and operating readiness.
Instrument
make the trajectory and feedback loop reliable
Validate
prove benefit, safety, workflow fit, and limits
Scale
expand only when the controls scale as well
ROADMAP PRINCIPLE · GATED BY EVIDENCE · UPDATED AS FACTS CHANGE
Chapter 08 · Roadmap and proof46 / 50
Part VIII · Plate 01 · Five-year roadmapOzyra.

A roadmap is a sequence
of evidence gates.

FIGURE 32
FIVE-YEAR ROADMAP

Dates are directional. The order matters more: instrumentation before claims, validation before broad deployment, and operating controls before network scale.

PHASE 01 · 2026
Instrument
Define priority use cases. Establish canonical trajectories, rights mapping, provenance, outcome protocols, and reliable feedback capture.
Exit · adequate data quality, workflow feasibility, participation, and governed outcome linkage.
PHASE 02 · 2027
Validate
Prospectively evaluate selected applications. Measure lift, burden, safety, human factors, subgroup performance, and operating cost.
Exit · comparator-based benefit and acceptable risk in the intended context.
PHASE 03 · 2028
Integrate
Embed validated applications in repeatable workflows. Formalize monitoring, support, quality systems, change control, and partner interfaces.
Exit · repeatable implementation and stable performance across more than one setting.
PHASE 04 · 2029
Expand
Broaden populations, sites, modalities, and use cases selectively. Test external validity, economics, equity, and governance capacity.
Exit · controls, benefit, and trust remain adequate as heterogeneity grows.
PHASE 05 · 2030
Platform
Offer reusable intelligence, evaluation, and workflow services through governed partner and product ecosystems.
Exit · portfolio-level utility with transparent accountability and sustainable delivery.
Roadmap note. This is a strategic sequence, not a commitment, forecast, financing plan, or assurance of regulatory authorization. Each phase should be revised as evidence and context change.
Chapter 08 · Evidence gates from instrumentation to platform47 / 50
Part VIII · Plate 02 · Stakeholder proof pointsOzyra.

Progress must be legible
to every stakeholder.

FIGURE 33
PROOF REGISTER

The same milestone can mean different things to different people. Ozyra should state whose question is being answered and what evidence would change the decision.

Patients + participants
Does this help me without taking control away?
Understandable purpose, proportionate burden, useful feedback, privacy and security, exercisable rights, support, and benefit-sharing where appropriate.
Proof · comprehension, experience, retention, requests, incidents, resolution
Clinicians + care teams
Does this improve a decision or workflow safely?
Relevant timing, reliable inputs, calibrated output, limitations, override, training, manageable workload, and outcome-linked evaluation.
Proof · use, override, time, burden, safety, clinical or functional lift
Care organizations
Can this be implemented and sustained?
Integration, service levels, governance, security, workforce readiness, quality improvement, economics, and clear accountability.
Proof · repeatable deployment, reliability, total cost, durable utility
Researchers + evidence partners
Is the evidence fit for the question?
Protocol, provenance, population, endpoints, missingness, bias controls, reproducibility, access route, and publication discipline.
Proof · data fitness, analysis validity, transparent limitations, reusable knowledge
Oversight + regulators
Are risks identified, controlled, and monitored?
Classification, quality system, documentation, validation, human oversight, transparency, monitoring, incident response, and change control.
Proof · complete dossier, traceable decisions, timely corrective action
Partners + team
Is the strategy executable and worth committing to?
Differentiated core, realistic dependencies, milestone ownership, operating capacity, partner model, learning velocity, and sustainable economics.
Proof · evidence gates met, risks reduced, capability and trust compound
A proof point should name the stakeholder, decision, evidence, threshold, owner, time window, uncertainty, and consequence of not meeting it.
Chapter 08 · Evidence for the people who shape and use the system48 / 50
Closing thesis · Volume 01Ozyra.
“Ozyra is building the governed infrastructure that helps care learn from every trajectory - with people, not from them.
OZYRA REVIEW · THE VALUE OF HEALTH INTELLIGENCE

The opportunity is not to accumulate the most records. It is to connect the right observations, actions, and outcomes; place useful intelligence inside accountable workflows; and prove that the system improves without weakening rights, trust, safety, or equity.

TEST · 01
Useful
Does the system improve a meaningful decision, workflow, experience, or outcome?
TEST · 02
Governed
Can its purpose, rights, evidence, risks, actions, and changes be explained and audited?
TEST · 03
Proven
Does the benefit persist across relevant people, settings, and time at proportionate burden?
THE STANDARD IS NOT MORE DATA. THE STANDARD IS RESPONSIBLE, MEASURABLE LEARNING.
Ozyra Review · Closing thesis49 / 50
Sources + method · Volume 01Ozyra.

Sources,
declared.

PRIMARY SOURCES
ACCESSED JUL · 2026
01 · WHO
Noncommunicable diseases · Fact sheet · 25 September 2025
02 · WHO
Rehabilitation · Fact sheet · 22 April 2024
03 · WHO
Ageing and health · Fact sheet · 1 October 2025
04 · OECD
Health at a Glance 2025 · Health expenditure in relation to GDP
05 · FDA
Digital Health Technologies for Remote Data Acquisition in Clinical Investigations · Final guidance · December 2023
06 · EUROPEAN COMMISSION
European Health Data Space Regulation (EU) 2025/327 · Implementation timeline
07 · WHO
Artificial Intelligence for Health · 27 May 2024
08 · WHO
Ethics and governance of artificial intelligence for health · 28 June 2021
09 · SCIENTIFIC DATA
Wilkinson et al. · The FAIR Guiding Principles for scientific data management and stewardship · 2016
10 · TEHDAS
European Health Data Space Data Quality Framework · Deliverable 6.1 · 18 May 2022
11 · NATURE CANCER
Morin et al. · Longitudinal EHR and real-world data for continuous pan-cancer prognostication · 2021
12 · NATURE
Moor et al. · Foundation models for generalist medical artificial intelligence · 2023
13 · PMLR
Ghorbani & Zou · Data Shapley: Equitable Valuation of Data for Machine Learning · 2019
14 · FDA
Real-World Data: Assessing Electronic Health Records and Medical Claims Data · Final guidance · July 2024
15 · EMA
Data Quality Framework for EU medicines regulation: application to Real-World Data · March 2026
16 · AHRQ
About Learning Health Systems · Agency for Healthcare Research and Quality · reviewed May 2019
17 · NATIONAL ACADEMY OF MEDICINE
Learning Health System · science, informatics, incentives, culture, and continuous improvement
18 · IMDRF
Good Machine Learning Practice for Medical Device Development: Guiding Principles · final · 29 January 2025
19 · NIST
Artificial Intelligence Risk Management Framework (AI RMF 1.0) · NIST AI 100-1 · January 2023
20 · EUROPEAN UNION
Regulation (EU) 2024/1689 laying down harmonised rules on artificial intelligence · Official Journal
21 · HL7
FHIR R5 · Fast Healthcare Interoperability Resources · specification version 5.0.0
22 · FDA / HEALTH CANADA / MHRA
Transparency for Machine Learning-Enabled Medical Devices: Guiding Principles · June 2024
Method note

Contextual numeric claims are source-backed. Ozyra-specific content is identified as thesis, framework, scenario, or roadmap. Synthetic examples and scale markers are illustrative; these frameworks are not clinical, legal, accounting, or regulatory instruments.

Analytical discipline

Quantitative valuation should disclose assumptions, ranges, uncertainty, comparables, constraints, and the bridge from verified lift to realizable value. This paper is not medical, legal, financial, or investment advice.

Ozyra Review · Volume 0150 / 50