Research program

Engineering Care Delivery

Engineering Care Delivery studies how care itself should be designed when technical capability becomes easier to produce but adoption, patient access, workflow, trust, reimbursement, accountability, and outcomes remain scarce. It treats care delivery as an engineered system rather than a passive buyer of technology.

How should care be designed when technical capability is easy to produce but adoption, access, workflow, trust, reimbursement, accountability and outcomes stay scarce?

Programs inside this one

Research notes

• Continuous traceability instead of reconstruction

  The chain from clinical need to post-market observation should be current every day, not rebuilt in the weeks before a submission or an audit. Reconstruction is where the errors enter.

• Decisions, Not Alerts

  One ranked stream of judgment calls, each carrying a recommendation, calibrated confidence, reasoning, evidence, the policy in force, the downside if wrong, the alternatives considered, a deadline and the role required to authorize it.

• Evidence as the Product

  A flight recorder for every recommendation and action: what happened, who initiated it, what data was used, which policy and model version were in force, what a human changed on review, and whether it was later judged helpful, neutral or harmful.

• Graduated Autonomy

  Autonomy is granted per workflow across four levels, from observe to operate within policy, against measured shadow-mode agreement, with drift monitoring and a rollback path. A single global autonomy score is unsafe.

• Migrating Off an Incumbent

  Import, reconcile, parallel run, readiness score. What cannot be reconciled is listed individually rather than rounded away, and shadow-mode disagreement clusters reveal where the operation is unready before anyone cuts over.

• Month-End Is an Evidence Problem

  Billing readiness presented exception-first: what is ready, what is blocked and why, what conflicts with policy, and what is documented but should not be billed. The system never asserts final eligibility and never submits a claim.

• Populations Without Tables

  Clustered groups of patients and a conversational explorer replace the filter-and-sort grid. Patients are grouped by what they need operationally, and the grouping itself can be interrogated.

• Regulatory execution as an AI workforce

  The regulated development system, described as work rather than paperwork: research, drafting, traceability, orchestration, evidence management, and audit support, with qualified professionals holding every judgment and approval.

• Reimbursement-first device discovery

  Find the reimbursed problem before designing anything. Coverage, coding, payment, and utilization are public, so whether anyone will pay is a research question with a documented answer, not a bet taken after two years of engineering.

• Simulation as an Operating Surface

  Enrollment changes, staff departures, program shifts, cadence and threshold changes and autonomy increases are run against the operation before they are made, with assumptions, uncertainty and risks shown rather than a single number.

• The Briefing as Home Screen

  The home screen is a narrative of what changed since the last review: what the system handled, what failed, ranked developments, one intervention worth making, and an evidence link on every claim.

• The commodity component assumption

  Do not design bespoke electronics unless a commodity part is demonstrably inadequate for the claim being made. The burden of proof sits with the custom design, the same way Zero Neo puts it on new software.

• The Dashboard Is a Symptom

  Tile walls, module sidebars and worklists are the visible residue of software that could not reason. Zero dashboard is not a visual preference; it is what an interface looks like once the ranking moves inside the system.

• The Executive Narrative Without KPI Tiles

  Leadership reporting written as prose that states the causal claim, with measures, believed drivers and confidence in support, and an explicit section on what remains unknown.

• The Labor Curve

  A rules-based operation has no mechanism to break the link between panel size and headcount. An AI-native one does, but the first two quarters of a migration cost more, not less, and the curve only bends where shadow-mode agreement supports widening autonomy.

• The Patient Story Replaces the Chart Page

  A narrative generated around the question currently being asked, with timeline lanes that draw missing data as missing, and explicit reasoning about which care program actually fits.

• The phone is part of the device

  Architectures where a smartphone or wearable is a component of the medical device, not a companion screen. It removes hardware and adds obligations, and the trade is only worth making with both sides written down.

• Typed Utterances and Honest Uncertainty

  Every statement the system makes carries a type: observation, inference, recommendation, automated action, human-approved action, unresolved uncertainty, or prohibited action. Missing data is rendered as missing, and standing constraints hold at every autonomy level.

• What is actually medically specialized

  Separate component availability from medical performance. The part being cheap says nothing about calibration, accuracy, reliability, safety, cybersecurity, manufacturing controls, or evidence. Confusing the two is how consumer-grade thinking hurts patients.

• What This Line of Research Does Not Claim

  No proven outcome improvement without a connected analytics warehouse; disagreement concentrating in recent medication changes; projections that remain projections. The limits stated plainly, in one place.

• Where the device thesis should be rejected

  The rejection criteria, written before we get attached to anything. Novel materials, exotic sensing physics, sterile manufacturing, life-sustaining risk, and trial-scale evidence requirements all put a category outside this thesis.

• 

  Adoption latency score

  This note shows how to estimate how long adoption will take, so a reader can plan around institutional clocks instead of assuming they disappear. The estimate covers three points: first use, repeat use, and routine use, with the source, range and confidence attached to each. Estimate the time from first evidence to first use, repeat use, and routine use without pretending institutional clocks disappear.

• Care delivery opportunity screen

  Test whether patient access, workflow control, payment, outcomes, technology absorption, and learning support a care business. The screen scores each dimension with evidence, keeps any fatal weakness visible, and ends in a decision to build, acquire, partner, license, monitor, or stop.

• Care delivery readiness map

  This note decides readiness by the slowest essential condition: a product is not ready for routine care because its technical component is ready. The map organizes the review but does not establish clinical safety, regulatory standing, coverage, or professional approval. A review of patient, caregiver, clinician, site, supply, data, payment, and exception readiness before an intervention enters routine care.

• Care model decision package

  Bring the hypothesis, evidence, economics, rights, uncertainty, failure case, and next decision into one reviewable record. The package lets a reviewer decide on a care model with every material claim and every missing answer in view, instead of an average that hides a fatal weakness.

• Cheapest next test

  This note says to pick the smallest observable test that could reverse the decision, and to prefer behavior over interest. A reader gains a cheaper way to settle the assumption that matters most, and no larger experiment should be funded when a smaller one can invalidate it. Reduce the most important uncertainty with the least time, money, access, and patient exposure reasonably required.

• Cross-signal scorecard

  Compare spending, science, translation, delivery, policy, rights, and commercial movement without hiding confidence. The scorecard records measure, source, confidence, and contrary evidence for each dimension, and a high composite score cannot repair a missing provider base, payment route, workflow owner, or credible mechanism.

• Evidence debt register

  Make stale, weak, unsupported, single-source, and unresolved claims visible before they quietly drive decisions. The register records each claim's evidence status, risk if wrong, owner, and decision blocked, so debt that can reverse a clinical, economic, rights, or adoption decision is resolved first.

• Evidence freshness register

  This note gives a register that sets a useful life, refresh schedule and decision trigger for each public signal, so stale evidence does not drive current decisions. Stale evidence remains in the historical record but loses authority over current decisions. Set the useful life, refresh schedule, and decision trigger for every public signal used in active research.

• Evidence ladder

  Rank evidence by the scarce real-world commitment it reflects, from public signal through replicated outcome. The ladder lets a reader state which rung supports each claim and what next commitment would materially change confidence, instead of treating all evidence as equal.

• Failure mechanism map

  This note decides where to work on a regulated intervention: on the failure mechanism that limits the outcome and can be changed. It classifies failures and traces each one to a measure, a responsible party and a smallest test, and stops when the mechanism is weak, immaterial or beyond practical control. Locate where a regulated intervention fails, what can be measured, and which party can change the binding condition.

• Health system condition map

  List every independent condition that must hold for an intervention to reach repeatable use. The map names the current state, evidence, owner, and stop rule for each condition, and identifies the binding condition that sets the adoption clock, which is where work should concentrate.

• Hospital adoption engineering

  This note decides when a pilot may advance: only when an operating owner accepts the work and repeat use survives beyond the sponsor. It maps stakeholders and tests activation, voluntary use and abandonment, so a reader can separate sponsored use from routine institutional use. Map the people, evidence, approvals, work, and support needed to move from sponsored pilot to routine institutional use.

• Outcome margin model

  This note gives a test for whether a redesigned care pathway lets its owner keep the value it creates. It rejects the theory when labor remains proportional, outcomes are not measurable, payment does not reward the change, or suppliers retain the decisive bargaining power. Test whether a redesigned care pathway can retain value created through lower cost, better outcomes, or changed risk.

• Public healthcare opportunity graph

  Connect conditions, populations, spending, providers, evidence, products, payment, rights, pathways, and commercial activity. Each edge states why two records are connected, the time period, and confidence. The graph is for finding mismatches and changes and generating hypotheses, not automatic commercial conclusions.

• Public signal evidence ledger

  This note finds that a public signal can justify investigation but is not a validated opportunity until external commitment or real-world evidence supports it. The ledger keeps sourced observations separate from calculations, interpretations, AI inferences, hypotheses and commercial claims. Keep sourced observations separate from calculations, interpretations, AI inferences, hypotheses, and commercial claims.

• Risk bearer map

  This note gives a map of who carries each risk and who controls it, with the rule that responsibility should not move away from control. It matters because a party asked to carry risk needs the information, authority and intervention path to manage it. Show who carries clinical, product, data, implementation, cybersecurity, financial, operational, and reputational risk.

• Technology absorption decision

  This note gives a rule for the build, buy, license, partner or consume choice: own a component when control changes the outcome, defensibility, safety, financeability or ability to learn, and consume it when it is interchangeable and dependence is manageable. Choose whether a care organization should build, buy, license, partner for, or consume a technical component.

• Therapy orchestration record

  This note gives a record that makes an adaptive treatment sequence testable, stage by stage, so a reader can check whether sequence or timing changes the outcome. The record supports workflow research and does not prescribe care. Specify the measurements, transitions, timing, decisions, and exceptions in an adaptive treatment sequence.

• Treatment readiness index

  Test whether a composite readiness measure can improve treatment timing or selection without hiding weak dimensions. The index is rejected when aggregation hides a fatal condition, the score cannot guide action, or performance does not transfer across settings, and it must show that acting on the score improves outcomes.

Next