Decision spine
Frames the controlling thesis, evidence status, buyer implications, and research-use limits.
Digital Health
Zemi Dossier
Research-use only
AI Clinical Validation & Digital Biomarkers
A decision package for evaluating whether AI and digital-biomarker tools can survive prospective validation, deployment drift, and payer or regulatory evidence standards.
Decision this dossier helps answer
Which AI or digital-biomarker claims are ready for prospective validation, and which are still retrospective performance stories?
Problem this dossier solves
Retrospective accuracy is not decision utility.
The dossier argues that most medical-AI failures are not model failures alone; they are validation-design failures. It identifies the minimum prospective evidence needed before a digital signal should change trial design, product strategy, or buyer confidence.
What is moving: current field evidence and platform activity
What is unresolved: decision-grade proof, endpoint logic, and failure modes
What Zemi solves: evidence-gated next-study strategy
Full decision package
118-page research report plus 33-sheet Evidence & Decision Workbook
Buy now
$3,500$5,500Launch price
Controlling thesis
Retrospective accuracy is not decision utility.
The dossier argues that most medical-AI failures are not model failures alone; they are validation-design failures. It identifies the minimum prospective evidence needed before a digital signal should change trial design, product strategy, or buyer confidence.
A Zemi Dossier turns uncertain emerging research into a decision package: what is known, what is inferred, what is hypothetical, what would falsify the thesis, and what should be built, funded, avoided, monitored, or validated next.
Research report preview
First 30 pages of the actual AI Clinical Validation & Digital Biomarkers research report.
The public page shows the real document style, density, evidence posture, and decision framing before a buyer requests access to the full Zemi Dossier package.
Research report
The research report is the argument. It is not the whole product.
The PDF research report is the structured argument: field status, evidence maturity, buyer implications, falsifiable hypotheses, power-calculated next studies, strategic recommendations, limitations, and traceability posture.
Inspectable infrastructure
Pairs the report with claim ledgers, evidence tables, power rows, audit surfaces, and decision tools.
Executable next studies
Links hypotheses to endpoint logic, assumptions, sample-size logic, budget, and timeline planning.
Research report table of contents
Public TOC generated from the source report.
The full research report is delivered as a licensed PDF. This table shows the decision structure without publishing the full product.
| Section | What it covers | Buyer value |
|---|---|---|
| Executive Summary | Condenses the dossier thesis, evidence posture, buyer implications, and next-study priorities. | Helps the buyer inspect the argument, evidence posture, and decision logic before relying on the package. |
| Scope and Evidence Integrity Statement | Defines the evidence window, source boundaries, research-use limits, and labeling discipline. | Lets the buyer audit support strength before relying on the package. |
| Decision Brief: Read This First | Turns the report into immediate buyer triage: what matters now, what is still uncertain, and what to do next. | Helps the buyer inspect the argument, evidence posture, and decision logic before relying on the package. |
| Decision Snapshot Overview | Maps findings to buyer actions: build, fund, avoid, monitor, or validate next. | Helps the buyer inspect the argument, evidence posture, and decision logic before relying on the package. |
| Buyer Decision Cheat Sheet | Maps findings to buyer actions: build, fund, avoid, monitor, or validate next. | Helps the buyer inspect the argument, evidence posture, and decision logic before relying on the package. |
| Source Quality Dashboard | Shows how strong the source base is and where conclusions depend on indirect or thinner support. | Lets the buyer audit support strength before relying on the package. |
| Current-Signal Refresh | Frames the field's June 2026 state, active programs, research fronts, and timing signals. | Helps the buyer inspect the argument, evidence posture, and decision logic before relying on the package. |
| Why 2026 Is the Inflection: The Prospective Verdict | Frames the field's June 2026 state, active programs, research fronts, and timing signals. | Helps the buyer inspect the argument, evidence posture, and decision logic before relying on the package. |
| The Four Axis Translation Model | Defines the governing argument and the evidence gates that would support or falsify it. | Helps the buyer inspect the argument, evidence posture, and decision logic before relying on the package. |
| The four axes | Summarizes a report component that ties field evidence to the dossier's decision logic. | Helps the buyer inspect the argument, evidence posture, and decision logic before relying on the package. |
| Why the axes multiply | Summarizes a report component that ties field evidence to the dossier's decision logic. | Helps the buyer inspect the argument, evidence posture, and decision logic before relying on the package. |
| The product is a bottleneck: formalizing the law | Defines the governing argument and the evidence gates that would support or falsify it. | Helps the buyer inspect the argument, evidence posture, and decision logic before relying on the package. |
| The cost geometry of the bottleneck | Summarizes a report component that ties field evidence to the dossier's decision logic. | Helps the buyer inspect the argument, evidence posture, and decision logic before relying on the package. |
| The Binding-Axis Classifier | Operationalizes the thesis as a reusable decision surface for comparing programs or evidence states. | Gives the buyer a reusable way to compare programs, risks, and evidence states. |
| The Bifurcation Theorem: Which Axis Binds by Category | Summarizes a report component that ties field evidence to the dossier's decision logic. | Helps the buyer inspect the argument, evidence posture, and decision logic before relying on the package. |
| Why AI Clinical Validation Stalled | Summarizes a report component that ties field evidence to the dossier's decision logic. | Helps the buyer inspect the argument, evidence posture, and decision logic before relying on the package. |
| Cross Domain Convergence Map | Shows how separate research signals converge, diverge, or interact across domains, mechanisms, or study layers. | Helps the buyer inspect the argument, evidence posture, and decision logic before relying on the package. |
| Cross-Domain Synthesis | Shows how separate research signals converge, diverge, or interact across domains, mechanisms, or study layers. | Helps the buyer inspect the argument, evidence posture, and decision logic before relying on the package. |
| Competing Interpretations Adjudicated | Compares alternative readings of the evidence and explains why the dossier weights one interpretation over another. | Helps the buyer inspect the argument, evidence posture, and decision logic before relying on the package. |
| Evidence Maturity Matrix and Category Comparison | Links claims to source support so users can separate what is known, inferred, and hypothetical. | Gives the buyer a reusable way to compare programs, risks, and evidence states. |
| Key Findings | Summarizes the most decision-relevant findings and their implications for the dossier thesis. | Helps the buyer inspect the argument, evidence posture, and decision logic before relying on the package. |
| Evidentiary Standard | Shows how strong the source base is and where conclusions depend on indirect or thinner support. | Helps the buyer inspect the argument, evidence posture, and decision logic before relying on the package. |
| Validity | Summarizes a report component that ties field evidence to the dossier's decision logic. | Helps the buyer inspect the argument, evidence posture, and decision logic before relying on the package. |
| Utility | Summarizes a report component that ties field evidence to the dossier's decision logic. | Helps the buyer inspect the argument, evidence posture, and decision logic before relying on the package. |
| Deployment | Summarizes a report component that ties field evidence to the dossier's decision logic. | Helps the buyer inspect the argument, evidence posture, and decision logic before relying on the package. |
| Further Key Findings and Their Convergence | Summarizes the most decision-relevant findings and their implications for the dossier thesis. | Helps the buyer inspect the argument, evidence posture, and decision logic before relying on the package. |
| What the Exceptions Teach: Anatomy of a Translatable Program | Supports inspection of the report's visual evidence, calculations, or supplemental decision material. | Helps the buyer inspect the argument, evidence posture, and decision logic before relying on the package. |
| Quantitative Evidence | Links claims to source support so users can separate what is known, inferred, and hypothetical. | Lets the buyer audit support strength before relying on the package. |
| Reading the Quantitative Evidence | Links claims to source support so users can separate what is known, inferred, and hypothetical. | Lets the buyer audit support strength before relying on the package. |
| Reading the numbers together | Summarizes a report component that ties field evidence to the dossier's decision logic. | Helps the buyer inspect the argument, evidence posture, and decision logic before relying on the package. |
| Mechanistic Insights and Failure Modes | Explains which biological signals matter for interpreting the field and discriminating between programs. | Helps the buyer inspect the argument, evidence posture, and decision logic before relying on the package. |
| Safety, Risk and Failure Modes | Identifies translational, regulatory, safety, or delivery constraints that shape whether the thesis can be tested. | Shows where the thesis could fail or require more validation before action. |
| Domain Deep Dives Cardiology Neurology Oncology | Explains which biological signals matter for interpreting the field and discriminating between programs. | Helps the buyer inspect the argument, evidence posture, and decision logic before relying on the package. |
| Prospective Evidence Standards and Minimal Viable Designs | Shows how strong the source base is and where conclusions depend on indirect or thinner support. | Lets the buyer audit support strength before relying on the package. |
| Non Obvious Early Pipeline Ai Opportunities | Connects research maturity to competitive position, program strategy, and diligence watchpoints. | Helps the buyer inspect the argument, evidence posture, and decision logic before relying on the package. |
| Competitive Landscape | Frames the field's June 2026 state, active programs, research fronts, and timing signals. | Connects research claims to practical diligence and monitoring decisions. |
Evidence & Decision Workbook
The workbook is the decision engine.
A locked public preview of selected workbook-derived sheets: sheet structure, decision modules, power rows, traceability, and audit posture. Licensed buyers receive the full workbook files.
Read-only workbook preview with clipped values and no formulas. Licensed buyers receive the full Evidence & Decision Workbook with editable assumptions and working files.
Workbook inventory
Delivered workbook sheets.
Full inventories belong on dossier pages, not the homepage. Sheet names below are extracted from the actual workbook file.
| # | Workbook sheet | What it does |
|---|---|---|
| 1 | Workbook Guide | Navigation layer for using the workbook, understanding sheet roles, and following the decision package. Public preview exposes Purpose, Buyer-auditable evidence base for. fields and clipped entries such as Reconciliation, 131 audited claims = 59., Grade. |
| 2 | Claim Ledger | Claim-by-claim ledger showing verdicts, source support, inference status, and decision relevance. Public preview exposes claim_id, claim_text, report_section, report_page fields and clipped entries such as K1, AI for Kounis syndrome (acute., Limitations &. |
| 3 | Evidence Table | Evidence-verification surface for checking sources, support quality, and the audit trail. Public preview exposes id, type, title, journal/source fields and clipped entries such as PMID:40305017, PMID, Generalizability of FDA-Approved.. |
| 4 | Source-Support Verdicts | Independent check on whether key claims are actually supported by cited sources. Public preview exposes claim_id, citation, verified_title, support_verdict fields and clipped entries such as K1, PMID:42052969, Decoding Kounis syndrome with.. |
| 5 | Source Quality Summary | Summary of source strength, gaps, and areas where conclusions depend on weaker evidence. Public preview exposes Total customer-facing claims., 59 fields and clipped entries such as Unique primary sources (deduped), ~46, Cited references (rendered in report.. |
| 6 | Category Scorecard | Executable decision surface for classifying programs, risks, gates, or evidence states. Public preview exposes Foundation ECG models, High discrimination;., Untested at operating point, Retrospective+external fields and clipped entries such as. |
| 7 | Therapeutic-Area Scorecard | Executable decision surface for classifying programs, risks, gates, or evidence states. Public preview exposes Area, Axis read, Key sources, Core gap fields and clipped entries such as Cardiology, Validity-illusion exemplar; best., DeepECG, PPG-PAD, AI-ECG.. |
| 8 | V-U-E-D Scorecard | Executable decision surface for classifying programs, risks, gates, or evidence states. Public preview exposes Axis, Question, What it covers, How to establish (not assume) fields and clipped entries such as Validity, Is the number real & durable?, Leakage,. |
| 9 | Figure Source Map | Traceability surface linking claims, figures, calculations, and workbook outputs back to sources. Public preview exposes figure_id, figure_title, figure_family, section_anchor fields and clipped entries such as Buyer decision matrix, Deterministic. |
| 10 | Figure Attempt Ledger (Internal | Figure and numeric reproduction log for validating calculations, chart inputs, or displayed values. Public preview exposes fig #, title, family, attempts fields and clipped entries such as Buyer decision matrix, Deterministic chart, Rendered. |
| 11 | Hypotheses Ledger | Maps hypotheses to evidence state, falsification logic, endpoints, and next-study design. Public preview exposes id, hypothesis, rationale (why non-obvious), falsification criterion fields and clipped entries such as H1, Among AI diagnostic/prognostic., The. |
| 12 | Most Valuable Next Studies | Prioritizes next studies that could change the decision, including endpoints and evidence gates. Public preview exposes id, study, question, design fields and clipped entries such as S1, AUROC<->net-benefit decoupling., Does the headline metric predict.. |
| 13 | Early-Pipeline AI Opportunities | Workbook surface supporting a specific evidence, decision, study-planning, or release-control workflow. Public preview exposes #, opportunity, what it does, evidence base fields and clipped entries such as AI-driven endpoint selection &., Choose. |
| 14 | Regulatory Tracker | Tracks regulatory, endpoint, reimbursement, or qualification requirements that shape testing strategy. Public preview exposes instrument, detail, status, what it changes fields and clipped entries such as FDA AI device list, ~1,451 cumulative (295 in 2025;. |
| 15 | Trial Registry Tracker | Tracks programs, trials, sponsors, or competitive context for diligence and monitoring. Public preview exposes NCT, title, status, type fields and clipped entries such as NCT06364267, Randomized Double Blind Phase II., RECRUITING. |
| 16 | Verification & Search Log | Release-discipline surface for revisions, unresolved issues, audit findings, and readiness tracking. Public preview exposes activity, method/source, outcome, date fields and clipped entries such as Citation resolution, NCBI E-utilities esummary+efetch, 80/80. |
| 17 | Excluded Claims | Disposition log for rejected, bounded, or quarantined claims so they do not leak into the decision thesis. Public preview exposes STR#9, RECOVERED, §18 Regulatory, EFPIA Mar 2025 EMA Qualification. fields and clipped entries such as STR#10, RECOVERED, §18. |
| 18 | Salvage Ledger | Disposition log for rejected, bounded, or quarantined claims so they do not leak into the decision thesis. Public preview exposes Broad cited evidence base, REUSE, 59 kept claims / ~74 sources are. fields and clipped entries such as Quantitative anchors,. |
| 19 | Power Calculations | Power-calculation layer for assumptions, endpoint logic, computed sample sizes, and sensitivity checks. Public preview exposes hypothesis_id, study_id, endpoint, test_type fields and clipped entries such as H1, S1, AUROC vs decision-curve net-benefit.. |
| 20 | BAC_Input_Template | Executable decision surface for classifying programs, risks, gates, or evidence states. Public preview exposes Evidence completeness (0-1), Fraction of the four axes with., ____ fields and clipped entries such as Regulatory clearance, cleared / not-cleared. |
| 21 | BAC_Feature_Score | Executable decision surface for classifying programs, risks, gates, or evidence states. Public preview exposes Axis (output class), Meaning (rubric: 3 = severe., Deficit 0-3 (ENTER), Weight (expert-init) fields and clipped entries such as VALIDITY-LIMITED,. |
| 22 | BAC_Class_Output | Executable decision surface for classifying programs, risks, gates, or evidence states. Public preview exposes Evidence completeness (ENTER 0-1), 0.8 fields and clipped entries such as Regulatory clearance (ENTER., cleared, Validity failure mode (ENTER if.. |
| 23 | BAC_Decision_Rules | Executable decision surface for classifying programs, risks, gates, or evidence states. Public preview exposes Class (binding axis), Recommended evidence move, Primary endpoint, Required safety/deployment readout fields and clipped entries such as. |
| 24 | BAC_Failure_Mode_Rules | Captures failure modes and constraints that could invalidate or narrow a program. Public preview exposes Entered form (any documented., Resolved failure class, Matched remedy (each demands a. fields and clipped entries such as LEAKAGE, LEAKAGE, Grouped /. |
| 25 | BAC_Minimum_Assay_Panel | Executable decision surface for classifying programs, risks, gates, or evidence states. Public preview exposes Evidence item, Axis, Tier fields and clipped entries such as External / multi-site validation., Validity, Required. |
| 26 | BAC_Archetype_Cards | Executable decision surface for classifying programs, risks, gates, or evidence states. Public preview exposes High-AUROC cleared diagnostic, no., 1,3,2,2 (cleared), UTILITY-LIMITED, Net-benefit study;. fields and clipped entries such as Single-site. |
| 27 | BAC_Validation_Log | Executable decision surface for classifying programs, risks, gates, or evidence states. Public preview exposes Comparator baseline, chronological evidence-tier alone. fields and clipped entries such as Calibration requirement, required (reliability across. |
| 28 | BAC_Traceability_Map | Traceability surface linking claims, figures, calculations, and workbook outputs back to sources. Public preview exposes BAC element, Maps to claim/finding, Evidence ID, Figure fields and clipped entries such as Validity axis + failure-mode resolver,. |
| 29 | Competitive Landscape | Tracks programs, trials, sponsors, or competitive context for diligence and monitoring. Public preview exposes Company / program, Asset / modality, Axis sold, Stage / maturity fields and clipped entries such as ArteraAI, Multimodal digital-pathology.,. |
| 30 | Score Basis | Explains how scores or ratings were assigned so the decision instrument can be inspected. Public preview exposes fig: evidence-maturity heat map, ordinal 0-4 per category x axis, expert synthesis from cited. fields and clipped entries such as fig: category. |
| 31 | Traceability Certificate | Traceability surface linking claims, figures, calculations, and workbook outputs back to sources. Public preview exposes claim_id, report_section, report_page, claim_text fields and clipped entries such as K1, Limitations & Evidence Gaps, AI for Kounis. |
| 32 | Version Registry | Release-discipline surface for revisions, unresolved issues, audit findings, and readiness tracking. Public preview exposes v1.0 (source build), June 2026, Original audited source dossier;., prior build fields and clipped entries such as Model, Four-Axis. |
| 33 | Audit Issues Log | Release-discipline surface for revisions, unresolved issues, audit findings, and readiness tracking. Public preview exposes #, Issue, Status, Resolution fields and clipped entries such as AI-1, Workbook genuinely executable?, RESOLVED. |
Power-calculated next studies
Power calculations are the execution bridge.
Power calculations are planning-grade research calculations for endpoint logic, assumptions, sample size, budget, and timeline. They do not prove hypotheses or constitute clinical protocol.
| Hypothesis | Primary endpoint | Test | Effect | α | Power | Computed n | Design n |
|---|---|---|---|---|---|---|---|
| H1 | AUROC vs decision-curve net-benefit rank correlation | Correlation (Fisher-z) | 0.7 | 0.05 | 0.90 | 57 | 200 |
| H2 | Leakage-stress score vs external AUROC drop (correlation) | Correlation (Fisher-z) | 0.5 | 0.05 | 0.85 | 33 | 40 |
| H3 | Drift-alarm lead-time before measured decay (days) | Two-sample t | 0.6 | 0.05 | 0.80 | 44 | 44 |
| H4 | Common-vs-rare AUROC degradation (rare-stratum AUC) | AUC (Hanley-McNeil, external) | 0.85 | 0.05 | 0.90 | 420 | 420 |
| H5 | Realized type-I error precision (simulation) | Single-proportion precision | 0.05 | 0.05 | 0.80 | 7299 | 7299 |
Release audit discipline
Audit posture is part of the product.
Public pages do not publish exact audit scores. They should state the release discipline, revision posture, and limitations without implying peer review or validated clinical guidance.
Research-use decision package
Not medical advice, clinical guidance, investment advice, or a clinical protocol.
Known / inferred / hypothetical
The package separates cited evidence from inference and frontier hypothesis generation.
Audit-gated and revision-controlled
Adversarial audit checks source support, formulas, figures, limitations, and release readiness.
Buy this dossier
Full decision package — 118-page report + 33-sheet workbook
Buy now
- Decision research report (118 pages, PDF)
- Evidence & Decision Workbook (33 sheets, .xlsx) — claims, source-support verdicts, power calculations, traceability, and audit log
- Live, editable power-calculated next studies
$3,500$5,500Launch price