Key Strengths
- Clear, jargon-free vision statement (HQ1) — "What if every healthcare billing team..." framing is accessible and compelling
- Strong "Why Now" with 8 market timing factors including technology convergence, regulatory mosaic, and AI adoption paradox
- Explicit "Why ARPA-H" argument — commercial incentive misalignment is well-articulated
- Go/No-Go gate at Month 4 with specific criteria — shows program management discipline
- Three Technical Areas (TA1/TA2/TA3) with clear scope boundaries and milestone mapping
- 10 risk rows including transition, programmatic, reputational, and regulatory — comprehensive risk awareness
- Federal keyword screen is clean — "parity" replaced with "nondiscrimination" even in MHPAEA context
- IP Table, BOE, and Payment Milestone Schedule now included per UMich template
- Quantified 3-5x improvement metric and 40pt overturn variance with methodology (n=47, propensity matched, κ=0.84)
- Broad Access expanded with safety-net provider commitments (FQHCs, Critical Access, rural); HQ10 safeguards in place
- TA dependency flow documented: TA1→TA2→TA3 with closed-loop feedback
- Partner selection strategy documented with target ICP profile (50+ bed BH provider, PHP/IOP/SUD, 835 data)
Key Weaknesses & Gaps
- Partner organization is still TBD — "[Partner Healthcare Organization - TBD during contracting]" weakens credibility
- All sub-awardee POCs are TBD — no named individuals beyond Stratum's team
- Statistical power discussion is Phase II deferred — reviewers may question Phase I sample size (60-80 outcomes)
- Phase I budget ($441K) is significantly below ARPA-H norms ($1M-$109M range, avg $26M) — needs increase
- No letters of support or intent referenced
- No explicit team qualifications section — relies on cover page POCs only
- Cost Sharing line shows "Partner In-Kind" but partner is TBD
▶ Heilmeier Questions (HQ1–HQ10)
88 / 100Evaluated against ARPA-H's "Hidden Questions Behind the Heilmeier Questions" — 10 questions with sub-questions and considerations that program managers use to evaluate proposals.
HQ1: What are you trying to do?
V11 opens with: "What if every healthcare billing team — regardless of size or expertise — could access the same decision intelligence..."
The vision statement is accessible, avoids jargon, and passes the "explain to a grandparent" test. Technical depth follows in subsequent paragraphs. The "precedent-as-software" framing is memorable and differentiating.
HQ2: How is it done today, and what are the limits?
Three-category taxonomy (workflow automation, retrospective analytics, documentation without outcome validation) plus updated Q1 2026 competitive landscape with 5 competitor categories. Now sourced from continuous competitive intelligence monitoring.
Thorough competitive mapping backed by Stratum's ongoing CI infrastructure. The "AI adoption paradox" framing (46% adoption, rising denial rates) is compelling. The comparison table is effective. The "horizontal breadth vs. vertical depth" strategic framing creates clear differentiation.
HQ3: Why now? Why ARPA-H?
8 "Why Now" factors + explicit "Why ARPA-H" (commercial incentive misalignment) + "Why not possible 5 years ago" (3 converging technologies).
This is V11's strongest section. The convergence argument (NLP maturity 2023-24, standardized 835 data, MHPAEA 2024 Final Rule) is specific and credible. The "ARPA-H bridges the gap" argument is well-constructed — public-interest infrastructure that no single vendor will fund.
HQ4: What's new in your approach?
Five "genuinely new" innovations listed: outcome-labeled precedent objects, evidence-combination granularity, traceability ledgers, math-first signal processing, human-governable design.
The "precedent-as-software" concept is genuinely novel. The evidence-combination granularity claim (40pt overturn variance) is now backed with methodology: n=47 cases, single MA payer, Q3-Q4 2025, propensity-matched cohort, dual-reviewer coding (κ=0.84). Structured evidence assemblies achieved 72% overturn vs 31% unstructured. This is the proposal's strongest empirical anchor.
HQ5: Who cares? What risks?
10 risk rows covering technical (evidence linking, policy ambiguity, outcome noise, drift), operational (integration, volume), strategic (transition, programmatic, reputational), and regulatory (MHPAEA enforcement uncertainty).
Comprehensive risk coverage. The addition of transition, programmatic, reputational, and regulatory risks shows maturity. The regulatory risk row explicitly states Stratum's value is "regulatory-agnostic" with state-level demand as independent driver. Each mitigation is specific and actionable.
HQ6: Deliverables & Timeline
6 milestones with TA mapping, exit criteria, and deliverables. 5 deliverable categories. BAA execution now at M3 (was M1). M1 refocused on technical foundation and partner engagement (LOI/MOU).
Milestone table is well-structured with clear exit criteria per UMich template guidance. The Go/No-Go gate at M4 shows disciplined program management. Moving BAA to M3 is strategically sound — partner identification begins pre-award, BAA execution is post-award within first 90 days. This is standard for ARPA-H OT contracts where clinical site is finalized during contracting/early execution.
HQ7: Cost & Budget
$441K total, 6 months, $73.5K per milestone payment. BOE formatted to ARPA-H standard categories. $45K partner in-kind cost sharing (9.3%).
Budget increased to $950K — right-sized for ARPA-H expectations. New breakdown: $450K labor (8 roles including Clinical Integration Lead and Data Analyst, 3,440 hours), $290K subawardees (5 subs including $120K clinical partner site coordination, $50K independent evaluator), $45K materials, $15K equipment, $30K travel, $52K ODC, $48K indirect (10% MTDC), $20K profit (2.1%). Weighted milestone payments ($140K-$175K) reflect execution intensity. Partner in-kind: $100K. Grand total project value: $1.05M. Phase II de-risk target updated to $3-5M.
HQ8: Midterm Check
Go/No-Go Gate at Month 4: time-to-proof-kit ≤45 min AND traceability completeness ≥80%. Pivot plan if not met.
Clear, quantitative gate criteria with an explicit fallback plan. This directly addresses the ARPA-H program management philosophy of "fail fast."
HQ9: Broad Access & Affordability
Exportable JSON/XML formats, open-standard design, reduced dependence on scarce expert billers, cross-facility transfer design, Phase II cost modeling planned. Now includes explicit safety-net provider commitments: FQHCs, Critical Access Hospitals, community BH centers, independent rural providers.
Strong coverage after expansion. Three specific commitments: (1) open-standard schemas at no licensing cost, (2) per-facility pricing viable for sub-50-bed facilities, (3) corpus sharing architecture. Framed as reducing taxpayer-funded administrative waste and strengthening small independent providers — language aligned with current administration priorities (fiscal efficiency, small business, rural communities) without DEI terminology.
HQ10: Misperception Safeguards
Addresses: not automating clinical decisions, symmetric information access, HIPAA compliance, traceability auditability, uncertainty markers, human review thresholds.
Proactively addresses the "gaming the system" perception. The symmetric information argument is sophisticated — "levels the playing field rather than advantaging one side."
▶ Technical Innovation & Differentiation
90 / 100Novelty of Core Concept
"Precedent-as-software" — converting expert judgment into versioned, governable, outcome-labeled data artifacts — is genuinely novel in healthcare RCM. No funded ARPA-H project or commercial product takes this approach. The closest analogues (PRECISE-AI, DIGIHEALS) operate in different domains.
Technical Architecture Clarity
TA1/TA2/TA3 decomposition is clean. Each TA has a clear purpose: data assets (TA1), adaptive learning (TA2), operational integration (TA3). Dependency flow now explicitly documented: TA1→TA2→TA3 with TA3 closing the loop back to TA1. Critical path identified through TA1. TA2 and TA3 operate in parallel post-initial artifacts, converging at M4 Go/No-Go gate.
Quantitative Claims
Strong claims throughout: $262B annual denial losses [1], 82% overturn rate [2], 41% at 10%+ denial rates [3]. The critical 40pt variance claim now includes full methodology: n=47 PHP/IOP denial cases, single MA payer, Q3-Q4 2025, retrospective cohort with propensity matching on denial type/diagnosis/acuity, dual-reviewer coding (κ=0.84). Structured evidence assemblies: 72% overturn vs 31% unstructured. The 3-5x improvement metric (from ~12% appeal rate to 40-60%) is bold but grounded in the appeal rate gap.
Competitive Moat Argument
The "horizontal breadth vs. vertical depth" framing is effective. The argument that residual high-complexity denials concentrate as competitors clear easy cases is strategically sound. The claim that provenance infrastructure "cannot be retrofit onto existing platforms" is bold but defensible.
Engineering & Scientific Basis
References are legitimate and relevant: W3C PROV-DM [5], Hollnagel resilience engineering [6], Nonaka & Takeuchi knowledge transfer [7], Montgomery statistical process control [8]. The connection between these foundations and the proposed system is well-articulated.
▶ UMich Template & ARPA-H Format Compliance
82 / 100Evaluated against University of Michigan ARPA-H resource collection: Solution Summary Template, Budget Development Resources, Payment Milestone Schedule, IP Table Template, Task Description Document.
Sub-Awardee Completeness
Sub-awardee entries remain TBD for names/POCs, but now include: (1) explicit justification — "Partner selection will be finalized during contracting phase per ARPA-H flexible teaming provisions," and (2) target clinical partner ICP — 50+ bed BH provider operating PHP/IOP/SUD programs, established RCM department, 835 electronic remittance data, ≥15% denial rates, Greater Seattle/PNW region. This is acceptable at Solution Summary stage — named partners are most critical at the Full Proposal invitation stage.
Page Length Concern
Duplicate Adoption Challenges section has been removed (~0.5 page recovered). However, new content was added (safety-net commitments, TA dependency flow, methodology detail, regulatory risk row, partner selection note). Net page impact should be monitored — regenerate DOCX after budget decision to verify 6-page compliance.
BOE Format
BOE now uses ARPA-H standard categories (Direct Labor, Labor Hours, Subawardees, Materials, Equipment, Travel, ODC, Profit, Cost Sharing). Cost Sharing row included. Format matches UMich Budget Development Resources template.
▶ Federal Keyword Screening
95 / 100Scanned against NSF 68-word list, HHS/Head Start ~200-word list, and PEN America 350+ word tracking list. V10 replacements (barriers→challenges, systemic→structural, unbiased→independent, institutional memory→operational memory, tribal knowledge→undocumented knowledge) are confirmed in V11.
| Category | Status | Detail |
|---|---|---|
| DEI Terminology | CLEAR | No instances of: equity, diversity, inclusion, belonging, marginalized, underserved, vulnerable, minority |
| Gender/Race Terms | CLEAR | No instances of: gender, female, women, racial, racism. "Race" appears only as substring in "traceability" — false positive |
| Systemic/Structural | CLEAR | "Systemic" replaced with "structural" in V10. V11 uses "structural" consistently |
| Barriers | CLEAR | "Barriers" replaced with "challenges" in V10. V11 uses "challenges" consistently |
| Bias/Unbiased | CLEAR | "Unbiased" replaced with "independent" in V10. No instances remain |
| Tribal/Indigenous | CLEAR | "Tribal knowledge" replaced with "undocumented knowledge" in V10 |
| Trauma | CLEAR | No instances of: trauma, trauma-informed, ACE, adverse childhood |
| Disability | CLEAR | No instances |
| Disparity/Inequity | CLEAR | No instances |
| Behavioral Health Context | CLEAR | "Behavioral health" used throughout — essential domain term. "Parity" replaced with "nondiscrimination" even in MHPAEA regulatory context. Clean. |
Overall Keyword Risk
V11 is clean across all three screening lists. The V10 replacements are properly carried through. The only monitoring item is "behavioral health" and "parity" — both are essential domain terms used in regulatory context, not flagged DEI terminology. No reviewer would flag these in a healthcare proposal.
▶ Proxy Awardee Pattern Alignment
78 / 100Evaluated against patterns extracted from 6 ARPA-H proxy awardees: PRECISE-AI, DIGIHEALS, ADVOCATE, MARCUS, Every Cure, and Duality Technologies.
PRECISE-AI
Relevance: High
Pattern: AI-augmented clinical decision support with provenance. V11 mirrors this with precedent objects + traceability ledgers.
DIGIHEALS
Relevance: Medium
Pattern: Infrastructure resilience in health systems. V11's "Resilient Systems" framing aligns well with RSO mission.
ADVOCATE
Relevance: Medium-High
Pattern: Patient advocacy through technology. V11's "symmetric information access" echoes this patient-centric framing.
Duality Technologies
Relevance: Medium
Pattern: For-profit small company winning ARPA-H funding. Validates Stratum's organizational type is not a disqualifier.
Every Cure
Relevance: Low-Medium
Pattern: Data-driven approach to healthcare system inefficiencies. Different domain but similar thesis of hidden information asymmetry.
MARCUS
Relevance: Low
Pattern: Health system monitoring and response. V11's drift detection has loose conceptual parallels.
Patterns V11 Adopts Well
Clear problem quantification — V11 leads with $262B annual loss, 82% overturn rate. Proxy awardees consistently lead with quantified impact.
Phased technical approach — All successful awardees use progressive complexity scaling. V11's Phase I→II→III progression mirrors this.
Public-interest framing — Exportable schemas, open standards, broad access. Matches ARPA-H's emphasis on health impact beyond the performer.
Phased technical approach — All successful awardees use progressive complexity scaling. V11's Phase I→II→III progression mirrors this.
Public-interest framing — Exportable schemas, open standards, broad access. Matches ARPA-H's emphasis on health impact beyond the performer.
Patterns V11 Could Strengthen
Named institutional partners — Successful awardees typically have named clinical partners or university collaborators at submission time. V11's TBD partners are a significant gap.
Team credentials section — Proxy awardees highlight PI/team qualifications prominently. V11 has no dedicated team qualifications section.
Preliminary data specificity — Awardees with "preliminary data" claims typically cite specific studies or pilot results with methodology. V11's "preliminary operational data" is vague.
Team credentials section — Proxy awardees highlight PI/team qualifications prominently. V11 has no dedicated team qualifications section.
Preliminary data specificity — Awardees with "preliminary data" claims typically cite specific studies or pilot results with methodology. V11's "preliminary operational data" is vague.
Adding even one named clinical partner and a brief team qualifications paragraph would significantly strengthen the proposal's alignment with proxy awardee patterns.
▶ Risk Coverage & Resilience Design
86 / 100Go/No-Go Gate Design
Month 4 gate with dual criteria (time-to-proof-kit ≤45 min AND traceability completeness ≥80%) plus explicit pivot plan ("schema refinement only"). This shows ARPA-H-compatible program management discipline. The criteria are measurable and the pivot is concrete.
Resilience Framing
The proposal explicitly maps to RSO mission attributes: stability, recoverability, adaptive learning. The stress testing plan (cross-experience-level consistency, volume spikes, simulated policy change) at M6 directly validates resilience claims.
Regulatory/Political Risk
Regulatory risk row now added: "Regulatory uncertainty (MHPAEA enforcement)" with mitigation citing regulatory-agnostic value proposition and state-level acceleration as independent demand driver. This directly addresses the ERIC challenge risk and demonstrates the proposal isn't dependent on any single regulatory outcome.
▶ Priority Actions for V12
Action Items| # | Action | Impact | Status |
|---|---|---|---|
| ✓ | Done | Done | |
| 2 | Name at least one clinical partner — even a letter of intent removes the biggest remaining credibility gap. Target ICP now documented. | Critical | External |
| 3 | Add team qualifications paragraph — PI experience, domain expertise, technical capabilities. 3-4 sentences. | High | Pending |
| 4 | Regenerate DOCX and verify page count — new content added; confirm 6-page compliance. | Medium | Pending |
| ✓ | Done | Done | |
| ✓ | Done | Done | |
| ✓ | Done | Done | |
| ✓ | Done | Done | |
| ✓ | Done | Done | |
| ✓ | Done | Done | |
| ✓ | Done | Done |