Skip to main content

Platform Evolution: Multi-Stakeholder Acceptance Optimization

Review (2026-03-19): This research has been critically reviewed. Applicable elements (MPAS-7 benchmark, NHI timing context, tiered output pattern, evidence-to-narrative gap) were extracted into the Consolidated Action Plan. Hypotheses H1-H4 and the literature clusters are NOT adopted — see the consolidated plan for rationale. A new research brief (Acceptance Validation Research Brief) replaces the remaining AutoResearchClaw pipeline stages with an implementation-driven validation approach.

Status: research-in-progress — Generated by AutoResearchClaw (stages 1–8 of 23 completed). Experiment design and validation stages pending. Findings require human review before adoption.

Source: AutoResearchClaw v0.3.1, ACP mode (Claude Sonnet via Mac subscription), March 18 2026. Artifacts at ~/dev/AutoResearchClaw/artifacts/sv0-platform/.

Research Question​

Given a multi-perspective acceptance deficit across 7 stakeholder roles (CISO, SecOps, UX, Enterprise Executive, CEO, Product QA, Security Auditor), how do you derive the minimal set of platform changes that maximize acceptance across all roles simultaneously?

This is the inverse problem of standard analyst frameworks (Gartner, Forrester): instead of scoring a finished product externally, derive design-time changes from measured acceptance gaps.

Why This Is Timely (2024–2026)​

  1. NHI governance emerged as a distinct category in 2025. OWASP NHI Top 10, CSA MAESTRO, MITRE ATLAS v4 all published. SecurityV0 is among the first platforms — evolution decisions now define category norms.

  2. Multi-axis evaluation became standard. SACR AI SOC (2025) introduced dual-axis scoring. Forrester CRQ Wave (Q2 2025) uses 3 dimensions. NIST CSF 2.0 (Feb 2024) added "Govern." The industry accepted single-score inadequacy.

  3. Evidence-based security replaced checkbox compliance. CISOs demand continuous control validation with demonstrable evidence. SecurityV0's SHA256 integrity-hashed evidence packs are architecturally aligned — but evidence is trapped in technical format.

Literature Synthesis​

Fifteen papers analyzed across five thematic clusters. Full synthesis at artifacts/sv0-platform/stage-07/synthesis.md.

Cluster 1: Multi-Stakeholder Framework Design​

Papers: Hindricks 2020 (ESC AF Guidelines), Visseren 2021 (ESC CVD Prevention), Hale 2021 (OxCGRT)

Clinical guideline frameworks demonstrate graded recommendation classes (I–III) with evidence levels (A–C), producing a two-dimensional signal that different consumers parse differently.

SV0 Implication: Adopt a dual-axis scoring model — action urgency × evidence confidence — so that CISOs see prioritized action items, SecOps see technical confidence, and Executives see comparative posture across business units.

Cluster 2: Tiered Output & Adoption-Driven Design​

Papers: Percie du Sert 2020 (ARRIVE 2.0), Page 2021 (PRISMA 2020)

Both ARRIVE 2.0 and PRISMA 2020 solved the same problem SecurityV0 faces: a strong evidence engine whose output was not being operationally consumed. The solution was to split output into an "Essential 10" tier (minimum actionable items) and a "Recommended Set" (context-dependent depth).

SV0 Implication:

  • Implement an "Essential Actions" view (5–10 critical items per persona) and a "Full Evidence" view.

  • Auto-generate visual evidence chains showing data source → finding → recommendation → action.

  • Each finding should have an expandable "Why this matters" section with worked examples.

Cluster 3: Model Validation & Operationalization​

Papers: Wynants 2020 (COVID-19 Prediction Models), Damschroder 2022 (Updated CFIR)

Most prediction models were at high risk of bias and not ready for clinical use despite being technically sophisticated — the gap was in validation, calibration, and real-world applicability.

SV0 Implication:

  • Surface validation metadata — show analysts not just findings but confidence intervals, data freshness, and coverage gaps.

  • Build iterative feedback mechanisms — allow analysts to flag false positives/negatives.

  • Maintain backward compatibility during platform evolution.

Cluster 4: AI Architecture & Scalable Processing​

Papers: Alzubaidi 2021, Zhou 2021 (Informer), Chen 2020, Amarasinghe 2020

SV0 Implication:

  • Apply selective attention to telemetry/log data — focus on highest-information-density events.

  • Persona-adaptive presentation: dynamically adjust dashboard content based on user role.

  • Connector-specific analysis pipelines — one-size-fits-all processing produces low-quality output.

Cluster 5: Crisis Response & Coordinated Action​

Papers: Holmes 2020, Vindegaard 2020

SV0 Implication:

  • Prioritize comprehensive data ingestion during incidents before recommending actions.

  • Automatically flag high-risk assets requiring differentiated response playbooks.

  • Build incident views that bridge SOC, IT, legal, and executive teams.

Identified Gaps​

  1. No validated framework for multi-persona acceptance scoring — CFIR comes closest but operates within a single domain.

  2. No systematic method for translating evidence to non-technical narratives — the evidence-to-board-deck translation gap.

  3. No partner/channel sellability patterns — how to package complex analysis for indirect sales.

  4. No feedback loop architecture — continuous validation in a software platform context.

Hypotheses​

Four synthesized hypotheses generated through multi-agent debate (pragmatist, innovator, contrarian perspectives). Full debate transcript at artifacts/sv0-platform/stage-08/hypotheses.md.

H1: Engine Completeness Is the Binding Constraint​

Core claim: Empty target_resource fields, count discrepancies, and partial spec implementations mean every presentation-layer investment is premature. Fix the engine first.

Test: Achieve 100% field population, then re-run 7-reviewer scoring with zero UI changes. If ≥3 of 7 scores improve by ≥10pp, the engine was the bottleneck.

Time-box: 2 weeks. If field population can't reach 95%+, the gaps are architectural, not implementational.

H2: Two-Product Architecture​

Core claim: Optimizing a single artifact for 7 personas is structurally unsound. Split into:

  • Analyst Workbench — optimized for speed, depth, technical precision. Scored by SecOps + API Auditor + Product QA + UX.

  • Opinionated Report Generator — outputs board-ready deliverables requiring zero partner rewriting. Scored by CISO + Executive + CEO + Partner/Sales.

Test: Survey 5 SI partners on demo-vs-report workflow. If ≥3 demo the platform live, reject this hypothesis.

H3: Cross-Source Disagreement as Primary Signal​

Core claim: Normalization destroys the signal that identifies novel threats. Cross-source disagreement IS the analytical insight — exploit it instead of smoothing it away.

Test: Jensen-Shannon divergence across 4 simulated source assessments on 50 scenarios. Target: ≥0.80 AUROC for novel vs routine threat distinction.

Dependency: H1 must pass first (need reliable multi-source output before measuring disagreement).

Competitive moat potential: No reviewed paper considers multi-source disagreement as a feature.

H4: Opinionated Verdicts Over Analytical Dimensionality​

Core claim: Single verdict + business impact outsells dual-axis metadata for non-technical buyers. "McKinsey-style verdicts, not PRISMA flow diagrams."

Test: A/B compare two report formats across LLM-simulated Executive/CEO/Partner personas. Format B (verdict + action + impact) should score ≥25% higher on purchase intent.

Proposed Execution Sequence​

Week 1-2:  H1 (Engine Completeness Gate) — time-boxed sprint
           H3 spike (30-min synthetic test of disagreement signal)
           H2 prerequisite (survey 5 SI partners on workflow)

Week 3:    Decision gate based on H1/H2/H3 results

Week 3-6:  If H1 passes → H2 build (two-product split)
           If H1 fails → engine completeness work continues
           H4 test (verdict vs dual-axis A/B) runs in parallel

Benchmark: MPAS-7 (Multi-Perspective Acceptance Score)​

Custom instrument — no external benchmark exists for this problem.

RoleMetricBaselineTarget
CISO Executive15-second executive clarity70%≥85%
SecOps AnalystDay-1 task completion70%≥80%
Product QASpec match8 partial, 2 missing≤2 partial, 0 missing
UX CriticIA grade + jargonB- / 23 termsA- / ≤5 terms
Security AuditorData consistencyMultiple issuesZero critical
Enterprise ExecSellability (1-5)1.8/5≥3.5/5
CEOItems accepted18/28 (64%)≥24/28 (86%)

Constraints​

  • Deterministic only — no ML, no probabilistic scoring
  • Read-only connector model preserved
  • CEO constraints: no speculation, no effort estimates, no telemetry creep, partner-first, cut > add, plain English

Next Action​

Status: research-in-progress

Decision needed from: Ivan (CTO)

Options:

  1. Adopt H1 — run 2-week engine completeness sprint, then proceed to H2/H3/H4
  2. Skip H1 — assume engine is sound, proceed directly to H2 (two-product architecture)
  3. Prioritize H4 — quick A/B test of verdict vs dual-axis format before any architecture changes
  4. Defer — revisit after pilot evaluation (OpenClaw vs ClaudeClaw) concludes

GitHub Issue: not yet created

Raw artifacts: ~/dev/AutoResearchClaw/artifacts/sv0-platform/