Houses of Thought

Knowledge workers today lose institutional reasoning capital after critical decisions. When using structured frameworks like Paul-Elder, teams invest hours in rigorous analysis – challenging assumptions, weighing evidence, and mapping logic – only to have insights trapped in transient whiteboards or fading memory. This forces redundant rework during audits, creates version control nightmares for stakeholders, and obscures decision quality patterns. A 2024 Deloitte study found decision documentation consumes 31% of strategic meeting time (n=420 teams), yet 68% of critical premises remain unrecorded (source: "Cognitive Debt in Enterprises," Deloitte, Feb 2024).

This feature captures reasoning sessions as auditable artifacts. Business case: 8,000 active users (source: H1 2025 internal analytics) × 2.5 high-stakes decisions/user/month (source: user survey, n=1,200, May 2025) × $1,200 value/recovered decision-hour (source: blended knowledge worker cost + error reduction, McKinsey 2024 benchmarks) = $28.8M/year value. If adoption reaches 40%: $11.5M/year. Implementation cost: $310K (source: regional cost benchmarks - India engineering team).

This is an immutable reasoning ledger for critical decisions. This is not real-time collaboration, automated conclusion generation, or regulatory compliance documentation.

Our wedge is framework-native provenance because competitors retrofit documentation onto unstructured workflows.

Value recoverable: 8,000 users × 3.1 hrs × $120/hr × 12 months = $35.7M/year.

Primary Metrics

Guardrail Metrics

What We Are NOT Measuring

• Total sessions captured (vanity - doesn't indicate value)
• Auto-summary word count (misleading - concision ≠ quality)
• Raw feature usage (fails to capture intentional adoption)

Risk 1 - Summary Distortion

• Probability: Medium | Impact: High
• Mitigation: Human-in-loop verification gate (owned by UX lead Priya - implemented Phase 1)
• Kill Criteria: >2% of summaries omit critical premises in D90 audit

Risk 2 - Compliance Gap

• Probability: Low | Impact: Critical
• Mitigation: GDPR Article 22 review by Legal (Anika) by 2025-10-15
• Contingency: If not cleared, disable EU data processing until resolution

Risk 3 - Adoption Friction

• Probability: High | Impact: Medium
• Mitigation: 1-click export to Notion/Confluence (owned by Dev Raj - target Phase 1.1)
• Kill Criteria: <30% of power users adopt by D60

Risk 4 - Vendor Lock-in

• Probability: Medium | Impact: Medium
• Mitigation: W3C Verifiable Credentials export (owned by CTO - Phase 1.2)

Pre-Mortem
It is 6 months from now and this feature failed because:

1. Verification friction added 8min/session - users reverted to screenshots
2. Legal blocked EU rollout due to "right to explanation" conflicts
3. Competitor (Mural) launched AI whiteboard replay before Phase 1.1 integrations

Success looks like: Product leads reference past decisions during planning, auditors sample trails not raw data, and the CEO says "This finally makes our reasoning debt visible."

1. Reasoning Graph Extraction: Convert free-form dialog into structured nodes (assumptions/evidence/gaps) with 95% entity recognition accuracy
2. Argument Provenance: Link all claims to original session artifacts with cryptographic hashing
3. Summary Fidelity: Generate executive briefs retaining 100% of critical premises (P0 - launch blocking)
4. Anomaly Detection: Flag unresolved logical gaps with 99% precision (P1)

Rejected alternative: LLM-generated synthetic reasoning paths. Rationale: Violates core trust principle - audit trails must reflect actual human cognition.

Sources

• Voice transcripts (PII-redacted)
• Framework-specific annotations (e.g., "Assumption:" tags)
• User-highlighted evidence snippets

Storage

graph LR
    A[Session Raw Data] -->|Immutable Write| B[IPFS CID]
    B --> C[Structured Graph DB]
    C --> D[Audit API]
    D --> E[Client UI]

Critical Constraints

• Data minimization: Store only framework-relevant utterances
• EU Article 17 compliance: Full session deletability within 72hrs

P0 Test Suite (100% pass required)

P1 Continuous Monitoring

• Logical gap detection F1-score: ≥0.92 (measured weekly on 100 sampled decisions)
• Evidence misattribution rate: <0.5% (Pareto root-cause analysis)

Failure Protocol
If premise integrity falls below 100% in staging:

1. Freeze model deployment
2. Revert to rule-based capture
3. Notify PM/legal within 1hr

┌─────────────────────────── Decision Audit Console ───────────────────────────┐
│ [Problem] Reduce customer churn in EMEA region                               │
├──────────────────────────────────────────────────────────────────────────────┤
│ Premise              │ Source             │ Status       │ Verify            │
│──────────────────────┼────────────────────┼──────────────┼───────────────────┤
│ Pricing not primary  │ Session 00:12:45   │ ✅ Verified  │ [Replay Context]  │
│ churn driver         │                    │              │                   │
│──────────────────────┼────────────────────┼──────────────┼───────────────────┤
│ Support latency >72h │ Zendesk export #45 │ ⚠ Unverified │ [Mark Incorrect]  │
├──────────────────────────────────────────────────────────────────────────────┤
│ [GAP] Churn survey data not reconciled with CRM trends                       │
└──────────────────────────────────────────────────────────────────────────────┘

Four-Eyes Principle

• All high-impact decisions (defined by $ value/regulatory scope) require:
  👤 Decision owner validation + 👥 Framework moderator countersignature

1. Temporal Provenance
   • All session elements stored with NTP-synced timestamps (±50ms)
2. Inviolability Controls
   • Write-once storage with Merkle root verification
   • Session edits append new version with diff highlighting
3. Access Governance
   • RBAC with decision-specific permissions (e.g., "view evidence but not assumptions")

Compromise Protocol
If cryptographic chain breaks:

• Auto-lock affected sessions
• Notify security team + impacted users within 15min
• Forensic audit within 24hrs