Technical Specification

How the engine works. Every detail.

This page documents the architecture, pipeline, and capabilities of the big4.cloud analysis engine. For enterprise evaluation, compliance review, or technical due diligence.

Architecture Overview

A fixed pipeline with no shortcuts, no manual intervention. Every stage produces traced output on an append-only blackboard.

TOPIC → [Classifier] → [Research 11 sources] → [Gate Loop 97 nodes] → [Cross-Reference] → [Clustering] → [Meta-Curator] → [Curator] → [Audit 10 stages] → DOSSIER

The engine processes every topic through this fixed pipeline. Each stage produces structured, traced output appended to an immutable blackboard. The blackboard is the single source of truth — all downstream stages read from it, none mutate prior entries.

Cognitive Gate Nodes (97)

Independent analytical modules that fire concurrently, each producing structured output appended to the blackboard.

Node Architecture

  • Each gate is an independent analytical module with its own prompt, schema, and validation
  • Gates fire concurrently — up to 48 in parallel
  • Each gate receives the topic + accumulated blackboard state as input
  • Each gate produces a structured output appended to the blackboard
  • Gates are encrypted at rest (AES-256-GCM)
  • ~70 universal gates (always fire) + ~27 domain-specific (activated by classifier)

Strategic Frameworks 20

Five Forces, MECE, Scenario Planning, Blue Ocean, Ansoff, BCG, Value Chain, PESTEL, SWOT, Diamond Model, Wardley Map, etc.

Risk & Premortem 12

Pre-mortem, Kill Criteria, Sensitivity Analysis, Monte Carlo, Black Swan, Risk Register, Tail Risk, etc.

Financial Analysis 15

Unit Economics, DCF, Comparable Transactions, LBO, Working Capital, Revenue Decomposition, Margin Analysis, etc.

Compliance & Regulatory 10

EU AI Act, GDPR, AML/KYC, Sanctions Screening, Export Control, SOC 2, PCI DSS, etc.

Technical Due Diligence 12

Architecture Review, Scalability Assessment, Security Posture, Tech Debt Scoring, Dependency Audit, etc.

Socratic & Lateral 12

Six Hats, SCAMPER, Provocation, Reversal, Analogy Transfer, Random Entry, Concept Fan, etc.

Validation & Quality 16

Cross-reference, Adversarial Challenge, Confidence Scoring, Reproducibility Check, Source Quality, etc.

Research Engine (11 Sources)

Research is planned before execution — the classifier determines which sources are relevant. Each source has a specialized parser/extractor. All sources queried in parallel.

Source Records Use Case
Prediction Markets (Polymarket) Real-time Calibration against real-money forecasts
Academic Papers 200M+ Scientific evidence, peer-reviewed claims
Patents 100M+ Prior art, IP landscape, freedom-to-operate
SEC Filings Full database Financial disclosures, M&A history
Company Records 200M+ Ownership, financials, corporate structure
Sanctioned Entities 2M+ AML, PEP, OFAC, UN, EU sanctions
DeFi / On-chain Real-time Crypto forensics, protocol analysis
Community Sentiment Reddit, HN Market perception, developer sentiment
Web Extraction Full-page Structured data from any URL
News Feed Real-time Breaking developments, market-moving events
Domain Databases Vertical Industry-specific datasets

Audience Simulation

Generates personas matching your target stakeholders — board members, investors, regulators, customers. They debate the dossier conclusions across multiple rounds.

How It Works

  • Personas generated from context (industry, role, known biases)
  • Multi-round adversarial debate — personas challenge each other
  • Personas can change position based on arguments presented
  • Async execution with polling (credits: 5 per simulation)

Output

  • Who changed their mind and what convinced them
  • Consensus score (0–100)
  • Adoption probability per stakeholder type
  • Ranked objections with severity
  • Recommended rebuttals for each objection

Cross-Reference & Contradiction Detection

Automatic post-loop phase that ensures logical consistency across all gate outputs.

  • Compares all gate outputs pairwise for logical consistency
  • Flags contradictions with severity: minor / major / critical
  • Resolution via 3-model panel (DeepSeek + Gemini + Qwen independently evaluate)
  • Majority vote determines resolution — dissent is recorded
  • Unresolvable contradictions are declared as such — not hidden

Embedding Clustering

Post gate-loop analysis that identifies redundant claims, thematic groups, and information density.

  • All blackboard entries embedded via nomic-embed-text
  • DBSCAN clustering (eps=0.30) — no predefined cluster count
  • Identifies redundant claims and thematic groups
  • Dispersion score per cluster — high dispersion signals disagreement
  • Curator receives cluster structure for better synthesis and deduplication

Reference & Citation System

Every factual claim linked to its source with cryptographic integrity verification.

  • Every factual claim linked to source document
  • SHA-256 hash per source document
  • Merkle root for complete source set — tamper-evident by construction
  • If a source changes post-delivery, hash mismatch is detected automatically
  • Fetch failures logged — attempted but unavailable sources shown for honesty
  • No silent omissions: if a source was queried and failed, you see it

Reproducibility

Consistency tracking across successive runs ensures deeper analysis never contradicts shallower analysis.

  • Delta classification between runs: STABLE, REFINEMENT, ADDITION, CONTRADICTION, DROP
  • Deeper runs must not contradict shallower runs
  • Contradictions between runs are flagged as critical events
  • Full diff available between any two runs on the same topic
  • Confidence scores track directional movement across runs

Security & Isolation

Enterprise-grade security by default. No configuration required.

Execution Tiers

Four tiers for different depth requirements. Auto-tier: the classifier selects appropriate tier if not specified.

Tier Duration Rounds Use Case
Superfast ~3 min 1 Quick screening, triage
Fast ~6 min 2 Standard analysis, most topics
Normal ~12 min 3 Full pipeline with research + audit
Deep ~45 min 5 Maximum depth, enterprise due diligence

Ready to evaluate?

Request a dossier See pricing

For enterprise evaluation, compliance documentation, or DPA — contact enterprise@big4.cloud