Mamut Lab: Investigation Memory Platform

Complex, Long-Running Investigations Require Memory

Resume multi-month investigations with full context. Compliance teams, researchers, and investigators pause/resume work across customs, FDA, legal, M&A, and research domains. Named after the Kikinda mammoth—preserved for 500,000 years, just like your investigation context.

The universal problem: Multi-month investigations lose context when personnel change, teams rotate, or work pauses for months. Current tools store documents but lose the reasoning that makes investigations defensible. 🚀 Q2 2027 Beta: Customs Compliance MVP — First validated application proving investigation memory across USA, China, Germany, and Serbia. GAO report GAO-20-182 documented the crisis: 35,000+ duty drawback claims ($2B+) lacked proper review because CBP cannot manage investigation context at scale.

WHERE WE ARE NOW

October 2025
  • Architecture Complete

    100+ research docs, MVP + future layers documented

  • Technology Stack Validated

    ArangoDB, OpenTelemetry, Go/Python - MVP foundation ready

  • Phase 1: Duty Drawback MVP

    IN PROGRESS - Focused Assessment response workflow with checkpoint-based methodology reconstruction

  • Q4 2026: Internal MVP Demo

    Duty drawback Focused Assessment workflow + knowledge-augmented checkpoints

  • Q2 2027: Private Beta Launch

    Limited beta (10-15 customers) with multi-jurisdiction support - Large importers + customs law firms

Architecture Complete → MVP Implementation: 15%

→ Read our customs compliance focus announcement | → Explore technical documentation

Join the Q2 2027 Beta Waitlist

Be among the first 10-15 customers for multi-jurisdiction support: USA + China + Germany + Serbia. Multi-jurisdiction importers, global customs law firms, multinational brokers. Get 50% early adopter discount for first 12 months and shape the product roadmap.

Join Beta Waitlist →

Opens in new tab • We'll only contact you with beta access updates • No spam

Why Now? Perfect Storm for Investigation Memory

📅

HS2027 Transition: January 1, 2027

Every importer faces ~150 product code changes. WCO Harmonized System transitions from HS2022 to HS2027 create split codes, deleted codes, and renumbered codes requiring manual review.

Risk: Multi-million dollar misclassification penalties during transition period. Manual spreadsheet tracking creates errors across jurisdictions.

Mamut Lab Q2 2027 beta launches just months after HS2027 transition—perfect timing to handle version management automatically.

🚨

Government-Documented Crisis

GAO Report GAO-20-182: U.S. Government Accountability Office found 35,000+ duty drawback claims ($2B+) lacked proper review as of August 2019.

Root Cause: CBP cannot manage investigation context at scale. Personnel rotation breaks investigation continuity. Teams waste 2-4 weeks reconstructing decisions during Focused Assessments.

→ Read GAO-20-182 Report

⚖️

Increasing Compliance Mandates

Focused Assessment 30-day deadlines (19 CFR § 163) require rapid methodology reconstruction from prior submissions.

EAPA investigations (300-360 days) span multiple personnel rotations—context loss guaranteed without investigation memory.

C-TPAT 4-year compliance history requirements demand temporal audit trails across regime changes.

ROI: $500K-$2M annual savings by avoiding consultant fees and misclassification penalties.

The timing is perfect: HS2027 transition creates urgency, GAO report validates the problem, and regulatory mandates are intensifying. The investigation memory crisis is now—not in 5 years.

Q2 2027 beta limited to 10-15 customs compliance customers. Early adopters get HS2027 transition support and 50% discount for first 12 months.

The Investigation Context Loss Crisis Across Domains

Multi-month investigations lose context when personnel change, teams rotate, or work pauses. Current systems store documents but lose the reasoning that makes investigations defensible. This problem spans customs compliance, FDA regulatory, legal investigations, M&A due diligence, and research:

Customs Compliance [Q2 2027 MVP]

Focused Assessments: 30-day deadline to reconstruct methodology from prior submissions (19 CFR § 163). Teams waste 2-4 weeks, $50K-$100K in consultant time. EAPA Investigations: 300-360 days with personnel rotation—context loss when investigators change. GAO Crisis: Report GAO-20-182 documented 35,000+ claims ($2B+) lacked review because CBP cannot manage investigation context at scale.

→ Read GAO-20-182

FDA Regulatory Investigations

Drug Approval Investigations: 2-3 year clinical trial data synthesis spanning preclinical → Phase I → Phase II → Phase III → NDA submission. When regulatory affairs staff rotate, new teams must reconstruct "why we chose endpoint X" or "why we excluded competitor data Y." Regulatory Defense: FDA Form 483 observations require reconstructing months of validation decisions. No system captures clinical hypothesis evolution or regulatory strategy reasoning chains.

Legal Compliance Investigations

False Claims Act Cases: Qui tam investigations span 2.5-3 years. When lead attorney leaves mid-case, successors rebuild understanding from 1000+ documents. Trade Fraud Defense: Multi-year CBP investigations requiring complete audit trail of classification decisions, valuation methodologies, origin determinations. No system captures "why we rejected Alternative X"—reasoning lives in people's heads.

M&A Due Diligence Investigations

Trade Compliance Due Diligence: 6-15 month delays common (McKinsey: 30% of top 50 acquisitions delayed). When deals pause, teams waste 2-4 weeks reconstructing tariff exposure analysis, penalty risk assessment, duty optimization opportunities. Post-Close Integration: 40% talent loss means new teams cannot access reasoning behind compliance recommendations. Stalled deals resurrect requiring full investigation restart.

Why Current Tools Fail Across Domains:

  • Customs Tools (Descartes, SAP GTM): Transaction processing (entry filing, screening), not investigation workflows. No reasoning capture for EAPA determinations, FA methodology, or classification decisions.
  • Regulatory Tools (Veeva, Medidata): Clinical trial data management, not regulatory investigation reasoning. Store trial results, don't capture "why we excluded competitor endpoint" or "why we chose dosing strategy X."
  • Legal Tools (Relativity, Everlaw): Document review and e-discovery, not investigation reasoning chains. Find documents matching keywords, don't preserve "why we rejected settlement offer Y" or "why expert opinion Z supports our case theory."
  • M&A Tools (DealRoom, Intralinks): Document sharing for due diligence, not investigation memory. Store financial models, don't capture "why tariff exposure assessment changed" or "why we downgraded compliance risk from High to Medium."
  • Generic Case Management (Case IQ, ServiceNow): Ticket tracking and document storage, not domain-specific investigation reasoning. No regulatory citation management, no hypothesis evolution tracking, no cross-domain knowledge synthesis.

The Universal Gap: All these tools store documents (the WHAT) but none preserve reasoning chains (the WHY). When investigations resume after months—CBP audits, FDA inspections, litigation discovery, M&A deal resurrections—the reasoning that makes investigations defensible is lost. Mamut Lab captures investigation memory: the decisions, hypotheses, contradictions, dead-ends, and confidence evolution that current tools ignore.

Use Cases: Investigation Memory Platform

Domain-agnostic platform for complex investigations requiring context preservation across months or years. Architecture supports customs compliance (Q2 2027 MVP), FDA regulatory, legal investigations, M&A due diligence, and research intelligence:

Phase 1: Customs Compliance [Q2 2027 MVP]

First validated application proving investigation memory capabilities. Beta launch Q2 2027 with multi-jurisdiction support (USA, China, Germany, Serbia).

1
🔨 PHASE 1

Duty Drawback Focused Assessments

Problem: Focused Assessment (FA) requests require 30-day response to reconstruct methodology from prior submissions (CBP's 5-year recordkeeping requirement per 19 CFR § 163). Teams waste 2-4 weeks, estimated $50K-100K in consultant time.

Platform Solution: Knowledge-Augmented Checkpoints capture methodology reasoning at claim preparation. One-click context restoration when FA arrives.

Innovations Applied:

  • Knowledge-Augmented Checkpoints: Methodology reconstruction in 2-3 days vs 2-4 weeks
  • Temporal Knowledge Substrate: "What did we know when?" for audit defense
  • Event Sourcing: Complete provenance for CBP review

ROI: Estimated $500K-$2M annual savings for large importers (10-20 FAs/year)

2
📋 PHASE 2

EAPA Investigation Management

Problem: 300-360 day investigations with personnel rotation. When investigators change, teams waste weeks reconstructing origin determination reasoning.

Platform Solution: Investigation timeline tracking with complete reasoning provenance. Team collaboration with shared context.

Innovations Applied:

  • Temporal Knowledge Substrate: Investigation timeline reconstruction
  • Knowledge-Augmented Checkpoints: Supplier verification reasoning preservation
  • Coordinated Space: Cross-functional team coordination (legal, operations, finance)

ROI: 60-80% reduction in investigation context reconstruction time

3
📋 Q2 2027 BETA

Multi-Jurisdiction Investigation Linking

Problem: 75% of Fortune 500 importers operate multi-jurisdiction supply chains (USA + China + Germany). Current systems cannot track investigations across multiple customs authorities (CBP, GACC, Zoll).

Platform Solution: WCO Framework Foundation enables investigation tracking across 182 member countries. Universal HS 6-digit codes provide shared classification base with jurisdiction-specific extensions.

Innovations Applied:

  • Multi-Jurisdiction Extension (ADR-010): Track investigations across USA, China, Germany, Serbia simultaneously
  • TKS Integration: HS code version mapping (HS2022 → HS2027) with automatic correlation tables
  • WCO Framework: 5,387 harmonized codes shared globally, jurisdiction extensions (HTS 10-digit, CN 8-digit, TARIC 10-digit)

ROI: 3x TAM expansion ($1.3B USA-only → $2.3B-$3.6B international). Unlock 75% of Fortune 500 multi-jurisdiction importers.

Phase 2: Adjacent Regulated Investigations [Post-MVP Expansion]

Same platform capabilities applied to FDA regulatory, legal compliance, and M&A due diligence. Domain configuration, not rearchitecture.

4
PHASE 2

FDA Regulatory Investigations

Problem: Drug approval investigations span 2-3 years (preclinical → Phase I/II/III → NDA submission). Regulatory affairs staff rotation causes context loss. FDA Form 483 observations require reconstructing months of validation decisions.

Platform Solution: Clinical trial decision provenance, regulatory strategy preservation across phases, hypothesis evolution tracking, FDA guidance synthesis with patent landscape analysis.

Innovations Applied:

  • Knowledge-Augmented Checkpoints: Clinical endpoint selection reasoning, competitor data exclusion rationale
  • Temporal Knowledge Substrate: Trial decision timeline, regulatory submission version control
  • Multi-Domain Synthesis: Clinical data + FDA guidance + patent landscape + market analysis

Market: $8B+ FDA regulatory compliance market. Same investigation memory drivers as customs (audit trails, reasoning provenance, multi-year history).

5
PHASE 2

Legal Compliance & Trade Fraud Defense

Problem: False Claims Act qui tam cases span 2.5-3 years. 25% attorney turnover annually ($9.1B cost to top 400 law firms). When lead attorney leaves mid-case, successors rebuild understanding from 1000+ documents.

Platform Solution: Legal reasoning preservation, case theory evolution tracking, rejected arguments documentation, settlement recommendation provenance, discovery document timeline.

Innovations Applied:

  • Knowledge-Augmented Checkpoints: Legal strategy reasoning, rejected settlement rationale, expert opinion integration
  • Temporal Knowledge Substrate: Litigation timeline, case law evolution, precedent analysis "what existed when we filed motion X"
  • Multi-Domain Synthesis: Case law + statutes + expert opinions + regulatory guidance

Market: $26-32B legal tech market + $12B+ regulatory compliance investigation market.

6
PHASE 2

M&A Trade Compliance Due Diligence

Problem: Deal delays of 6-15 months common (McKinsey: 30% of top 50 acquisitions delayed). 40% talent loss post-close. When deals pause then resurrect, teams waste 2-4 weeks reconstructing tariff exposure analysis, penalty risk assessment, duty optimization.

Platform Solution: Due diligence decision history, valuation assumption tracking, compliance risk assessment reasoning, regulatory penalty estimation provenance. Investigation resurrects with full context when deals restart.

Innovations Applied:

  • Knowledge-Augmented Checkpoints: Valuation model reasoning, compliance risk downgrade rationale, integration plan assumptions
  • Temporal Knowledge Substrate: Due diligence timeline, financial model version history, risk assessment evolution
  • Multi-Domain Synthesis: Financial models + regulatory risk + market analysis + compliance audits

Market: $5B+ M&A due diligence software market. Many customs compliance customers also have M&A teams → cross-sell opportunity.

Phase 3: Research Intelligence [Architecture Complete, Implementation TBD]

Long-term expansion to PhD research, industrial R&D, and technical due diligence. Architecture documented (ADR-004, ADR-009 primary examples). Implementation depends on Phase 1-2 validation.

7
PHASE 3

PhD Research & Academic Investigations

Problem: Multi-year research investigations requiring hypothesis evolution tracking, dead-end documentation, cross-domain synthesis (academic + market + technical + regulatory). ADR-009 primary example: Drug target investigation spanning 6+ months.

Platform Solution: Research hypothesis tracking, literature review synthesis, experiment decision provenance, contradiction management across papers/patents/reports.

Market: $3B+ research-intensive knowledge worker segment. Post-MVP expansion after customs/FDA validation.

8
PHASE 3

Industrial R&D & Technical Due Diligence

Problem: VCs conducting technical due diligence, industrial R&D scientists evaluating technology strategies, technical strategists assessing competitive landscape—all require multi-month investigations with temporal continuity.

Platform Solution: Technical evaluation reasoning, patent landscape analysis, competitor technology assessment, regulatory risk synthesis across academic papers + patents + market intelligence.

Market: VC technical due diligence, corporate R&D strategy, technology assessment consulting.

Platform Strategy: Disciplined 3-phase expansion. Phase 1 (2027): Prove investigation memory with customs compliance MVP (10-15 customers, $1.5M ARR). Phase 2 (2028): Expand to adjacent regulated investigations—FDA, legal, M&A (same platform capabilities, domain configuration). Phase 3 (2029+): Research intelligence for PhD research, industrial R&D, technical due diligence.

Why This Works: 6 core innovations (Knowledge-Augmented Checkpoints, Temporal Knowledge Substrate, Neurosymbolic Verification, Multi-Domain Synthesis, Darwin-Godel, Graduated Autonomy) are domain-agnostic by design. Architecture documented in ADR-004, ADR-005, ADR-007, ADR-009. Customs proves platform viability; FDA/legal/M&A validate universality; research demonstrates long-term vision.

MVP Philosophy: Ship proven customs compliance workflows (Duty Drawback Focused Assessment response + checkpoint-based investigation memory + multi-jurisdiction tracking) in Q2 2027 beta. Validate with 10-15 customers. Then expand to additional workflows (EAPA management, C-TPAT monitoring) and adjacent regulated industries based on customer feedback and beta traction.

Explore complete architecture (MVP + research tracks) →

Why Mamut Lab vs. Alternatives

Enterprise customs compliance teams have four options for managing multi-year investigations. Here's how investigation memory compares to existing approaches:

Capability Mamut Lab Descartes/SAP Consultants Build In-House
Investigation Memory ✓ Multi-year context ✗ Transaction-focused Limited (per engagement) Requires 3-5 years dev
Multi-Jurisdiction Support ✓ 4 countries (Q2 2027) Limited (USA focus) Per engagement Must build from scratch
Focused Assessment Response ✓ 2-3 days Manual (2-4 weeks) $50K-$100K per FA After 3-5 years dev
HS2027 Transition Support ✓ Automatic (Jan 1, 2027) Manual spreadsheet tracking Project-based ($100K+) Custom development needed
Annual Cost $400K-$2M $200K-$500K (est.) $500K-$2M $2M-$5M dev + ongoing
Knowledge Retention ✓ Permanent reasoning capture ✗ Document storage only Lost after engagement Only if documented
ROI Timeline Year 1 (beta) Varies Per project 3-5 years

vs. Descartes/SAP GTM

Their focus: Transaction processing—clearing shipments, calculating duties, filing entries.

Our differentiation: Investigation memory—capturing WHY decisions were made, not just WHAT transactions occurred.

Many customers use both: Descartes for daily transactions + Mamut Lab for investigation memory.

vs. Consultant Model

Their model: $50K-$100K per Focused Assessment. $500K-$2M annual retainer for large importers.

Our advantage: $400K/year platform < $500K-$2M consultant fees, with permanent knowledge retention.

Consultants leave, platform persists. Investigation memory survives personnel changes.

vs. Build In-House

Their reality: 3-5 year development timeline, $2M-$5M cost, ongoing maintenance burden.

Our timing: Q2 2027 beta access at $200K/year (50% beta discount). Start realizing ROI in Year 1.

Opportunity cost: Distract engineering team from core business vs. deploy proven platform.

The Clear Choice for Investigation Memory

Transaction processing tools store WHAT happened. Mamut Lab captures WHY decisions were made—the reasoning that makes investigations defensible.

When CBP asks "How did you determine this classification 18 months ago?" only Mamut Lab can answer in 2-3 days instead of 2-4 weeks.

Platform Innovations: Investigation Memory Across Domains

Domain-agnostic investigation memory platform with six core innovations. Architecture supports customs compliance (Q2 2027 MVP), FDA regulatory, legal investigations, M&A due diligence, and research:

Innovation 1: Knowledge-Augmented Checkpoints

Platform Capability: Capture WHY decisions were made, not just WHAT was executed. First system to preserve reasoning provenance alongside execution state. Enables pause/resume investigations after months with full context restoration.

  • Reasoning Provenance: Decision logic, alternatives considered, rejection reasons, confidence evolution
  • Evidence Linkage: Complete chain from sources to conclusions with domain-specific citations
  • Context Restoration Goal: 15-30 min vs 20-40 hours (50-100x improvement - research target)

Multi-Domain Applications:

  • Customs [Q2 2027 MVP]: Focused Assessment methodology reconstruction (2-3 days vs 2-4 weeks, $50K-100K savings). EAPA investigation handoff when personnel rotate.
  • FDA Regulatory: Clinical trial decision provenance, regulatory strategy preservation across Phase I → II → III, NDA submission reasoning chains.
  • Legal: Case theory evolution, rejected settlement rationale, expert opinion integration across multi-year litigation.
  • M&A: Due diligence decision history, valuation assumption tracking, compliance risk assessment reasoning when deals resume after months.

→ Read technical deep dive | → Architecture docs (ADR-009)

Innovation 2: Temporal Knowledge Substrate

Platform Capability: Versioned knowledge graphs with Git-like time-travel queries. Complete investigation timeline reconstruction at any past point. Bi-temporal model (business time + system time) enables "what did we know when?" queries across multi-year investigations.

  • Time-Travel Queries: Instant state restoration to any past investigation point
  • Branching Exploration: Test alternative hypotheses without affecting main investigation
  • Contradiction Management: Track conflicting evidence, resolution reasoning, confidence evolution
  • Provenance Tracking: Full lineage from sources to conclusions with decision timestamps

Multi-Domain Applications:

  • Customs [Q2 2027 MVP]: 4-year C-TPAT compliance history, EAPA timeline reconstruction, audit defense ("what did we know when CBP review started?"). HS code temporal versioning (HS2022 → HS2027) handled automatically.
  • FDA Regulatory: Clinical trial decision timeline, regulatory submission version control, audit trail for FDA Form 483 responses showing decision evolution across years.
  • Legal: Litigation timeline tracking, discovery document provenance, case law evolution showing "what precedents existed when we filed motion X."
  • M&A: Due diligence timeline, valuation model version history, compliance assessment evolution showing decision basis when deals resurrect months later.

→ Temporal Knowledge Substrate architecture (ADR-005)

Innovation 3: Neurosymbolic Verification

Platform Capability: Layered verification pyramid combining neural pattern recognition (multi-model AI) with symbolic reasoning (domain rules) and formal proof capability. Ensures critical decisions are explainable, auditable, and verifiable against regulatory requirements.

  • Multi-Model Consensus: 2-3 specialized models analyze independently, flag disagreements for human review
  • Symbolic Rule Validation: Verify decisions against domain-specific regulatory rules and constraints
  • Explainable Audit Trails: Complete reasoning chain from input to conclusion with confidence scores
  • Formal Proof Generation: Mathematical guarantees for critical decisions (research track, post-MVP)

Multi-Domain Applications:

  • Customs [Q2 2027 MVP]: HTS classification verification against General Rules of Interpretation (GRI 1-6). Prevent multi-million dollar penalties (Morrison: £4.7M, OtterBox: $4.3M). Multi-model consensus flags classification disagreements.
  • FDA Regulatory: Clinical endpoint validation against FDA guidance, efficacy claim verification, adverse event classification against MedDRA hierarchy.
  • Legal: Case law precedent matching, legal argument validation against statutory requirements, settlement recommendation verification.
  • M&A: Compliance risk scoring validation, tariff exposure calculation verification, regulatory penalty estimation against historical enforcement data.

→ Neurosymbolic verification architecture (ADR-007)

Innovation 4: Darwin-Godel Self-Improvement [POST-MVP RESEARCH]

Platform Capability: Pattern learning from past investigations to optimize future workflows. System discovers better investigation strategies automatically by analyzing what worked across thousands of cases.

  • Pattern Discovery: Identify recurring investigation triggers, red flags, and resolution patterns
  • Investigation Intelligence: Recommend checkpoint timing, evidence collection strategies, workflow optimizations
  • Cross-Domain Learning: Patterns from one domain (customs) inform strategies in others (M&A, FDA)

Multi-Domain Applications:

  • Customs: Reduce Focused Assessment triggers through proactive best practices. Identify classification dispute patterns → preventive guidance.
  • FDA: Clinical trial design optimization based on approval patterns. Regulatory strategy recommendations from submission history analysis.
  • Legal: Settlement recommendation timing based on similar case outcomes. Discovery strategy optimization from past litigation patterns.
  • M&A: Due diligence workflow optimization based on deal closure patterns. Risk flag prioritization from historical assessment accuracy.

Status: Post-MVP research track. Conceptual design complete, requires MVP deployment + significant data corpus. No delivery timeline.

Innovation 5: Graduated Autonomy Framework

Platform Capability: 5-level autonomy framework with mandatory skill preservation protocols. Prevents automation complacency (70% trust paradox from aviation research) while enabling AI efficiency gains.

  • L0-Manual: AI disabled for skill maintenance (monthly manual practice required)
  • L1-Augmented: AI suggests, human executes all actions
  • L2-Collaborative: AI analyzes, human validates conclusions
  • L3-Delegated: AI executes workflows, human verifies reasoning chains
  • L4-Autonomous: AI handles routine tasks, flags anomalies for human review

Multi-Domain Applications:

  • Customs [Q2 2027 MVP]: L3 for Focused Assessment draft responses, L1 for origin determination (always human-led). Monthly L0 practice prevents skill degradation.
  • FDA Regulatory: L3 for regulatory submission document assembly, L1 for clinical endpoint selection. Critical regulatory strategy decisions always human-led.
  • Legal: L3 for discovery document review, L1 for case theory development. Settlement decisions always require human judgment (L0-L1 only).
  • M&A: L3 for financial model validation, L1 for compliance risk assessment. Deal valuation and go/no-go decisions always human-controlled.

Research Foundation: Aviation human factors research on automation-induced skill degradation applied to high-stakes AI systems.

Innovation 6: Multi-Domain Synthesis Engine

Platform Capability: Integrate knowledge across 7 domain types: academic literature, market intelligence, technical specifications, regulatory guidance, scientific research, financial data, and legal precedents. Enables cross-domain reasoning that current tools cannot provide.

  • 7 Domain Types: Academic (PubMed, arXiv), Market (industry reports), Technical (patents, GitHub), Regulatory (FDA, CBP), Scientific (clinical trials), Financial (10-K, market data), Legal (case law, statutes)
  • Cross-Domain Reasoning: Synthesize insights across domains (e.g., academic research + regulatory guidance + market analysis → strategic recommendation)
  • Contradiction Detection: Flag conflicts across sources (FDA guidance vs. published research, regulatory precedent vs. current practice)
  • Provenance Chains: Complete lineage showing which domains informed which conclusions

Multi-Domain Applications:

  • Customs [Q2 2027 MVP]: Synthesize CBP rulings + WCO guidance + HTS rules + jurisdiction regulations. Example: Classification decision requires technical specs + regulatory precedents + competitor filings.
  • FDA Regulatory: Integrate clinical trials + FDA guidance + patent landscape + competitive market analysis → regulatory strategy. Example: Drug approval requires efficacy data + regulatory precedent + market positioning.
  • Legal: Synthesize case law + statutes + expert opinions + regulatory guidance → legal strategy. Example: Trade fraud defense requires customs regulations + precedent analysis + industry practice.
  • M&A: Integrate financial models + regulatory risk + market analysis + compliance audits → valuation. Example: Tariff exposure requires trade data + regulatory history + supply chain analysis.

→ Multi-Domain Synthesis Engine architecture (ADR-007)

Platform Synthesis: No competitor offers this domain-agnostic investigation memory platform. **Knowledge-Augmented Checkpoints** preserve reasoning provenance. **Temporal Knowledge Substrate** enables time-travel queries across years. **Neurosymbolic Verification** ensures regulatory correctness. **Multi-Domain Synthesis Engine** integrates knowledge across 7 domain types. **Darwin-Godel Self-Improvement** learns from past investigations. **Graduated Autonomy** prevents automation complacency. This architecture works across customs compliance (Q2 2027 MVP), FDA regulatory, legal investigations, M&A due diligence, and research—without major rearchitecture.

Research Value: 100+ architecture documents (1000+ hours) documenting these innovations. This intellectual depth creates the competitive moat. → Browse all research docs on GitHub

Multi-Jurisdiction Architecture (ADR-010)

Investigation Model Extension: Optional jurisdiction_context field enables multi-jurisdiction tracking without breaking changes. Extends existing Investigation abstraction instead of creating separate Jurisdiction service.

Key Design Decision Benefits:

  • Backward Compatible: 847 existing USA investigations continue working. Zero-downtime deployment for Q2 2027 international launch
  • Saves ~2,000 LOC: Extend existing Investigation model (70% existing architecture) vs. creating separate Jurisdiction service (70% new code)
  • Cross-Jurisdiction Analytics: Single schema enables queries across all jurisdictions. "How many investigations involve HS code 847130 globally?"
  • TKS Perfect Fit: HS code temporal versioning (HS2022 → HS2027) maps directly to TKS valid_from/valid_until semantics

Implementation: 4-phase rollout over 9 months (USA baseline → China → Germany → Serbia). → Read ADR-010 full specification | → Technical implementation guide

Built on Coordinated Space Architecture

Breakthrough innovations require robust infrastructure. 6-component execution substrate using proven technologies:

Context Management

Shared investigation state with temporal versioning. Event sourcing + CQRS patterns provide complete audit trails.

Customs: EAPA state tracking, Focused Assessment queue monitoring

Model Registry

Tool capability catalog with intelligent routing. Matches tasks to specialized models with fallback chains.

Customs: Document analysis orchestration

Task Executor

Durable workflows with checkpoint-based resumability. Temporal.io-inspired patterns for long-running investigations.

Customs: 30-day Focused Assessment workflow, 300-360 day EAPA management

Protocol Adapters

Universal integration layer. REST, gRPC, MQTT, Kafka support.

Customs: CBP ACE integration, enterprise DMS

Observability

OpenTelemetry-based monitoring with distributed tracing. Investigation progress dashboards.

Customs: Focused Assessment deadline tracking, compliance metrics

Knowledge Persistence

ArangoDB multi-model storage (graphs + documents + vectors). Foundation for checkpoints and temporal knowledge.

Customs: Audit trails, 4-year compliance history

Technology Stack: Go (orchestrator), Python (cognition & LLMs), Rust (performance services), ArangoDB (persistence), NATS JetStream (messaging), OpenTelemetry (observability)

Design Philosophy: Use proven, production-ready technologies. Innovate at synthesis layer (how they're combined), not infrastructure layer. → Full implementation plan

HS2027 Transition Support (Critical for Q1 2027)

Time-Sensitive Value Driver: WCO HS codes change every 5 years. On January 1, 2027, HS2022 transitions to HS2027 affecting ~150 product codes per average importer.

Code Change Types:

  • Split codes: 9405.40 → 9405.41 (LED lamps) + 9405.42 (other lamps) - requires manual product review and selection
  • Deleted codes: 4202.11 → null (URGENT reclassification required, no automatic mapping)
  • Renumbered codes: 8542.31 → 8542.33 (automatic mapping with TKS correlation tables)
  • Merged codes: 8517.61 + 8517.62 → 8517.61 (automatic consolidation)

Platform Solution: TKS generates HS2027 transition reports 6 months in advance (June 2026). Flags products requiring manual review. Prevents multi-million dollar misclassification penalties during version transition. Automatic cutover on Jan 1, 2027 - investigations created before use HS2022, investigations after use HS2027.

Competitive Advantage: No competitor handles HS version transitions automatically. Manual spreadsheet tracking creates errors. TKS bi-temporal versioning maps directly to WCO 5-year revision cycles—saves ~2,000 LOC of custom temporal logic.

Ongoing Research

Formal Proof Integration (Exploratory): Investigating Lean4/Coq integration for mathematical guarantees in neurosymbolic verification. This is bleeding-edge research with no implementation timeline—orchestrating formal provers with LLMs is an unsolved problem. → Research notes

Research Philosophy: Core innovations (5 breakthroughs above) are architecturally sound and implementable with current technology. Formal proof integration is exploratory research that may enhance neurosymbolic verification if breakthroughs occur.

Who We're Designing For

Professionals conducting complex, long-running investigations across regulated domains. Platform supports customs compliance (Q2 2027 MVP), FDA regulatory, legal, M&A, and research:

Customs Compliance Professionals [Q2 2027 MVP]

Target Segments: Large multi-jurisdiction importers (Fortune 500 manufacturers), customs law firms (EAPA/duty drawback representation), customs brokerage firms (multi-client compliance), in-house trade compliance teams, customs consultants.

Use Cases: EAPA investigations (300-360 days), Focused Assessment responses (30-day deadlines), C-TPAT monitoring (4-year cycles), multi-jurisdiction tracking (USA, China, Germany, Serbia), classification reviews.

Why Investigation Memory: Personnel rotation breaks context. CBP audits require reconstructing decisions made months prior. GAO-20-182: 35,000+ claims ($2B+) lacked review due to investigation context loss.

FDA Regulatory Affairs Teams

Target Segments: Pharmaceutical/biotech regulatory affairs, clinical operations teams, regulatory consultants, CMC (Chemistry, Manufacturing, Controls) specialists, regulatory strategists.

Use Cases: Drug approval investigations (2-3 years: preclinical → Phase I/II/III → NDA), FDA Form 483 response preparation, regulatory strategy preservation across trial phases, clinical trial decision provenance.

Why Investigation Memory: Regulatory staff rotation common. FDA inspections require explaining "why we chose endpoint X" or "why we excluded competitor data Y" from decisions made months ago. Investigation context loss delays submissions.

Legal Compliance & Trade Fraud Teams

Target Segments: International trade law firms, qui tam/False Claims Act practitioners, regulatory defense attorneys, in-house legal compliance, trade fraud investigators.

Use Cases: Trade fraud litigation (2.5-3 years), False Claims Act qui tam cases, CBP regulatory defense, customs penalty mitigation, classification dispute litigation, settlement strategy.

Why Investigation Memory: 25% attorney turnover annually ($9.1B cost to top 400 firms). When lead attorney leaves mid-case, successors rebuild from 1000+ documents. No system captures "why we rejected settlement offer Y."

M&A Due Diligence Specialists

Target Segments: M&A teams at Fortune 500 importers, private equity trade compliance due diligence, investment banks (regulatory risk assessment), trade compliance consultants, integration planning teams.

Use Cases: Trade compliance due diligence, tariff exposure analysis, penalty risk assessment, duty optimization opportunities, post-close integration planning, compliance system audits.

Why Investigation Memory: 6-15 month deal delays common (McKinsey: 30% of top 50 acquisitions delayed). 40% talent loss post-close. Deals pause then resurrect requiring complete due diligence reconstruction.

Research Professionals

Target Segments: PhD researchers (multi-year investigations), industrial R&D scientists, technical due diligence specialists (VCs), competitive intelligence analysts, technology strategists.

Use Cases: Academic research investigations (6+ months), industrial R&D evaluations, patent landscape analysis, technical due diligence for acquisitions, competitive technology assessment.

Why Investigation Memory: Research spans months to years with hypothesis evolution, dead-end documentation, cross-domain synthesis (academic + market + technical + regulatory). ADR-009 primary example: Drug target investigation.

Policy & Regulatory Analysts

Target Segments: Trade associations (regulatory advocacy), government policy analysts, regulatory consultants, compliance advisory firms, think tanks (trade policy research).

Use Cases: Regulatory comment preparation, policy position tracking, regulatory impact assessments, cross-jurisdiction regulatory analysis, precedent tracking for advocacy.

Why Investigation Memory: Multi-year advocacy campaigns require tracking policy evolution, regulatory comments, stakeholder positions. Personnel changes break institutional knowledge of "why we recommended approach X."

Market Opportunity: Investigation Memory Platform

Multi-domain platform addressing investigation context loss crisis across customs compliance, FDA regulatory, legal, M&A, and research. No competitor offers domain-agnostic investigation memory infrastructure.

Platform Total Addressable Market

$30B+

Investigation Memory Across Domains:

  • Customs Compliance (Beachhead): $2.3B-$3.6B TAM (multi-jurisdiction via WCO framework, 182 countries)
  • FDA Regulatory Compliance: $8B+ TAM (pharmaceutical/biotech regulatory operations, clinical trial management)
  • Legal Compliance & Litigation: $12B+ TAM (e-discovery, regulatory defense, compliance monitoring, $26-32B legal tech)
  • M&A Due Diligence: $5B+ TAM (trade compliance, regulatory risk, integration planning)
  • Research Intelligence: $3B+ TAM (PhD research, industrial R&D, technical due diligence)

Platform TAM reflects investigation memory need across regulated domains. No unified solution exists today.

Phased Market Entry Strategy

3-Phase Expansion

Phase 1 (2027): Customs Compliance MVP

  • $2.3B-$3.6B TAM (multi-jurisdiction support via WCO framework)
  • 10-15 beta customers: multi-jurisdiction importers, customs law firms, brokers
  • Target: $1.5M ARR (Year 1), prove investigation memory platform viability

Phase 2 (2028): Adjacent Regulated Investigations

  • Add FDA regulatory ($8B TAM), legal compliance ($12B TAM), M&A due diligence ($5B TAM)
  • Same platform capabilities, domain configuration not rearchitecture
  • Target: $10M ARR (Year 2), validate multi-domain platform thesis

Phase 3 (2029+): Research Intelligence

  • Add PhD research, industrial R&D, technical due diligence ($3B TAM)
  • Target: $35M ARR (Year 3), demonstrate platform universality

Year 3 Revenue Trajectory

$35M ARR

Multi-Domain Revenue Mix (2029):

  • Customs Compliance: $20M ARR (30 enterprise customers @ avg $667K/year)
  • FDA Regulatory: $8M ARR (10 pharma/biotech teams @ avg $800K/year)
  • Legal & M&A: $5M ARR (8 law firms + 5 M&A teams @ avg $385K/year)
  • Research Intelligence: $2M ARR (early adopters, land-and-expand)

Enterprise Pricing Tiers:

  • Customs multi-jurisdiction: $400K-$2M/year (jurisdiction count, investigation volume)
  • FDA regulatory: $500K-$2M/year (clinical trial pipeline, submission volume)
  • Legal/M&A: $300K-$1M/year (case load, due diligence frequency)

Conservative penetration: <1% of $30B+ platform TAM. Disciplined expansion, validated by customer traction.

Competitive Advantage & Moat

12-18 Month Lead

Technical Moat (Platform Architecture):

  • Domain-Agnostic Design: 6 core innovations (checkpoints, temporal substrate, neurosymbolic, multi-domain synthesis, self-improvement, graduated autonomy) work across all domains
  • Architecture Depth: 100+ research documents (1000+ hours), ADR-004/005/007/009 prove multi-domain viability
  • Extension Mechanism: New domains via configuration, not rearchitecture (Domain Context Provider pattern)

Go-to-Market Moat (Customs First):

  • Founder domain expertise → 12-18 month customs market lead
  • WCO framework integration → 182 countries, 98% of global trade (no competitor offers)
  • GAO-20-182 documented crisis → validated problem, clear ROI ($500K-$2M annual savings)

Customs proves platform works; FDA/legal/M&A validate universality. Platform advantages prevent "customs-only" limitation.

Why Now? Four Powerful Market Drivers

Regulatory Compliance

EU AI Act (2024), US state AI transparency laws, financial services audit requirements. Explainable AI shifting from nice-to-have to mandatory. $16.2B explainable AI market by 2028.

AI Cost Optimization

Organizations spending $50K-500K/month on LLM APIs. Multi-model orchestration with smart routing can reduce costs 40-60% while improving accuracy through specialized model selection.

Code Quality Crisis

"Code slop" from AI assistants creating technical debt. Organizations need verified, explainable automation—not fast but fragile output. Neurosymbolic approach provides formal guarantees.

Research Continuity Gap

3.2M researchers globally using tools like Elicit, Consensus, Semantic Scholar—but these are session-based. Multi-year investigations need temporal knowledge persistence. No existing solution.

Go-To-Market Strategy

Focused beachhead approach starting with customs compliance (Q2 2027), expanding to adjacent regulated industries. Compliance creates urgency, premium pricing, and clear ROI.

Phase 1: Customs Compliance (Q2 2027 Launch)

Multi-Jurisdiction Importers

Primary Target

Large importers ($500M+ annual imports) operating across USA-China-Germany-Serbia trade lanes. German automotive (BMW, Volkswagen), electronics manufacturers (Samsung), pharmaceutical companies. Cannot track investigations across multiple customs authorities (CBP, GACC, Zoll, Serbia Customs).

Why This Beachhead?

  • Government-documented crisis: GAO-20-182 found 35,000+ claims ($2B+) lacked review as of August 2019
  • High willingness to pay ($400K/year Multi-Jurisdiction tier, $800K-$2M Global Enterprise tier)
  • Clear ROI: $500K-$2M annual savings (10-20 Focused Assessments/year at $50K-100K each)
  • 75% of Fortune 500 importers operate multi-jurisdiction supply chains - USA-only tools unusable

Acquisition Strategy

  • WCO conferences & customs trade shows (WCO Global Summit, NCBFAA Annual Conference)
  • Direct outreach to Fortune 500 trade compliance VPs (automotive, electronics, pharma)
  • Partnerships with Big 4 customs advisory (Deloitte Trade, EY Customs, KPMG Trade & Customs)
  • Pilot programs with 2-3 multi-jurisdiction importers (Q2 2027 beta)

Success Metrics

Year 1 (Q2-Q4 2027): 2-3 multi-jurisdiction pilots, $1.5M ARR. Year 2 (2028): 10 enterprise importers, $8.75M ARR. Year 3 (2029): 30 enterprise customers, $34M ARR.

Global Customs Law Firms

Secondary Target

Global law firms with customs practices: Sandler Travis (9 global offices), Baker McKenzie, Hogan Lovells. Handle clients operating across USA-China-Germany-Serbia. USA EAPA experience doesn't transfer to EU anti-dumping or China origin audits. No unified case management across jurisdictions.

Pain Points

  • Regional systems don't integrate (CBP ACE, GACC ECIQ, EU ICS2)
  • Language barriers for China customs correspondence (GACC documents in Chinese)
  • Client expects seamless global service, but investigation workflows are jurisdiction-specific
  • Attorney turnover creates context loss across multi-year EAPA/AD/CVD cases

Platform Value & Pricing

Single investigation management platform across CBP + GACC + Zoll + Serbia Customs. Translate workflows (EAPA → 原产地核查 → Ursprungsüberprüfung). Multi-language UI (English, Chinese, German, Serbian).

Pricing: $500K-$2M annual subscription (matter-based or firm-wide license). Target: 15+ global law firms by Year 2.

Entry Timing

Year 1-2 (2027-2028). Co-launch with multi-jurisdiction importers. Law firms represent importers—solving both pain points simultaneously creates network effects.

Multinational Customs Brokers

Tertiary Target

Multinational brokers: C.H. Robinson, Kuehne+Nagel, DHL Global Forwarding. Operate in 10+ countries but regional systems don't integrate. Cannot offer unified investigation management to global clients (e.g., automotive OEM importing across USA-China-Germany-Serbia).

Platform Value & Pricing

API integration for broker TMS/WMS systems. White-label options for client-facing portals. Multi-tenant architecture (broker manages multiple importer clients).

Pricing: $1M-$3M enterprise licensing (based on client count + investigation volume). Target: 5 multinational brokers by Year 3.

Entry Timing

Year 2-3 (2028-2029). After establishing credibility with importers and law firms. Brokers require white-label/API capabilities (Phase 3 features).

Why We'll Win: 5 Competitive Moats

Domain Expertise

100+ research documents on neurosymbolic AI, continual learning, temporal knowledge—represents 1000+ hours creating knowledge moat competitors can't replicate overnight.

Unique Architecture

Only platform purpose-built for multi-year customs compliance investigations with knowledge-augmented checkpoints + temporal investigation substrate + neurosymbolic verification. Architecture combining investigation memory, regulatory knowledge integration, and audit-ready provenance is novel IP for trade compliance.

Network Effects

Temporal Knowledge Substrate creates data moat: as users build investigations over months/years, switching costs become prohibitive. Knowledge graphs compound in value.

Regulatory Timing

First-mover advantage in explainable agentic AI as EU AI Act enforcement begins. Compliance becomes mandatory 2025-2026—perfect timing for our beta launch.

Focus & Speed

While Microsoft/Google have 100 priorities, we're 100% dedicated to this problem. Solo → small team agility allows faster iteration than enterprise vendors.

About Mamut Lab: Platform Vision with Customs-First Strategy

Investigation memory platform for complex, long-running tasks across domains. Starting with customs compliance (Q2 2027 MVP) to prove platform viability, then expanding to FDA regulatory, legal, M&A, and research intelligence.

Platform Vision

Core Thesis: Investigation memory is a universal problem across regulated domains. Complex investigations lose context when personnel change, teams rotate, or work pauses for months. Current tools store documents but lose the **reasoning** that makes investigations defensible.

Architecture Foundation:

  • Domain-Agnostic Design: 6 core innovations (Knowledge-Augmented Checkpoints, Temporal Knowledge Substrate, Neurosymbolic Verification, Multi-Domain Synthesis, Darwin-Godel, Graduated Autonomy) work across all domains
  • 100+ Architecture Documents: ADR-004, ADR-005, ADR-007, ADR-009 prove multi-domain viability. Primary example in ADR-009 is **PhD drug research**, not customs.
  • Extension Mechanism: Domain Context Provider pattern enables new domains via configuration, not rearchitecture

GitHub Repository: Browse Platform Architecture →

Why Customs First?

Strategic Beachhead Market:

  • Founder Domain Expertise: 12 years C#/.NET enterprise software (regulatory compliance, audit trail architecture) + deep customs compliance research (100+ docs analyzing CBP workflows, WCO framework)
  • Validated Problem: GAO-20-182 documented 35,000+ claims ($2B+) lacked review due to investigation context loss. Government-documented crisis.
  • Clear ROI: $500K-$2M annual savings for large importers (10-20 Focused Assessments/year @ $50K-$100K each)
  • Multi-Jurisdiction Advantage: WCO framework (182 countries, 98% of global trade) enables 3x TAM expansion ($1.3B USA → $2.3B-$3.6B international). No competitor offers this.

Platform Advantage: Customs proves investigation memory works. Then expand to adjacent regulated investigations (FDA, legal, M&A) using same platform—domain configuration, not rearchitecture.

Disciplined Execution Timeline

3-Phase Platform Expansion:

  • Q4 2026: Internal MVP Demo (Duty Drawback Focused Assessment workflow)
  • Q2 2027: Customs Beta Launch (10-15 customers: multi-jurisdiction importers, law firms, brokers)
  • 2028 (Phase 2): FDA regulatory + legal compliance + M&A due diligence expansion. Validate multi-domain platform thesis.
  • 2029+ (Phase 3): Research intelligence (PhD research, industrial R&D, technical due diligence)

Architecture Status (October 2025):

  • Core platform architecture complete (domain-agnostic)
  • Customs domain specification complete (WCO framework integration, multi-jurisdiction model)
  • MVP implementation 15% complete (Duty Drawback workflow)

Platform vs. Product Positioning: Mamut Lab is a **platform for investigation memory** with customs compliance as the **first validated application**. The architecture genuinely supports multi-domain expansion (ADR-004, ADR-007, ADR-009 prove this). Customs proves platform viability; FDA/legal/M&A validate universality; research demonstrates long-term vision.

Transparency: All architecture documentation, implementation plans, and design decisions publicly available on GitHub. Enterprise buyers and investors can verify platform technical depth before committing. This is not vaporware—100+ documents (1000+ hours) demonstrate serious engineering.

Beta Program (Q2 2027): Limited to 10-15 customs compliance customers to prove investigation memory platform works. Success metrics: Investigation context restoration time, personnel handoff efficiency, audit preparation speed. Contact for beta waitlist ->

Research Foundations

Standing on the shoulders of giants. The concept experiments with established academic research instead of claiming novel invention—now expanded with research intelligence integration:

Knowledge-Augmented Checkpoints [BREAKTHROUGH - OCT 2025]

Revolutionary checkpoint architecture combining execution state with reasoning provenance—the first system to capture WHY decisions were made, not just WHAT was executed.

Why it matters: Reduces context restoration time from 20-40 hours to 15-30 minutes (50-100x improvement - research goal). Customs compliance teams can pause/resume EAPA investigations seamlessly after personnel changes, respond to Focused Assessments with full investigation history, and maintain audit-ready documentation across multi-year cases. No competitor offers this capability.

-> Knowledge-Augmented Checkpoints Architecture | -> Read the announcement blog post

Temporal Knowledge Persistence [NEW - OCT 2025]

Versioned knowledge graphs with Git-like branching, time-travel queries, and provenance tracking inspired by event sourcing and knowledge graph evolution research.

Why it matters: Foundation for knowledge-augmented checkpoints. Enables multi-year investigations with full context restoration after extended breaks. Understanding evolution tracked, contradictions managed, dead ends prevented.

-> Temporal Knowledge Substrate

Automation Complacency Prevention [NEW - OCT 2025]

Graduated autonomy framework with mandatory skill preservation protocols based on aviation and process control human factors research (70% trust paradox).

Why it matters: 5-level framework (Manual → Autonomous) with monthly manual practice, verification rate monitoring, and red team exercises preserves human expertise as automation increases.

-> Graduated Autonomy Design

Neurosymbolic AI [CORE - ENHANCED]

Researching how neural pattern recognition and symbolic reasoning can produce explainable, verifiable decisions. Now enhanced with multi-model ensemble verification and formal proof generation for research claims.

Why it matters: Layered verification pyramid (source → consensus → symbolic → formal) improves error detection through ensemble approaches. Mathematical guarantees for critical decisions.

-> Neurosymbolic integration

Continual Learning

Task-incremental, domain-incremental, and class-incremental learning mechanisms studied to reduce catastrophic forgetting.

Why it matters: Goal: learn from workflows without discarding prior knowledge. Darwin-Godel layer learns better investigation strategies over time.

-> Continual learning architecture

Dual-Process Cognition

Hybrid Type 1 (fast, reactive) and Type 2 (slow, deliberative) decision-making modeled on cognitive science literature.

Why it matters: Goal: balance speed with thoughtful reasoning for complex tasks. Research workflows optimized for exploration vs exploitation.

-> Dual-process engine design

Synthetic Data Generation [CORE]

Research into axiom-based training data aiming for mathematical guarantees. Goal: generate rich examples from finite symbolic rules without manual labeling.

Why it matters: Success would mean privacy-conscious, better-understood training data for continual learning experiments.

-> Synthetic data generation

Darwin-Godel Self-Improvement

Research goal: explore self-modifying architectures inspired by Godel machines and evolutionary algorithms—now learns investigation strategies.

Why it matters: Aspirational path toward continuous capability expansion without manual intervention. Research workflows improve automatically over time.

-> Self-improvement architecture

Event-Driven CQRS/Event Sourcing

Architecture draft uses event sourcing with Command Query Responsibility Segregation to maximise auditability—foundation for Temporal Knowledge Substrate versioning.

Why it matters: Enables time-travel debugging and immutable audit trails for compliance if implemented. TKS extends event sourcing with semantic versioning.

-> Architecture overview

100+ Architecture Documents [PUBLIC]

Complete investigation memory system design documented publicly—from customs workflow analysis to multi-jurisdiction investigation models. Includes ADR-010 (jurisdiction extension), WCO HS framework integration, and temporal knowledge substrate specifications.

Why it matters: Enterprise buyers can evaluate the technical approach before committing. Transparent documentation demonstrates serious engineering, not vaporware.

-> Browse all docs on GitHub

Transparent Development: All architecture decisions, workflow designs, and implementation plans documented in Architecture Decision Records (ADRs). Enterprise buyers can verify technical depth before beta participation. October 2025 update: Multi-jurisdiction extension (ADR-010) completed, enabling Q2 2027 international beta launch.

Frequently Asked Questions

Is Mamut Lab a platform or a customs compliance product?

Answer: Investigation memory platform with customs compliance as the first validated application (Q2 2027 MVP).

Platform Architecture Evidence:

  • Domain-Agnostic Core: 6 innovations (Knowledge-Augmented Checkpoints, Temporal Knowledge Substrate, Neurosymbolic Verification, Multi-Domain Synthesis, Darwin-Godel, Graduated Autonomy) work across all domains—documented in ADR-004, ADR-005, ADR-007, ADR-009
  • Multi-Domain Design: ADR-009's **primary example is PhD drug research (FDA-regulated domain)**, not customs. ADR-007 explicitly supports 7 domain types: academic, market, technical, regulatory, scientific, financial, legal
  • Extension Mechanism: Domain Context Provider pattern enables new domains via configuration, not rearchitecture. Adding FDA compliance requires domain config, not platform rebuild

Why Customs First?

  • Founder Expertise: 12 years enterprise software + deep customs research (100+ docs) = 12-18 month market lead
  • Validated Problem: GAO-20-182 documented crisis (35,000+ claims/$2B+ lacked review)
  • Clear ROI: $500K-$2M annual savings for large importers
  • Multi-Jurisdiction Advantage: WCO framework (182 countries) → 3x TAM expansion ($1.3B → $2.3B-$3.6B)

Platform Expansion Path:

  • Phase 1 (2027): Customs compliance MVP proves investigation memory platform works
  • Phase 2 (2028): Add FDA regulatory, legal compliance, M&A due diligence—same platform, domain config
  • Phase 3 (2029+): Research intelligence (PhD, industrial R&D, technical due diligence)
  • Total Platform TAM: $30B+ across customs ($3.6B) + FDA ($8B) + legal ($12B) + M&A ($5B) + research ($3B)

Platform vs. Product: If this were a customs product, the architecture would be tightly coupled to CBP workflows and HTS codes. Instead, core components are domain-agnostic (ADR-009 proves this with FDA example), and WCO HS framework is a **separable customs domain module**, not core platform dependency. Customs validates the platform; FDA/legal/M&A validate universality.

What are Knowledge-Augmented Checkpoints?

MVP Version (Q4 2026): Simplified checkpoint architecture capturing execution state + basic reasoning context

  • Execution State: Task progress, variables, call stack (standard checkpoint features)
  • Basic Reasoning Context: Recent decisions, key insights, confidence levels stored in knowledge graphs
  • Resume Capability: Restore task + context summary on resume (not full cognitive immersion)
  • Versioning: Git-like branching for safe exploratory execution

Future Vision (Post-MVP Research Track): Full cognitive context with reasoning provenance

  • The Goal: Capture WHY decisions were made, reasoning chains, confidence evolution, contradictions, dead-ends
  • Context Restoration Target: 15-30 minutes vs. 20-40 hours (requires validation)
  • Dead-End Prevention: Rejected approaches documented with failure reasons
  • Provenance: Complete lineage from sources to conclusions

MVP Scope: We're shipping basic resumability first, validating market demand, then iterating toward the full vision. The "50-100x faster restoration" claim is a research goal, not an MVP guarantee.

What's the Temporal Knowledge Substrate (TKS)? [NEW]

A versioned knowledge graph system that tracks understanding evolution over months/years for long-running research investigations:

  • Time-Travel Queries: View understanding state at any past timestamp
  • Provenance Tracking: Full reasoning chain history with source citations
  • Contradiction Management: Detect and track conflicting information over time
  • Dead-End Prevention: Failed exploration paths tracked to prevent redundant work
  • Git-like Branching: Create exploratory investigation branches with rollback safety

Implementation: ArangoDB multi-model database with custom versioning layer extending event sourcing with semantic understanding.

What's Graduated Autonomy? [NEW]

A 5-level autonomy framework with mandatory skill preservation protocols to prevent automation complacency:

  • Level 0 - Manual: AI disabled (skill maintenance mode)
  • Level 1 - Augmented: AI suggests, human executes
  • Level 2 - Collaborative: AI analyzes, human validates
  • Level 3 - Delegated: AI conducts workflows, human verifies reasoning
  • Level 4 - Autonomous: AI handles routine, flags surprises

Skill Preservation: Monthly manual mode, verification rate monitoring (≥40% threshold), quarterly red team exercises, and interleaved practice prevent the 70% trust paradox from aviation research.

What's the Investigation Memory architecture?

Mamut Lab's platform architecture enables multi-year customs compliance investigation tracking:

  • Investigation Lifecycle Management: Pause/resume EAPA investigations across months with complete personnel handoff context
  • Regulatory Knowledge Integration: Synthesize CBP rulings + WCO guidance + HTS General Rules of Interpretation + jurisdiction-specific regulations
  • Investigation Workflows: Focused Assessment response, origin determination analysis, C-TPAT compliance monitoring, classification review
  • Workspace Isolation: Rollback-safe exploratory classification analysis with audit trail preservation
  • Compliance Safety: Citation verification, regulatory change monitoring, multi-jurisdiction consistency checking

Status: Core architecture complete, MVP implementation (Duty Drawback Focused Assessment workflow) targeting Q4 2026 internal demo.

What's Darwin-Godel learning?

Status: Long-term research goal, NOT an MVP feature

Exploratory research direction investigating whether a self-improving architecture could evolve its own capabilities:

  • Observing what works in workflows and discovering better patterns
  • Evolutionary algorithms optimizing task execution strategies
  • Self-modifying architecture (with safety constraints)

Reality check: This is an open research problem inspired by Godel machines and evolutionary computation. It requires major theoretical breakthroughs and is NOT included in the MVP or beta timeline. Success depends on: (1) MVP proving market fit, (2) team scaling, (3) significant research investment. No delivery commitment.

What's neurosymbolic reasoning?

MVP Version: Basic multi-model orchestration and knowledge graphs (feasible)

  • Model Registry: Match tasks to specialized models (document extraction, HTS classification reasoning, regulatory analysis)
  • Simple Consensus: Compare outputs from 2-3 models for critical customs decisions (classification, valuation, origin determination)
  • Knowledge Graphs: Store investigation reasoning context with ArangoDB (proven tech)
  • Basic Explainability: Traceable decision chains for CBP audit defense without formal proofs

Research Track: Formal verification with theorem provers (bleeding-edge)

  • Formal Verification: Lean4/Coq subprocess integration (unsolved research problem)
  • Symbolic Validation: Z3, SymPy for mathematical guarantees
  • Layered Verification Pyramid: Source → Consensus → Symbolic → Formal proof

Reality check: Orchestrating formal theorem provers (Lean4/Coq) with LLM-driven agents is an active research problem. The MVP delivers practical multi-model coordination. Formal verification remains a long-term research goal with no delivery timeline.

What's the Coordinated Space architecture?

A proposed enterprise-grade task execution infrastructure with six specialized components—now enhanced for research workflows:

  • Context Management System for shared state (with temporal versioning)
  • Model Registry for tool capability discovery and intelligent selection
  • Task Executor for long-running tasks with checkpoint-based resumability and temporal knowledge accumulation
  • Protocol Adapters for system integration
  • Observability Platform for monitoring
  • Darwin-Godel Machine for self-improvement (learns investigation strategies)

These components are designed to work together to enable both long-running task execution and multi-year research investigations with transparency.

What's your background and team structure?

Mamut Lab is founded by an enterprise software engineer with domain expertise in customs compliance:

  • 12 years C#/.NET enterprise backend experience: Production-grade systems for regulated industries, audit trail architecture, compliance workflows
  • 100+ documented architecture files: Complete investigation memory system design, WCO framework integration, multi-jurisdiction specifications publicly available on GitHub
  • Customs compliance focus: Deep research into CBP workflows (Focused Assessments, EAPA investigations, C-TPAT), WCO Harmonized System framework, trade compliance software requirements
  • Transparent development: All technical decisions, architecture designs, and implementation plans documented in Architecture Decision Records (ADRs)

Team scaling: Current focus is MVP validation with initial customers. Post-beta team expansion will include customs compliance domain experts, enterprise sales, and customer success—funded by beta customer revenue and potential venture investment.

When will the beta be ready?

Customs Compliance MVP Timeline:

  • Q4 2026: Internal MVP Demo - Duty Drawback Focused Assessment workflow with checkpoint-based methodology reconstruction
  • Q2 2027: Private Beta Launch - Limited beta (10-15 customers: multi-jurisdiction importers, customs law firms, brokerage firms) with multi-jurisdiction support (USA, China, Germany, Serbia)
  • Post-Beta Expansion - Additional customs workflows (EAPA investigation management, C-TPAT monitoring), then adjacent regulated industries

MVP Scope Reality: The MVP delivers practical customs compliance capabilities with simplified checkpoint architecture. The "50-100x faster context restoration" referenced earlier is a long-term research goal, not an MVP guarantee. Beta features include: investigation timeline tracking, checkpoint-based resume capability, multi-jurisdiction coordination, and audit trail generation—proven technologies adapted for customs compliance, not breakthrough research.

Contact info@mamutlab.net for beta waitlist.

How much will it cost?

Enterprise-focused pricing for customs compliance (annual contracts):

  • Multi-Jurisdiction Tier: $400K/year for importers managing 2-3 jurisdictions (USA + China or USA + Germany)
  • Global Enterprise Tier: $800K-$2M/year for Fortune 500 importers with all 4 jurisdictions (USA + China + Germany + Serbia), unlimited investigations, dedicated support
  • Law Firm License: $500K-$2M/year for international trade law firms (matter-based or firm-wide licensing, multi-client investigation management)
  • Broker Enterprise: $1M-$3M/year for customs brokerage firms (multi-tenant architecture, white-label capabilities, client-branded portals)
  • Custom Deployment: Private cloud or on-premises deployment with compliance certifications, SLA guarantees, and regulatory audit support

Beta participants receive 50% discount for first 12 months. Pricing reflects multi-year investigation storage (4+ years for C-TPAT compliance), cross-jurisdiction synchronization, HS code version management, and audit-ready documentation generation for CBP proceedings.

Can I see the documentation?

Yes, transparency is core to the approach:

-> Browse 100+ docs on GitHub

Documentation covers investigation memory architecture, customs compliance workflows, implementation plans, design decisions, and multi-jurisdiction extension specifications (ADR-010) created October 2025.

What customs compliance use cases does this support?

Mamut Lab is purpose-built for customs compliance investigations requiring long-term memory:

  • Phase 1 (MVP): Duty Drawback Focused Assessment responses - reconstruct methodology from prior submissions within 30-day CBP deadlines
  • Phase 2: EAPA investigation management - 300-360 day timelines with personnel rotation, origin determination workflows, supplier verification tracking
  • Phase 3: C-TPAT compliance monitoring - 4-year rolling compliance history, regulatory change monitoring, one-click audit packages
  • Multi-Jurisdiction (Q2 2027): Track investigations across USA (CBP), China (GACC), Germany (Zoll), and Serbia Customs with WCO HS framework alignment
  • Future Expansion: HTS classification reviews, valuation disputes, trade fraud investigations (False Claims Act qui tam cases)

Platform roadmap: Start with customs compliance (MVP validation), then expand to adjacent regulated investigations (FDA compliance, financial services audits) based on market traction.

Why "Mamut Lab"?

Named after the Kikinda mammoth discovered in Serbia—preserved for 500,000 years, symbolizing investigation context preservation:

  • Historical Note: Originally named "TrueNames" (October 2025 rebrand to avoid confusion with TrueName.ai)
  • Mammoth Metaphor: Just as mammoths were preserved in permafrost for millennia, Mamut Lab preserves investigation reasoning, methodology, and context across multi-year compliance cases
  • Serbia Connection: Founder's heritage + multi-jurisdiction vision (Serbia is one of four initial jurisdictions: USA, China, Germany, Serbia)
  • Mission: Build investigation memory infrastructure that outlasts personnel changes, organizational restructuring, and regulatory transitions

See blog post: "Why Mamut Lab? A 500,000-Year Story of Preservation" for the complete story.

Join the Q2 2027 International Beta Waitlist

Limited to 10-15 enterprise customers. Multi-jurisdiction support: USA + China + Germany + Serbia. WCO framework foundation (182 countries). 50% early adopter discount. Shape international roadmap.

Reserve Your Beta Spot →

Opens in new tab • Beta updates only • No spam • Contact us

Want to follow development in real-time? Track progress on GitHub →