PDTI Portfolio · Anchor 1
Forensic graph analysis on public records. Math finds patterns; humans interpret them. Every claim hashed back to a source document — auditable end-to-end. No chatbot synthesis.
Engineering Substrate
Active cases: unregistered-lobbying litigation · fabricated-study federal-grant fraud (False Claims Act) · ~10,500 entities, 1M+ derived edges, sealed exhibit chain.
How EMI Works
Source documents
FEC · LDA · SEC EDGAR · GDELT · Congress.gov · court records · web · structured extraction
Sealed evidence objects
Five canonical types — Entity, Document, Event, Claim, Source Record. Cryptographic hash on every write.
Analytical methods
Bipartite overlap · structural-hole analysis · spectral community detection · Granger causality · transfer entropy · cycle detection
Cognoscenti surface
Twelve-pattern signal library — surfaces magic-number persistence, citation laundering, methodology fabrication, amplifier networks.
Attorney-readable brief
Sealed exhibit packets. Every claim hashed back to a source document. Auditable end-to-end.
Product Lines
Six product lines configured on a shared engineering substrate. Each line is sealed-evidence, attorney-defensible, and auditable from output back to source document. Three are active today; three are scoped for next-cycle development.
Sealed evidence bundles for attorneys handling complex multi-matter cases. Cross-matter pattern detection across FEC filings, lobbying disclosures, SEC documents, court records, and news. Output: cited, hashed, attorney-defensible briefs. Active reference case: unregistered-lobbying litigation.
MVP activeForensic graph analysis on federal-program funding chains — provenance traces from programmatic justification back to source documents. Cycle detection on closed-loop fund flows. Active reference case: fabricated-study federal-grant fraud (False Claims Act, NGO chain).
Brief in late draftSame architecture, different lens — pattern detection for in-house compliance teams at pharma, financial-services, and regulated-industry organizations. Catches structural patterns conventional firms produce for $50K–$500K per engagement.
Pilot inboundTemporal-signal layer monitoring filing velocities, prediction-market order books, sentiment shifts, and structural anomalies across event streams. Output: signal briefs identifying patterns ahead of consensus. Anchor selection Q3 2026.
Engineering gatedCounterparty due diligence, supply-chain risk surfacing, and competitive structure mapping for boards, fund managers, and corporate development teams. Highest-revenue, lowest-political-toxicity commercialization target.
Post-MVPLighter-primitive configuration for narrative-pattern detection across earned media — news, press coverage, social posts. Identifies coordinated messaging, amplifier networks, and reputation-shift inflection points. Lower-priority near-term.
QueuedArchitecture
Five evidence object types. Eight live connectors. Nineteen analytical methods. Every Neo4j write carries source URL, fetch time, payload hash, run ID, and connector version. Cryptographic provenance on every claim.
5
Entity · Document · Event · Claim · Source Record. Five canonical types. Each carries a cryptographic hash and full provenance back to source.
8
FEC · LDA lobbying · SEC EDGAR · GDELT · Congress.gov · court records · web · structured extraction. Each connector versioned and hash-validated.
19
Bipartite overlap · Burt structural-hole analysis · Louvain/Leiden community detection · Granger causality · transfer entropy · Tarjan cycle detection · link prediction.
Sample methods · selected
Methodology
Two engines. Three named primitives. The system surfaces structural patterns in primary-source documents; humans interpret them. Math-based pattern detection — not chatbot synthesis.
Engine 01
Retrospective and stationary analysis. Resolves who, what, when, with what relationships, using which evidence. Output: case-theory-grade investigative deliverables, primary-source-anchored, with cryptographic exhibit provenance.
Active reference cases: unregistered-lobbying litigation · fabricated-study federal-grant fraud (False Claims Act, NGO chain).
Engine 02
Forward and temporal analysis. Surfaces signal change over time before consensus forms. Output: temporal-signal dashboards, alpha-grade event reports, early-warning indicators. Operates on the same engineering substrate, different analytical surface.
Reference cases: Post-hoc validation includes correctly-called Virginia DA primary, Romania election, Netherlands election. Anchor selection Q3 2026.
PRIMITIVE 01
Soft-cluster-to-anchor resolution. When a corpus has many mentions distributed across unstructured documents that all point at one load-bearing source, the system identifies that single source and pulls its content as exhibit-grade material.
Worked example: "Alliance for Public Health" appears in three independent citation chains under three different formal names. Anchor Convergence resolves all three to the single legal entity and surfaces the canonical IBBS PWID 2020 source document.
PRIMITIVE 02
Retrieval resilience across mirrors. When a primary URL fails — state.gov, GAO, FOIA portals, archived government records — the system walks a per-source ladder of alternate hosting locations until content resolves and validates.
Worked example: PEPFAR COP source URLs are fragile. The connector walks state.gov → GAO → House Appropriations → govinfo → amfAR copsdata mirror → Wayback CDX until the document resolves with a hash match.
PRIMITIVE 03
Pattern library for surfacing soft signals within a corpus. Twelve patterns v1.0 — magic-number persistence, citation-chain laundering, methodology fabrication, internal inconsistency, amplifier networks, prevalence-table mismatch, and others.
Worked example: The "1.7% / 350,000" figure persisting across UN, UNODC, UNAIDS, PEPFAR documents while being internally inconsistent with IBBS source data — flagged by the magic-number persistence pattern.
Methodology
Two engines. Three named primitives. The system surfaces structural patterns in primary-source documents; humans interpret them. Math-based pattern detection — not chatbot synthesis.
Engine 01
Retrospective and stationary analysis. Resolves who, what, when, with what relationships, using which evidence. Output: case-theory-grade investigative deliverables, primary-source-anchored, with cryptographic exhibit provenance.
Active reference cases: unregistered-lobbying litigation · fabricated-study federal-grant fraud (False Claims Act, NGO chain).
Engine 02
Forward and temporal analysis. Surfaces signal change over time before consensus forms. Output: temporal-signal dashboards, alpha-grade event reports, early-warning indicators. Operates on the same engineering substrate, different analytical surface.
Reference cases: Post-hoc validation includes correctly-called Virginia DA primary, Romania election, Netherlands election. Anchor selection Q3 2026.
PRIMITIVE 01
Soft-cluster-to-anchor resolution. When a corpus has many mentions distributed across unstructured documents that all point at one load-bearing source, the system identifies that single source and pulls its content as exhibit-grade material.
Worked example: "Alliance for Public Health" appears in three independent citation chains under three different formal names. Anchor Convergence resolves all three to the single legal entity and surfaces the canonical IBBS PWID 2020 source document.
PRIMITIVE 02
Retrieval resilience across mirrors. When a primary URL fails — state.gov, GAO, FOIA portals, archived government records — the system walks a per-source ladder of alternate hosting locations until content resolves and validates.
Worked example: PEPFAR COP source URLs are fragile. The connector walks state.gov → GAO → House Appropriations → govinfo → amfAR copsdata mirror → Wayback CDX until the document resolves with a hash match.
PRIMITIVE 03
Pattern library for surfacing soft signals within a corpus. Twelve patterns v1.0 — magic-number persistence, citation-chain laundering, methodology fabrication, internal inconsistency, amplifier networks, prevalence-table mismatch, and others.
Worked example: The "1.7% / 350,000" figure persisting across UN, UNODC, UNAIDS, PEPFAR documents while being internally inconsistent with IBBS source data — flagged by the magic-number persistence pattern.
Get in touch
EMI's brief-engine is operating now. Litigation matters, FCA cases, compliance pattern detection, and signal analysis are accepting engagements. Methodology and bundle artifacts available for technical review under NDA.
Email Project_Insight
Project_Insight@embergraphintel.com — engagement inquiries, technical review under NDA, brief requests. The engine name is the contact line.
About PDTI
EMI's parent — Pioneering Deep Tech Institute — and the multi-anchor portfolio.
Methodology essays
Long-form analysis, sanitized case studies, technical explainers in the PDTI Journal.
info@PioneeringDeep.Tech
© 2026 Pioneering DeepTech Institute (PDTI) . All rights reserved.
