TLDR
The synthesis engine generates three tiers of narrative product -- executive summary, evidentiary chronology, and methodology disclosure -- totaling approximately 3,600 words across three markdown files, each designed for a different audience and each compliant with Federal Rules of Evidence Rule 1006 (PAPER TRAIL Project, 2026a). Every factual claim links to its provenance chain, making the products usable as litigation summaries of voluminous records.
Why Three Tiers
FRE 1006 permits a party to present a summary, chart, or calculation to prove the content of voluminous writings that cannot themselves be conveniently examined in court (Federal Rules of Evidence, Rule 1006). A corpus of 2.1 million documents qualifies. But a single summary document cannot serve all audiences. An executive reading a two-page brief needs different information than a litigator building an evidentiary argument, who needs different information than an opposing expert challenging the methodology.
The three-tier structure addresses this by separating content by audience and purpose (PAPER TRAIL Project, 2026a). Each tier is a standalone document that cross-references the others. Together, they form a complete package that satisfies both FRE 1006 and the Daubert standard for admissible expert testimony.
Tier 1: Executive Summary
The executive summary is approximately 338 words -- readable in under two minutes (PAPER TRAIL Project, 2026b). It states the scope of the analysis (2.1 million documents, 2.38 million entities, 224 wire transfers), the three primary hypotheses (spoliation, willful blindness, asset concealment), and the key finding that all three hypotheses are strongly favored over their alternatives but that no evidence chain crosses the 0.75 confidence threshold required for formal classification as a finding.
The executive summary does not contain methodology, error rates, or caveats beyond a single statement that the corpus is estimated to be 63.7% complete. These details are in Tiers 2 and 3. The purpose of Tier 1 is to answer the question "what did you find?" in the time it takes to read a single page (PAPER TRAIL Project, 2026b).
Every number in Tier 1 links to a specific export file or database query. The 2.38 million entity count links to the entity table. The 224 wire transfers link to the wire_transfers table. The ACH scores link to the ACH summary export. An executive who wants to verify any claim can follow the provenance chain without reading the full evidentiary chronology.
Tier 2: Evidentiary Chronology
The evidentiary chronology is the longest product at approximately 2,585 words, organized chronologically with inline citations to source documents (PAPER TRAIL Project, 2026c). It walks through the evidence domain by domain: corporate formation, banking relationships, wire transfer patterns, FedEx shipping network, flight operations, and institutional responses.
Each section follows a consistent structure: what the data shows, which specific documents or records support the claim, what the adjusted confidence score is, and what alternative explanations exist. The chronological organization allows a reader to track the evolution of the network over time -- from corporate formation in the early 2000s through the 2005 FedEx cessation, the 2013-2018 Deutsche Bank relationship, the 2019 arrest and death, and the posthumous estate management.
The evidentiary chronology is the tier most directly useful for litigation. Under FRE 1006, a summary must be made available for examination and copying by other parties. The chronology's citation density ensures that every claim can be challenged, verified, or rebutted by reference to the underlying records (PAPER TRAIL Project, 2026c).
Tier 3: Methodology Disclosure
The methodology disclosure is approximately 684 words covering the processing pipeline, error rates, Monte Carlo robustness testing, and Bayesian Belief Network results (PAPER TRAIL Project, 2026d). It is written for a technical audience -- an opposing expert or a court-appointed special master evaluating the reliability of the analysis.
This tier discloses the eight-stage compound error calculation (8.4% worst-case correct rate), the Monte Carlo confirmation that all ACH verdicts are 100% stable across 5,000 iterations, and the BBN posterior probabilities (Spoliation P=0.90, Willful Blindness P=0.77, Asset Concealment P=0.69). It also discloses the limitations: the naive Bayes independence assumption, the sensitivity to corpus completeness, and the classification of all evidence chains as leads rather than findings.
The Daubert standard requires that expert testimony be based on sufficient facts, reliable principles, and reliable application of those principles to the facts (Daubert v. Merrell Dow Pharmaceuticals, 1993). Tier 3 addresses each requirement: the facts are the 2.1 million documents, the principles are the statistical methods (PELT, Leiden, Splink, Chao1, ACH, BBN, Monte Carlo), and the application is documented in the script source code with measured error rates at each stage (PAPER TRAIL Project, 2026d).
Provenance and Reproducibility
All three tiers are generated programmatically from the synthesis engine's exports, not written by hand (PAPER TRAIL Project, 2026a). This means they are reproducible: running the narrative generation module on the same data produces the same documents. If the underlying data changes -- new documents are released, entity resolution improves, wire transfers are re-parsed -- the narratives can be regenerated to reflect the updated state.
Each claim in the narrative carries a provenance tag linking to the specific CSV row, database record, or export file that supports it. This chain -- from narrative claim to export file to database query to source document -- is the backbone of FRE 1006 compliance. The summary is not a substitute for the underlying records; it is a structured index into them.
References
Daubert v. Merrell Dow Pharmaceuticals, Inc., 509 U.S. 579 (1993).
Federal Rules of Evidence, Rule 1006 (Summaries to Prove Content).
PAPER TRAIL Project. (2026a). Narrative generation module [Software]. Script 25b, app/scripts/25_cross_domain_synthesis.py
PAPER TRAIL Project. (2026b). Tier 1: Executive summary [Data set]. _exports/synthesis/narrative_tier1.md
PAPER TRAIL Project. (2026c). Tier 2: Evidentiary chronology [Data set]. _exports/synthesis/narrative_tier2.md
PAPER TRAIL Project. (2026d). Tier 3: Methodology disclosure [Data set]. _exports/synthesis/narrative_tier3.md
This investigation is part of the SubThesis accountability journalism network.