TLDR
Every analytical result in the pipeline receives a confidence grade from a military-origin system for rating how reliable a source is and how credible the information is (the NATO Admiralty Code), adjusted downward by an estimate of how complete our data is. Because the corpus is only 63.7% complete, every confidence score is reduced — and all 9 cross-domain results currently classify as investigative leads rather than confirmed findings (PAPER TRAIL Project, 2026a).
The Two-Axis System
The NATO Admiralty Code is a standardized intelligence grading system used by military and law enforcement agencies worldwide (NATO, 2016). It evaluates information on two independent axes.
Source reliability rates how trustworthy the source is, regardless of the specific information it provides. An "A" rating means completely reliable — a government court filing, a regulatory consent order, an enacted statute. A "B" means usually reliable — corpus-derived data from a validated pipeline. Lower ratings descend through "C" (fairly reliable), "D" (not usually reliable), "E" (unreliable), to "F" (reliability cannot be judged) (NATO, 2016).
Information credibility rates the specific piece of information, regardless of its source. A "1" rating means confirmed by other independent sources. A "2" means probably true — consistent with other information but not independently verified. The scale descends to "6" (truth cannot be judged) (NATO, 2016).
A result graded A1 comes from a completely reliable source and is independently confirmed. A result graded B3 comes from a usually reliable source but the information has only possible truth value. The two axes are independent: a reliable source can report uncertain information, and an unreliable source can coincidentally report confirmed facts.
The Completeness Adjustment
The raw Admiralty grade would be sufficient for a complete corpus. But this corpus is not complete. A statistical species-richness estimator (the Chao1 method, borrowed from ecology where it estimates the total number of species in a habitat from a partial sample) calculated 1,290,141 total entities from 821,633 observed — meaning 36.3% of entities remain undetected (PAPER TRAIL Project, 2026b).
The cross-domain synthesis engine (Script 25b) implements an adjustment based on how complete we estimate our data to be (the Chao1 adjustment), using the formula: adjusted confidence = metric confidence x completeness estimate. The metric confidence (derived from the Admiralty grade) is multiplied by the completeness estimate. At 63.7% completeness, even a perfect A1 source grading is reduced to 63.7% of its theoretical maximum confidence (PAPER TRAIL Project, 2026a).
This is deliberate conservatism. When you know that more than a third of the entities in the system have never been observed, any conclusion about relationships, patterns, or absence of evidence must account for the possibility that the unobserved entities would change the picture. The completeness adjustment forces this accounting into every confidence score (PAPER TRAIL Project, 2026c).
The Finding Threshold
The pipeline distinguishes between Findings (confidence above 0.75 after the completeness adjustment) and Leads (below 0.75). This threshold reflects the minimum confidence at which an analytical result should be treated as established rather than investigative (PAPER TRAIL Project, 2026a).
Currently, all 9 cross-domain results from the synthesis engine classify as leads. None reach the 0.75 threshold (PAPER TRAIL Project, 2026a). The highest-confidence results — the Deutsche Bank compliance failures, the corporate ownership chains confirmed against public registries — approach but do not cross the line when the completeness adjustment is applied.
This means that despite the analytical pipeline processing 2.1 million documents, extracting 2.38 million entities, computing 29.5 million relationships, and running 6 cross-domain synthesis modules, the system's own confidence framework says: these are leads, not findings. Follow them, do not cite them.
How Sources Map to Grades
The pipeline's four-tier verification system maps directly onto the Admiralty reliability axis (PAPER TRAIL Project, 2026d):
| Verification Tier | Admiralty Reliability | Examples |
|---|---|---|
| T1 (Government primary) | A | Court filings, consent orders, enacted statutes |
| T2 (Corpus-derived) | B | Wire transfers parsed from bank reports, extracted entities, FedEx records |
| T3 (Journalism) | C | Miami Herald investigation, media reports |
| T4 (Estimation) | D-E | Statistical projections, model outputs |
A wire transfer parsed from the TD Bank suspicious activity report (T2 source) receives a B reliability grade. The specific claim that "$2 million was wired to Representation Trust on August 8, 2019" receives a credibility grade based on whether other independent evidence confirms it. If only the bank report documents this wire, the grade might be B2 (usually reliable source, probably true). If the will filing date independently confirms activity on August 8, the credibility might rise to B1 (confirmed by independent source) (PAPER TRAIL Project, 2026a).
Independence Verification
A subtle but critical feature of the grading system is independence verification. If two evidence streams both derive from the same scanned document, they are not independent — they should not both contribute to raising a confidence grade. The synthesis engine checks for circular evidence: claims that appear to be confirmed by multiple sources but actually trace back to a single original document (PAPER TRAIL Project, 2026c).
When circular evidence is detected, the dependent sources receive a penalty that reduces their combined contribution to the confidence score. This prevents a single document from inflating confidence by appearing in multiple analytical pipelines (PAPER TRAIL Project, 2026c).
The Epistemic Honesty
The grading system, combined with the completeness adjustment, produces an uncomfortable outcome: nothing in the pipeline currently qualifies as a finding. Every cross-domain result is a lead. This is by design (PAPER TRAIL Project, 2026a).
The system could easily produce findings by lowering the threshold, removing the completeness adjustment, or relaxing the independence requirements. Any of those changes would generate impressive-sounding confirmed results. They would also be dishonest. The corpus is 63.7% complete. More than a third of the entities are estimated to exist but have never been observed. Under those conditions, claiming findings would overstate what the evidence supports (PAPER TRAIL Project, 2026b).
The 898 cross-domain entity profiles, 38 evidence chain nodes, 3 hypothesis-testing matrices, and 9 investigative leads represent what can honestly be said given partial data. They are starting points for investigation, not endpoints for accusation. The Admiralty Code ensures that this distinction is quantified, not merely stated.
References
NATO. (2016). Allied joint doctrine for intelligence, counter-intelligence and security (AJP-2). NATO Standardization Office.
PAPER TRAIL Project. (2026a). Cross-domain synthesis — admiralty grading and findings classification (Script 25b) [Data]. _exports/synthesis/findings_vs_leads.csv
PAPER TRAIL Project. (2026b). Chao1 validation — corpus completeness estimate (Script 23) [Data]. _exports/validation/chao1_summary.json
PAPER TRAIL Project. (2026c). Cross-domain synthesis methodology — independence verification and confidence adjustment [Technical report]. research/CROSS_DOMAIN_SYNTHESIS.md
PAPER TRAIL Project. (2026d). Episode references and verification tier definitions [Data]. communications/ep02_slides/references.md