Research

Methodology, not magic.

FRIDAY Sovereign Intelligence ships with a published methodology surface. We document how convergence is detected, how signals are scored, and how false positives are controlled. Customers can audit the methodology, not just consume the output.

01Convergence theory

Convergence is the technical core of the platform. The thesis is that independent signal layers reaching the same conclusion about an entity within a defined window is a stronger signal than any single layer, including its strongest.

Independence is the prerequisite. Two signals derived from the same data source are not independent. The platform engineers every layer as a separate ingest path with separate provenance.
The window matters. Convergence within 72 hours is a different signal than convergence over a quarter. Windows are configurable per pattern.
Layer count is meaningful. Two-layer convergence is interesting; three-layer is rare; four-plus is the strongest tier we observe in production. Convergence score is reported with the layer count, not abstracted into a single number.

02Signal independence

We do not chain inferences. A signal that is itself the output of another signal layer doesn't count toward convergence. This rules out a class of false positives that plague platforms trained on derived features.

Source independence. Each layer has its own ingest path, its own provenance chain, and is auditable independently.
Layer documentation. Every layer's data source, ingest method, and processing pipeline is documented and available to customers under contract.
No black-box composition. Convergence is computed from explicit signals, not from a learned representation that fuses them implicitly.

03False-positive control

False positives are the failure mode that destroys analyst trust. The platform controls them through threshold engineering and explicit reporting.

Per-pattern thresholds. Detection thresholds are pattern-specific and tunable per tenant. There is no universal threshold.
Confidence reporting. Every flag is reported with its confidence interval and the specific signals that drove it. Analysts see the evidence, not just the conclusion.
Verification windows. Predictions are timestamped with verification windows. After the window closes, outcomes are recorded against the prediction.

04Accuracy posture

We do not publish a single accuracy number. Anyone publishing a single accuracy number for a multi-domain intelligence platform is either confused or selling something.

Per-pattern measurement. Accuracy is measured per pattern, per window, per data layer.
Production-tracked outcomes. Verification windows close in production and outcomes are recorded. Customers can audit the historical record for any pattern they care about.
NIST 800-53 alignment. Audit methodology for accuracy tracking is being aligned with NIST 800-53 AU controls. Formal certification work is in progress.

05Patent portfolio

Core platform methods are protected by a portfolio of pending and granted patents. The portfolio is not the product, but it does mean the methods we describe in customer engagements are ours and stay ours.

Convergence detection architecture. Methods for fusing independent signal layers into entity-level findings.
Causal compression. Methods for reducing signal volume while preserving causal structure, used in the platform's core pipeline.
Signal independence engineering. Methods for guaranteeing independence properties across ingest paths.

Discuss methodology under NDA

Detailed methodology discussions are scheduled with qualified buyers during procurement evaluation. Subject to NDA.

Request Methodology Briefing