Two Filings · One Closed Loop
From event capture to AI oversight.
Patent 1 creates the deterministic financial evidence: every wire, every accrual, every reconciliation packaged with cryptographic proof. Patent 2 governs how the AI that reasons over that evidence is trained, monitored, and held to the same audit-grade discipline. Together they form a closed loop — not two parallel filings. This is governed domain AI, not a chatbot bolted onto accounting software.
Patent 1 · Deterministic Evidence Chain
Three filed inventions, engineered as one statutory architecture.
Filed under Track One Prioritized Examination. The first two inventions establish settlement verification and FASB 91 EIR computation; the third unifies them under a single persistent identifier threading the Evidence Spine.
Patent 1 · Invention A
Automated settlement verification.
When a wire instruction arrives for a private credit transaction, the system verifies it against the loan’s current state — outstanding balance, accrued interest, fee schedules, reserves — and generates a deterministic verification token. Corroboration is scored across independent data sources. No back-and-forth email confirmations.
- Verification
- Deterministic, cryptographic
- Corroboration
- Multi-source, scored
- Settlement
- Wire + ledger attestation
Independent claim with supporting dependent claims
Patent 1 · Invention B
Precision interest-rate computation with audit trail.
Effective interest rates are computed using the Newton-Raphson iterative solver — the method required by FASB ASC 310-20 for loan origination fee amortization. Every intermediate step is recorded on a hash-chain provenance ledger. An auditor can reconstruct any number, on any historical date, to the cent.
- Solver tolerance
- ≤10-10
- Standard
- FASB ASC 310-20
- Provenance
- Hash-chain across iterations
Independent claim with supporting dependent claims
Patent 1 · Invention C
Multi-layer Evidence Spine architecture.
All financial data is organized along the Evidence Spine — origination, components, lifecycle events, attestation, settlement, reconciliation, daily position — linked by a single persistent facility identifier. Audit evidence under PCAOB AS 1105 is a built-in property of the architecture, not an after-the-fact reconstruction. Five-role attestation per material event requires at least two of Operator, Agent, Administrator, Borrower, Fund Manager.
- Architecture
- Multi-layer Evidence Spine
- Identifier
- One persistent facility identifier
- Attestation
- Five-role, hash-chain gated
Independent claim with supporting dependent claims
Patent 2 · AI Governance Layer
Four filed inventions governing the AI layer above the platform.
Filed under Track One Prioritized Examination. The methodology Enveriti uses to train its own domain-specific AI is itself the patent — a closed-loop discipline that makes AI training auditable, capability-preserving, and reproducibly governed.
Patent 2 · Invention A
Multi-dimensional corpus classification.
Every training record is classified along multiple orthogonal dimensions, producing a coverage matrix that surfaces hidden gaps — such as thousands of records explaining one process but zero teaching the model to diagnose a failure in it. Targeted record generation fills those gaps under governance.
- Classification
- Multi-dimensional, orthogonal
- Output
- Coverage matrix with gap surfacing
- Remediation
- Governed targeted generation
Independent claim with supporting dependent claims
Patent 2 · Invention B
Deterministic coverage replay.
Modern training frameworks shuffle data and report only aggregates — it becomes impossible to know which records the model has seen. This invention deterministically replays training to produce a per-record coverage map, without modifying the underlying framework.
- Mechanism
- External, deterministic
- Framework intrusion
- Read-only
- Output
- Per-record coverage map
Independent claim with supporting dependent claims
Patent 2 · Invention C
Multi-signal training oversight.
Multiple training-quality signals are constructed at every checkpoint. Composite indicators — drawn from disciplines that institutional finance already uses for quantitative oversight — fire only when multiple signals agree, producing high-specificity stop decisions.
- Signals
- Multi-channel, composite
- Decision rule
- Multi-indicator agreement
- Specificity
- Calibrated to zero false-positive stops
Independent claim with supporting dependent claims
Patent 2 · Invention D
Intelligence and capability retention monitor.
During domain-specific training, the model is continuously tested against a held-out general-knowledge evaluation set across multiple categories. If performance degrades, the system flags it and intervenes. Domain expertise without losing general intelligence.
- Evaluation
- Held-out general-knowledge
- Categories
- Reasoning · Math · Code · Language · Common-sense
- Action
- Automated flag + intervention
Independent claim with supporting dependent claims
See the platform that runs on this architecture.
The demo environment spans 30 facilities, 4 funds, and nearly 120,000 events — running on 426 backend routes across 38 domains. Every number deterministically computed.