Technical Registry · Institutional IP Doctrine

Architecture & IP Registry

// THE askOdin PROTOCOL STACK

askOdin is not a software interface. It is the deterministic verification infrastructure that secures the gap between the data room and the partner meeting.

Conventional AI platforms layer an LLM over documents to summarize narratives. We enforce strict separation of extraction from evaluation. The semantic layer reads the unstructured human narrative; the core architecture — patent-pending, mathematically deterministic — evaluates the structural physics.

Four U.S. provisional patents and IPOS Section 34 clearance secure the architecture.

4 U.S. Prov. Patents

§34 IPOS Cleared

0 Values the Model Sees

110,000+ Benchmark Corpus

// The Deterministic Split

Extraction is LLM. Evaluation is Go. Separation guarantees the provenance.

Column 01 · The Trap

The Probabilistic Facade

Standard AI diligence software layers an LLM over a data room and expects it to verify a business model. This is structurally invalid. Generative models are probabilistic; they optimize for narrative fluency, not structural physics. They cannot be reliably audited.

Column 02 · The Architecture

The askOdin Architecture

askOdin structurally isolates extraction from evaluation.

1. Extraction: Isolated LLM layers parse unstructured narrative solely to map variables into mathematical constraints.
2. Evaluation: The RUNE Protocol executes deterministically. It compiles structured variables through 40+ mathematical checkpoints, validating claims against a benchmark corpus of 110,000+ Clarity Scores.

The Provenance Ledger

Every evaluated claim generates a cryptographic, per-claim hash. If an Investment Committee or LP challenges a Clarity Score, the compiler deterministically replays the execution log. The output is mathematically reconstructible and legally defensible.

// THE PROTOCOL STACK

The Protocol Stack

SYSTEM RUNE // LAYER 01

RUNE Protocol™

U.S. PATENT PENDING 63/948,559

Layer: Domain-Specific Intermediate Representation Compiler
Function: The core interrogation protocol. RUNE strips unstructured financial narratives and translates them into a directed acyclic graph. It executes a 3-pass compilation (Syntax, Logical Physics, Regulatory Compliance) to automatically propagate Brittleness Scores and flag Compile-Time Errors before execution.

CLAIM 01

Compile-Time Error Detection

Identifies logical physics violations — e.g. market_share > 100% — before a single token of analysis is generated, not after.

CLAIM 02

Uncertainty Propagation

Mathematically cascades the semantic ambiguity of a source claim through every downstream calculation, so brittle assumptions cannot hide inside a confident-looking number.

UNSTRUCTURED INPUT

"We expect to capture 140% of the addressable market by FY27."

↓ rune compile

{

"var": "market_share",

"value": 1.40,

"constraint": "<= 1.0",

"brittleness": 0.73,

"status": "VIOLATION"

}

SYSTEM RAVEN // LAYER 02 SEALED

RAVEN Protocol™

U.S. PROV. PATENT NO. 63/994,876

Layer: Cross-Document Triangulation Engine
Function: The verification layer for heterogeneous data rooms. RAVEN runs adversarial cross-document triangulation across pitch decks, financial models, and supporting evidence to surface internal contradictions between documents — e.g., a narrative assertion in a PDF checked against the embedded formula in an Excel model.
Note: The architectural mechanics of RAVEN's triangulation engine are protected under U.S. Provisional Patent No. 63/994,876 and are not publicly disclosed. askOdin licenses the verification layer, not the source code.

CLAIM 01

Contradiction Preservation

When a deck and a model disagree, the contradiction is preserved and surfaced — never smoothed into a plausible-sounding reconciliation the way a single-context-window model would.

CLAIM 02

Cross-Document Provenance

Every reconciled figure is traced back to its origin across heterogeneous formats — PDF, spreadsheet, memo — so an allocator can audit exactly which two sources collided.

THREAD A · PDF

deck.pdf · p.7

Revenue

$5.0M

THREAD B · XLSX

model.xlsx · C12

Revenue

$1.2M

KILL SHOT Δ $3.8M unreconciled across documents

SYSTEM NORN // LAYER 03

NORN Protocol™

U.S. PROV. PATENT NO. 64/011,252

Layer: Temporal Semantic Drift Detector
Function: Chronological state tracking. NORN evaluates chronologically sequential documents to isolate the divergence between a Presentation Score (narrative assertion density) and a Clarity Score (structural logic integrity). It mathematically detects Narrative Inflation — when rhetoric escalates while the underlying business physics degrade.

CLAIM 01

Latent Drift Calculation

Flags Narrative Inflation: the presentation gets louder ΔP ≥ 0 while the underlying structure quietly degrades ΔC < 0.

CLAIM 02

Retroactive Graph Weighting

Uses confirmed structural failures to adjust the relational decay weights of future deal-flow queries — the corpus learns from every resolved outcome.

SEMANTIC DRIFT · Q1→Q3

— Presentation— Clarity

ΔPresentation +0.31 ΔClarity −0.44

SYSTEM JUDGE // LAYER 04

JUDGE Protocol™

U.S. PROV. PATENT NO. 64/017,488 IPOS §34 NATIONAL SECURITY CLEARANCE (ISSUED 2026-03-26)

Layer: Asymmetric Non-Parametric Calibrator & Runtime Circuit Breaker
Function: The deterministic immune system. General LLMs optimize for persuasion; askOdin compiles for physics. If an underlying language model generates a novel probabilistic hallucination, JUDGE intercepts the payload, calculates an intra-document semantic delta, dynamically synthesizes a new logical constraint, and atomically hot-loads the rule into active memory. It permanently neutralizes specific logic failures across all enterprise pipelines.
Note: IPOS Section 34 clearance (issued March 26, 2026) establishes the JUDGE architecture as legally-vetted, sovereign-grade infrastructure.

CLAIM 01

Rule Auto-Compiler

Evaluates intra-document semantic deltas and synthesizes new constraints on the fly — the engine writes its own guardrails as it reads.

CLAIM 02

Atomic Hot-Loading

Hot-loads new constraints across thousands of concurrent execution threads with zero downtime, without altering the underlying neural network.

judge · runtime.go HOT-LOAD

payload := llm.Hallucinated()

rule := compile(

"revenue.deck == revenue.xlsx"

)

mu.Lock()

rules["r_4821"] = rule

mu.Unlock()

✓ constraint live · 0ms downtime

// Output Layer

The Provenance Ledger™

Every execution of the askOdin stack issues a mathematically traceable artifact. We do not issue probabilistic "recommendations." We record directly to the Provenance Ledger — reconstructing the exact logic and mathematical physics necessary to force institutional conviction.

This is the standard of care required to execute modern fiduciary duty.

For Founders

Deploy Crucible

Execute your presentation through the exact protocol stack an Investment Committee demands.

Initiate Diagnostic

Aggregate View

Read the Q1 Macro Audit

The State of Venture Physics: Q1 2026. The protocol stack applied at scale across the 110,000+ score benchmark corpus.

Examine the Corpus Data