AI Judgment Infrastructure™
Hope is not an asset class.
Audit the arithmetic before you commit capital. askOdin Clarity tests the claims in a CIM or an investment memo against the financial model behind it, cell by cell, before the committee votes.
- Guidance EV
- $140.0M
- Supported EV
- $96.0M
- Gap
- −$44.0M · −31.4%
“One-time” $1.8M ERP cost recurs FY23–FY25.
Synthetic data room. Read the case →
THE LAST INSTITUTIONAL BLIND SPOT
Every critical institutional function has a standard. Except judgment.
askOdin is AI judgment infrastructure: the missing standard, measuring and verifying investment reasoning before capital moves.
The memo is written and the partners agree, partly because dissenting inside a partnership carries a social cost. When the deal breaks, it is rarely bad luck. It is an unexamined assumption someone copied off a slide, and nobody in the room could say who verified it.
The decision to allocate millions of dollars still depends on reasoning that cannot be independently measured or verified. Judgment remains the last institutional capability without a common standard.
No enterprise closes its books on an unstandardized ledger. No bank settles a payment on informal trust. But an investment committee will wire twenty million dollars on a number nobody in the room can source.
TWO DESKS · ONE COMPILER
One deterministic compiler. A desk for each kind of deal.
Language models summarize deal documents. askOdin tests the claims in them against the numbers supplied with them, and reports where they disagree before capital is committed.
Private equity & M&A · pre-LOI
The add-back, tested against the databook
Every add-back in a sell-side CIM is tested against the seller's own Excel databook, in minutes, before the LOI.
- PE-001 · a “one-time” cost the databook shows recurring
- PE-002 · owner pay above a replacement salary
- PE-003 · run-rate savings inside trailing EBITDA
- PE-004/005 · reported and adjusted EBITDA against the databook and the seller's QoE
- Supported EV at the CIM's own multiple, and a leverage scenario on your terms
Venture & CVC · pre-IC
The memo, tested against the model
The pitch deck and the investment memo are scored on the Clarity Framework and checked against the financial model behind them.
- Five scored pillars, from problem definition to deal structure
- Three audit checks that cap a pillar: business physics, unit economics, market evidence
- Contradictions between the deck and the model, each tied to both sources
- A kill shot that floors the score on a terminal contradiction
- A Defensible Audit Log for every run
THE STANDARD
A common language for investment reasoning.
Investment outcomes contain uncertainty. Investment reasoning should not. The Clarity Framework™ evaluates the quality of reasoning in a venture deal before capital is committed. Buyout CIMs are scored on the PE desk's own terms: five add-back rules and a seven-factor base.
Will this investment succeed?
Is our reasoning rigorous enough, and do the numbers supplied support the price?
Each pillar is scored 0–20. Three audit checks — business physics, unit economics, market evidence — can cap any pillar independently of its own score. Formatting, wording, and delivery are scored at zero.
EVIDENCE
Reasoning can be measured.
Each example scores the reasoning in the documents, never the outcome. Theranos is a real company's investor memo; Apex is a synthetic buyout data room we built to show the PE desk at work.
Apex Industrial — Buyout CIM
PE & M&A DESK · SYNTHETIC DATA ROOM
Three of four add-backs failed against the seller's own databook. At the CIM's 10.0x multiple, $140.0M of guidance supports $96.0M. Debt sized on the CIM would sit at 5.8x supported EBITDA against a 4.5x ceiling the analyst set.
Read the case study →Theranos — 2013 Investor Memo
DO NOT PROCEED
A hardware physics violation: fingerstick draw volume cannot satisfy a multi-analyte assay claim. The narrative was immaculate; the reasoning did not hold.
Read the terminal audit →WHY THIS STANDARD IS DEFENSIBLE
Standards require enforcement.
Without enforcement, a standard is only a guideline. askOdin's enforcement mechanism is a deterministic compiler — it measures reasoning rather than generating opinions. Unlike a probabilistic language model, the evaluation runs the same rules every time: the same recorded variables return the same score under the same version, and a past verdict is checked from its record, not re-sampled.
The mechanism is benchmarked against the Judgment Graph™ — a corpus of more than 110,000 evaluated investment narratives built on public deal data. Four U.S. provisional patents protect the architecture. These are not product features; they are the institutional safeguards that make the standard defensible.
askOdin Verify → — how a deterministic compiler produces the same verdict twice, and why a language model cannot.
RUNE PROTOCOL · DETERMINISTIC JUDGMENT COMPILER
Compiling reasoning into auditable logic.
RUNE Protocol™ compiles unstructured natural language into an executable, logic-validated dependency graph. Marketing polish is stripped; the core business assertions are isolated and scored for brittleness.
LLMs optimize for persuasion. askOdin compiles for physics.
U.S. PATENT PENDING 63/948,559- read
- Apex_CIM.pdf · 4 add-backs, p.7
- map
- Apex_Databook.xlsx · labels to cells, no values seen
- rule
- PE-001 · Northgate Systems · Opex detail!B4:E4
- test
- FY23 1.59 · FY24 1.70 · FY25 1.81, each ≥ half of 1.80
RAVEN PROTOCOL · MULTI-DOCUMENT TRIANGULATION
Adversarial triangulation.
RAVEN Protocol™ cross-checks every qualitative claim against the raw financial model. It is the verification layer for heterogeneous data rooms — divergent data points are flagged as conflicts, never silently reconciled.
The architectural mechanics of RAVEN's triangulation engine are protected under U.S. Provisional Patent No. 63/994,876 and are not publicly disclosed.
THE OUTPUT
A Defensible Audit Log.
A deterministic, time-stamped audit trail. Every finding carries an anchor back to its exact source — the page, the cell and the fingerprint of the file — a Defensible Audit Log™ your Investment Committee and LPs can check.
See a full run: the Apex case Synthetic buyout · every finding traced Read →THE PRODUCT
Operationalize institutional judgment.
askOdin Clarity applies the Clarity Framework™ consistently across analysts, partners, and investment committees. It is how an institution operationalizes the standard.
Before the data room moves: retention, residency, and the compliance roadmap.
- Evaluate investment theses consistently, regardless of who runs the analysis.
- Surface reasoning gaps and unsupported add-backs before capital is committed — in minutes, not weeks.
- Compare opportunities objectively, on a single standard.
- Preserve institutional knowledge as a defensible, reproducible record.
// FOR FOUNDERS
Pre-IC stress-testing for founders.
askOdin Crucible runs a founder's deck and model through the same framework Clarity applies for investors, in under three minutes, before the partner meeting. Accelerators run whole cohorts through it. Every evaluation is scored against the Judgment Graph's benchmark corpus, never used to train a model.
// THE CATEGORY, DEFINED
What is askOdin?
The category
What is AI Judgment Infrastructure?
The deterministic verification layer for private capital. Unlike information infrastructure (which retrieves and organizes data), judgment infrastructure interrogates the logical coherence of an investment thesis — compiling each claim against the documents supplied with it before capital is deployed; a benchmark corpus of 110,000+ Clarity Scores on public deal data puts the result in context.
What is deterministic due diligence?
Due diligence whose verdict is reproducible and audit-defensible months later, because the same recorded variables under the same Framework version always return the same score, and the record keeps them. It is the opposite of a probabilistic LLM summary, which can change on every run and cannot be defended to an investment committee or a regulator.
Is there an AI rating agency for startups or private deals?
Not in the sense of a regulated rating agency. What askOdin provides is a standard for the reasoning: the Clarity Score, a deterministic 0–100 measure of whether an investment narrative holds up against its own evidence, backed by a Defensible Audit Log. Credit ratings and audited accounts give other asset classes an independent reference point; private-market investment reasoning has lacked one.
How it differs
How is askOdin different from ChatGPT?
ChatGPT summarizes what a deck says (probabilistic). askOdin compiles whether the physics are sound (deterministic) and emits a reproducible Clarity Score and a Defensible Audit Log — not prose.
How is askOdin different from PitchBook, Affinity, or AlphaSense?
Those are retrieval, CRM, and market-intelligence layers. askOdin is the judgment layer above them: retrieval retrieves, analytics aggregates, workflow automates — we compile judgment.
How it works, and who builds it
How does askOdin work?
In two passes. An extraction model reads the deck, memo and model and pulls out the variables they state; the RAVEN Protocol cross-checks them across documents. A deterministic engine then does all of the scoring and emits a Clarity Score with a Defensible Audit Log — every finding traced to its source. Extraction is a model; evaluation is deterministic.
Is askOdin a real company, and who founded it?
Yes. askOdin Pte. Ltd. is headquartered in Singapore, founded by YekSoon Lok. It is built on four U.S. provisional patents — RUNE, RAVEN, NORN, and JUDGE — and is a member of the NVIDIA Inception Program.
Private equity
Does askOdin replace a quality of earnings report?
No. A buy-side QoE is commissioned after the LOI, once you have exclusivity, and tests the numbers against the general ledger, samples transactions and sets the working-capital peg. askOdin does none of that. It runs before the IOI and the LOI and tests the CIM's add-backs against the seller's own databook, so you bid on earnings the documents support and spend the QoE budget confirming a bridge you have already tested.
If the language model never sees a number, how are the numbers checked?
The model has two jobs: reading the CIM's claims in its own words, and mapping the databook's labels to sheet and cell coordinates. It does both without seeing a value. Deterministic Go code then reads every number from those cells and applies fixed rules, with thresholds set in the engine before any data room was run. The arithmetic never passes through the model.
What happens to our confidential deal documents?
No Training Use: extraction runs against an external model API under a paid-tier agreement that prohibits training, and the deterministic engine does the analysis. Processing runs under Stateless API Orchestration and Strict Data Sovereignty, and raw documents are held no longer than 30 days under a documented retention ceiling. The ceiling is documented policy; until its automated enforcement ships, deletion is executed on request. The detail, control by control, is on the security page.
Judgment is the last unscalable asset.
Public markets have GAAP, statutory audit and regulated clearing. Private capital has run on persuasion and unverified spreadsheets. The next generation of investment firms will compete on institutional judgment. askOdin audits judgment before capital moves.