// THE TERM, DEFINED
What is deterministic due diligence?
Deterministic due diligence™ is 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 — the opposite of a probabilistic LLM summary that changes on every run and cannot be defended to an investment committee or a regulator.
That is the whole idea, and it is not a small one. Most "AI due diligence" today is a language model reading a deck and writing a confident paragraph back. Ask it the same question tomorrow and you may get a different answer — with no record of why either was produced. askOdin treats diligence the way an auditor treats a ledger: a fixed process, a cited source for every line, and a finding you can reconstruct exactly, on demand, long after the decision.
// DETERMINISTIC vs. PROBABILISTIC
The difference is not quality. It is physics.
A probabilistic model and a deterministic engine are not two grades of the same tool. They are two different operations. LLMs optimize for persuasion. askOdin compiles for physics. Here is where the gap shows up.
CLAIM 01
Reproducibility
Deterministic: the same recorded variables under the same Framework version return the same score, today or in eighteen months, and the record shows them.
Probabilistic LLM: the same prompt samples a new answer on every run. There is no canonical finding to defend.
CLAIM 02
Traceability
Deterministic: every conclusion traces to a cited source line and a named forensic dimension.
Probabilistic LLM: the model cannot tell you why it concluded what it concluded. The reasoning is a black box even to itself.
CLAIM 03
Defensibility
Deterministic: the output is a forensic record you can hand to an IC, an LP, or a regulator.
Probabilistic LLM: the output is persuasive prose. Persuasion is not a defense; it is the thing under audit.
CLAIM 04
Failure mode
Deterministic: when the math conflicts, the engine flags the conflict and floors the score. Silence is impossible.
Probabilistic LLM: optimizes for a fluent answer, so it smooths over contradictions rather than surfacing them.
Read the full comparison: deterministic vs. probabilistic AI →
// HOW askODIN COMPILES IT
Extract. Evaluate. Anchor.
Determinism is not a marketing claim; it is an architectural decision. askOdin separates the act of reading from the act of judging, so that the judging never depends on a model's mood. Three stages, in order.
Stage 01 · Extraction
The RUNE Protocol™ extracts.
RUNE compiles the messy, persuasive source — pitch deck, memo, data room — into structured, machine-checkable claims. This is the only stage where natural language is involved, and its job is narrow: turn prose into discrete, sourced assertions. Nothing is judged yet.
U.S. PATENT PENDING 63/948,559Stage 02 · Evaluation
A deterministic Go engine evaluates.
The extracted claims run through a fixed, rule-governed engine — not a language model — that tests them against five scored pillars and three audit checks and a benchmark corpus of 110,000+ Clarity Scores™ built on public deal data. The output is a Clarity Score on a 0–100 scale. Because the engine is deterministic, the same claims always resolve to the same score. There is no sampling, no temperature, no run-to-run drift.
Stage 03 · Anchoring
The Defensible Audit Log™ anchors it.
Every run is written to a sealed record bound to the SHA-256 fingerprint of the files read: the extracted variables, the Framework version, the checks applied, the evidence cited, and the verdict reached. Months later, you check the finding from that record — same recorded variables, same version, same score. Extraction runs on an external model and can change, so a past result is read from its record, not proven by hoping a fresh run matches. That is what makes the diligence defensible — not that it was confident, but that it can be shown.
Cross-document contradictions across a heterogeneous data room are handled separately by the RAVEN Protocol™, askOdin's adversarial triangulation layer. The temporal drift between what a company claimed last year and what it claims today is caught by the NORN Protocol™ (U.S. Prov. Patent No. 64/011,252). And the JUDGE Protocol™ acts as a runtime circuit breaker that floors the score the moment a structural conflict is detected — U.S. Prov. Patent No. 64/017,488. Four patent-pending protocols, one deterministic spine.
The architectural mechanics of RAVEN's triangulation engine are protected under U.S. Provisional Patent No. 63/994,876 and are not publicly disclosed.
// WHY IT MATTERS
Defensibility is the product.
Fiduciary defensibility. Fiduciary duty is not a standard of being right in hindsight — it is a standard of being able to show your work. When the math is run by a deterministic engine and cited line by line, the decision stands on a record, not on a partner's recollection of a good meeting.
Reproducibility. A finding you cannot reproduce is an opinion with a timestamp. A finding whose recorded variables return the same score under the same Framework version is evidence. That distinction is the entire difference between a probabilistic summary and a deterministic audit.
An audit trail you can show. When an LP, an auditor, or a regulator asks why capital moved, you read the analysis as it stood on the day of the decision. The Defensible Audit Log is built for precisely that question; whether it satisfies a given regulator is the fund's determination.
Venture capital is the last unaudited asset class. askOdin provides the infrastructure to close the gap.
See the category. Then deploy it.
THE CATEGORY
AI Judgment Infrastructure
Deterministic due diligence is one output of a larger category: the infrastructure that compiles judgment instead of summarizing it. See where the term sits.
Read the category →CLARITY FOR FUNDS
Deploy Deterministic Diligence
The enterprise deal-flow workspace for VCs, PE, and family offices. Forensic audit and IC-ready memos — every one on a Defensible Audit Log.
Request Deal Team Access →