Home How It Works Research Security Pricing Partners FAQ About Contact
Sign In Start Free Review
Back to home

The Research Behind TriageCounsel

Every legal finding in TriageCounsel is backed by peer-reviewed deterministic methods. Our architecture was accepted for presentation at the International Conference on Computational Science (ICCS 2026) and published in Springer LNCS.

Presented at ICCS 2026
Published in Springer LNCS
Peer Reviewed
Independent Research

Deterministic Execution Frameworks for Hybrid Symbolic–Probabilistic Computational Pipelines

Santhosh Guntupalli

International Conference on Computational Science (ICCS 2026) · Springer Lecture Notes in Computer Science (LNCS)

Abstract. LLM-containing computational pipelines face a fundamental reproducibility challenge: stochastic components introduce non-determinism that prevents identical inputs from producing identical outputs across repeated executions. This paper presents a deterministic execution framework for hybrid symbolic–probabilistic pipelines that enforces execution invariance by isolating deterministic modules from stochastic components. The architecture employs a deterministic symbolic engine for all state transitions and decision logic, while LLM components operate only as non-authoritative, post-hoc explainers of pre-computed deterministic outputs. We evaluate the framework on a corpus of 115 structured text documents, demonstrating 100% execution determinism and 100% traceability across 300 repeated runs with zero output variance.

DOI and Springer page will be available upon publication.

Published benchmarks from the ICCS 2026 paper

100%
Deterministic Output
0%
Pure LLM Determinism
300
Repeated Executions
100%
Traceability Rate

Hybrid System vs. Pure LLM Baseline

Metric TriageCounsel (Hybrid) Pure LLM
Determinism Rate 100.0% 0.0%
Traceability Rate 100.0% 0.0%
Ungrounded Outputs / Doc 0.00 0.43
Output Variance (300 runs) Zero 18.3 distinct sets/doc
Avg. Execution Time / Doc 0.41s 7.3s
Documents Evaluated 115 115

How the deterministic framework works

The system enforces a strict separation between deterministic risk detection and probabilistic language generation. The LLM never determines legal risk — it only explains deterministic findings in plain language.

Execution Pipeline

1
Document Ingestion & Normalization
2
Deterministic Symbolic Engine
3
Context Logic & Suppression
4
Deterministic Output (Structured State)
5
LLM Explanation Layer (constrained, non-authoritative)
Attorney Review

Key Design Principles

Deterministic-First

All risk detection is performed by versioned, rule-based logic. No probabilistic inference, no statistical weighting, no machine-learned parameters.

LLM Lockdown

The LLM never sees the full contract. It receives only pre-validated deterministic findings and generates explanations from those.

Clause Anchoring

Every finding is tied to exact character offsets and textual excerpts. Reviewers can independently verify each risk indicator.

Safe Degradation

If the LLM is unavailable or fails, all deterministic findings remain intact and actionable. The LLM is never a single point of failure.

Why deterministic execution matters for legal analysis

Pure LLM Systems

Different results on every run for the same document
Can hallucinate clauses that don't exist in the contract
No verifiable reasoning path for auditors
Cannot defend past outputs when challenged

TriageCounsel (Deterministic Hybrid)

Identical outputs guaranteed for identical inputs
Zero hallucinated findings — every risk linked to source text
Every decision traceable to a versioned rule ID
Historical analyses remain auditable as rules evolve

Frequently asked questions

What is ICCS?

The International Conference on Computational Science (ICCS) is one of the leading conferences in computational science, published through Springer's Lecture Notes in Computer Science (LNCS) series. ICCS 2026 covers advances in computational methods across scientific and engineering domains.

What does "deterministic" mean in this context?

Deterministic means the same input always produces the same output. When you upload the same contract twice, TriageCounsel produces identical findings every time. Pure LLM systems cannot guarantee this — they produce different results on different runs, even with identical inputs.

Does TriageCounsel use AI at all?

Yes. TriageCounsel uses AI (LLM) to generate plain-language explanations of findings. However, the AI never decides what is risky — that is done entirely by the deterministic rule engine. The AI only explains what the deterministic engine has already found.

How were the 300 repeated runs conducted?

15 documents (8 public, 7 synthetic) were each analyzed 20 times across multiple random seeds and computational environments. The hybrid system produced zero output variances across all 300 runs. The pure LLM baseline showed output differences in 100% of cases, with an average of 18.3 distinct output sets per document.

Can I read the full paper?

The paper will be available through Springer LNCS upon publication. DOI and direct links will be posted here once available.

See the research in action

Upload a contract and see deterministic, evidence-backed findings — the same method validated in our ICCS 2026 paper.

Start Free Review