← Back to Patent Claims
Patent Claim 031 All Patents →

Deterministic LLM Verdict with Numeric Drift Guard

Bounding a language model's numeric output by an independently-computed deterministic floor so a civil-evidentiary verdict cannot drift beyond a configured tolerance

Patent Claim JIL Sovereign July 2026 Claim 31 of 157

Executive Summary

A language model is a probabilistic system wearing a deterministic-looking interface. Successive invocations with identical input can still produce different numeric extractions, different formatting, or in the worst case a different verdict - which is fatal for any compliance output meant to be independently reproduced by a third party. JIL Sovereign's deterministic verdict module does not try to make the model itself perfectly deterministic; instead, it bounds whatever the model produces against a floor computed entirely without the model.

The system invokes the language model under a fully-hashed fixed configuration - temperature zero, a fixed seed, a fixed system-prompt hash, a fixed model-identifier hash. Its response is parsed against a strict structured schema; every numeric field in that schema is compared against a deterministic floor independently computed by a non-language-model process (typically a direct database aggregation) over the same input data. If a field falls outside a configured tolerance of its floor, or parsing fails outright, the invocation is retried under the identical configuration up to a bounded count. If every retry fails, the verdict falls back to the deterministic floor with the language-model contribution explicitly marked as suppressed - a fallback that is itself a fully valid attestation product.

Core Innovation: Numeric output from a language model is never trusted on its own - it is validated against an independently-computed, non-LLM deterministic floor within a configured tolerance, with bounded retries and a documented fallback, so the verdict's numeric content cannot drift beyond a known, reproducible bound.

Problem Statement

Structured-output features (temperature zero, JSON-schema-constrained generation, fixed seeds) reduce a language model's stochasticity but do not eliminate the specific failure mode that matters for compliance verdicts: numeric drift. A model might correctly recognize a value but format it differently across runs, or - more seriously - emit a number that flatly contradicts ground truth already established by a deterministic query over the same data. Retry-on-parse-failure patterns common in LLM application frameworks catch schema violations but do nothing to catch a syntactically valid number that is simply wrong relative to an independently-known fact.

Self-consistency decoding (sample many times, take the majority) and critic-model verification (a second model checks the first) both attempt to improve reliability, but neither anchors the result to a deterministic, non-probabilistic ground truth, and neither produces a signed, third-party-replayable manifest of the invocation. For a verdict intended to survive adversarial replay under Claim 26's reproducibility manifest, the numeric portion needs a bound that does not itself depend on the model behaving consistently.

Temp 0
Required sampling temperature for every gated invocation
3
Typical bounded retry count before fallback
100%
Of numeric schema fields checked against a non-LLM floor

Why Existing Approaches Fall Short

  • Structured output / schema validation alone: catches parse failures, not numerically-wrong-but-well-formed values.
  • LangChain-style retry parsers: retry on parse failure only; no drift guard against an independent ground truth.
  • Self-consistency decoding: majority-vote across N samples is still purely probabilistic and not bound to deterministic ground truth.
  • Critic / verify-then-generate models: the verifier is itself an LLM, not a deterministic non-LLM check, and produces no signed replay manifest.

Technical Architecture

Pipeline Components

ComponentFunction
Configuration loaderReads provider, model, version, endpoint, temperature (required zero), fixed seed, and system-prompt text; hashes the full configuration.
Invocation engineEmits the request under the loaded configuration; captures the verbatim response.
Schema parserStrictly parses the response against a structured schema with named numeric, enumerated, and string fields; any absent or mistyped field is a parse failure.
Deterministic-floor computerComputes, per numeric field, a floor via a non-LLM process over the same input data - typically a structured-query aggregation.
Tolerance evaluatorCompares each parsed numeric field to its floor against a configured absolute or relative tolerance.
Retry controllerRepeats the invocation, under identical configuration, up to a bounded count on parse failure or tolerance violation.
Fallback moduleProduces a verdict from the deterministic floor with a suppressed-LLM-contribution annotation if the retry bound is exhausted.
Manifest writerRecords configuration hash, schema hash, floor results, tolerance spec, retry log, and outputs into the verdict-replay manifest.

Retry and Fallback Logic

Each numeric field's schema declaration specifies whether the deterministic floor acts as a lower bound (model output must be at or above) or an upper bound (model output must be at or below), and the tolerance may be expressed as an absolute bound, a relative fraction, or a combination. A bounded retry count - typically three, capped at five in the preferred embodiment - exists specifically to catch transient stochastic anomalies that survive even a temperature-zero, fixed-seed configuration.

Composability

The configuration hash, schema hash, floor results, and retry log all feed directly into the seven-hash-class reproducibility manifest of Claim 26, and the model identity used in each invocation is separately sealed under Claim 32's model-card mechanism - together the three claims give a language-model-mediated verdict the same replay guarantees as a fully deterministic one.

Composable with Claims 26 and 32: this claim's deterministic-floor hashes populate Claim 26's manifest directly, and every invocation gated by this mechanism is separately sealed by Claim 32's model-card system.

Prior Art Differentiation

ApproachBounds numeric output vs. ground truthBounded retryDocumented fallbackReplayable manifest
Structured output (temp=0, schema)NoNoNoNo
LangChain retry parsersNoYes (parse only)NoNo
Self-consistency decodingNo (majority vote only)NoNoNo
Critic / verify-then-generateVerifier is itself an LLMVariesNoNo
JIL SovereignYes, non-LLM floorYesYes, suppressed-contribution flagYes

The defensible combination is the deterministic-floor-plus-tolerance-evaluator pairing: no prior art bounds language-model numeric output against an independently-computed, non-LLM floor with a documented, still-valid fallback verdict when the bound cannot be met.

Patent Claim

Independent Claim 31: A deterministic language-model verdict system, comprising: a configuration loader configured to read at least a language-model provider identifier, model identifier, model version, endpoint, a sampling temperature equal to zero, a fixed sampling seed, and a system-prompt text, and to compute a configuration hash over the foregoing; an invocation engine configured to emit a request under the loaded configuration and capture a verbatim response; a schema parser configured to parse the response against a structured schema specifying named numeric, enumerated, and string fields, wherein absence or mistyping of any required field is a parse failure; a deterministic-floor computer configured to compute, for each numeric field, a floor via a non-language-model computation over the same input data as the invocation; a tolerance evaluator configured to compare each parsed numeric field against its floor per a configured absolute or relative tolerance; a retry controller configured to repeat the invocation, under the identical configuration, up to a bounded retry count upon parse failure or tolerance violation; and a fallback module configured to produce a verdict from the deterministic floor, annotated with a suppressed-language-model-contribution indicator, if the retry bound is exhausted; wherein the resulting verdict, whether language-model-mediated or fallback, is recorded together with the configuration hash and floor results in a manifest supporting independent third-party replay.