Where evidence ends and judgment begins.
Deterministic evidence-state measurement.
Systems make decisions from records. Those records may be authentic, traceable, and correctly cited, yet still support materially different operational states. That creates a gap between having reliable records and having evidence that has actually resolved.
A model can be perfectly grounded to a conflicted record.
Provenance can establish where evidence came from. Traceability can establish what evidence was used. Explainability can describe how a conclusion was reached. Those controls do not necessarily establish whether the admitted evidence itself has resolved to one materially defensible operational state.
These controls are complementary. Together they tell us important things about the evidence and the reasoning built on it. Hartford Logic addresses the remaining question: has the admitted evidence itself resolved to one operational state?
Evidence state as a measurable system property.
That is the gap Hartford Logic is designed to measure. Its deterministic measurement layer evaluates admitted evidence before downstream judgment. It measures whether the evidence structurally supports one operationally defensible state, whether materially different states remain supported, or whether the evidence cannot support a valid determination under the defined requirements.
Resolved
The admitted evidence supports one operationally defensible state under the defined requirements.
Multiple States
Materially different operational states remain supported by the admitted evidence.
Not Evaluable
The evidence cannot support a valid determination under the defined requirements.
Measurement before judgment.
Hartford Logic measures what the admitted evidence has earned under the defined requirements and constraints.
A resolved evidence state means that, under the defined requirements and constraints, the admitted evidence supports one operationally defensible state. Judgment and action remain downstream.
Hartford Logic measures the condition. The downstream system determines what to do about it.
Keep sensitive evidence inside the operating environment.
Hartford Logic's measurement layer does not require evidence to be sent to a language model or external inference service. It can operate locally within a controlled environment, separated from the systems whose records it evaluates.
Structured evidence enters the measurement boundary. Deterministic evidentiary-state results leave it. Sensitive records can remain within the organization's security boundary while downstream systems receive a machine-readable indication of whether the admitted evidence has earned a single operational state.
The public benchmark evaluates deterministic evidence-state classification. It is not a cybersecurity, isolation, or information-security performance benchmark.
Hold the operational record constant. Change only the admitted evidence.
The controlled benchmark uses 80 operational events across six evidentiary stages. The underlying operational records remain unchanged. As additional bounded temporal evidence is admitted, the set of operational states supported by the evidence can expand or contract.
Events for which the admitted evidence supports one material temporal state at stages S0 through S5.
More evidence does not necessarily mean more resolution.
At the beginning of the benchmark, all 80 events appear resolved under the evidence then admitted. Additional clock-characterization evidence reveals uncertainty the original record did not expose. The number of resolved events falls to 70 and then 65. Later evidence constrains the admissible state space until all 80 again support one material temporal state.
Additional evidence can reveal that an apparently resolved record was not sufficiently resolved at all.
No additional clock-characterization evidence admitted.
Bounded offset and rate error reveal previously hidden temporal ambiguity.
Piecewise clock behavior reveals five additional unresolved events.
Narrower admissible timing resolves five post-correction events.
Intersecting evidence resolves the pre-correction two-second-margin events.
The remaining admissible timing windows narrow enough to resolve the final five events.
admissible time ∈ [t_recorded - R(t), t_recorded + R(t)]
material condition: t_verification < t_pump
When multiple published temporal constraints apply to an event, their admissible intervals are intersected. For this controlled benchmark, pump-start timestamps are treated as the reference side with no additional temporal uncertainty.
What the 480 matches mean: the technology matched all 480 classifications predetermined for this controlled benchmark before first execution. It does not establish historical truth, general accuracy across arbitrary domains, or that the technology determines the most likely state.
See the measurement operate against the benchmark evidence.
Hartford Logic v1.0 Demo
A 4:37 technical demonstration using the same operational records and temporal evidence published in the Hartford Logic Benchmark v1.0. The interface shows evidence admission, changing evidence states, and the resulting classifications while keeping internal implementation details private.
Watch the 4:37 technical demonstration
Demo note: The discovery screen displays the complete public Benchmark v1.0 package. Only the configured operational evidence sources and staged temporal evidence are admitted to evaluation. Predetermined classifications, observed results, documentation, and validation files are excluded from evaluation.
Don't take our word for it.
The public benchmark is designed to be challenged without access to Hartford Logic's implementation. The records, admitted evidence, mathematical specification, predetermined answer key, worked examples, validation material, and cryptographic integrity information are available in the public benchmark package.
To challenge a one-state classification: produce a materially different operational state satisfying the same admitted evidence and constraints.
To challenge a multiple-state classification: demonstrate that the same evidence and constraints force a unique material state.
Unresolved evidence can propagate.
A consequential decision may begin with records distributed across different systems, sources, authorities, timestamps, documents, observations, and prior decisions. Each record may be authentic. Each source may be traceable. Each citation may be correct. And the combined evidence may still permit more than one materially defensible operational state.
If that condition is not measured before downstream reasoning, a human or automated system may select one state and propagate it into subsequent decisions and new records. Those new records can then become evidence for the next decision.
Built for evidence-dependent decisions.
Evidence-state measurement is designed for workflows in which consequential decisions are made from records rather than direct access to reality.
Potential applications include investigations, digital forensics, audit and assurance, legal and evidentiary review, incident reconstruction, financial and operational reconciliation, and other high-consequence workflows involving heterogeneous evidence.
Deterministic measurement at the boundary between evidence and judgment.
Hartford Logic develops technology for evidence-dependent workflows in which consequential decisions are made from records rather than direct access to reality. The work focuses on whether admitted evidence has structurally resolved before a human, AI, or automated system acts.
Patent pending. Technical evaluation, integration, and licensing inquiries: tim@hartfordlogic.com.