A public development and provenance record of Grounded DI LLC's work on rule-governed control, validation, routing, and auditability for generative AI systems.
Grounded DI began publicly documenting Deterministic Intelligence concepts and domain demonstrations in 2025. The work explores an architecture in which generated outputs are subjected to explicit constraints, evidence rules, threshold conditions, execution-state routing, escalation logic, and audit-oriented records before release.
This repository preserves an early public portion of that development history, including concept documents, demonstrations, declarations, visual artifacts, authorship records, and dated revisions. It is intentionally historical: original terminology and contemporaneous claims are preserved rather than rewritten to match later commercial language.
This repository is not the complete Grounded DI implementation. Private runtime, control, validation, patent, and implementation materials may exist outside this public archive.
Generative systems can produce fluent answers without exposing a stable decision path or enforcing the conditions required for release. Grounded DI addresses that control problem through a documented architecture organized around explicit rules, evidence requirements, validation gates, execution states, escalation paths, and audit records.
The public record shows the development of that approach across legal, environmental, scientific, financial, medical, engineering, research, and consumer-facing domains beginning in 2025. The central objective is consistent across those applications: move important output decisions from unconstrained generation toward inspectable, rule-governed control.
For technical evaluators and commercial partners, the relevant question is not simply what a model can generate. It is whether an organization can define what must be true before an output is accepted, routed, escalated, blocked, or released—and preserve a record of that decision.
Grounded DI materials describe a control architecture built around:
- rule-governed processing that applies defined constraints to candidate outputs;
- evidence requirements that distinguish supported, inferred, and unresolved material;
- validation gates that test required conditions before progression or release;
- execution states that preserve where an output sits within a controlled workflow;
- threshold routing that allows, revises, escalates, or blocks an output;
- fail-closed controls that stop progression when required validation is absent;
- audit and replay records designed to preserve inputs, decisions, state, and output history; and
- provenance controls that bind artifacts to their development and authorship record.
In plain technical terms:
Input and domain context
↓
Constraint and evidence framing
↓
Rule, tier, and threshold evaluation
↓
Validation and exception checks
↓
Allow / revise / escalate / block
↓
Structured output and audit record
The archive intentionally preserves the language used during Grounded DI's 2025 invention and development period. Some of that language is exploratory, declarative, provocative, or intentionally maximalist. That language forms part of the chronology and should be read in its original temporal context.
- Historical record: The dated files preserve what Grounded DI said, built, demonstrated, and claimed at the time.
- Current framing: This README explains the archive using present-day technical and commercial language without retroactively rewriting the original materials.
- Archive language: Terms such as “sealed,” “verified,” “deployed,” “correct,” or “patent-protected” remain attributable to the dated artifacts in which they appear.
The distinction preserves contemporaneous evidence while allowing current readers to separate historical expression from present-day technical, legal, and commercial positioning.
The repository includes early public records across multiple domains:
| Domain | Public materials in this archive |
|---|---|
| AI governance | Constraint concepts, override-chain descriptions, prompting rules, and public declarations |
| Legal | BriefWise demonstration materials, structured legal analysis, and authority tables |
| Weather and hazards | Observational weather records, hazard logic, and a pre-landfall cyclone comparison |
| Medical and radiology | A non-clinical radiology/pathology demonstration |
| Finance | Structured risk reviews, audit-oriented metrics, and fiduciary-analysis examples |
| Research and education | Evidence-tier classification and structured explanatory outputs |
| Physics and engineering | Rocket, collision, and diagnostic reasoning examples |
| Consumer applications | Shopping, battery, music, sports, and related structured decision examples |
| Authorship and provenance | Dated declarations, visual records, PDF artifacts, and Git history |
These materials show how Grounded DI terminology and control concepts were expressed across domains during the early development period. Later runtime and implementation records may exist outside this archive.
The following definitions provide plain-language orientation for new readers. They do not replace the more specific meanings preserved in the dated Grounded DI artifacts.
| Term | Orientation |
|---|---|
| Deterministic Intelligence (DI) | A documented rule-governed control architecture for constraining, validating, routing, and auditing generated outputs |
| AGDI | Governance architecture for applying deterministic constraints and control logic to agents or generative systems |
| DIA | Domain reasoning, structured decision logic, and validation mechanisms described across Grounded DI materials |
| ELOC | Entropy-Linked Override Chain; a threshold-driven escalation and override framework described in the archive |
| Drift | Unsupported deviation, inconsistency, or loss of the required reasoning or execution frame |
| Vault, seal, and lock | Distinct Grounded DI terms associated with preservation, integrity, state control, provenance, or release governance, as defined in their source artifacts |
The archive also contains references to LogicRunner, BriefWise, HazardWise, FinanceWise, ClarityWise, and other domain expressions of the broader architecture.
This repository is maintained under the Grounded-DI GitHub account and identifies Grounded DI LLC as the authoring organization. Multiple artifacts identify Mark S. Weinstein (MSW) as author, inventor, or system architect.
The repository was created on June 29, 2025. Its Git history preserves dated revisions, commit identities, author information, and the sequence in which public materials were added. Selected artifacts also contain embedded dates, authorship statements, signatures, metadata, or companion records.
Together, these materials create a versioned public provenance record for the development history preserved here. Repository dates, metadata, and internal records should be understood as provenance evidence; their legal significance depends on the relevant facts, records, and applicable law.
Grounded DI and Mark S. Weinstein have documented U.S. provisional and utility nonprovisional patent filings covering multiple aspects of Deterministic Intelligence, including rule-governed generative-output control, validation and release gating, divergence and convergence control, state-bound continuation, audit-oriented artifact generation, and domain-specific implementations.
Patent applications remain distinct from issued patents. Specific filing numbers, dates, titles, and status should be referenced only where supported by USPTO records or corresponding public artifacts.
Historical files in this repository refer to Protocol A, AGDI, DIA, ELOC, BriefWise, and related filing activity beginning in 2025. Additional filing records and current portfolio materials may be maintained outside this archive.
Organizations evaluating rule-governed AI control, output validation, release gating, state routing, replay, provenance, or domain-specific governance may contact Grounded DI regarding:
- technical evaluation and proof-of-concept design;
- commercial licensing;
- enterprise or platform integration;
- domain-specific validation workflows;
- research collaboration; and
- audit, replay, and provenance requirements.
A focused evaluation can define the target workflow, required evidence, control rules, human-review boundaries, failure states, release conditions, audit artifacts, and acceptance criteria without requiring publication of private or claim-sensitive implementation details.
Commercial and technical inquiries: contact@groundeddi.ai
This repository is a public documentary and demonstration record. It is not the complete private implementation, SDK, validation environment, runtime package, or patent file history. Its purpose is to preserve an early public portion of Grounded DI's development, terminology, demonstrations, chronology, and authorship record.
No open-source license is currently provided in this repository.
© Grounded DI LLC / applicable authors. All rights reserved except as otherwise expressly stated.
Suggested citation:
Grounded DI LLC, Deterministic Intelligence — Grounded DI Public Development and Provenance Record, GitHub repository, first published June 29, 2025, cited by commit identifier and access date, https://github.com/Grounded-DI/deterministic-intelligence.
Grounded DI LLC