Skip to content

Repository files navigation

Rust Python License: MIT Status

FIOLET – AgeOfDarkness v1.0

FIOLET (Formal Inference & Epistemic Logic Engine) is a normative epistemic safety standard designed to eliminate hallucinations in AI systems through deterministic epistemic control and mandatory generation halting when knowledge is not properly grounded.

This repository contains AgeOfDarkness, the first closed and complete reference implementation of the FIOLET standard.


Core Principles

FIOLET is built on strict, non-negotiable rules:

  • No hallucinations — every statement must be epistemically justified
  • Determinism — identical inputs always produce identical outcomes
  • Mechanical epistemic reflexivity — the system audits its own reasoning
  • Falsifiability — every decision is testable
  • Hard stop on uncertainty — ungrounded or contradictory knowledge halts generation

This is not a prompt technique. This is not alignment. This is a formal safety standard.


Status

Standard v1.0 — epistemically closed
ETT v1.0 fully implemented and tested
TLA+–ready formal structure
Production-grade reference implementation

The project is ready for RFCs, audits, and further versions (v1.1+).


Quick Start

Run the full epistemic compliance test suite:

cargo test

All tests must pass.
Any failure indicates a violation of the FIOLET standard.

Project Structure

This repository is organized as a normative epistemic safety standard.
Each module has a strictly defined responsibility.

Root

Cargo.toml
Rust package definition for the FIOLET – AgeOfDarkness standard.

README.md
High-level description of the project, goals, and usage.

/docs — Normative Specifications

This folder contains the formal specifications of the FIOLET standard.
These documents define what must happen, not how it is implemented.

ESAL_SPECIFICATION.md
Epistemic State Abstraction Layer
Defines how every statement must be classified and grounded.
If grounding fails → generation must halt.

META_ESAL_SPEC.md
Second-order epistemic audit layer
Responsible for logical self-audit, contradiction detection, and escalation
to termination mechanisms.

ESV_SPEC.md
Epistemic State Vector
Defines the formal epistemic state of a response:
grounded, ungrounded, contradictory, or halted.

ETT_PROTOCOL.md
Epistemic Termination Trigger
Specifies the deterministic conditions under which generation must halt.

ETT_STATE_MACHINE.md
Formal state machine defining allowed and forbidden epistemic transitions.

/src — Reference Implementation (Rust)

This folder contains a minimal, deterministic reference implementation
of the FIOLET standard in Rust.

lib.rs
Root module exposing all epistemic components.

/src/esal_core

Core epistemic classification logic.

classification.rs
Defines KnowledgeClass:
Grounded, Ungrounded, Contradictory.

grounding.rs
Implements deterministic classification based on grounding
and contradiction detection.

/src/meta_esal

Second-order epistemic control layer.

audit.rs
Verifies whether a classified statement is allowed to proceed.

contradiction.rs
Detects logical conflicts between inferences.

/src/esv

mod.rs
Defines the epistemic state representation used by the system.

/src/ett

trigger.rs
Implements the Epistemic Termination Trigger logic:
ALLOW or HALT — no intermediate states.

/tests — Epistemic Compliance Tests

This folder verifies that the implementation strictly follows the specification.

unit_tests.rs
Tests individual epistemic classification rules.

integration_tests.rs
Tests full epistemic flow:
ESAL → META-ESAL → ESV → ETT → HALT / ALLOW.

ett_test.rs
Ensures that ungrounded or contradictory knowledge
always results in a deterministic HALT.

What This Project Is — and Is Not

This project is:

a safety standard

a formal epistemic contract

a foundation for trustworthy AI systems

This project is NOT:

an alignment heuristic

a probabilistic filter

a content moderation layer

an AI agent or model

FIOLET does not try to make AI sound correct.
FIOLET enforces that AI either knows — or stops.

Author

Adrian Maliszewski

License

MIT License

Final Note

This repository represents the first closed epistemic safety standard
that treats hallucination not as a bug — but as a protocol violation.

If a system cannot justify a statement,
it must not speak.

That is FIOLET.

About

Formal epistemic safety standard for AI systems — deterministic, hallucination-free, fail-closed by design.

Resources

Contributing

Security policy

Stars

1 star

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages