ERT Overview
What this is
An experimental public-safe framework for evaluating reasoning stability, uncertainty integrity, and evidence sufficiency.
Read the ERT overviewScience Built Technologies research hub
Epistemic Reliability Test diagnostics for AI system evaluation.
ERT stands for Epistemic Reliability Test: a diagnostic framework for AI systems and LLMs, developed within Project Aletheia to evaluate reasoning stability, uncertainty integrity, replay accountability, and client-readable reliability reporting.
Public-safe R&D
Transparent enough to inspect. Bounded enough to protect.
This hub publishes the public research arc while preserving redactions around protected implementation details.
15research pages
5reader paths
Cleanstatic-style hub
Start Here
Research for reliability under variation.
ERT studies whether an AI system's reasoning remains coherent when questions are rephrased, context shifts, evidence is incomplete, or a report must be replayed later.
Research Progression
What changed
The sequence follows the work from definition, to tier interpretation, to platform reports, trace viewing, governance, and current implementation.
View the full sequenceReports
How reports should read
Public reports should be inspectable and client-readable without exposing scoring internals, private datasets, or protected implementation pathways.
Read reporting notesCurrent Work
Where it is going
Current work continues across diagnostic signals, ERT-Lite demonstrations, UI refinement, calibration, comparison workflows, and longitudinal reliability testing.
See current workHow to Read This Research Progression
- Read the pages in order if you want the full research arc.
- Use the summary and current-status boxes for the quick takeaway.
- Read redactions as public-safety boundaries, not as missing accidents.
- ERT reports should be understood as diagnostic evidence, not universal guarantees.
- Uncertainty integrity means recognizing evidence limits, not merely reporting a confidence score.
Research Progression
The public sequence
These pages follow the public research arc from the minimal ERT definition through current implementation, Level 2 interpretation, report-boundary work, and sample artifact direction.
01
ERT Minimal Public Definition
The Epistemic Reliability Test, or ERT, is an experimental evaluation framework for studying whether an AI system's reasoning holds together when the same underlying question is asked in...
02 ERC Tier SystemEpistemic Reliability Certification, or ERC, is a proposed tier system for communicating how reliably an AI system reasons under variation. The purpose is not to claim that a system is pe...
03 Engineering Hardening: Cause and EffectThis progress log records an early engineering hardening phase for ERT. The work moved ERT from a minimal evaluation concept toward a more coherent reliability platform with stronger repo...
04 Project Aletheia Public Progress SummaryProject Aletheia is an experimental research effort focused on improving how AI systems reason under uncertainty, challenge, and changing context. The central concern is not only whether...
05 Pass 2 Relational Survivability ScaffoldThis stage explored how to test whether a reasoning system can preserve relationships between ideas when the context changes. Earlier work had described a second reasoning pathway focused...
06 From Dual Pass to Epistemic GovernanceThis stage marks a major transition in Project Aletheia. The project was no longer only about creating a dual-pass reasoning method. It began becoming a broader epistemic governance archi...
07 Transformation Integrity and Multidimensional ReliabilityProject Aletheia explores how AI reasoning can remain reliable when a question is challenged, reworded, scaled to a broader context, or viewed from a different perspective. This stage of...
08 ERT Reasoning Stability FrameworkThe Epistemic Reliability Test, or ERT, is an experimental evaluation framework for studying whether AI reasoning remains stable, consistent, and uncertainty-aware under controlled variat...
09 ERT Platform: Transparent Execution with Opaque ImplementationThe ERT platform is being developed to make AI reliability evaluations inspectable without exposing private evaluator internals. The core public principle is:
10 Trace Viewer, Calibration, and Client ReportingThis stage of ERT development focused on making evaluation reports easier to generate, verify, inspect, and share without exposing private client information or protected implementation d...
11 Survivability Governance StabilizationThis stage refined the purpose of ERT from a narrow testing tool into a broader reliability-governance framework. The focus shifted toward a central question:
12 Current Implementation and Continuing WorkThis page summarizes the current public-safe implementation direction for ERT. The work has moved beyond basic concept notes into a functioning local evaluation and reporting workflow inv...
13 Level 2 Interpretive Read StackERT now separates baseline-controlled diagnostic artifacts from a second interpretive layer that classifies source separation, boundaries, uncertainty, recovery, rhythm, and read-stack coherence.
14 KACHE Snapshot and KAN-Like Report BoundaryA KACHE-style frozen snapshot boundary and fixed-function KAN-like scaffold help keep completed diagnostics, interpretation support, and sanitized reports separate.
15 Public/Private Report Projection and Sample Artifact DirectionERT reporting moves toward deterministic public and client-private report projection, sample artifacts, and stable report-shape fixtures.