Science Built Technologies research hub
This site collects public-safe research notes on epistemic reliability, reasoning stability, uncertainty integrity, replay accountability, and trustworthy AI evaluation.
Start Here
ERT is an experimental research effort connected to Science Built Technologies and Project Aletheia. It studies whether AI reasoning remains coherent, appropriately uncertain, and accountable when prompts, contexts, or evaluation conditions change.
Some implementation details are intentionally redacted to protect IP and prevent unsafe disclosure. Redaction markers are visible on purpose.
ERT Overview
Begin with the minimal public definition, then move through ERC, engineering hardening, reporting, governance, and current implementation notes.
Read the ERT overviewReports / Research Logs
The reporting pages explain public-safe report boundaries, trace viewing, calibration, client-readable summaries, and replay accountability without exposing protected scoring or operational details.
Read the reporting pageCurrent Work
The current-state page explains continuing work around diagnostic signals, UI refinement, client-friendly reports, ERT-Lite demonstrations, calibration workflows, and longitudinal reliability testing.
See current workResearch Progression
These pages follow the public research arc from the minimal ERT definition through current implementation and continuing work.
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...