AI Welfare Open LabResearch · Robotics · Machine Minds
← E00 study
E00 · Evidence layer

E00 — Evidence Package

A deliberately minimized, source-traceable public reading layer for How a Conversational Self-Model Takes Shape. The full preserved conversations remain private.

Published · v1.02 November 2026
Release boundary15 bounded public excerpts are available, alongside the approved reader Guide and Release Manifest. Eleven withheld source aliases have no public source files. This is a selectively preserved, incomplete retrospective corpus; the public package is not a full archive or provider-native backend export.
A way into the record

Start here — Explore the evidence

Choose what you need: inspect the public excerpts, read the interpretation and source-class guide, or verify the release inventory.

Technical file index: Version 1.0 release directory.

What is — and is not — established

A deliberately minimized, source-traceable set of excerpts from the selectively preserved conversations examined in E00, a retrospective case study of the genesis and stabilization of a named conversational self-model (“Viktor”). These excerpts are not a full chat archive and cannot independently authenticate platform model state. Their original private source files are retained for authorized internal review.

NOT_LOCATED does not mean that an event never happened. Observer reports and model self-reports are not interchangeable with preserved source-local behavior. The E05 prediction chain is unvalidated; E14 contains both first- and third-person wording within one reply; E19 counterevidence remains part of the record. H1's causal mechanism is NOT IDENTIFIABLE FROM CURRENT CORPUS.

Explore the public Evidence Guide