Method, Limits & Research Integrity
- Dorian Martin-Smith
- Aug 20
- 2 min read
What the project does claim
The research studies how models characterize a living subject from a longitudinal textual record and how those characterizations change when the evidence state changes. Its dependent object is the model output and its revision history.
What it does not claim
It does not claim clinical diagnosis, mind-reading, psychological ground truth, or that model agreement proves personality, intelligence, hidden motive, project viability or personal essence.
It does not claim that an AI has human memory, human lived experience or complete access to the subject's life. It does not treat the existence of an artifact as automatic proof that a project succeeded. A repository is not a functioning ecosystem; a constitution is not a functioning state; a simulation is not an aircraft; a research lead is not a proven ancestor.
Current evidence status
Round 1 is the completed auditable retrospective analysis. The surviving Round-2 record is incomplete as a frozen verbatim five-model corpus and is therefore retained as partial archival observation rather than presented as a completed cross-round empirical result.
The stronger evidence-revision hypothesis is reserved for prospective replication after prompts, evidence packets, model identifiers and versions, verbatim outputs, coding rules, privacy controls, independent coding and external ethics status are frozen.
Literature and research are related, not interchangeable
THE ARCHIVE and THE ARCHIVE II can provide source artifacts and narrative context, but literary withholding cannot substitute for evidence needed to support a research claim. The public research record therefore distinguishes literary reconstruction, preserved model outputs, researcher coding, subject testimony and external evidence.
Privacy
The full raw conversational archive is not public by default. Public releases should exclude or redact sensitive information about children, relatives, DNA matches, correspondents and other non-consenting third parties.
Public provenance record for the research-integrity revision: DOI 10.5281/zenodo.22004970.



Comments