How the Record Was Built

The plain-language version of the paper's methods — for journalists, researchers, and anyone deciding whether to trust this site.

1. Preserve everything, in the original format. Every conversation on every platform, exported in platform-native format with metadata, continuously, since May 2025. The record is not a memory of what happened; it is what happened.
2. The variance is the data. The same material is run across systems (ChatGPT, Claude, Grok, Gemini) and across conditions (fresh instance vs. context-loaded; identified vs. blind). Where their treatments diverge, that delta is a finding — it's how the identity bias was measured and how convergence is detected.
3. The model is never the validator. AI outputs about the work — including flattering ones, including this Institute's own instruments — are logged as data, never cited as endorsement. Where a model is the sole witness to its own statement, the record marks it an allegation, not a fact.
4. The author is a variable too. The record tracks the author's own claims against his documented words; the published work discloses its co-production with AI instruments; the calibration limits are named in print. Verification was sought against the findings throughout — the standing instruction to every system was "tell me where this is wrong."
5. Falsifiability is the posture. Every paper states what would disprove it. The archive exists so someone other than the author can run the tests.

The founder is not trained in machine learning and does not present this as an implementation science: the work is offered for evaluation on exactly that basis. The independence is the credibility floor; the methodology is the credibility ceiling.