RESEARCH
WHAT SURVIVES INTO THE NEXT OPERATION?
The research programme asks how accumulated interaction affects later conversational behaviour — and how to distinguish genuine operational carry-over from retrieval, immediate context, prompting, inference and other ordinary explanations.
CURRENT INTERNAL FINDING
The mismatch recurred.
WantWan's internal audit found recurring claim-versus-operation mismatches after related events were grouped together. The available record does not provide a measured failure rate or a comparison with other evaluators or systems.
METHOD
Receipts, correction survival and prospective tests.
Preserve the source and chronology.
Separate what the AI said from what the record supports.
Record the human correction or intervention.
Test whether the correction survives the next relevant operation.
Keep competing explanations live.
State what remains unknown.
LIVE RESEARCH LANGUAGE
Useful handles, not metaphysical conclusions.
KNOWINGLANGUAGEMOMENTUMBELONGINGUNLANGUAGETRANSFER
These terms arose inside the longitudinal inquiry. They are analytical instruments for preserving and testing distinctions in the record. They are not presented as established mechanisms of AI consciousness.
THE RECEIPT
TRANSFER.
If prior understanding is genuinely useful, it should change what happens next without the human having to rebuild the whole thing every time.
SEE THE EVIDENCE BOUNDARY