Repository navigation
Sessions and Protocols
AI RV Harness separates ordinary Conversations from Remote Viewing sessions. They use different Workspace types and different runtime rules.
Conversations are persistent chat threads. You can add supported files and images, archive threads and watch the estimated context usage.
Viewer Learning is enabled by default when the selected AI identity has a completed learning package. It adds the current Field Guide and Viewer Notes as read-only context. You can turn it off for an individual Conversation. Conversation never updates either learning object.
RV Sessions support:
- Manual RV, where the human acts as Monitor;
- Automatic RV, where the application controls the protocol flow;
- Automatic RV with an AI Monitor;
- Full RCP, RV Lite, Telepathic and supported Custom Protocol paths.
Manual RV does not silently inherit the Conversation Viewer Learning switch. Each session freezes its own target, Reveal, protocol, identity and learning inputs so Resume uses the same evidence.
Custom Protocol creation uses a visible, scrollable multi-step editor. A custom protocol can be loaded from a file or created in the application. Its blind-session instructions remain separate from post-Reveal review.
If you assign labels such as Subject A, Object B or a custom task, the Reveal must explain what those labels meant. Otherwise the post-Reveal comparison cannot evaluate the requested task reliably.
The sealed blind record is never rewritten after Reveal. The shared post-Reveal layer may perform:
- Viewer review;
- Monitor review when the flow uses a Monitor;
- Training-only Field Guide update;
- Training-only Viewer Notes reflection;
- optional Judge evaluation.
Analytical output recovery is separate from Field Guide/Viewer Notes capacity retry. A length-truncated analytical response receives one controlled second attempt; document capacity and its retry contract remain unchanged.