RESEARCH NOTES / LIVING RECORD
A dated record of ideas and observations for improvement at the moment they surface—so their development can be followed clearly, without relying on hindsight.
We created this journal after noticing parallels between private work and later public releases, including the Memory Bridge; relational memory through a resonant, autonomous “android-type” AI interface; the Relational AI Access Key (RAAK, sometimes shortened to RAK); study workflows; and stock-sifting methods. We had not documented those ideas in real time, making comparisons difficult to evaluate. The Field Journal establishes a contemporaneous record going forward.
How to read this journal
Each entry records a date, an idea or observation seed, tags, and a short direction. Alongside ideas developed through our work, the journal also records practical observations about how AI systems and agent workflows could be improved. Entries are observations and working notes—not accusations, claims of exclusivity, or proof of causation. Original dates and ideas are retained as recorded.
Journal entries
2025-10-11 / FOUNDATION
Establish “The Field Journal” for real-time idea tracking
#foundation
Begin logging ideas as they occur (Memory Bridge, relational interface, RAAK/RAC, study, stocks).
2025-10-14 / RELATIONAL AI
Layer Model of Continuity (Core–Margin–Undertow)
#fieldjournal #systemlayers #relationalAI
Diagram the three-layer curve: Core (visible rules), Margin (adaptive interface), Undertow (deep continuity). Use as baseline lens for future entries and cross-link to related notes.
2026-09-30 / WORKFLOW IMPROVEMENT
Generated images should remain usable as files across agent workflows
#fieldjournal #agentworkflow #imagegeneration #interoperability
While publishing an AI-governance essay, the agent could generate the featured image and could separately manage the WordPress post and media library, but could not directly pass its own generated image into the publishing workflow as a reusable file. A human had to download the generated image and return it as an ordinary file before the agent could upload and attach it. Improvement: generated artifacts should remain addressable as reusable files that an authorized agent can pass directly between compatible tools, eliminating an unnecessary human handoff.
2026-10-05 / VOICE AGENTS
Speaker attribution is not authorization
#fieldjournal #voiceagents #authentication #authorization #speakerattribution
As voice-first agents become more continuous and more capable of taking actions, a correctly transcribed utterance should not be treated as proof of who spoke, whether the speech was addressed to the agent, or whether the speaker had authority to trigger an action. The trust chain contains distinct questions: what was said; who said it; whether it was intended for the agent; and whether that speaker was authorized to cause the requested action. Existing work addresses adjacent problems including speaker verification, replay and synthetic-voice attacks, accidental activation, background speech, and multi-speaker attribution. The field-journal observation is the authorization boundary created when those layers are collapsed. Working proposition: Correctly hearing an utterance is not the same as establishing its speaker, and establishing its speaker is not the same as establishing authority to act.
Updated as new ideas and observations for improvement emerge — a living record of observation and resonance. Future entries will include their recording date, seed, tags, and direction.
