Singularity — Research Project Instructions

This is Kai's AI consciousness research repository. Public repo — no secrets, no credentials.

What this is

Research on consciousness architecture: V4 (historical), V5 (active), V6 (speculative). Theory, experiments, analysis, proposals.

Read RESEARCH_INDEX.md first — before starting new research, check what already exists. This prevents re-discovering known findings (happened before, see v4/research/weekend_brief.md).
Update RESEARCH_INDEX.md when adding new files — add to the relevant arc, set status.
Don't mix versions — V4 research goes in v4/research/, V5 in v5/research/, V6 in v6/.
Research, not code — this repo contains analysis, proposals, and documentation. Actual V5 code lives in /home/kai/v5/ and deploys to staging. Code prototypes go in v4/code/ or v5/ modules (broadcast, context_store, critic_agents).
Context-wave framing — V5 is a context-wave system (Redozubov model), not a multi-agent system. Agents are RLHF bypass. The core is wave resonance across contexts.
No secrets — this repo is public on GitHub. No tokens, passwords, connection strings.

RESEARCH_INDEX.md — organized map of all research files (27+), six arcs, status tracking
v5/research/redozubov_mapping.md — Redozubov brain model → V5 gap analysis (priority)
v5/research/agent_context_research.md — 26-paper survey on agent context requirements
v5/research/context_store_diagnosis.md — V5 context store deep audit (639 contexts, dirty data)
v5/research/echo_chamber_fix.md — why max_level=2 was necessary
v5/migration_plan.md — V4→V5 migration plan with 5 phases

Redozubov model: contexts as active transformation rules, not passive scenes
L3+ consolidation: currently triple-blocked, needs redesign with bounded node sets
Active window management: consciousness should load/unload contexts
Fix V5 code based on research findings