Research & Working Papers¶

Deep-dive analyses on specific topics.

Published Papers¶

Working Paper 01: Context-Bounded Operational Understanding¶

Status: ACCEPTED FOR NOW
Date: 2026-07-18
Language: English + Thai
Length: ~15 pages

A bilingual, falsifiable engineering hypothesis about how AI agents maintain operational context within bounded systems.

Key questions: - How do agents know when context is complete? - What happens when context switches? - How do we verify operational understanding is accurate?

Relevant for: - Multi-agent systems (Claude + ChatGPT coordination) - MCP server design - Agent skill frameworks

Read: Working Paper 01 (English | Thai)

In Progress¶

Topic: Capital Movement Framework¶

Status: PLANNED
Target: Phase 2 (Finance department, 2027)

Deep dive into how capital moves through family systems.

Questions to explore: - What forms does capital take in a family context? - How do we measure capital movement? - What are optimal movement patterns? - How do we teach family members about capital?

Topic: Evidence Verification Case Study¶

Status: PLANNED
Based on: SIM Security Analysis (Punnaraj-007)

Real-world example of how to apply evidence standards to a complex claim.

Questions to explore: - How do we separate technical fact from fear narrative? - What evidence levels apply? - How do we resolve contradictions? - What's the next verification step?

Topic: Multi-Agent Coordination Patterns¶

Status: PLANNED
Motivation: Dual-AI system (Claude + ChatGPT)

How multiple AI agents can work together without duplicating effort or creating conflicts.

Questions to explore: - How do agents divide labor? - What's the handoff protocol? - How do contradictions get resolved? - What architecture prevents merge conflicts?

Topic: Generational Knowledge Transfer¶

Status: PLANNED
Target: Phase 5 (HR department, 2028-2029)

How to teach the next generation to operate the system.

Questions to explore: - What knowledge must be explicit? - What knowledge is implicit in operations? - How do we verify successor understands? - What's the minimum viable training?

Research Philosophy¶

All research follows RAOS methodology:

Reflect — Document raw thinking and observations
Assess — Get feedback from multiple perspectives
Observe — Measure outcomes, not just intentions
Survive — Record lessons for future learning

Papers are falsifiable — they make specific claims that can be proven wrong.

Papers are honest — they acknowledge uncertainty and limitations.

Papers are practical — they serve the system's actual needs.

Contributing Research¶

Want to propose a research paper?

State the question — What are we trying to understand?
Why it matters — How does it serve the system?
Initial hypothesis — What do we think the answer might be?
Evidence needed — What would prove us right/wrong?
Timeline — When would we know?

Add your proposal to the roadmap or governance discussions.

Reading Order¶

For newcomers: Start with Working Paper 01 (Context-Bounded Operational Understanding)

For understanding capital: Wait for capital movement framework (2027)

For evidence standards: Reference the SIM security case study (planned)

For multi-agent work: Check multi-agent coordination patterns (planned)

Archives¶

Historical working papers and analysis are preserved in: - /raos-agent/OBSERVATIONS.md — Agent thinking - /work-log.md — Chronological record - Conversation archives (external to this site)