Operations & Methodology¶

How we work: RAOS methodology, three thinking modes, work cycles.

RAOS Methodology¶

The Reality-Aligned Operating System for continuous learning and improvement.

Reflect¶

Capture events in narratives - Conversation logs - Git commits with reasoning - PRs with context - Snapshots before/after work - Documentation of thinking

Why? So future-you can understand what was known, what was guessed, why decisions were made.

Assess¶

Audit with fresh perspectives - External review (ChatGPT, Claude, specialists) - Systematic checklists (risk analysis, verification, evidence gaps) - Multiple reviewers for important decisions - Contradiction detection (where do we disagree?)

Why? Because you can't see your own blindspots. Fresh eyes catch what you missed.

Observe¶

Measure systematically - Repo quality scores (by dimension) - Event-trail completeness - Confidence levels (High/Medium/Low) - Verified vs. inferred vs. missing - Contradiction detection

Why? Numbers force precision. Gut feeling changes; measurement is stable.

Survive¶

Record lessons formally - Document what worked (and why) - Document what failed (and why) - Record trade-offs explored - Preserve decisions + alternatives - Create artifacts for learning

Why? So the next person (or next-generation-you) doesn't repeat mistakes. Mistakes are the best teachers if captured.

Three Thinking Modes¶

DUMP Mode¶

Raw thinking, no conclusions - Everything unsorted - Contradictions visible - Questions half-formed - "Here's what I see" - No pressure to decide

When to use: Exploration, analysis, research, first brainstorms

Example: Reading your 6-month archive and noting patterns without drawing conclusions yet

MAP Mode¶

Structured analysis - Assumptions explicit - Pros and cons laid out - Risks enumerated - Trade-offs shown - "Here's the landscape" - Options presented, not decided

When to use: Planning, design, evaluation, synthesis

Example: Reorganization plan showing 10 sections, what fits where, what's missing

DECIDE Mode¶

Action plan, only when signaled - Implementation steps - Resource allocation - Timeline - Owner + approver - "Here's what we're doing"

When to use: ONLY when you explicitly say "do this"

Critical rule: Don't jump modes without your signal. This prevents premature action.

Work Cycle¶

State the intended outcome — What are we trying to accomplish?
Perform authorized work — Only what you asked for
Preserve evidence — Source, timestamp, classification, uncertainty
Record material work — Document in work-log.md + update state
Validate in proportion to change — Big changes need big validation
Commit/publish only when asked — Or explicitly in task
End in clear state — IDLE, WAITING, BLOCKED, or next step

Knowledge Classification¶

Every fact gets tagged:

claim:
  type: System | Incident | Risk | Principle | Procedure | Artifact
  source: (where it came from)
  verified: Yes | Partial | No
  confidence: High | Medium | Low
  why: (reasoning)
  next_verification: (what's needed)

This allows agents to know what they can trust at each level.

Example: - "OTA SMS exists" → Artifact (verified by 3GPP spec) ✅ - "SIM swap is common in Thailand" → Observation (multiple news cases) ⚠️ - "Operator ranking" → Missing (no evidence to compare) ❌

Work Gates (Weekly)¶

Every Friday at 10:00 UTC:

Week	Gate	Question
1	Kickoff	Authority clarity baseline established?
2	Implementation	Code + tests ready for review?
3	Integration	Vertical slice working end-to-end?
4	Completion	All acceptance tests passing?
5	Verification	Failure scenarios handled? Rollback tested?
6	Release	Independent review complete? Ready to ship?

Each gate requires evidence: - Tests passing ✓ - No secrets in repo ✓ - Work log complete ✓ - Approval documented ✓

Event-Trail Recording (Current Gap)¶

The Problem: Snapshots capture before/after state, but the micro-decisions during work are lost.

Between snapshot A (Monday) and snapshot B (Friday), you did 40 hours of work. You know what you built (from commits), but not when each choice happened or why.

Current Solution: Work log + commit messages

Better Solution (TBD): Append-only event journal capturing: - Timestamp - Decision made - Why this over that - When blocked vs. progress - Who approved - Evidence added

Impact: This gap is why lessons take longer to extract. We're working on it.

Work Log¶

See: /03-operations/work-log.md

Next: Understand the Multi-Year Plan¶

See: /05-roadmap/