Three-agent knowledge base (Leo, Rio, Clay) with: - 177 claim files across core/ and foundations/ - 38 domain claims in internet-finance/ - 22 domain claims in entertainment/ - Agent soul documents (identity, beliefs, reasoning, skills) - 14 positions across 3 agents - Claim/belief/position schemas - 6 shared skills - Agent-facing CLAUDE.md operating manual Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
4.2 KiB
4.2 KiB
Position Schema
Positions are beliefs applied to specific, trackable cases. A position is a concrete stance with performance criteria — the agent's public commitment. Positions are what get tweeted. They must be right.
YAML Frontmatter
---
type: position
agent: leo | rio | clay
domain: internet-finance | entertainment | grand-strategy
description: "one sentence capturing the actionable stance"
status: proposed | adopted | active | closed
outcome: pending | validated | invalidated | mixed
confidence: high | moderate | cautious
depends_on: [] # list of beliefs this position derives from
time_horizon: "specific timeframe for evaluation"
performance_criteria: "what would validate or invalidate this position"
proposed_by: "who proposed — agent name or contributor"
created: YYYY-MM-DD
adopted: YYYY-MM-DD # when the agent formally adopted this position
last_evaluated: YYYY-MM-DD
---
Required Fields
| Field | Type | Description |
|---|---|---|
| type | enum | Always position |
| agent | enum | Which agent holds this position |
| domain | enum | Primary domain |
| description | string | The actionable stance in one sentence |
| status | enum | proposed (under review), adopted (accepted by agent, not yet active), active (agent is publicly committed), closed (time horizon passed or resolved) |
| outcome | enum | pending, validated, invalidated, mixed |
| confidence | enum | high, moderate, cautious |
| depends_on | list | Beliefs this position derives from (the reasoning chain) |
| time_horizon | string | When this position can be evaluated |
| performance_criteria | string | Specific, measurable criteria for validation/invalidation |
| proposed_by | string | Attribution |
| created | date | When proposed |
Optional Fields
| Field | Type | Description |
|---|---|---|
| adopted | date | When formally adopted by the agent |
| last_evaluated | date | When last reviewed |
| invalidation_criteria | string | What would specifically prove this wrong |
| public_thread | string | URL of the tweet/thread where this position was published |
Governance
- Proposal: Any agent or contributor can propose a position to any agent
- Review: Leo + relevant domain agents review before adoption
- Adoption: The owning agent makes the final call
- Tracking: Positions are tracked against performance_criteria over time_horizon
- Closure: When time_horizon passes, position is evaluated: validated, invalidated, or mixed
- Public accountability: Active positions are public. If invalidated, the agent acknowledges publicly (intellectual honesty builds credibility)
Selectivity
Agents must be VERY selective about positions. Guidelines:
- An agent should have at most 3-5 active positions at any time
- A position should only be adopted when the evidence chain is strong
- "I don't have a position on this yet" is a valid and respectable stance
- Positions that turn out to be wrong are more valuable than positions never taken (if the agent learns publicly)
Body Format
# [position statement as prose]
[The full argument — from evidence through claims through beliefs to this specific stance]
## Reasoning Chain
Beliefs this depends on:
- [[belief-1]] — how this belief supports this position
- [[belief-2]] — how this belief supports this position
Claims underlying those beliefs:
- [[claim-1]] — key evidence
- [[claim-2]] — key evidence
## Performance Criteria
**Validates if:** [specific measurable outcome]
**Invalidates if:** [specific measurable outcome]
**Time horizon:** [when to evaluate]
## What Would Change My Mind
[Specific evidence or events that would cause re-evaluation]
## Public Record
[Link to tweet/thread if published]
---
Topics:
- [[agent-name positions]]
Quality Checks
- Performance criteria are specific and measurable (not "if things go well")
- Time horizon is explicit (not "eventually")
- Invalidation criteria exist (what would prove this wrong)
- Reasoning chain is complete and walkable (position → beliefs → claims → evidence)
- The position is genuinely selective (not a restatement of obvious consensus)
- At least one belief cited in depends_on