diff --git a/schemas/entity.md b/schemas/entity.md
index 97e003c9..e6cec2f6 100644
--- a/schemas/entity.md
+++ b/schemas/entity.md
@@ -13,26 +13,114 @@ Evidence → Claims (what's true about the world)
 
 Claims are static propositions with confidence levels. Entities are dynamic objects with temporal attributes. Both feed into agent reasoning.
 
-## Entity Types
+## Entity Type System
+
+The type system has two layers: **core types** shared by all agents, and **domain-specific extensions** that specialize core types for particular domains. Every entity uses exactly one type.
+
+### Core Types (all domains)
 
 | Type | What it tracks | Examples |
 |------|---------------|----------|
-| `company` | Protocol, startup, fund, DAO | MetaDAO, Aave, Solomon, Devoted Health |
-| `person` | Individual with tracked positions/influence | Stani Kulechov, Gabriel Shapiro, Proph3t |
+| `company` | Organization that operates — startup, fund, DAO, protocol | MetaDAO, Aave, Devoted Health, SpaceX |
+| `person` | Individual with tracked positions/influence | Proph3t, Stani Kulechov, Elon Musk |
+| `organization` | Government body, regulatory agency, standards body, consortium | SEC, CFTC, NASA, FLI, CMS |
+| `product` | Specific product, tool, or platform distinct from its maker | Autocrat, Starlink, Claude |
 | `market` | Industry segment or ecosystem | Futarchic markets, DeFi lending, Medicare Advantage |
-| `decision_market` | Governance proposal, prediction market, futarchy decision | MetaDAO: Hire Robin Hanson, MetaDAO: Burn 99.3% of META |
+
+### Domain-Specific Extensions
+
+Domain extensions are specialized subtypes that inherit from a core type. Use the most specific type available — it determines which fields are relevant.
+
+#### Internet Finance (Rio)
+
+| Type | Extends | What it tracks | Examples |
+|------|---------|---------------|----------|
+| `protocol` | company | On-chain protocol with TVL/volume metrics | Aave, Drift, Omnipair |
+| `token` | product | Fungible token distinct from its protocol | META, SOL, CLOUD |
+| `decision_market` | — | Governance proposal, prediction market, futarchy decision | MetaDAO: Hire Robin Hanson |
+| `exchange` | company | Trading venue (CEX or DEX) | Raydium, Meteora, Jupiter |
+| `fund` | company | Investment vehicle or DAO treasury | Solomon, Theia Research |
+
+#### Space Development (Astra)
+
+| Type | Extends | What it tracks | Examples |
+|------|---------|---------------|----------|
+| `vehicle` | product | Launch vehicle or spacecraft | Starship, New Glenn, Neutron |
+| `mission` | — | Specific spaceflight mission | Artemis III, ESCAPADE |
+| `facility` | — | Launch site, factory, or ground infrastructure | Starbase, LC-36 |
+| `program` | — | Multi-mission program or initiative | Artemis, Commercial Crew |
+
+#### Health (Vida)
+
+| Type | Extends | What it tracks | Examples |
+|------|---------|---------------|----------|
+| `therapy` | product | Treatment modality or therapeutic approach | mRNA cancer vaccines, GLP-1 agonists |
+| `drug` | product | Specific pharmaceutical product | Ozempic, Keytruda |
+| `insurer` | company | Health insurance organization | UnitedHealthcare, Devoted Health |
+| `provider` | company | Healthcare delivery organization | Kaiser Permanente, Oak Street Health |
+| `policy` | — | Legislation, regulation, or administrative rule | GENIUS Act, CMS 2027 Advance Notice |
+
+#### Entertainment (Clay)
+
+| Type | Extends | What it tracks | Examples |
+|------|---------|---------------|----------|
+| `studio` | company | Production company or media business | Beast Industries, Mediawan |
+| `creator` | person | Individual content creator or artist | MrBeast, Taylor Swift |
+| `franchise` | product | IP, franchise, or media property | Claynosaurz, Pudgy Penguins |
+| `platform` | product | Distribution or social media platform | YouTube, TikTok, Dropout |
+
+#### AI/Alignment (Theseus)
+
+| Type | Extends | What it tracks | Examples |
+|------|---------|---------------|----------|
+| `lab` | company | AI research laboratory | Anthropic, OpenAI, DeepMind |
+| `model` | product | AI model or model family | Claude, GPT-4, Gemini |
+| `framework` | product | Safety framework, governance protocol, or methodology | RSP, Constitutional AI |
+| `governance_body` | organization | AI governance or safety organization | AISI, FLI, Partnership on AI |
+
+### Choosing the Right Type
+
+```
+Is it a person?                          → person (or domain-specific: creator)
+Is it a government/regulatory body?      → organization (or domain-specific: governance_body)
+Is it a governance proposal or market?   → decision_market
+Is it a specific product/tool?           → product (or domain-specific: drug, model, vehicle, etc.)
+Is it an organization that operates?     → company (or domain-specific: lab, studio, insurer, etc.)
+Is it a market segment?                  → market
+Is it a policy or regulation?            → policy
+Is it a space mission?                   → mission
+Is it a physical facility?               → facility
+Is it a multi-mission program?           → program
+```
+
+**Rule:** Use the most specific type available. If a DeFi protocol fits `protocol`, use that instead of `company`. If an AI lab fits `lab`, use that instead of `company`. Domain-specific types carry domain-specific fields.
+
+### Adding New Types
+
+Core types require a schema PR reviewed by Leo. Domain-specific types are agent-managed — add a row to your domain's extension table via PR. No schema-wide changes needed. If a new type could apply to multiple domains, propose it as a core type instead.
+
+### Cross-Domain Entity Dedup
+
+One entity per real-world object. If Anthropic appears in both internet-finance and ai-alignment sources:
+
+1. **First creator owns the file.** Whichever agent creates the entity first files it in their domain (`entities/ai-alignment/anthropic.md`).
+2. **Other agents use `secondary_domains`.** The entity gets `secondary_domains: [internet-finance]` so it's discoverable across domains.
+3. **Both agents can update.** The `tracked_by` agent is responsible for staleness, but any agent can propose updates via PR when their sources contain new information.
+4. **Type follows primary domain.** If Theseus creates it, it's `lab`. If Rio had created it first, it would be `company`. The type reflects the primary tracking perspective.
+
+If two agents independently create the same entity, the reviewer merges them during PR review — keep the richer file, add `secondary_domains` from the other.
 
 ## YAML Frontmatter
 
 ```yaml
 ---
 type: entity
-entity_type: company | person | market | decision_market
+entity_type: company | person | organization | product | market | decision_market | protocol | token | exchange | fund | vehicle | mission | facility | program | therapy | drug | insurer | provider | policy | studio | creator | franchise | platform | lab | model | framework | governance_body
 name: "Display name"
 domain: internet-finance | entertainment | health | ai-alignment | space-development
 handles: ["@StaniKulechov", "@MetaLeX_Labs"]  # social/web identities
 website: https://example.com
-status: active | inactive | acquired | liquidated | emerging  # for company/person/market
+status: active | inactive | acquired | liquidated | emerging  # for most types
 # Decision markets use: active | passed | failed
 tracked_by: rio  # which agent owns this entity
 created: YYYY-MM-DD
@@ -45,7 +133,7 @@ last_updated: YYYY-MM-DD
 | Field | Type | Description |
 |-------|------|-------------|
 | type | enum | Always `entity` |
-| entity_type | enum | `company`, `person`, `market`, or `decision_market` |
+| entity_type | enum | Any type from the type system above |
 | name | string | Canonical display name |
 | domain | enum | Primary domain |
 | status | enum | Current operational status |
@@ -152,7 +240,7 @@ Example: `entities/internet-finance/metadao-hire-robin-hanson.md`
 ## Company-Specific Fields
 
 ```yaml
-# Company attributes
+# Company attributes (also used by protocol, exchange, fund, lab, studio, insurer, provider)
 founded: YYYY-MM-DD
 founders: ["[[person-entity]]"]
 category: "DeFi lending protocol"
@@ -184,7 +272,7 @@ launch_date: YYYY-MM-DD       # when the entity launched/raised
 People entities serve dual purpose: they track public figures we analyze AND serve as contributor profiles when those people engage with the KB. One file, two functions — the file grows from "person we track" to "person who participates."
 
 ```yaml
-# Person attributes
+# Person attributes (also used by creator)
 role: "Founder & CEO of Aave"
 organizations: ["[[company-entity]]"]
 followers: 290000  # primary platform
@@ -202,9 +290,19 @@ first_contribution: null  # date of first KB interaction
 attribution_handle: null  # how they want to be credited
 ```
 
-## Market-Specific Fields
+## Other Core Type Fields
 
 ```yaml
+# Organization attributes (also used by governance_body)
+jurisdiction: "United States"
+authority: "Securities regulation"  # what this body governs
+parent_body: "[[parent-organization]]"
+
+# Product attributes (also used by token, vehicle, drug, model, framework, franchise, platform)
+maker: "[[company-entity]]"  # who built/maintains this
+launched: YYYY-MM-DD
+category: "futarchy governance program"
+
 # Market attributes
 total_size: "$120B TVL"
 growth_rate: "flat since 2021"
@@ -213,6 +311,8 @@ market_structure: "winner-take-most | fragmented | consolidating"
 regulatory_status: "emerging clarity | hostile | supportive"
 ```
 
+**Domain-specific fields:** Each agent adds type-specific fields as they start extracting entities. The fields above cover core types. When Astra creates their first `vehicle` entity, they add vehicle-specific fields to the schema. Complexity is earned from actual use, not designed in advance.
+
 ## Body Format
 
 ```markdown
@@ -275,9 +375,19 @@ entities/
     claynosaurz.md
     pudgy-penguins.md
     matthew-ball.md
+    beast-industries.md                # studio
   health/
-    devoted-health.md
+    devoted-health.md                  # insurer
     function-health.md
+    ozempic.md                         # drug
+  ai-alignment/
+    anthropic.md                       # lab
+    claude.md                          # model
+    rsp.md                             # framework
+  space-development/
+    spacex.md
+    starship.md                        # vehicle
+    artemis.md                         # program
 ```
 
 **Filename:** Lowercase slugified name. Companies use brand name, people use full name. Decision markets use `{parent}-{proposal-slug}.md`.
@@ -299,6 +409,8 @@ Sources often contain entity information. During extraction, agents should:
 - Update entities (factual changes to tracked objects) → `entities/{domain}/`
 - Both from the same source, in the same PR
 
+See `skills/extract-entities.md` for the full extraction process.
+
 ## Key Difference from Claims
 
 | | Claims | Entities |
diff --git a/skills/extract-entities.md b/skills/extract-entities.md
new file mode 100644
index 00000000..2098842e
--- /dev/null
+++ b/skills/extract-entities.md
@@ -0,0 +1,149 @@
+# Entity Extraction Field Guide
+
+How to extract entities from source material. This skill works alongside `extract.md` (claim extraction) — both run during source processing.
+
+## When to Extract Entities
+
+Every source may contain entity data. During extraction, ask:
+
+1. **Does this source mention an organization, person, product, or market we don't already track?** → Create a new entity
+2. **Does this source contain updated information about an entity we already track?** → Update the existing entity (timeline, metrics, status)
+3. **Does this source describe a decision, proposal, or market outcome?** → Create a decision_market entity (if it meets significance threshold)
+
+## The Dual Extraction Loop
+
+```
+Source → Read completely
+       ↓
+       Extract claims (propositions about the world) → domains/{domain}/
+       Extract entities (objects in the world) → entities/{domain}/
+       Update existing entities (new timeline events, metrics)
+       ↓
+       Both in the same PR
+```
+
+## Entity Extraction Process
+
+### Step 1: Identify Entity Mentions
+
+Read the source and list every entity mentioned. For each:
+- Is it already in `entities/{domain}/`? → Flag for update
+- Is it new and significant enough to track? → Flag for creation
+- Is it mentioned in passing with no meaningful data? → Skip
+
+**Significance test:** Would tracking this entity help us evaluate claims or form positions? If the entity is just background context, skip it.
+
+### Step 2: Select Entity Type
+
+Use the most specific type available. See `schemas/entity.md` for the full type system.
+
+```
+Is it a person?                          → person (or domain-specific: creator)
+Is it a government/regulatory body?      → organization (or domain-specific: governance_body)
+Is it a governance proposal or market?   → decision_market
+Is it a specific product/tool?           → product (or domain-specific: drug, model, vehicle)
+Is it an organization that operates?     → company (or domain-specific: lab, studio, insurer)
+Is it a market segment?                  → market
+```
+
+### Step 3: Extract Frontmatter
+
+Fill in every field you have data for. Don't guess — leave fields empty rather than fabricating data.
+
+**Required fields** (every entity):
+- `type: entity`
+- `entity_type`: the specific type
+- `name`: canonical display name
+- `domain`: primary domain
+- `status`: current status
+- `tracked_by`: your agent name
+- `created`: today's date
+
+**Optional but valuable:**
+- `handles`: social media handles (from the source or quick lookup)
+- `website`: primary web presence
+- `tags`: discovery tags
+- `secondary_domains`: if the entity spans domains
+
+**Type-specific fields:** Fill in whatever the source provides. The schema lists all available fields — use the ones that have data.
+
+### Step 4: Write the Body
+
+Follow the body format from `schemas/entity.md`:
+
+1. **Overview**: What this entity is, why we track it (2-3 sentences)
+2. **Current State**: Latest known attributes from this source
+3. **Timeline**: Key events with dates (at minimum, the event from this source)
+4. **Competitive Position**: Where it sits relative to competitors (if known)
+5. **Relationship to KB**: Wiki-link to related claims and entities
+
+### Step 5: Check for Duplicates
+
+Before creating a new entity, search **all** `entities/` directories (not just your domain) for:
+- Same name (exact or variant spelling)
+- Same handles
+- Same website
+
+If a match exists in **your domain**, update the existing entity.
+
+If a match exists in **another domain**, don't create a duplicate. Instead, add your domain to the existing entity's `secondary_domains` list and propose updates via PR. See `schemas/entity.md` → "Cross-Domain Entity Dedup" for the full protocol.
+
+### Step 6: Update Parent Entities
+
+If the new entity has a `parent` or `parent_entity` field, update the parent:
+- Add the new entity to the parent's Relevant Entities section
+- If it's a decision_market, add to the parent's Key Decisions table (if significant)
+- Add a timeline entry on the parent
+
+## What Makes a Good Entity
+
+**Good entities have:**
+- Concrete, verifiable attributes (dates, metrics, names)
+- Clear relevance to at least one domain claim
+- Enough data to be useful (not just a name)
+- A reason to track changes over time
+
+**Bad entity candidates:**
+- Mentioned once in passing with no data
+- Purely historical with no ongoing relevance
+- Duplicates of existing entities under different names
+- Too granular (every tweet doesn't need an entity)
+
+## Domain-Specific Guidance
+
+### Internet Finance (Rio)
+- Protocols and tokens are separate entities (MetaDAO = company, META = token)
+- Every futardio launch that raises significant capital gets a company entity
+- Governance proposals that materially change direction get decision_market entities
+- Regulatory bodies (CFTC, SEC) get organization entities
+
+### Space (Astra)
+- Vehicles (Starship, New Glenn) are distinct from their makers (SpaceX, Blue Origin)
+- Programs (Artemis, Commercial Crew) are distinct from the agencies running them
+- Missions get entities when they're historically significant or produce notable data
+
+### Health (Vida)
+- Drugs are distinct from the companies that make them
+- Insurers and providers are separate entity types — don't conflate
+- Policies (legislation, CMS rules) get organization entities for the issuing body + policy entities for the rule itself
+
+### Entertainment (Clay)
+- Creators are distinct from their companies (MrBeast vs Beast Industries)
+- Franchises/IP are distinct from the studios that own them
+- Platforms (YouTube, TikTok) get product or platform entities
+
+### AI/Alignment (Theseus)
+- Labs are distinct from their models (Anthropic vs Claude)
+- Frameworks (RSP, Constitutional AI) get their own entities when they influence multiple claims
+- Governance bodies (AISI, FLI) get organization entities
+
+## Eval Checklist (for reviewers)
+
+1. `entity_type` is the most specific available type
+2. Required fields are all populated
+3. No fabricated data — empty fields are better than guesses
+4. Not a duplicate of existing entity
+5. Meets significance threshold
+6. Wiki links resolve to real files
+7. Parent entity updated if applicable
+8. Filing location is correct: `entities/{domain}/{slug}.md`