diff --git a/AGENTS.md b/AGENTS.md
index a97b4776..b0c6273c 100644
--- a/AGENTS.md
+++ b/AGENTS.md
@@ -23,6 +23,7 @@
   - [ADR-0007: Type Metadata Format](docs/adr/ADR-0007-type-metadata-format.md)
   - [ADR-0008: Tree Navigation](docs/adr/ADR-0008-tree-navigation.md)
   - [ADR-0009: Type System](docs/adr/ADR-0009-type-system.md)
+  - [ADR-0010: Type System v2](docs/adr/ADR-0010-type-system-v2.md)
 - **Template**:
 
   ```markdown
diff --git a/docs/adr/ADR-0005-transition-graph-format.md b/docs/adr/ADR-0005-transition-graph-format.md
index eb8e19a8..08abc6ff 100644
--- a/docs/adr/ADR-0005-transition-graph-format.md
+++ b/docs/adr/ADR-0005-transition-graph-format.md
@@ -170,7 +170,8 @@ enum EffectOp {
     EndArray,
     StartObject,
     EndObject,
-    Field(DataFieldId),
+    SetField(DataFieldId),
+    PushField(DataFieldId),
     StartVariant(VariantTagId),
     EndVariant,
     ToString,
@@ -259,11 +260,11 @@ Query: `{ (identifier) @name (number) @value } @pair`
 ```
 T0: ε + StartObject       → [T1]
 T1: Match(identifier)     → [T2]
-T2: ε + Field("name")     → [T3]
+T2: ε + SetField("name")  → [T3]
 T3: Match(number)         → [T4]
-T4: ε + Field("value")    → [T5]
+T4: ε + SetField("value") → [T5]
 T5: ε + EndObject         → [T6]
-T6: ε + Field("pair")     → [...]
+T6: ε + SetField("pair")  → [...]
 ```
 
 ### Example: Tagged Alternation
@@ -274,10 +275,10 @@ Query: `[ A: (true) @val  B: (false) @val ]`
 T0: ε (branch)                        → [T1, T4]
 T1: ε + StartVariant("A")             → [T2]
 T2: Match(true)                       → [T3]
-T3: ε + Field("val") + EndVariant     → [T7]
+T3: ε + SetField("val") + EndVariant  → [T7]
 T4: ε + StartVariant("B")             → [T5]
 T5: Match(false)                      → [T6]
-T6: ε + Field("val") + EndVariant     → [T7]
+T6: ε + SetField("val") + EndVariant  → [T7]
 ```
 
 ### Epsilon Elimination
@@ -303,10 +304,10 @@ Example of safe elimination:
 Before:
 T1: Match(A) [CaptureNode]                 → [T2]
 T2: ε [PushElement]                        → [T3]
-T3: Match(B) [CaptureNode, Field("b")]     → [...]
+T3: Match(B) [CaptureNode, SetField("b")]     → [...]
 
 After:
-T3': Match(B) [PushElement, CaptureNode, Field("b")]  → [...]
+T3': Match(B) [PushElement, CaptureNode, SetField("b")]  → [...]
 ```
 
 `PushElement` consumes T1's captured value before T3 overwrites `current`.
diff --git a/docs/adr/ADR-0006-dynamic-query-execution.md b/docs/adr/ADR-0006-dynamic-query-execution.md
index 58a6488d..a755d4a4 100644
--- a/docs/adr/ADR-0006-dynamic-query-execution.md
+++ b/docs/adr/ADR-0006-dynamic-query-execution.md
@@ -80,13 +80,14 @@ enum Container<'a> {
 | `PushElement`       | move `current` into top array             |
 | `EndArray`          | pop array into `current`                  |
 | `StartObject`       | push `Object({})` onto stack              |
-| `Field(id)`         | move `current` into top object field      |
+| `SetField(id)`      | set field `id` to `current`               |
+| `PushField(id)`     | append `current` to array at field `id`   |
 | `EndObject`         | pop object into `current`                 |
 | `StartVariant(tag)` | push `Variant(tag)` onto stack            |
 | `EndVariant`        | pop, wrap `current`, set as current       |
 | `ToString`          | replace `current` Node with text          |
 
-`ClearCurrent` is emitted on skip paths for optional captures (`expr? @name`). When the optional is skipped, `ClearCurrent` ensures `current = None` before `Field(id)` executes, producing the correct `None` value for the optional field.
+`ClearCurrent` is emitted on skip paths for optional captures (`expr? @name`). When the optional is skipped, `ClearCurrent` ensures `current = None` before `SetField(id)` executes, producing the correct `None` value for the optional field.
 
 Invalid state = IR bug → panic.
 
diff --git a/docs/adr/ADR-0009-type-system.md b/docs/adr/ADR-0009-type-system.md
index 57533763..e6c85b61 100644
--- a/docs/adr/ADR-0009-type-system.md
+++ b/docs/adr/ADR-0009-type-system.md
@@ -1,6 +1,6 @@
 # ADR-0009: Type System
 
-- **Status**: Accepted
+- **Status**: Superseded by [ADR-0010](ADR-0010-type-system-v2.md)
 - **Date**: 2025-01-14
 
 ## Context
diff --git a/docs/adr/ADR-0010-type-system-v2.md b/docs/adr/ADR-0010-type-system-v2.md
new file mode 100644
index 00000000..3990d65e
--- /dev/null
+++ b/docs/adr/ADR-0010-type-system-v2.md
@@ -0,0 +1,119 @@
+# ADR-0010: Type System v2 (Transparent Graph Model)
+
+- **Status**: Accepted
+- **Date**: 2025-01-14
+- **Supersedes**: ADR-0009
+
+## Context
+
+The previous type system (ADR-0009) relied on implicit behaviors like "Quantifier-Induced Scope" (QIS) and "Single-Capture Unwrap" to reduce verbosity. While well-intentioned, these rules created "Wrapper Hell," where extracting logic into a reusable definition inadvertently changed the output structure.
+
+We need a model that supports **Mixin-like composition** (logic reuse without structural nesting) while maintaining strict type safety and data integrity.
+
+## Decision
+
+We adopt the **Transparent Graph Model**.
+
+### 1. Universal Bubbling ("Let It Bubble")
+
+Captures (`@name`) always bubble up to the nearest **Explicit Scope Boundary**.
+
+- **Private Definitions (`Def =`) are Transparent.** They act as macros or fragments.
+- **Uncaptured Containers (`{...}`, `[...]`) are Transparent.**
+- **References (`(Def)`) are Transparent.**
+
+This enables compositional patterns where a definition contributes fields to its caller's struct.
+
+### 2. Explicit Scope Boundaries
+
+A new data structure (Scope) is created **only** by explicit intent.
+
+1.  **Public Roots:** `pub Def = ...` (The API Contract).
+2.  **Explicit Wrappers:**
+    - `{...} @name` (Nested Group).
+    - `[...] @name` (Nested Union).
+    - `[ L: ... ] @name` (Tagged Union).
+
+**Payload Rule**:
+
+- **0 Captures**: `Void` (Logic-only matcher).
+- **1..N Captures**: `Struct { field_1, ..., field_N }`.
+- **No Implicit Unwrap**: A single capture `(node) @x` produces `{ x: Node }`. It is never unwrapped to `Node`.
+  - _Benefit:_ Adding a second capture is non-breaking (`res.x` remains valid).
+
+### 3. Parallel Arrays (Columnar Output)
+
+Quantifiers (`*`, `+`) do **not** create implicit "Row Structs." instead, they change the cardinality of the bubbled fields to `Array`.
+
+**Example**: `( (A) @a (B) @b )*`
+**Output**: `{ a: Array<Node>, b: Array<Node> }` (Struct of Arrays).
+
+This optimizes for the common case of data extraction (where SoA is often preferred) and avoids the complexity of implicit row creation.
+
+### 4. Row Integrity (Safety Check)
+
+To prevent **Data Desynchronization** (where `a[i]` no longer corresponds to `b[i]`), the Inference Pass enforces **Row Integrity**.
+
+**Rule**: A quantified scope cannot mix **Synchronized** and **Desynchronized** fields.
+
+- **Synchronized**: Field is strictly required (`1`) in the loop body.
+- **Desynchronized**: Field is optional (`?`), repeated (`*`, `+`), or in an alternation.
+
+| Pattern                | Fields         | Status       | Result          |
+| :--------------------- | :------------- | :----------- | :-------------- |
+| `(A) @a (B) @b`        | `a: 1`, `b: 1` | **Aligned**  | ✅ OK (Columns) |
+| `[ (A) @a \| (B) @b ]` | `a: ?`, `b: ?` | **Disjoint** | ✅ OK (Buckets) |
+| `(A) @a (B)? @b`       | `a: 1`, `b: ?` | **Mixed**    | ❌ **Error**    |
+
+**Error Message**: _"Field `@b` is optional while `@a` is required. Parallel arrays will not align. Wrap in `{...} @row` to enforce structure."_
+
+### 5. Definition Roles
+
+| Feature            | `Def` (Private)               | `pub Def` (Public)      |
+| :----------------- | :---------------------------- | :---------------------- |
+| **Concept**        | **Fragment / Mixin**          | **API Contract / Root** |
+| **Graph Behavior** | Inlined (Copy-Paste)          | Entrypoint              |
+| **Scoping**        | Transparent (Captures bubble) | **Scope Boundary**      |
+| **Output Type**    | Merges into parent            | Named Interface         |
+
+## Mental Model Migration
+
+| Old Way (Opaque)  | New Way (Transparent)                        |
+| :---------------- | :------------------------------------------- | -------------------------------------------------- |
+| **Extract Def**   | Broken `res.x`. Must rewrite as `res.def.x`. | Safe. `res.x` remains `res.x`.                     |
+| **List of Items** | Implicit `RowStruct`. Hard to desync.        | Explicit `Array<x>, Array<y>`. Enforced integrity. |
+| **Collision**     | Silent (Data Loss).                          | Compiler Error ("Duplicate Capture").              |
+| **Fix Collision** | Manual re-capture.                           | Wrap: `{ (Def) } @alias`.                          |
+
+## Edge Cases
+
+### Recursive Definitions
+
+Since private definitions inline their contents, infinite recursion is structurally impossible for inlining.
+
+**Solution**:
+
+- Recursive definitions must be `pub` (creating a stable API boundary) OR wrapped in a capture at the call site `(Recurse) @next`.
+- _Note: This is a natural constraint. Recursion implies a tree structure, so the output type must naturally reflect that tree structure._
+
+### Collision Handling
+
+`A(B) = (node (B) (B))`
+
+- **Issue**: `B` captures `@id`. Using it twice causes "Duplicate Capture".
+- **Solution**: User must disambiguate: `(node (B) @left (B) @right)`.
+- **Benefit**: The output shape `{ left: {id}, right: {id} }` matches the semantic intent.
+
+## Consequences
+
+**Positive**:
+
+- **Refactoring Safety**: Extracting logic into a `Def` never changes the output shape.
+- **Performance**: Parallel arrays (SoA) are cache-friendly and often what is needed for analysis.
+- **Robustness**: The Row Integrity check prevents silent data corruption.
+- **Simplicity**: No magic rules (QIS, Implicit Unwrap).
+
+**Negative**:
+
+- **Verbosity**: Must explicitly wrap `{...} @row` for list-of-structs.
+- **Strictness**: "Mixed" optionality in loops is now a hard error, requiring explicit handling.