feat(archetype): scaffold lance-graph-archetype crate (DU-2.1..2.6)

.claude/board/AGENT_LOG.md

@@ -0,0 +1,12 @@
+# AGENT_LOG
+
+Append-only log of agent sessions. Prepend new entries at the top.
+
+---
+
+## 2026-04-24T16:30 — Archetype scaffold v2 (sonnet, claude/archetype-crate-scaffold)
+
+**D-ids:** DU-2.1..2.6
+**Commit:** `816a7c0`
+**Tests:** 12 pass
+**Outcome:** Shipped `lance-graph-archetype` crate scaffold: Component + Processor traits (Arrow-backed), World meta-state with tick/fork/at_tick stubs, CommandBroker FIFO queue, ArchetypeError (thiserror). Added to root workspace members. No compile errors; 12 unit tests green.

.claude/plans/archetype-scaffold-v1.md

@@ -0,0 +1,90 @@
+# Archetype Transcode Crate Scaffold — v1
+
+> **Status:** In progress (2026-04-24)
+> **Owner:** @archetype-specialist, @truth-architect
+> **Scope:** NEW crate `crates/lance-graph-archetype/`; deps only on `lance-graph-contract`, `arrow`, `lance` (peer-dep, optional).
+> **Depends on:** ADR-0001 Decision 1 (transcode-not-bridge). No runtime dependency on upstream Python.
+
+## Goal
+
+Flip `lance-graph-archetype` from "does-not-exist" to "scaffolded-and-locked." Ship the 6 foundational trait/struct files per ADR-0001 Decision 1. No runtime behaviour yet — this is the LOCKED-MAPPING-INCOMPLETE → LOCKED-AND-SCAFFOLDED pivot.
+
+## Deliverables
+
+- **DU-2.1** — `crates/lance-graph-archetype/Cargo.toml` + `src/lib.rs` + workspace `members` entry in root `Cargo.toml`.
+- **DU-2.2** — `src/component.rs: pub trait Component { fn arrow_field() -> arrow::datatypes::Field; fn type_id() -> &'static str; }` plus a test-only `MockComponent` impl asserting trait-object construction.
+- **DU-2.3** — `src/processor.rs: pub trait Processor { fn matches(schema: &arrow::datatypes::Schema) -> bool; fn process(batch: arrow::record_batch::RecordBatch) -> Result<arrow::record_batch::RecordBatch, ArchetypeError>; }`.
+- **DU-2.4** — `src/world.rs: pub struct World { tick: u64, dataset_uri: String }` with `new() / tick() / current_tick() / fork(&self, branch: &str) / at_tick(&self, tick: u64)` methods. `fork()` and `at_tick()` return `Err(ArchetypeError::Unimplemented { method: "..." })` stubs — docstrings tie to ADR-0001:61-72 / 95.
+- **DU-2.5** — `src/command_broker.rs: pub struct CommandBroker { queue: Vec<Command>, ... }` + `pub enum Command { Spawn, Despawn, Update }` — channel-based drain interface with `submit() / drain()` method stubs.
+- **DU-2.6** — `src/error.rs: pub enum ArchetypeError { Unimplemented { method: &'static str }, SchemaMismatch { ... }, LanceIo(...) }` with `thiserror::Error` impl.
+
+## Non-goals (explicit)
+
+- Runtime World tick behaviour — stubs only.
+- `AsyncProcessor` (Python async equivalent) — future follow-up.
+- Entity=`PersonaCard` wiring — DU-2.7, later PR.
+- Lance dataset integration beyond the `dataset_uri: String` placeholder — the `fork()` → `lance::checkout(branch)` wiring is DU-2.8.
+
+## Acceptance criteria
+
+- `cargo check -p lance-graph-archetype` compiles cleanly.
+- `cargo test -p lance-graph-archetype` — minimum 4 tests pass (one per core trait + one per stub-returns-Unimplemented).
+- `cargo test --workspace` — no regressions in other crates.
+- Root `Cargo.toml` workspace.members updated.
+- `STATUS_BOARD.md` DU-2 row status: Queued → In progress.
+- Verdict flip in `.claude/plans/unified-integration-v1.md §6`: Archetype row `LOCKED-MAPPING-INCOMPLETE` → `LOCKED-AND-SCAFFOLDED`.
+- `.claude/board/INTEGRATION_PLANS.md` — prepend entry pointing to this plan file.
+- `.claude/board/LATEST_STATE.md § Contract Inventory` — add a new block for `lance-graph-archetype` naming the shipped types.
+- `.claude/board/EPIPHANIES.md` — prepend short FINDING entry noting scaffold landed.
+
+## Architecture notes
+
+Per ADR-0001 Decision 1 (`.claude/adr/0001-archetype-transcode-stack.md:14-102`): this crate defines its OWN Rust interface. It does NOT mirror the Python `VangelisTech/archetype` API. The Python repo is a DESIGN SPEC, not a runtime dependency. "Upstream Python API unstable" is NOT a blocker.
+
+Per ADR-0001 Decision 3 (`adr/0001-archetype-transcode-stack.md:320-334`): BBB invariant bans `Vsa16kF32` / `RoleKey` / `NarsTruth` / `BlackboardEntry` from crossing the membrane. Archetype types defined in this crate are INSIDE-BBB; they do NOT appear on `CognitiveEventRow`. The scalar projection for "archetype tick happened" is already covered by `CognitiveEventRow.cycle_fp_hi/lo` + `MetaWord`.
+
+Mapping (locked, do not re-litigate):
+
+| ECS concept | lance-graph-contract type | This crate |
+|---|---|---|
+| Entity | `contract::persona::PersonaCard` | imported, not redefined |
+| World | `contract::a2a_blackboard::Blackboard` (runtime) + `World { dataset_uri, tick }` (archetype meta) | the latter is new here |
+| Tick | `contract::collapse_gate::GateDecision` fire | imported, not redefined |
+| Component | trait in this crate | **DU-2.2** |
+| Processor | trait in this crate | **DU-2.3** |
+| CommandBroker | struct in this crate | **DU-2.5** |
+
+## File layout
+
+```
+crates/lance-graph-archetype/
+  Cargo.toml
+  src/
+    lib.rs              # pub use component::*; etc.
+    component.rs        # trait Component
+    processor.rs        # trait Processor
+    world.rs            # struct World
+    command_broker.rs   # struct CommandBroker, enum Command
+    error.rs            # enum ArchetypeError (thiserror)
+```
+
+## Test layout
+
+Each module gets a `#[cfg(test)] mod tests` with at minimum one test. Minimum 4 tests total:
+
+1. `component::tests::mock_component_has_arrow_field`
+2. `processor::tests::trait_object_is_constructable`
+3. `world::tests::fork_returns_unimplemented`
+4. `world::tests::tick_increments`
+
+## Dependencies
+
+```toml
+[dependencies]
+lance-graph-contract = { path = "../lance-graph-contract" }
+arrow = { workspace = true }
+thiserror = { workspace = true }
+
+[dev-dependencies]
+# nothing initially
+```

Cargo.toml

@@ -7,6 +7,7 @@ members = [
     "crates/lance-graph-contract",
     "crates/neural-debug",
     "crates/lance-graph-callcenter",
+    "crates/lance-graph-archetype",
 ]
 exclude = [
     # Python bindings (upstream-inherited, opt-in via --manifest-path)

crates/lance-graph-archetype/Cargo.toml

@@ -0,0 +1,19 @@
+[package]
+name = "lance-graph-archetype"
+version = "0.1.0"
+edition = "2021"
+description = "Archetype transcode scaffold (ECS-style types transcoded to Arrow/Lance). Per ADR-0001 Decision 1: defines its OWN Rust interface (not a mirror of upstream Python). Inside-BBB only — types in this crate never cross the CognitiveEventRow membrane."
+license = "Apache-2.0"
+keywords = ["lance", "graph", "archetype", "ecs", "transcode"]
+
+[dependencies]
+lance-graph-contract = { path = "../lance-graph-contract" }
+# NOTE: plan specified { workspace = true } for arrow/thiserror, but this
+# workspace has no shared [workspace.dependencies] table today; using
+# explicit versions consistent with the rest of the codebase (arrow 57,
+# thiserror 2). See PR description for the single-line deviation.
+arrow = "57"
+thiserror = "2"
+
+[dev-dependencies]
+# nothing initially

crates/lance-graph-archetype/src/command_broker.rs

@@ -0,0 +1,99 @@
+//! `CommandBroker` — a channel-based queue for deferred world mutations.
+//!
+//! Per ADR-0001 Decision 1, `CommandBroker` is the archetype-side
+//! equivalent of Bevy's `Commands` / the Python ECS `CommandBroker`:
+//! a FIFO queue of world mutations that accumulates during a
+//! Processor pass and is drained by the World at tick boundaries.
+//!
+//! Stub-only at this stage — `submit` accepts commands, `drain` returns
+//! what was submitted in order. Actual application-to-World logic lands
+//! in DU-2.7 together with Entity wiring.
+
+/// A deferred world-mutation command. The three variants cover the
+/// ECS-standard operations (spawn a new entity, despawn an existing
+/// one, or update components on one). Payloads are opaque at the
+/// scaffold stage; DU-2.7 will parameterise them over concrete
+/// `Component` types.
+#[derive(Debug, Clone, PartialEq, Eq)]
+pub enum Command {
+    /// Spawn a new entity. The `u64` is a placeholder for the
+    /// eventually-to-be-typed component bundle identifier.
+    Spawn(u64),
+
+    /// Despawn an entity by its integer ID.
+    Despawn(u64),
+
+    /// Update the entity identified by the first `u64` with the
+    /// component-bundle identifier in the second `u64`.
+    Update(u64, u64),
+}
+
+/// FIFO queue of deferred commands.
+///
+/// Used by Processors to schedule world mutations without mutating the
+/// World mid-pass (which would break the Arrow-batch transcode model).
+/// The World's tick driver calls `drain` at tick boundaries and applies
+/// the commands in order. The scaffold uses a `Vec<Command>`; DU-2.7
+/// may upgrade to a `std::sync::mpsc::channel` for multi-processor
+/// concurrency.
+#[derive(Debug, Default, Clone)]
+pub struct CommandBroker {
+    queue: Vec<Command>,
+}
+
+impl CommandBroker {
+    /// Construct an empty broker. No allocation is performed until the
+    /// first `submit`.
+    pub fn new() -> Self {
+        Self { queue: Vec::new() }
+    }
+
+    /// Enqueue a command. O(1) amortised.
+    pub fn submit(&mut self, cmd: Command) {
+        self.queue.push(cmd);
+    }
+
+    /// Drain all queued commands in insertion order. Returns an owned
+    /// `Vec<Command>`; the broker is empty afterwards. O(n).
+    pub fn drain(&mut self) -> Vec<Command> {
+        std::mem::take(&mut self.queue)
+    }
+
+    /// Read the current queue length without draining.
+    pub fn len(&self) -> usize {
+        self.queue.len()
+    }
+
+    /// `true` iff no commands are queued.
+    pub fn is_empty(&self) -> bool {
+        self.queue.is_empty()
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn new_broker_is_empty() {
+        let b = CommandBroker::new();
+        assert_eq!(b.len(), 0);
+        assert!(b.is_empty());
+    }
+
+    #[test]
+    fn submit_and_drain_preserves_order() {
+        let mut b = CommandBroker::new();
+        b.submit(Command::Spawn(1));
+        b.submit(Command::Update(1, 7));
+        b.submit(Command::Despawn(1));
+        assert_eq!(b.len(), 3);
+
+        let drained = b.drain();
+        assert_eq!(
+            drained,
+            vec![Command::Spawn(1), Command::Update(1, 7), Command::Despawn(1)]
+        );
+        assert!(b.is_empty());
+    }
+}

crates/lance-graph-archetype/src/component.rs

@@ -0,0 +1,73 @@
+//! The `Component` trait — ECS-style component definition transcoded to Arrow.
+//!
+//! Per ADR-0001 Decision 1 (Archetype Transcode, not bridge): a `Component`
+//! is a Rust-side type that declares how it projects into an Arrow
+//! `Field`. The transcode surface is Arrow because every downstream
+//! consumer of this crate lands in a Lance dataset; the `arrow_field`
+//! method is what a `Processor` keys its `matches(schema)` check against.
+//!
+//! This trait deliberately stays Sized-agnostic at the scaffold stage —
+//! only associated functions, no self-receiver. Implementors declare
+//! static metadata (field shape, type ID) and the runtime machinery
+//! lives elsewhere.
+
+use arrow::datatypes::Field;
+
+/// An ECS-style component that knows how to project itself into an Arrow
+/// `Field`. Components do not carry row data at this stage — they declare
+/// SHAPE. Row data flows through `RecordBatch`es handed to `Processor`.
+///
+/// **BBB-invariant:** component types defined by implementors live
+/// INSIDE-BBB. They do not cross the external membrane (see
+/// `lance_graph_contract::external_membrane`). The scalar projection
+/// "a component tick happened" is carried by `CognitiveEventRow`'s
+/// existing columns (`cycle_fp_hi/lo`, `MetaWord`); this crate does
+/// not extend that row.
+pub trait Component {
+    /// Arrow field descriptor for this component. Called once at
+    /// `Processor::matches` time, not per-row. Implementors should
+    /// return a `Field` with a stable name and dtype.
+    fn arrow_field() -> Field;
+
+    /// Stable string identifier for this component type. Used by the
+    /// `CommandBroker` drain path to address entities-by-component
+    /// without relying on Rust's `TypeId` (which is not stable across
+    /// builds). Convention: `"<crate>::<type>"`.
+    fn type_id() -> &'static str;
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use arrow::datatypes::DataType;
+
+    /// Test-only component used to assert that the trait is implementable
+    /// and that its metadata is reachable without constructing a value.
+    struct MockComponent;
+
+    impl Component for MockComponent {
+        fn arrow_field() -> Field {
+            Field::new("mock_component", DataType::Int64, false)
+        }
+
+        fn type_id() -> &'static str {
+            "lance_graph_archetype::tests::MockComponent"
+        }
+    }
+
+    #[test]
+    fn mock_component_has_arrow_field() {
+        let field = MockComponent::arrow_field();
+        assert_eq!(field.name(), "mock_component");
+        assert_eq!(field.data_type(), &DataType::Int64);
+        assert!(!field.is_nullable());
+    }
+
+    #[test]
+    fn mock_component_type_id_is_stable() {
+        assert_eq!(
+            MockComponent::type_id(),
+            "lance_graph_archetype::tests::MockComponent"
+        );
+    }
+}

crates/lance-graph-archetype/src/error.rs

@@ -0,0 +1,73 @@
+//! Error type for the archetype transcode crate.
+//!
+//! Per ADR-0001 Decision 1, this crate defines its own error surface rather
+//! than mirroring the Python `VangelisTech/archetype` exceptions. The
+//! variants below are scoped to the scaffold (DU-2.1..2.6) — Lance I/O
+//! wiring for `World::fork` / `World::at_tick` is deliberately parked
+//! behind DU-2.8.
+
+use thiserror::Error;
+
+/// Top-level error type for archetype transcode operations.
+///
+/// All fallible methods in this crate return `Result<T, ArchetypeError>`.
+/// The `Unimplemented` variant is used for stubs that will be wired in
+/// follow-up deliverables; see `World::fork` / `World::at_tick` for the
+/// canonical example.
+#[derive(Debug, Error)]
+pub enum ArchetypeError {
+    /// A stub method that has not yet been wired. The `method` field names
+    /// the specific method (for example, `"World::fork"`). Once the
+    /// corresponding deliverable (DU-2.7 / DU-2.8) lands, the variant
+    /// stays but the call site no longer returns it.
+    #[error("archetype method `{method}` is not yet implemented (scaffold stub)")]
+    Unimplemented {
+        /// Fully-qualified method name, for example `"World::fork"`.
+        method: &'static str,
+    },
+
+    /// A `Processor::process` invocation received a `RecordBatch` whose
+    /// schema does not match what the processor declared via `matches`.
+    /// The `expected` / `actual` fields are human-readable descriptions;
+    /// no Arrow schema equality is defined at the scaffold stage.
+    #[error("schema mismatch: expected {expected}, got {actual}")]
+    SchemaMismatch {
+        /// Human-readable description of the expected schema.
+        expected: String,
+        /// Human-readable description of the actual schema that arrived.
+        actual: String,
+    },
+
+    /// Placeholder for Lance dataset I/O errors. Once DU-2.8 wires
+    /// `lance::checkout(branch)` into `World::fork`, the inner type will
+    /// be upgraded from `String` to `lance::Error`. Today it carries a
+    /// bare message — no `lance` dependency on this PR per the plan's
+    /// Non-goals section.
+    #[error("lance I/O error: {0}")]
+    LanceIo(String),
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn unimplemented_carries_method_name() {
+        let err = ArchetypeError::Unimplemented { method: "World::fork" };
+        let msg = format!("{err}");
+        assert!(msg.contains("World::fork"));
+        assert!(msg.contains("not yet implemented"));
+    }
+
+    #[test]
+    fn schema_mismatch_formats() {
+        let err = ArchetypeError::SchemaMismatch {
+            expected: "Schema{a: Int32}".to_string(),
+            actual: "Schema{a: Utf8}".to_string(),
+        };
+        let msg = format!("{err}");
+        assert!(msg.contains("expected"));
+        assert!(msg.contains("Int32"));
+        assert!(msg.contains("Utf8"));
+    }
+}