From 092b1d2a24bdbd60a2699bb9f324955c0ebebe7d Mon Sep 17 00:00:00 2001
From: jgrusewski <jgrusewski@localhost>
Date: Sat, 7 Mar 2026 21:49:36 +0100
Subject: [PATCH] docs(ml): update plan for 6-crate split (DQN/PPO separate)

Split ml-rl into ml-dqn and ml-ppo for 3-way parallel compilation
and better sccache hit rate. Extract TradingAction to ml-core to
enable full DQN/PPO independence.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
---
 .../plans/2026-03-07-ml-crate-split-design.md |  51 ++--
 docs/plans/2026-03-07-ml-crate-split-plan.md  | 243 ++++++++++++------
 2 files changed, 196 insertions(+), 98 deletions(-)

diff --git a/docs/plans/2026-03-07-ml-crate-split-design.md b/docs/plans/2026-03-07-ml-crate-split-design.md
index 1bb8df69a..bf0c949ec 100644
--- a/docs/plans/2026-03-07-ml-crate-split-design.md
+++ b/docs/plans/2026-03-07-ml-crate-split-design.md
@@ -14,7 +14,8 @@ Split `ml` into sub-crates that compile in parallel, reducing:
 
 ```
 ml-core (shared types, traits, utilities, features, data)
-    ├─→ ml-rl (DQN + PPO + cuda_pipeline)          ── parallel ──┐
+    ├─→ ml-dqn (DQN + cuda_pipeline)               ── parallel ──┐
+    ├─→ ml-ppo (PPO)                                ── parallel ──┤
     ├─→ ml-supervised (TFT, Mamba, Liquid, etc.)    ── parallel ──┤
     │                                                              ↓
     └──────────────────────────→ ml-infra (hyperopt, ensemble, trainers)
@@ -22,7 +23,7 @@ ml-core (shared types, traits, utilities, features, data)
                                    ml (thin facade, re-exports)
 ```
 
-`ml-rl` and `ml-supervised` compile **fully in parallel** — they have zero dependencies on each other.
+`ml-dqn`, `ml-ppo`, and `ml-supervised` compile in **3-way parallel** — they have zero dependencies on each other. This also maximizes sccache hit rate in CI: editing DQN never invalidates PPO's cache entry.
 
 ## Crate Assignment (Every Module Accounted For)
 
@@ -78,19 +79,29 @@ Shared infrastructure that 2+ other sub-crates depend on.
 - `dqn/action_space.rs` (245 lines) → `ml-core/src/action_space.rs`
 - `dqn/order_router.rs` (175 lines) → `ml-core/src/order_router.rs`
 
+**Extracted from dqn/agent.rs → ml-core:**
+- `TradingAction` enum (Buy/Sell/Hold) — used by PPO tests, must be in ml-core for DQN/PPO independence
+
 **NOT extracting:**
 - `dqn/circuit_breaker.rs` — already a re-export of `common/circuit_breaker.rs` (which stays in ml-core)
 - `liquid/FixedPoint` — only used by mamba (both in ml-supervised), no extraction needed
 
-### ml-rl (~44K lines)
+### ml-dqn (~35K lines)
 
-Reinforcement learning models. Depends on: `ml-core`.
+DQN reinforcement learning. Depends on: `ml-core`.
+
+- `dqn/` (33,116 lines minus ~2,500 extracted = ~30,600 lines)
+- `cuda_pipeline/` (4,431 lines) — GPU data pipeline for DQN trainers
+
+### ml-ppo (~14K lines)
+
+PPO reinforcement learning. Depends on: `ml-core`. **Zero dependency on ml-dqn.**
 
-- `dqn/` (33,116 lines minus 2,273 extracted = ~30,843 lines)
 - `ppo/` (14,182 lines)
-- `cuda_pipeline/` (4,431 lines) — GPU data pipeline for DQN/PPO trainers
 
-**Import changes:**
+After extraction of shared items to ml-core, PPO's only remaining DQN import was `TradingAction` (test-only) — now in ml-core.
+
+### Import changes (ml-dqn and ml-ppo):
 - `crate::dqn::mixed_precision::training_dtype` → `ml_core::mixed_precision::training_dtype`
 - `crate::dqn::xavier_init::linear_xavier` → `ml_core::xavier_init::linear_xavier`
 - `crate::dqn::portfolio_tracker` → `ml_core::portfolio_tracker`
@@ -122,7 +133,7 @@ Supervised/generative models. Depends on: `ml-core`. **Zero dependency on ml-rl.
 
 ### ml-infra (~67K lines)
 
-Training infrastructure. Depends on: `ml-core`, `ml-rl`, `ml-supervised`.
+Training infrastructure. Depends on: `ml-core`, `ml-dqn`, `ml-ppo`, `ml-supervised`.
 
 - `hyperopt/` (19,934 lines) — Bayesian optimization, adapters for all 10 models
 - `ensemble/` (13,744 lines) — inference adapters, confidence, voting
@@ -133,8 +144,8 @@ Training infrastructure. Depends on: `ml-core`, `ml-rl`, `ml-supervised`.
 - `walk_forward.rs` (485 lines)
 
 **Import changes:**
-- `crate::dqn::DQN` → `ml_rl::dqn::DQN`
-- `crate::ppo::PPO` → `ml_rl::ppo::PPO`
+- `crate::dqn::DQN` → `ml_dqn::dqn::DQN`
+- `crate::ppo::PPO` → `ml_ppo::ppo::PPO`
 - `crate::tft::TemporalFusionTransformer` → `ml_supervised::tft::TemporalFusionTransformer`
 - (and so on for all model types)
 - Shared items: `crate::X` → `ml_core::X`
@@ -173,11 +184,11 @@ Re-exports everything from sub-crates. Contains modules that cross sub-crate bou
 **Re-exports (preserves external API):**
 ```rust
 // ml/src/lib.rs
-pub use ml_core::*;          // All types, errors, traits
-pub use ml_rl::dqn;          // DQN module
-pub use ml_rl::ppo;          // PPO module
-pub use ml_rl::cuda_pipeline; // GPU pipeline
-pub use ml_supervised::tft;   // All supervised models
+pub use ml_core::*;           // All types, errors, traits
+pub use ml_dqn::dqn;          // DQN module
+pub use ml_dqn::cuda_pipeline; // GPU pipeline
+pub use ml_ppo::ppo;           // PPO module
+pub use ml_supervised::tft;    // All supervised models
 pub use ml_supervised::mamba;
 // ... etc for all model modules
 pub use ml_infra::hyperopt;
@@ -198,7 +209,7 @@ pub use ml_infra::trainers;
 | `mimalloc-allocator` | `ml-core` | allocator |
 | `simd` | `ml-core` | SIMD features |
 | `gc` | `ml-core` | GC features |
-| `nccl` | `ml-rl` | multi-GPU for RL training |
+| `nccl` | `ml-dqn` | multi-GPU for RL training |
 
 ## External Consumer Migration
 
@@ -209,13 +220,15 @@ All external consumers (`trading_service`, `ml_training_service`, etc.) continue
 | Phase | Time | Parallelism |
 |-------|------|-------------|
 | ml-core | ~15s | serial (first) |
-| ml-rl + ml-supervised | ~13s | **2-way parallel** |
+| ml-dqn + ml-ppo + ml-supervised | ~12s | **3-way parallel** |
 | ml-infra | ~18s | serial (after models) |
 | ml facade | ~3s | serial (last) |
-| **Total clean** | **~49s** | vs 55s current |
-| **Edit in dqn (incremental)** | **~13s** | vs 55s current |
+| **Total clean** | **~48s** | vs 55s current |
+| **Edit in dqn (incremental)** | **~12s** | vs 55s current |
+| **Edit in ppo (incremental)** | **~6s** | vs 55s current |
 | **Edit in tft (incremental)** | **~10s** | vs 55s current |
 | **Edit in features (incremental)** | **~10s** | vs 55s current |
+| **CI sccache (PPO edit)** | **~6s** | only ml-ppo rebuilt |
 
 ## Risks
 
diff --git a/docs/plans/2026-03-07-ml-crate-split-plan.md b/docs/plans/2026-03-07-ml-crate-split-plan.md
index 096592d71..ba5f788e8 100644
--- a/docs/plans/2026-03-07-ml-crate-split-plan.md
+++ b/docs/plans/2026-03-07-ml-crate-split-plan.md
@@ -2,9 +2,9 @@
 
 > **For Claude:** REQUIRED SUB-SKILL: Use superpowers:executing-plans to implement this plan task-by-task.
 
-**Goal:** Split the monolithic `ml` crate (260K lines, 55s compile) into 5 crates that compile in parallel, reducing incremental rebuilds from 55s to ~10-15s.
+**Goal:** Split the monolithic `ml` crate (260K lines, 55s compile) into 6 crates that compile in parallel, reducing incremental rebuilds from 55s to ~6-12s.
 
-**Architecture:** 4 sub-crates (`ml-core`, `ml-rl`, `ml-supervised`, `ml-infra`) + thin `ml` facade that re-exports everything. External consumers see no API changes.
+**Architecture:** 5 sub-crates (`ml-core`, `ml-dqn`, `ml-ppo`, `ml-supervised`, `ml-infra`) + thin `ml` facade. DQN, PPO, and supervised models compile in 3-way parallel. External consumers see no API changes.
 
 **Tech Stack:** Rust workspace, Cargo features, `pub use` re-exports.
 
@@ -39,8 +39,10 @@ Note the exact test count and build time for comparison after split.
 **Files:**
 - Create: `crates/ml-core/Cargo.toml`
 - Create: `crates/ml-core/src/lib.rs`
-- Create: `crates/ml-rl/Cargo.toml`
-- Create: `crates/ml-rl/src/lib.rs`
+- Create: `crates/ml-dqn/Cargo.toml`
+- Create: `crates/ml-dqn/src/lib.rs`
+- Create: `crates/ml-ppo/Cargo.toml`
+- Create: `crates/ml-ppo/src/lib.rs`
 - Create: `crates/ml-supervised/Cargo.toml`
 - Create: `crates/ml-supervised/src/lib.rs`
 - Create: `crates/ml-infra/Cargo.toml`
@@ -50,7 +52,7 @@ Note the exact test count and build time for comparison after split.
 **Step 1: Create crate directories**
 
 ```bash
-mkdir -p crates/ml-core/src crates/ml-rl/src crates/ml-supervised/src crates/ml-infra/src
+mkdir -p crates/ml-core/src crates/ml-dqn/src crates/ml-ppo/src crates/ml-supervised/src crates/ml-infra/src
 ```
 
 **Step 2: Add workspace members**
@@ -58,7 +60,8 @@ mkdir -p crates/ml-core/src crates/ml-rl/src crates/ml-supervised/src crates/ml-
 In root `Cargo.toml`, add to `[workspace] members`:
 ```toml
 "crates/ml-core",
-"crates/ml-rl",
+"crates/ml-dqn",
+"crates/ml-ppo",
 "crates/ml-supervised",
 "crates/ml-infra",
 ```
@@ -66,7 +69,8 @@ In root `Cargo.toml`, add to `[workspace] members`:
 And add workspace dependency entries:
 ```toml
 ml-core = { path = "crates/ml-core" }
-ml-rl = { path = "crates/ml-rl" }
+ml-dqn = { path = "crates/ml-dqn" }
+ml-ppo = { path = "crates/ml-ppo" }
 ml-supervised = { path = "crates/ml-supervised" }
 ml-infra = { path = "crates/ml-infra" }
 ```
@@ -77,9 +81,9 @@ Each sub-crate gets a `Cargo.toml` with:
 - Same edition/rust-version as `ml`
 - Dependencies subset (only what its modules need)
 - Feature flags relevant to its modules
-- `ml-core` as dependency for `ml-rl`, `ml-supervised`, `ml-infra`
+- `ml-core` as dependency for `ml-dqn`, `ml-ppo`, `ml-supervised`, `ml-infra`
 
-See the detailed Cargo.toml specs in Tasks 5-8.
+See the detailed Cargo.toml specs in Tasks 4-10.
 
 **Step 4: Create stub lib.rs for each**
 
@@ -97,7 +101,7 @@ Expected: Compiles (empty crates are valid)
 **Step 6: Commit**
 
 ```bash
-git add crates/ml-core crates/ml-rl crates/ml-supervised crates/ml-infra Cargo.toml
+git add crates/ml-core crates/ml-dqn crates/ml-ppo crates/ml-supervised crates/ml-infra Cargo.toml
 git commit -m "feat(ml): scaffold sub-crate directory structure for ml split"
 ```
 
@@ -107,7 +111,7 @@ git commit -m "feat(ml): scaffold sub-crate directory structure for ml split"
 
 ### Task 3: Extract Shared Items from dqn/ to Top-Level Modules
 
-These items live in `dqn/` but are used by PPO, trainers, ensemble, hyperopt — they belong in `ml-core`.
+These items live in `dqn/` but are used by PPO, trainers, ensemble, hyperopt — they belong in `ml-core`. Also extract `TradingAction` from `dqn/agent.rs` to enable DQN/PPO independence.
 
 **Files:**
 - Move: `crates/ml/src/dqn/mixed_precision.rs` → `crates/ml/src/mixed_precision.rs`
@@ -115,6 +119,8 @@ These items live in `dqn/` but are used by PPO, trainers, ensemble, hyperopt —
 - Move: `crates/ml/src/dqn/portfolio_tracker.rs` → `crates/ml/src/portfolio_tracker.rs`
 - Move: `crates/ml/src/dqn/action_space.rs` → `crates/ml/src/action_space.rs`
 - Move: `crates/ml/src/dqn/order_router.rs` → `crates/ml/src/order_router.rs`
+- Modify: `crates/ml/src/dqn/agent.rs` — move `TradingAction` enum to new file
+- Create: `crates/ml/src/trading_action.rs` — `TradingAction` enum (Buy/Sell/Hold)
 - Modify: `crates/ml/src/dqn/mod.rs` — replace module declarations with re-exports
 - Modify: `crates/ml/src/lib.rs` — add `pub mod` for new top-level modules
 - Modify: All files importing these via `crate::dqn::*` path
@@ -130,7 +136,11 @@ mv dqn/action_space.rs action_space.rs
 mv dqn/order_router.rs order_router.rs
 ```
 
-**Step 2: Update dqn/mod.rs**
+**Step 2: Extract TradingAction from dqn/agent.rs**
+
+Create `crates/ml/src/trading_action.rs` with the `TradingAction` enum (copied from `dqn/agent.rs`). In `dqn/agent.rs`, replace the enum with `pub use crate::trading_action::TradingAction;`. This is the only PPO→DQN dependency remaining after the other extractions — making it shared enables full DQN/PPO independence.
+
+**Step 3: Update dqn/mod.rs**
 
 Replace the old `pub mod` declarations with re-exports so existing `crate::dqn::X` paths still work:
 ```rust
@@ -146,9 +156,10 @@ pub use crate::xavier_init;
 pub use crate::portfolio_tracker;
 pub use crate::action_space;
 pub use crate::order_router;
+pub use crate::trading_action::TradingAction;
 ```
 
-**Step 3: Add modules to ml/lib.rs**
+**Step 4: Add modules to ml/lib.rs**
 
 Add these lines to `lib.rs`:
 ```rust
@@ -157,11 +168,12 @@ pub mod xavier_init;
 pub mod portfolio_tracker;
 pub mod action_space;
 pub mod order_router;
+pub mod trading_action;
 ```
 
-**Step 4: Fix internal imports in moved files**
+**Step 5: Fix internal imports in moved files**
 
-In each moved file, change `use crate::dqn::` imports to `use crate::` where referencing peer modules (e.g., `mixed_precision.rs` may import from `dqn` internals — fix those paths).
+In each moved file, change `use crate::dqn::` imports to `use crate::` where referencing peer modules.
 
 Specific fixes needed:
 - `action_space.rs`: has no `crate::dqn::` imports (only std/serde) — no changes needed
@@ -169,6 +181,7 @@ Specific fixes needed:
 - `mixed_precision.rs`: uses `candle_core` only — no changes needed
 - `portfolio_tracker.rs`: check for `crate::dqn::` imports and fix
 - `order_router.rs`: imports `crate::dqn::action_space::*` → change to `crate::action_space::*`; imports `crate::common::action::*` → keep as-is
+- PPO test files using `crate::dqn::TradingAction` → change to `crate::trading_action::TradingAction`
 
 **Step 5: Verify compilation**
 
@@ -383,16 +396,16 @@ Re-export from ml facade for backward compatibility."
 
 ---
 
-## Phase 3: Build ml-rl (RL Models)
+## Phase 3: Build ml-dqn and ml-ppo (RL Models — Separate Crates)
 
-### Task 6: Write ml-rl Cargo.toml
+### Task 6: Write ml-dqn Cargo.toml
 
 **Files:**
-- Create: `crates/ml-rl/Cargo.toml`
+- Create: `crates/ml-dqn/Cargo.toml`
 
 **Step 1: Write Cargo.toml**
 
-Dependencies: `ml-core` + `candle-core`, `candle-nn` + RL-specific deps (`crossbeam`, etc.).
+Dependencies: `ml-core` + `candle-core`, `candle-nn` + DQN-specific deps (`crossbeam`, etc.).
 
 Feature flags:
 ```toml
@@ -406,41 +419,39 @@ nccl = ["cuda"]
 
 ---
 
-### Task 7: Move RL Modules to ml-rl
+### Task 7a: Move DQN + cuda_pipeline to ml-dqn
 
 **Files:**
-- Move: `crates/ml/src/dqn/` → `crates/ml-rl/src/dqn/`
-- Move: `crates/ml/src/ppo/` → `crates/ml-rl/src/ppo/`
-- Move: `crates/ml/src/cuda_pipeline/` → `crates/ml-rl/src/cuda_pipeline/`
-- Modify: `crates/ml-rl/src/lib.rs`
+- Move: `crates/ml/src/dqn/` → `crates/ml-dqn/src/dqn/`
+- Move: `crates/ml/src/cuda_pipeline/` → `crates/ml-dqn/src/cuda_pipeline/`
+- Modify: `crates/ml-dqn/src/lib.rs`
 - Modify: `crates/ml/src/lib.rs`
 
 **Step 1: Move modules**
 
 ```bash
-mv crates/ml/src/dqn crates/ml-rl/src/
-mv crates/ml/src/ppo crates/ml-rl/src/
-mv crates/ml/src/cuda_pipeline crates/ml-rl/src/
+mv crates/ml/src/dqn crates/ml-dqn/src/
+mv crates/ml/src/cuda_pipeline crates/ml-dqn/src/
 ```
 
-**Step 2: Write ml-rl/src/lib.rs**
+**Step 2: Write ml-dqn/src/lib.rs**
 
 ```rust
 // Same clippy attributes as ml-core
 pub mod dqn;
-pub mod ppo;
 pub mod cuda_pipeline;
 ```
 
-**Step 3: Fix all imports in ml-rl**
+**Step 3: Fix all imports in ml-dqn**
 
-In every file under `crates/ml-rl/src/`:
+In every file under `crates/ml-dqn/src/`:
 - `use crate::MLError` → `use ml_core::MLError`
 - `use crate::mixed_precision` → `use ml_core::mixed_precision`
 - `use crate::xavier_init` → `use ml_core::xavier_init`
 - `use crate::portfolio_tracker` → `use ml_core::portfolio_tracker`
 - `use crate::action_space` → `use ml_core::action_space`
 - `use crate::order_router` → `use ml_core::order_router`
+- `use crate::trading_action` → `use ml_core::trading_action`
 - `use crate::common::` → `use ml_core::common::`
 - `use crate::cuda_compat::` → `use ml_core::cuda_compat::`
 - `use crate::gradient_accumulation::` → `use ml_core::gradient_accumulation::`
@@ -455,57 +466,42 @@ In every file under `crates/ml-rl/src/`:
 - `use crate::PRECISION_FACTOR` → `use ml_core::PRECISION_FACTOR`
 - `use crate::{MLError, MLResult}` → `use ml_core::{MLError, MLResult}`
 
-Keep `use crate::dqn::` and `use crate::ppo::` as-is (these reference ml-rl's own modules).
+Keep `use crate::dqn::` as-is (references ml-dqn's own dqn module).
+Keep `use crate::cuda_pipeline::` as-is (references ml-dqn's own cuda_pipeline module).
 
-Remove the re-exports from `dqn/mod.rs` added in Task 3 (they re-exported from `crate::` which was ml, now those modules are in ml-core):
+Update the re-exports from `dqn/mod.rs` added in Task 3:
 ```rust
-// REMOVE these from dqn/mod.rs:
-// pub use crate::mixed_precision;
-// pub use crate::xavier_init;
-// etc.
-// REPLACE WITH:
+// REPLACE crate:: references WITH ml_core::
 pub use ml_core::mixed_precision;
 pub use ml_core::xavier_init;
 pub use ml_core::portfolio_tracker;
 pub use ml_core::action_space;
 pub use ml_core::order_router;
+pub use ml_core::trading_action::TradingAction;
 ```
 
-Also fix PPO re-exports in `ppo/mod.rs`:
-```rust
-// ppo/circuit_breaker.rs re-exports from crate::common::circuit_breaker
-// Change to: ml_core::common::circuit_breaker
-```
+cuda_pipeline imports:
+- `crate::dqn::mixed_precision` → `ml_core::mixed_precision`
+- `crate::dqn::replay_buffer_type::GpuBatch` → `crate::dqn::replay_buffer_type::GpuBatch` (stays in ml-dqn ✓)
 
-Handle cross-RL deps:
-- PPO imports `crate::dqn::circuit_breaker` → becomes `crate::dqn::circuit_breaker` (still works via dqn re-export) OR `ml_core::common::circuit_breaker` (direct)
-- PPO imports `crate::dqn::mixed_precision::training_dtype` → `ml_core::mixed_precision::training_dtype`
-- PPO imports `crate::dqn::portfolio_tracker` → `ml_core::portfolio_tracker`
-- PPO imports `crate::dqn::xavier_init::linear_xavier` → `ml_core::xavier_init::linear_xavier`
-- cuda_pipeline imports `crate::dqn::mixed_precision` → `ml_core::mixed_precision`
-- cuda_pipeline imports `crate::dqn::replay_buffer_type::GpuBatch` → `crate::dqn::replay_buffer_type::GpuBatch` (stays in ml-rl)
-
-Watch for: `use crate::trainers::DQNTrainer` in dqn/ — this is a reverse dep. If it exists, it's likely in test code. If in non-test code, it needs to be removed/refactored (trainers will be in ml-infra).
-
-Watch for: `use crate::hyperopt::adapters::PPOTrainer` in ppo/ — same issue. Check if test-only.
+Watch for: `use crate::trainers::DQNTrainer` in dqn/ — reverse dep, likely test-only. Remove or cfg(test) gate with a comment.
 
 **Step 4: Update ml/src/lib.rs**
 
 Add re-exports:
 ```rust
-pub use ml_rl::dqn;
-pub use ml_rl::ppo;
-pub use ml_rl::cuda_pipeline;
+pub use ml_dqn::dqn;
+pub use ml_dqn::cuda_pipeline;
 ```
 
 Add dependency to `ml/Cargo.toml`:
 ```toml
-ml-rl = { workspace = true }
+ml-dqn = { workspace = true }
 ```
 
 **Step 5: Verify compilation**
 
-Run: `SQLX_OFFLINE=true cargo check -p ml-rl 2>&1 | tail -10`
+Run: `SQLX_OFFLINE=true cargo check -p ml-dqn 2>&1 | tail -10`
 Run: `SQLX_OFFLINE=true cargo check -p ml 2>&1 | tail -10`
 
 **Step 6: Fix errors iteratively**
@@ -520,7 +516,89 @@ Run: `SQLX_OFFLINE=true cargo test -p ml --lib 2>&1 | tail -5`
 
 ```bash
 git add -A
-git commit -m "refactor(ml): move DQN, PPO, cuda_pipeline to ml-rl crate"
+git commit -m "refactor(ml): move DQN + cuda_pipeline to ml-dqn crate"
+```
+
+---
+
+### Task 7b: Write ml-ppo Cargo.toml and Move PPO
+
+**Files:**
+- Create: `crates/ml-ppo/Cargo.toml`
+- Move: `crates/ml/src/ppo/` → `crates/ml-ppo/src/ppo/`
+- Modify: `crates/ml-ppo/src/lib.rs`
+- Modify: `crates/ml/src/lib.rs`
+
+**Step 1: Write ml-ppo Cargo.toml**
+
+Dependencies: `ml-core` + `candle-core`, `candle-nn`. NO dependency on `ml-dqn`.
+
+Feature flags:
+```toml
+[features]
+default = ["cuda"]
+cuda = ["ml-core/cuda"]
+```
+
+**Step 2: Move PPO**
+
+```bash
+mv crates/ml/src/ppo crates/ml-ppo/src/
+```
+
+**Step 3: Write ml-ppo/src/lib.rs**
+
+```rust
+// Same clippy attributes as ml-core
+pub mod ppo;
+```
+
+**Step 4: Fix all imports in ml-ppo**
+
+Same pattern as ml-dqn: `use crate::X` → `use ml_core::X` for all shared items.
+
+Key changes:
+- `use crate::dqn::circuit_breaker` → `use ml_core::common::circuit_breaker` (direct path, no dqn dependency)
+- `use crate::dqn::mixed_precision::training_dtype` → `use ml_core::mixed_precision::training_dtype`
+- `use crate::dqn::portfolio_tracker` → `use ml_core::portfolio_tracker`
+- `use crate::dqn::xavier_init::linear_xavier` → `use ml_core::xavier_init::linear_xavier`
+- `use crate::dqn::TradingAction` → `use ml_core::trading_action::TradingAction` (test-only)
+
+Fix `ppo/circuit_breaker.rs`:
+```rust
+// Change: pub use crate::common::circuit_breaker::*
+// To: pub use ml_core::common::circuit_breaker::*
+```
+
+Watch for: `use crate::hyperopt::adapters::PPOTrainer` in ppo/ — reverse dep, likely test-only.
+
+**Step 5: Update ml/src/lib.rs**
+
+```rust
+pub use ml_ppo::ppo;
+```
+
+Add dependency: `ml-ppo = { workspace = true }`
+
+**Step 6: Verify compilation**
+
+Run: `SQLX_OFFLINE=true cargo check -p ml-ppo 2>&1 | tail -10`
+Run: `SQLX_OFFLINE=true cargo check -p ml 2>&1 | tail -10`
+
+**Step 7: Verify DQN and PPO are independent**
+
+Run: `SQLX_OFFLINE=true cargo tree -p ml-ppo --depth 1 | grep ml-dqn`
+Expected: NO output (ml-ppo must NOT depend on ml-dqn)
+
+**Step 8: Run tests**
+
+Run: `SQLX_OFFLINE=true cargo test -p ml --lib 2>&1 | tail -5`
+
+**Step 9: Commit**
+
+```bash
+git add -A
+git commit -m "refactor(ml): move PPO to ml-ppo crate (independent of ml-dqn)"
 ```
 
 ---
@@ -642,13 +720,13 @@ transformers and flash_attention infrastructure."
 **Files:**
 - Create: `crates/ml-infra/Cargo.toml`
 
-Dependencies: `ml-core`, `ml-rl`, `ml-supervised` + `argmin` (for hyperopt), model-specific deps.
+Dependencies: `ml-core`, `ml-dqn`, `ml-ppo`, `ml-supervised` + `argmin` (for hyperopt), model-specific deps.
 
 Feature flags:
 ```toml
 [features]
 default = ["cuda"]
-cuda = ["ml-core/cuda", "ml-rl/cuda", "ml-supervised/cuda"]
+cuda = ["ml-core/cuda", "ml-dqn/cuda", "ml-ppo/cuda", "ml-supervised/cuda"]
 ```
 
 **Step 1: Commit**
@@ -692,10 +770,11 @@ pub mod walk_forward;
 
 **Step 3: Fix all imports**
 
-This is the most complex crate — it imports from all three sub-crates:
-- `use crate::dqn::DQN` → `use ml_rl::dqn::DQN`
-- `use crate::dqn::DQNConfig` → `use ml_rl::dqn::DQNConfig`
-- `use crate::ppo::PPO` → `use ml_rl::ppo::PPO`
+This is the most complex crate — it imports from all four model sub-crates:
+- `use crate::dqn::DQN` → `use ml_dqn::dqn::DQN`
+- `use crate::dqn::DQNConfig` → `use ml_dqn::dqn::DQNConfig`
+- `use crate::ppo::PPO` → `use ml_ppo::ppo::PPO`
+- `use crate::cuda_pipeline::*` → `use ml_dqn::cuda_pipeline::*`
 - `use crate::tft::*` → `use ml_supervised::tft::*`
 - `use crate::mamba::*` → `use ml_supervised::mamba::*`
 - `use crate::liquid::*` → `use ml_supervised::liquid::*`
@@ -704,7 +783,6 @@ This is the most complex crate — it imports from all three sub-crates:
 - `use crate::xlstm::*` → `use ml_supervised::xlstm::*`
 - `use crate::tgnn::*` → `use ml_supervised::tgnn::*`
 - `use crate::tlob::*` → `use ml_supervised::tlob::*`
-- `use crate::cuda_pipeline::*` → `use ml_rl::cuda_pipeline::*`
 - All shared items: `use crate::X` → `use ml_core::X`
 
 **Step 4: Update ml/src/lib.rs**
@@ -749,7 +827,7 @@ git commit -m "refactor(ml): move hyperopt, ensemble, trainers to ml-infra crate
 
 **Step 1: Update ml/Cargo.toml**
 
-- Add: `ml-core`, `ml-rl`, `ml-supervised`, `ml-infra` as dependencies
+- Add: `ml-core`, `ml-dqn`, `ml-ppo`, `ml-supervised`, `ml-infra` as dependencies
 - Remove: all dependencies that are now only used by sub-crates
 - Keep: dependencies used by remaining facade modules (`inference.rs`, `validation/`, `risk/`, etc.)
 
@@ -760,7 +838,8 @@ Every module path that external consumers use (e.g., `ml::dqn::DQNConfig`, `ml::
 **Step 3: Fix remaining facade module imports**
 
 Modules still in ml facade (`inference.rs`, `validation/`, `risk/`, `regime/`, etc.) need import updates:
-- `use crate::dqn::` — these now come from `ml_rl` via re-export, so `use crate::dqn::` still works ✓
+- `use crate::dqn::` — these now come from `ml_dqn` via re-export, so `use crate::dqn::` still works ✓
+- `use crate::ppo::` — same via re-export from `ml_ppo` ✓
 - `use crate::tft::` — same via re-export from `ml_supervised` ✓
 - `use crate::ensemble::` — same via re-export from `ml_infra` ✓
 - `use crate::MLError` — from `ml_core` via `pub use ml_core::*` ✓
@@ -804,7 +883,8 @@ Expected: Same test count as baseline (2506+), 0 failures
 Run each in parallel:
 ```bash
 SQLX_OFFLINE=true cargo test -p ml-core --lib 2>&1 | tail -5
-SQLX_OFFLINE=true cargo test -p ml-rl --lib 2>&1 | tail -5
+SQLX_OFFLINE=true cargo test -p ml-dqn --lib 2>&1 | tail -5
+SQLX_OFFLINE=true cargo test -p ml-ppo --lib 2>&1 | tail -5
 SQLX_OFFLINE=true cargo test -p ml-supervised --lib 2>&1 | tail -5
 SQLX_OFFLINE=true cargo test -p ml-infra --lib 2>&1 | tail -5
 ```
@@ -859,23 +939,25 @@ Run: `SQLX_OFFLINE=true cargo test -p common --lib 2>&1 | tail -5`
 
 **Step 1: Update MEMORY.md**
 
-Update the Architecture section to reflect the new 5-crate ML structure. Update the ML Model Training section with new crate paths. Remove any stale references to `crates/ml/src/dqn/mixed_precision.rs` etc.
+Update the Architecture section to reflect the new 6-crate ML structure. Update the ML Model Training section with new crate paths. Remove any stale references to `crates/ml/src/dqn/mixed_precision.rs` etc.
 
 **Step 2: Final commit**
 
 ```bash
 git add -A
-git commit -m "refactor(ml): complete crate split — 5 crates, parallel compilation
+git commit -m "refactor(ml): complete crate split — 6 crates, 3-way parallel compilation
 
 Split the monolithic ml crate (260K lines, 55s compile) into:
 - ml-core: shared types, traits, features, data, checkpoint, safety
-- ml-rl: DQN + PPO + cuda_pipeline
+- ml-dqn: DQN + cuda_pipeline
+- ml-ppo: PPO (independent of ml-dqn)
 - ml-supervised: TFT, Mamba, Liquid, TGNN, TLOB, KAN, xLSTM, Diffusion
 - ml-infra: hyperopt, ensemble, trainers, benchmark, deployment
 - ml: thin facade re-exporting everything
 
+3-way parallel compilation (ml-dqn + ml-ppo + ml-supervised).
 Zero API changes for downstream consumers.
-Incremental rebuild after model edit: ~55s → ~10-15s."
+Incremental rebuild after model edit: ~55s → ~6-12s."
 ```
 
 ---
@@ -886,20 +968,23 @@ Incremental rebuild after model edit: ~55s → ~10-15s."
 |------|-------------|-----------|------------|
 | 1 | Verify clean baseline | verify | — |
 | 2 | Create sub-crate dirs + workspace | scaffold | 1 |
-| 3 | Extract shared items from dqn/ | refactor | 2 |
+| 3 | Extract shared items from dqn/ + TradingAction | refactor | 2 |
 | 4 | Write ml-core Cargo.toml | config | 2 |
 | 5 | Move core modules to ml-core | **large** | 3, 4 |
-| 6 | Write ml-rl Cargo.toml | config | 4 |
-| 7 | Move RL modules to ml-rl | **large** | 5, 6 |
+| 6 | Write ml-dqn Cargo.toml | config | 4 |
+| 7a | Move DQN + cuda_pipeline to ml-dqn | **large** | 5, 6 |
+| 7b | Write ml-ppo Cargo.toml + move PPO | **large** | 5 |
 | 8 | Write ml-supervised Cargo.toml | config | 4 |
 | 9 | Move supervised models | **large** | 5, 8 |
-| 10 | Write ml-infra Cargo.toml | config | 6, 8 |
-| 11 | Move infra modules | **large** | 7, 9, 10 |
+| 10 | Write ml-infra Cargo.toml | config | 6, 7b, 8 |
+| 11 | Move infra modules | **large** | 7a, 7b, 9, 10 |
 | 12 | Clean up ml facade | medium | 11 |
 | 13 | Full verification | verify | 12 |
 | 14 | Update common dev-dep | small | 12 |
 | 15 | Update memory + final commit | small | 13 |
 
+Note: Tasks 7a, 7b, and 9 can execute in parallel (they move independent module sets).
+
 ## Critical Invariants (Verify After Every Task)
 
 1. **No stubs**: Every function must be wired to its real implementation