diff --git a/checkpoint.json b/checkpoint.json
new file mode 100644
index 000000000..02c2329fe
--- /dev/null
+++ b/checkpoint.json
@@ -0,0 +1,52 @@
+{
+  "timestamp": "2026-03-15T22:54:07.295492Z",
+  "current_node": "toolchain",
+  "completed_nodes": [
+    "start",
+    "toolchain"
+  ],
+  "node_retries": {
+    "start": 1,
+    "toolchain": 1
+  },
+  "context_values": {
+    "graph.goal": "# Random Edge Selection\n\n## Context\n\nEdge selection is currently fully deterministic — ties broken by highest weight then lexical node ID. This adds a `selection=\"random\"` node attribute so outgoing edges use weighted-random tiebreaking instead. The cascade priority (conditions → preferred label → suggested next → unconditional → fallback) is unchanged; randomness only replaces the pick-one-from-candidates step. Validation rejects `selection=\"random\"` combined with conditional edges.\n\n## Approach: Red/Green TDD\n\nWrite failing tests first, then write the minimum code to make them pass. Each step below is a red→green cycle. Run `cargo test -p fabro-workflows` after each change.\n\n---\n\n### Step 1: RED — `Node::selection()` accessor tests\n\nAdd tests to `graph/types.rs` `mod tests`:\n- `node_selection_default`: `Node::new(\"x\").selection()` == `\"deterministic\"` → fails (method doesn't exist)\n- `node_selection_random`: node with `selection=\"random\"` attr → `selection()` == `\"random\"` → fails\n\n### Step 2: GREEN — `Node::selection()` accessor\n\nAdd to `Node` impl in `graph/types.rs` (~after `backend()`, line 236):\n```rust\npub fn selection(&self) -> &str {\n    self.str_attr(\"selection\").unwrap_or(\"deterministic\")\n}\n```\n\n---\n\n### Step 3: RED — `weighted_random` tests\n\nAdd tests to `engine.rs` `mod tests`:\n- `weighted_random_empty_returns_none` → fails (function doesn't exist)\n- `weighted_random_single_edge` → fails\n- `weighted_random_zero_weight_all_selected`: 200 trials, both targets seen at least once → fails\n- `weighted_random_high_weight_dominates`: weight 100 vs 1, 500 trials, heavy >90% → fails\n\n### Step 4: GREEN — `weighted_random` function\n\nAdd after `best_by_weight_then_lexical` (~line 386) in `engine.rs`:\n- `weight <= 0` treated as 1 for probability calculation\n- Uses `rand::thread_rng()` (already imported at line 11)\n- Manual weighted sampling loop\n\n---\n\n### Step 5: RED — `select_edge` with `selection` param tests\n\nUpdate all ~8 existing `select_edge` tests to pass `\"deterministic\"` as 5th arg → fails (signature mismatch).\n\nAdd new tests:\n- `select_edge_random_returns_some_edge`: two unconditional edges, `selection=\"random\"` → fails\n- `select_edge_random_preferred_label_still_wins`: preferred label overrides random → fails\n\n### Step 6: GREEN — `select_edge` signature + `pick_edge` dispatcher\n\n**a)** Add `pick_edge` dispatcher:\n```rust\nfn pick_edge<'a>(edges: &[&'a Edge], selection: &str) -> Option<&'a Edge> {\n    match selection {\n        \"random\" => weighted_random(edges),\n        _ => best_by_weight_then_lexical(edges),\n    }\n}\n```\n\n**b)** Add `selection: &str` param to `select_edge`, replace 3 `best_by_weight_then_lexical` calls (lines 411, 442, 446) with `pick_edge(..., selection)`.\n\n**c)** Update 2 call sites to pass `node.selection()`:\n- Line 1548 (skip by StageStart hook)\n- Line 1759 (normal execution)\n\n---\n\n### Step 7: RED — validation rule tests\n\nAdd tests to `validation/rules.rs` `mod tests`:\n- `random_selection_no_conditions_clean`: random node + unconditional edges → no diagnostics → fails (rule doesn't exist)\n- `random_selection_with_conditions_errors`: random node + conditional edge → 1 Error diagnostic → fails\n- `deterministic_selection_with_conditions_ok`: deterministic node + conditional edges → no diagnostics → fails\n\n### Step 8: GREEN — validation Rule 23\n\nAdd `RandomSelectionNoConditionsRule` struct + `LintRule` impl:\n- Iterates nodes with `selection()==\"random\"`, checks outgoing edges for conditions\n- Error severity, message: `Node '{id}' has selection=\"random\" but also has conditional edges; random selection and conditions cannot be combined`\n- Fix: `Remove the condition attributes from outgoing edges, or remove selection=\"random\" from the node`\n- Register in `built_in_rules()` vec (line 35)\n\n---\n\n### Step 9: Docs (no TDD cycle)\n\n**`docs/workflows/transitions.mdx`:**\n- Update intro paragraph (line 6): `\"deterministic by default\"`, link to random selection section\n- Add \"Random selection\" section after \"Weight tiebreaking\" (~after line 157) with DOT example and `<Note>` about validation constraint\n\n**`docs/reference/dot-language.mdx`:**\n- Add `selection` row to \"All nodes\" attribute table (~line 191)\n\n---\n\n### Step 10: Final verification\n\n1. `cargo test -p fabro-workflows` — all tests pass\n2. `cargo clippy --workspace -- -D warnings` — clean\n3. `cargo fmt --check --all` — formatted\n\n## Files\n\n| File | Change |\n|---|---|\n| `lib/crates/fabro-workflows/src/graph/types.rs` | `selection()` accessor + tests |\n| `lib/crates/fabro-workflows/src/engine.rs` | `weighted_random`, `pick_edge`, modify `select_edge` sig + call sites + tests |\n| `lib/crates/fabro-workflows/src/validation/rules.rs` | Rule 23 + register + tests |\n| `docs/workflows/transitions.mdx` | Random selection section |\n| `docs/reference/dot-language.mdx` | `selection` attribute row |\n",
+    "command.stderr": "",
+    "current.preamble": "Goal: # Random Edge Selection\n\n## Context\n\nEdge selection is currently fully deterministic — ties broken by highest weight then lexical node ID. This adds a `selection=\"random\"` node attribute so outgoing edges use weighted-random tiebreaking instead. The cascade priority (conditions → preferred label → suggested next → unconditional → fallback) is unchanged; randomness only replaces the pick-one-from-candidates step. Validation rejects `selection=\"random\"` combined with conditional edges.\n\n## Approach: Red/Green TDD\n\nWrite failing tests first, then write the minimum code to make them pass. Each step below is a red→green cycle. Run `cargo test -p fabro-workflows` after each change.\n\n---\n\n### Step 1: RED — `Node::selection()` accessor tests\n\nAdd tests to `graph/types.rs` `mod tests`:\n- `node_selection_default`: `Node::new(\"x\").selection()` == `\"deterministic\"` → fails (method doesn't exist)\n- `node_selection_random`: node with `selection=\"random\"` attr → `selection()` == `\"random\"` → fails\n\n### Step 2: GREEN — `Node::selection()` accessor\n\nAdd to `Node` impl in `graph/types.rs` (~after `backend()`, line 236):\n```rust\npub fn selection(&self) -> &str {\n    self.str_attr(\"selection\").unwrap_or(\"deterministic\")\n}\n```\n\n---\n\n### Step 3: RED — `weighted_random` tests\n\nAdd tests to `engine.rs` `mod tests`:\n- `weighted_random_empty_returns_none` → fails (function doesn't exist)\n- `weighted_random_single_edge` → fails\n- `weighted_random_zero_weight_all_selected`: 200 trials, both targets seen at least once → fails\n- `weighted_random_high_weight_dominates`: weight 100 vs 1, 500 trials, heavy >90% → fails\n\n### Step 4: GREEN — `weighted_random` function\n\nAdd after `best_by_weight_then_lexical` (~line 386) in `engine.rs`:\n- `weight <= 0` treated as 1 for probability calculation\n- Uses `rand::thread_rng()` (already imported at line 11)\n- Manual weighted sampling loop\n\n---\n\n### Step 5: RED — `select_edge` with `selection` param tests\n\nUpdate all ~8 existing `select_edge` tests to pass `\"deterministic\"` as 5th arg → fails (signature mismatch).\n\nAdd new tests:\n- `select_edge_random_returns_some_edge`: two unconditional edges, `selection=\"random\"` → fails\n- `select_edge_random_preferred_label_still_wins`: preferred label overrides random → fails\n\n### Step 6: GREEN — `select_edge` signature + `pick_edge` dispatcher\n\n**a)** Add `pick_edge` dispatcher:\n```rust\nfn pick_edge<'a>(edges: &[&'a Edge], selection: &str) -> Option<&'a Edge> {\n    match selection {\n        \"random\" => weighted_random(edges),\n        _ => best_by_weight_then_lexical(edges),\n    }\n}\n```\n\n**b)** Add `selection: &str` param to `select_edge`, replace 3 `best_by_weight_then_lexical` calls (lines 411, 442, 446) with `pick_edge(..., selection)`.\n\n**c)** Update 2 call sites to pass `node.selection()`:\n- Line 1548 (skip by StageStart hook)\n- Line 1759 (normal execution)\n\n---\n\n### Step 7: RED — validation rule tests\n\nAdd tests to `validation/rules.rs` `mod tests`:\n- `random_selection_no_conditions_clean`: random node + unconditional edges → no diagnostics → fails (rule doesn't exist)\n- `random_selection_with_conditions_errors`: random node + conditional edge → 1 Error diagnostic → fails\n- `deterministic_selection_with_conditions_ok`: deterministic node + conditional edges → no diagnostics → fails\n\n### Step 8: GREEN — validation Rule 23\n\nAdd `RandomSelectionNoConditionsRule` struct + `LintRule` impl:\n- Iterates nodes with `selection()==\"random\"`, checks outgoing edges for conditions\n- Error severity, message: `Node '{id}' has selection=\"random\" but also has conditional edges; random selection and conditions cannot be combined`\n- Fix: `Remove the condition attributes from outgoing edges, or remove selection=\"random\" from the node`\n- Register in `built_in_rules()` vec (line 35)\n\n---\n\n### Step 9: Docs (no TDD cycle)\n\n**`docs/workflows/transitions.mdx`:**\n- Update intro paragraph (line 6): `\"deterministic by default\"`, link to random selection section\n- Add \"Random selection\" section after \"Weight tiebreaking\" (~after line 157) with DOT example and `<Note>` about validation constraint\n\n**`docs/reference/dot-language.mdx`:**\n- Add `selection` row to \"All nodes\" attribute table (~line 191)\n\n---\n\n### Step 10: Final verification\n\n1. `cargo test -p fabro-workflows` — all tests pass\n2. `cargo clippy --workspace -- -D warnings` — clean\n3. `cargo fmt --check --all` — formatted\n\n## Files\n\n| File | Change |\n|---|---|\n| `lib/crates/fabro-workflows/src/graph/types.rs` | `selection()` accessor + tests |\n| `lib/crates/fabro-workflows/src/engine.rs` | `weighted_random`, `pick_edge`, modify `select_edge` sig + call sites + tests |\n| `lib/crates/fabro-workflows/src/validation/rules.rs` | Rule 23 + register + tests |\n| `docs/workflows/transitions.mdx` | Random selection section |\n| `docs/reference/dot-language.mdx` | `selection` attribute row |\n\n",
+    "command.output": "cargo 1.94.0 (85eff7c80 2026-01-15)\n",
+    "internal.retry_count.start": 1,
+    "outcome": "success",
+    "thread.start.current_node": "toolchain",
+    "failure_signature": "",
+    "internal.fidelity": "compact",
+    "graph.rankdir": "LR",
+    "internal.thread_id": "start",
+    "internal.retry_count.toolchain": 1,
+    "internal.node_visit_count": 1,
+    "current_node": "toolchain",
+    "internal.run_id": "01KKSV9P020MWAMJ44XB7M93HW",
+    "failure_class": "",
+    "graph.model_stylesheet": "\n            * { backend: api; model: claude-opus-4-6;}\n        "
+  },
+  "logs": [],
+  "node_outcomes": {
+    "toolchain": {
+      "status": "success",
+      "context_updates": {
+        "command.output": "cargo 1.94.0 (85eff7c80 2026-01-15)\n",
+        "command.stderr": ""
+      },
+      "notes": "Script completed: command -v cargo >/dev/null || { curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh -s -- -y && sudo ln -sf $HOME/.cargo/bin/* /usr/local/bin/; }; cargo --version 2>&1",
+      "duration_ms": 55
+    },
+    "start": {
+      "status": "success",
+      "duration_ms": 0
+    }
+  },
+  "next_node_id": "preflight_compile",
+  "node_visits": {
+    "start": 1,
+    "toolchain": 1
+  }
+}
\ No newline at end of file
diff --git a/nodes/start/status.json b/nodes/start/status.json
new file mode 100644
index 000000000..f62b21991
--- /dev/null
+++ b/nodes/start/status.json
@@ -0,0 +1,6 @@
+{
+  "status": "success",
+  "notes": null,
+  "failure_reason": null,
+  "timestamp": "2026-03-15T22:54:07.230807+00:00"
+}
\ No newline at end of file
diff --git a/nodes/toolchain/script_invocation.json b/nodes/toolchain/script_invocation.json
new file mode 100644
index 000000000..d68c414c4
--- /dev/null
+++ b/nodes/toolchain/script_invocation.json
@@ -0,0 +1,5 @@
+{
+  "command": "command -v cargo >/dev/null || { curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh -s -- -y && sudo ln -sf $HOME/.cargo/bin/* /usr/local/bin/; }; cargo --version 2>&1",
+  "language": "shell",
+  "timeout_ms": null
+}
\ No newline at end of file
diff --git a/nodes/toolchain/script_timing.json b/nodes/toolchain/script_timing.json
new file mode 100644
index 000000000..3ec21f9cf
--- /dev/null
+++ b/nodes/toolchain/script_timing.json
@@ -0,0 +1,5 @@
+{
+  "duration_ms": 54,
+  "exit_code": 0,
+  "timed_out": false
+}
\ No newline at end of file
diff --git a/nodes/toolchain/status.json b/nodes/toolchain/status.json
new file mode 100644
index 000000000..7f2d9f758
--- /dev/null
+++ b/nodes/toolchain/status.json
@@ -0,0 +1,6 @@
+{
+  "status": "success",
+  "notes": "Script completed: command -v cargo >/dev/null || { curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh -s -- -y && sudo ln -sf $HOME/.cargo/bin/* /usr/local/bin/; }; cargo --version 2>&1",
+  "failure_reason": null,
+  "timestamp": "2026-03-15T22:54:07.295263+00:00"
+}
\ No newline at end of file