Implement self-improvement learning loop with trade analysis and adaptive config tuning

.claude/ralph/prd.json

@@ -17,7 +17,7 @@
         "Typecheck passes"
       ],
       "priority": 1,
-      "passes": false,
+      "passes": true,
       "notes": "Foundation story. .claude/ already exists with settings.local.json — add to it, don't overwrite. CLAUDE.md doesn't exist yet."
     },
     {
@@ -34,7 +34,7 @@
         "Typecheck passes"
       ],
       "priority": 2,
-      "passes": false,
+      "passes": true,
       "notes": "State file enables session continuity. ContinuousScanner already has _run_smart_loop() at src/scanner/automation/continuous.py:257."
     },
     {
@@ -51,7 +51,7 @@
         "Typecheck passes"
       ],
       "priority": 3,
-      "passes": false,
+      "passes": true,
       "notes": "sync_closed_trades_rl() is at src/scanner/execution.py. Each journal entry has full context: gates, agents, regime, outcome."
     },
     {
@@ -68,7 +68,7 @@
         "Typecheck passes"
       ],
       "priority": 4,
-      "passes": false,
+      "passes": true,
       "notes": "This is the evolution mechanism. Patterns that repeat become rules that actively gate or adjust trading behavior."
     },
     {
@@ -85,7 +85,7 @@
         "Typecheck passes"
       ],
       "priority": 5,
-      "passes": false,
+      "passes": true,
       "notes": "This closes the loop: trade outcome → learning → rule → config change → better trades. Bounds prevent runaway tuning."
     },
     {
@@ -102,7 +102,7 @@
         "Typecheck passes"
       ],
       "priority": 6,
-      "passes": false,
+      "passes": true,
       "notes": "Per-pair tuning is critical. EUR_USD 13.6 ATR behaves differently from USD_JPY 27.3 ATR. The scan phase calculates SL/TP in engine.py around line 1397."
     },
     {
@@ -119,7 +119,7 @@
         "Typecheck passes"
       ],
       "priority": 7,
-      "passes": false,
+      "passes": true,
       "notes": "This is the meta-learning dashboard. Without metrics, we can't know if the learning system itself is improving."
     },
     {
@@ -135,7 +135,7 @@
         "Typecheck passes"
       ],
       "priority": 8,
-      "passes": false,
+      "passes": true,
       "notes": "Observations build context even from non-trades. This data can later feed into a more sophisticated meta-learner."
     },
     {
@@ -152,7 +152,7 @@
         "Typecheck passes"
       ],
       "priority": 9,
-      "passes": false,
+      "passes": true,
       "notes": "self_refine.py exists at project root. This wires LLM reasoning into the learning loop for qualitative insights."
     },
     {
@@ -168,7 +168,7 @@
         "Typecheck passes"
       ],
       "priority": 10,
-      "passes": false,
+      "passes": true,
       "notes": "Context discipline. Files must stay concise or they become noise. Archive aggressively."
     },
     {
@@ -186,7 +186,7 @@
         "Typecheck passes"
       ],
       "priority": 11,
-      "passes": false,
+      "passes": true,
       "notes": "buddy_scanner.py already has CLI with scan/watch/trade subcommands. Add 'learn' alongside."
     },
     {
@@ -206,7 +206,7 @@
         "Typecheck passes"
       ],
       "priority": 12,
-      "passes": false,
+      "passes": true,
       "notes": "This is the integration story. All components from US-002 through US-010 get wired into the continuous loop here."
     }
   ]

.claude/ralph/progress.txt

@@ -1,11 +1,47 @@
 # Ralph Progress - Buddy Self-Improvement Loop
 
-## Stories: 0/12 completed
-
-## Context
-- ML Engine FX trading bot (Buddy)
-- Currently: scan → trade → RL feedback loop works
-- Missing: persistent learning, cross-session memory, adaptive config
-- Key files: src/scanner/automation/continuous.py, src/scanner/execution.py, src/scanner/agents.py
-- Trade journal: trained_data/trade_journal_rl.json (5 trades, 3 won / 2 lost)
-- Account: OANDA practice, NAV ~$102,580
+## Stories: 12/12 completed
+
+## Completed
+- US-001: Bootstrap .claude/ learning infrastructure ✓ (CLAUDE.md, learnings.md, rules/)
+- US-002: Session state engine ✓ (src/scanner/automation/state_engine.py — StateEngine class)
+- US-003: Trade outcome analyzer ✓ (src/scanner/automation/learning_engine.py — LearningEngine.analyze_trade)
+- US-004: Rule promotion ✓ (LearningEngine.check_promotions — 3+ observations → promoted rule)
+- US-005: Adaptive config tuner ✓ (src/scanner/automation/config_tuner.py — ConfigTuner class)
+- US-006: Per-pair SL/TP adaptation ✓ (LearningEngine.update_pair_sl_tp + engine.py reads pair_sl_tp_config.json)
+- US-007: Improvement tracker ✓ (src/scanner/automation/improvement_tracker.py — ImprovementTracker class)
+- US-008: Observation logging ✓ (src/scanner/automation/observation_log.py — ObservationLog class)
+- US-009: LLM deep analysis ✓ (LearningEngine.deep_analyze_loss + enable_llm_trade_analysis config flag)
+- US-010: Consolidation guards ✓ (LearningEngine.audit/consolidate — archive promoted, cap files)
+- US-011: Learn CLI command ✓ (cli/learn_commands.py + main.py dispatch — buddy learn [--analyze|--promote|--consolidate|--report|--status])
+- US-012: Full learning loop integration ✓ (continuous.py _run_learning_loop — wires all components into watch mode)
+
+## New Files Created
+- src/scanner/automation/state_engine.py — StateEngine for cross-session state
+- src/scanner/automation/learning_engine.py — LearningEngine for trade analysis + promotion + consolidation
+- src/scanner/automation/config_tuner.py — ConfigTuner for adaptive config from rules
+- src/scanner/automation/improvement_tracker.py — ImprovementTracker for meta-learning metrics
+- src/scanner/automation/observation_log.py — ObservationLog for market pattern capture
+- cli/learn_commands.py — CLI handler for `buddy learn` command
+
+## Modified Files
+- src/scanner/automation/continuous.py — Added _run_learning_loop(), observation logging, config tuning
+- src/scanner/engine.py — Per-pair SL/TP override from pair_sl_tp_config.json
+- src/scanner/config.py — Added enable_llm_trade_analysis flag + smart profile enablement
+- cli/argparser.py — Added 'learn' command + arguments
+- main.py — Wired handle_learn into dispatch table
+
+## Architecture
+```
+Scanner (engine.py) → Agents (agents.py) → Gates → Execution (execution.py) → OANDA
+     ↑                                                        ↓
+     ├── Config Tuner ← Rules ← Learnings ← RL Feedback ←── Trade Outcomes
+     ├── Observation Log (non-trade patterns)                  ↓
+     ├── Per-pair SL/TP config                    Learning Engine (analyze_trade)
+     └── State Engine (session continuity)         ↓
+                                             Rule Promotion (3+ observations)
+                                                   ↓
+                                             Config Tuner (apply_to_config)
+                                                   ↓
+                                             Improvement Tracker (meta-metrics)
+```

.claude/state.json

@@ -15,15 +15,25 @@
     "Minimum R:R ratio gate (1.2:1)",
     "Scan cycle JSONL logging",
     "Per-pair performance tracking",
-    "Bootstrap .claude/ learning infrastructure"
+    "Bootstrap .claude/ learning infrastructure",
+    "US-002: Session state engine (StateEngine class)",
+    "US-003: Trade outcome analyzer (LearningEngine.analyze_trade)",
+    "US-004: Rule promotion (3+ observations → promoted rules)",
+    "US-005: Adaptive config tuner (rules → config changes)",
+    "US-006: Per-pair adaptive SL/TP multipliers",
+    "US-007: Improvement tracker and meta-learning metrics",
+    "US-008: Observation logging for market patterns",
+    "US-009: LLM deep analysis for losing trades (>$100)",
+    "US-010: Learnings consolidation and anti-proliferation",
+    "US-011: buddy learn CLI command",
+    "US-012: Full learning loop integration into ContinuousScanner"
   ],
-  "next": "Run next scan cycle with improved position sizing and analyze results",
+  "next": "Run scan cycle in watch mode to validate full learning loop end-to-end",
   "open_questions": [
     "EUR_USD model predicted LONG but pair was bearish — is the model stale or is this a regime issue?",
-    "Should we increase min_confidence threshold given EUR_USD losses at 67% confidence?",
-    "Walk-forward orchestrator (US-005 from old PRD) still deferred — needs train-joint run first"
+    "Should we increase min_confidence threshold given EUR_USD losses at 67% confidence?"
   ],
-  "last_updated": "2026-03-17T06:15:00Z",
+  "last_updated": "2026-03-18T00:00:00Z",
   "portfolio_snapshot": {
     "nav": 102580.84,
     "open_trades": 0,
@@ -33,5 +43,6 @@
     "session_losses": 2,
     "win_rate": 0.60
   },
-  "improvement_focus": "Self-improvement loop infrastructure (learnings → rules → config adaptation)"
+  "improvement_focus": "Full self-improvement loop live — monitoring learning extraction and rule promotion",
+  "scan_cycle_count": 0
 }

cli/argparser.py

@@ -41,6 +41,7 @@
     "monitor",
     "find-candles",
     "transfer",
+    "learn",
 ]
 
 CLI_EPILOG = """
@@ -947,6 +948,41 @@ def _add_correlation_transfer_arguments(parser: argparse.ArgumentParser) -> None
     parser.add_argument("--transfer-epochs", type=int, default=50, help=argparse.SUPPRESS)
 
 
+def _add_learn_arguments(parser: argparse.ArgumentParser) -> None:
+    """Add learning loop CLI arguments (buddy learn command)."""
+    learn_group = parser.add_argument_group("Learn", "Self-improvement learning loop commands")
+    learn_group.add_argument(
+        "--analyze",
+        action="store_true",
+        default=False,
+        help="For learn: run trade outcome analysis on all unanalyzed journal entries",
+    )
+    learn_group.add_argument(
+        "--promote",
+        action="store_true",
+        default=False,
+        help="For learn: run promotion check and promote qualifying patterns to rules",
+    )
+    learn_group.add_argument(
+        "--consolidate",
+        action="store_true",
+        default=False,
+        help="For learn: run learnings consolidation (archive promoted, group by category)",
+    )
+    learn_group.add_argument(
+        "--report",
+        action="store_true",
+        default=False,
+        help="For learn: print improvement tracker report (total trades, win rate, trends)",
+    )
+    learn_group.add_argument(
+        "--status",
+        action="store_true",
+        default=False,
+        help="For learn: print current state from .claude/state.json and recent learnings",
+    )
+
+
 def create_argument_parser() -> argparse.ArgumentParser:
     """Create and configure the main argument parser.
 
@@ -979,6 +1015,7 @@ def create_argument_parser() -> argparse.ArgumentParser:
     _add_wandb_arguments(parser)
     _add_multi_pair_arguments(parser)
     _add_correlation_transfer_arguments(parser)
+    _add_learn_arguments(parser)
 
     return parser