"""Phase 2: engine LOO+1 (re-enable disabled engines on new snapshot) + risk_budget scaling. The 5 disabled engines were LOO-rejected on the OLD ftb_fix_v2 snapshot. On NEW pead_dualconv_ftb_fix_v2_probe snapshot they may behave differently. """ import json import subprocess import shutil import copy from pathlib import Path BASE = "configs/experiments/return_max_long_v8.11_composed_gld_tqqq_bmc30_cl65_vol40_rmin1_cap96.json" SWEEP_DIR = Path("configs/experiments/_tune_v811_p2") SWEEP_DIR.mkdir(exist_ok=True) SNAPSHOT = "pead_dualconv_ftb_fix_v2_probe" def load_base(): with open(BASE) as f: return json.load(f) def save(name: str, cfg: dict) -> str: cfg["experiment_name"] = name path = SWEEP_DIR / f"{name}.json" path.write_text(json.dumps(cfg, indent=2)) return str(path) def make_loo_plus(eid_to_enable: str) -> dict: """Re-enable a specific disabled engine.""" cfg = load_base() for e in cfg["strategy_engines"]: if e.get("engine_id") == eid_to_enable and "_disabled_reason" in e: del e["_disabled_reason"] return cfg raise ValueError(f"Engine {eid_to_enable} not found or not disabled") def make_loo_minus(eid_to_disable: str) -> dict: """Disable an enabled engine.""" cfg = load_base() for e in cfg["strategy_engines"]: if e.get("engine_id") == eid_to_disable and "_disabled_reason" not in e: e["_disabled_reason"] = "phase2_loo_minus_test" return cfg raise ValueError(f"Engine {eid_to_disable} not found or already disabled") def make_rb_scale(scale_dict: dict) -> dict: """Scale engine_risk_budget_pct for specific engines.""" cfg = load_base() for e in cfg["strategy_engines"]: eid = e.get("engine_id") if eid in scale_dict and "_disabled_reason" not in e: if "engine_risk_budget_pct" in e: e["engine_risk_budget_pct"] *= scale_dict[eid] return cfg def run_backtest(config_path: str) -> dict: cmd = ["python3", "-m", "apps.backtester.run", "--manifest", config_path, "--start", "2022-01-01", "--split", "all", "--initial-equity", "10000", "--snapshot-id", SNAPSHOT] cache = Path(f"data/parquet/{SNAPSHOT}/.runtime_cache") if cache.exists(): shutil.rmtree(cache) proc = subprocess.run(cmd, capture_output=True, text=True, timeout=200) tail = proc.stdout.split("\n")[-12:] run_id = return_pct = sqs_total = trades = None for line in tail: if "Run complete:" in line: run_id = line.split("Run complete:")[1].strip() elif "Total return:" in line: try: return_pct = float(line.split(":")[1].strip().rstrip("%")) except: pass elif "SQS:" in line: try: sqs_total = float(line.split("SQS:")[1].strip().split(" ")[0]) except: pass elif "Trades:" in line: try: trades = int(line.split(":")[1].strip()) except: pass mdd = sharpe = pf = None if run_id: m_path = Path(f"runs/{run_id}/metrics/metrics_summary.json") if m_path.exists(): m = json.loads(m_path.read_text()) mdd = m["max_drawdown_pct"] sharpe = m["sharpe_ratio"] pf = m["profit_factor"] return {"return": return_pct, "trades": trades, "mdd": mdd, "sharpe": sharpe, "pf": pf, "sqs": sqs_total, "run_id": run_id} DISABLED_ENGINES = [ "next_open_long_unknown_material_patient", "next_open_long_guidance_unknown_orderly", "next_open_long_unknown_material_orderly", "next_open_long_other_material_mixed_orderly", "next_open_long_unknown_ome_orderly", ] ENABLED_ENGINES = [ "reaction_close_long_core", "reaction_close_long_residual_lowclose_gap_d3", "reaction_close_long_residual_smallcap_gap", "reaction_close_long_extreme_orderly", "next_open_long_unknown_inline_hivol", "next_open_long_earnings_mixed_inline_orderly", "next_open_long_megacap_material_contract_orderly", "next_open_long_bullish_raised_recovery_broad_oneoff", ] def main(): baseline = {"return": 6839, "mdd": 9.78, "sharpe": 3.53, "pf": 8.26, "sqs": 84.8, "trades": 329} print(f"BASELINE v8.11: ret={baseline['return']}% MDD={baseline['mdd']}% Sharpe={baseline['sharpe']} PF={baseline['pf']} SQS={baseline['sqs']} trades={baseline['trades']}", flush=True) print("=" * 120, flush=True) results = [] # Phase 2A: re-enable each disabled engine print("\n=== Phase 2A: Re-enable disabled engines (LOO+1) ===", flush=True) for i, eid in enumerate(DISABLED_ENGINES): name = f"L_plus_{i:02d}_{eid[:30]}" cfg = make_loo_plus(eid) path = save(name, cfg) print(f"\n>>> [{i+1}/{len(DISABLED_ENGINES)}] +{eid}", flush=True) try: r = run_backtest(path) r["op"] = f"+{eid}" if r["return"] is None: print(f" FAILED", flush=True); continue d_ret = r["return"] - baseline["return"] d_sqs = (r["sqs"] or 0) - baseline["sqs"] d_mdd = (r["mdd"] or 0) - baseline["mdd"] print(f" ret={r['return']:.0f}% (Δ{d_ret:+.0f}) MDD={r['mdd']:.2f}% (Δ{d_mdd:+.2f}) Sharpe={r['sharpe']:.2f} PF={r['pf']:.2f} SQS={r['sqs']:.1f} (Δ{d_sqs:+.1f}) trades={r['trades']}", flush=True) results.append(r) except Exception as e: print(f" FAILED: {e}", flush=True) # Phase 2B: scale top engine risk budgets print("\n=== Phase 2B: Risk budget scaling ===", flush=True) rb_tests = [ ("core_x12", {"reaction_close_long_core": 1.2}), ("core_x08", {"reaction_close_long_core": 0.8}), ("residuals_x2", {"reaction_close_long_residual_lowclose_gap_d3": 2.0, "reaction_close_long_residual_smallcap_gap": 2.0}), ("extreme_x15", {"reaction_close_long_extreme_orderly": 1.5}), ("next_open_x3", {"next_open_long_unknown_inline_hivol": 3.0, "next_open_long_earnings_mixed_inline_orderly": 3.0, "next_open_long_megacap_material_contract_orderly": 3.0, "next_open_long_bullish_raised_recovery_broad_oneoff": 3.0}), ("hivol_x4", {"next_open_long_unknown_inline_hivol": 4.0}), ] for i, (label, scale_dict) in enumerate(rb_tests): name = f"RB_{label}" cfg = make_rb_scale(scale_dict) path = save(name, cfg) print(f"\n>>> RB scale: {label} {scale_dict}", flush=True) try: r = run_backtest(path) r["op"] = f"RB:{label}" if r["return"] is None: print(f" FAILED", flush=True); continue d_ret = r["return"] - baseline["return"] d_sqs = (r["sqs"] or 0) - baseline["sqs"] d_mdd = (r["mdd"] or 0) - baseline["mdd"] print(f" ret={r['return']:.0f}% (Δ{d_ret:+.0f}) MDD={r['mdd']:.2f}% (Δ{d_mdd:+.2f}) Sharpe={r['sharpe']:.2f} PF={r['pf']:.2f} SQS={r['sqs']:.1f} (Δ{d_sqs:+.1f}) trades={r['trades']}", flush=True) results.append(r) except Exception as e: print(f" FAILED: {e}", flush=True) print("\n" + "=" * 120, flush=True) print("SUMMARY (sorted by SQS, then return):", flush=True) results.sort(key=lambda x: (x.get("sqs") or 0, x.get("return") or 0), reverse=True) print(f"{'op':50s} {'return':>10s} {'MDD':>8s} {'Sharpe':>8s} {'PF':>6s} {'SQS':>7s} {'trades':>7s}", flush=True) print(f"{'BASELINE':50s} {baseline['return']:>9.0f}% {baseline['mdd']:>7.2f}% {baseline['sharpe']:>8.2f} {baseline['pf']:>6.2f} {baseline['sqs']:>7.1f} {baseline['trades']:>7d}", flush=True) for r in results: print(f"{r['op'][:48]:50s} {r['return']:>9.0f}% {r['mdd']:>7.2f}% {r['sharpe']:>8.2f} {r['pf']:>6.2f} {r['sqs']:>7.1f} {r['trades']:>7d}", flush=True) if __name__ == "__main__": main()