fix: derive crypto baseline from exact return tail

Co-Authored-By: Codex <noreply@openai.com>

Author: Pigbibi

.github/workflows/drift-check.yml

@@ -49,26 +49,35 @@ jobs:
             echo "::error::SNAPSHOT_REPOSITORY_TOKEN is required for lifecycle input artifact access"
             exit 1
           fi
-          gh api --paginate --slurp             "/repos/QuantStrategyLab/CryptoLivePoolPipelines/actions/artifacts?per_page=100"             > "${RUNNER_TEMP}/snapshot-artifacts.json"
+          gh api --paginate --slurp \
+            "/repos/QuantStrategyLab/CryptoLivePoolPipelines/actions/artifacts?per_page=100" \
+            > "${RUNNER_TEMP}/snapshot-artifacts.json"
+          gh api --paginate --slurp \
+            "/repos/QuantStrategyLab/CryptoLivePoolPipelines/actions/workflows/publish-lifecycle-inputs.yml/runs?branch=main&status=success&per_page=100" \
+            > "${RUNNER_TEMP}/trusted-snapshot-runs.json"
           python - <<'PY' > "${RUNNER_TEMP}/snapshot-artifact-selection.txt"
           import json
           import os
           from pathlib import Path
           pages = json.loads((Path(os.environ["RUNNER_TEMP"]) / "snapshot-artifacts.json").read_text())
+          run_pages = json.loads((Path(os.environ["RUNNER_TEMP"]) / "trusted-snapshot-runs.json").read_text())
           artifacts = [item for page in pages for item in page.get("artifacts", [])]
+          trusted_run_ids = {run["id"] for page in run_pages for run in page.get("workflow_runs", [])}
           candidates = [
               item for item in artifacts
               if not item.get("expired")
               and str(item.get("name", "")).startswith("crypto-lifecycle-inputs-")
               and item.get("workflow_run", {}).get("head_branch") == "main"
+              and item.get("workflow_run", {}).get("id") in trusted_run_ids
           ]
           if not candidates:
               raise SystemExit("no trusted crypto lifecycle input artifact is available")
           selected = max(candidates, key=lambda item: item["created_at"])
           print(selected["id"], selected["workflow_run"]["id"])
           PY
           read -r artifact_id workflow_run_id < "${RUNNER_TEMP}/snapshot-artifact-selection.txt"
-          gh api "/repos/QuantStrategyLab/CryptoLivePoolPipelines/actions/runs/${workflow_run_id}"             > "${RUNNER_TEMP}/snapshot-workflow-run.json"
+          gh api "/repos/QuantStrategyLab/CryptoLivePoolPipelines/actions/runs/${workflow_run_id}" \
+            > "${RUNNER_TEMP}/snapshot-workflow-run.json"
           python - <<'PY'
           import json
           import os
@@ -85,7 +94,8 @@ jobs:
           if mismatches:
               raise SystemExit(f"lifecycle input provenance check failed: {mismatches}")
           PY
-          gh api "/repos/QuantStrategyLab/CryptoLivePoolPipelines/actions/artifacts/${artifact_id}/zip"             > "${RUNNER_TEMP}/snapshot-artifact.zip"
+          gh api "/repos/QuantStrategyLab/CryptoLivePoolPipelines/actions/artifacts/${artifact_id}/zip" \
+            > "${RUNNER_TEMP}/snapshot-artifact.zip"
           python - <<'PY'
           import os
           import zipfile

scripts/run_walk_forward_backtest.py

@@ -8,6 +8,7 @@
 import hashlib
 import json
 import tempfile
+from dataclasses import replace
 from datetime import date
 from pathlib import Path
 from typing import Any
@@ -21,6 +22,7 @@
     SUPPORTED_PROFILES,
     build_backtest_runner,
 )
+from crypto_strategies.backtest.live_pool_simulator import _performance_metrics
 from crypto_strategies.strategies.crypto_equity_combo import PROFILE_NAME as CRYPTO_EQUITY_COMBO_PROFILE
 
 DEFAULT_WINDOWS: tuple[tuple[date, date], ...] = (
@@ -177,6 +179,26 @@ def _write_return_matrix(
     frame.reset_index().to_csv(output_path, index=False)
 
 
+def _baseline_from_return_tail(full_result: Any, returns: pd.Series) -> Any:
+    tail = returns.tail(DRIFT_BASELINE_HORIZON_DAYS)
+    metrics = _performance_metrics(tail)
+    max_drawdown = float(metrics["Max Drawdown"])
+    cagr = float(metrics["CAGR"])
+    return replace(
+        full_result,
+        sharpe_ratio=float(metrics["Sharpe"]),
+        calmar_ratio=abs(cagr / max_drawdown) if max_drawdown else None,
+        max_drawdown=max_drawdown,
+        cagr=cagr,
+        volatility=float(metrics["Annualized Volatility"]),
+        win_rate=float(metrics["Win Rate"]),
+        total_return=float(metrics["total_return"]),
+        start_date=tail.index.min().date(),
+        end_date=tail.index.max().date(),
+        observation_count=int(metrics["Trading Days"]),
+    )
+
+
 def run_walk_forward(
     *,
     profile: str,
@@ -214,7 +236,7 @@ def run_walk_forward(
     )
     full_start = min(start for start, _ in windows)
     baseline_end = max(end for _, end in windows)
-    return_matrix_runner.run(
+    full_window_raw = return_matrix_runner.run(
         profile,
         copy.deepcopy(baseline_params),
         start_date=full_start,
@@ -223,19 +245,7 @@ def run_walk_forward(
     full_window_returns = return_matrix_runner.last_daily_returns
     if len(full_window_returns) < DRIFT_BASELINE_HORIZON_DAYS:
         raise ValueError("full-window returns do not cover the 126-day drift baseline")
-    baseline_start = full_window_returns.index[-DRIFT_BASELINE_HORIZON_DAYS].date()
-    baseline_runner = _build_runner(
-        profile=profile,
-        panel=shared_panel,
-        market_history=shared_market_history,
-        synthetic_days=synthetic_days,
-    )
-    baseline_raw = baseline_runner.run(
-        profile,
-        copy.deepcopy(baseline_params),
-        start_date=baseline_start,
-        end_date=baseline_end,
-    )
+    baseline_raw = _baseline_from_return_tail(full_window_raw, full_window_returns)
     with tempfile.TemporaryDirectory(prefix=f"{profile}_wf_", dir=target_root) as scratch_dir:
         scratch_orchestrator = BacktestOrchestrator(store=PerformanceStore(local_root=Path(scratch_dir)))
         scratch_orchestrator.register_runner(

tests/test_drift_workflow_config.py

@@ -8,6 +8,7 @@ def test_drift_workflow_wires_real_pipeline_inputs_and_preflight_bundle() -> Non
     assert "needs: preflight_backtests" in workflow
     assert "Download latest trusted lifecycle inputs" in workflow
     assert "gh api --paginate --slurp" in workflow
+    assert "trusted-snapshot-runs.json" in workflow
     assert "crypto-lifecycle-inputs-" in workflow
     assert '"path": ".github/workflows/publish-lifecycle-inputs.yml"' in workflow
     assert '"conclusion": "success"' in workflow

tests/test_run_walk_forward_backtest.py

@@ -17,7 +17,7 @@
 
 import scripts.run_walk_forward_backtest as walk_forward
 import crypto_strategies.backtest.orchestrator_runner as orchestrator_runner
-from scripts.run_walk_forward_backtest import _baseline_param_set_id, run_walk_forward
+from scripts.run_walk_forward_backtest import _baseline_from_return_tail, _baseline_param_set_id, run_walk_forward
 
 
 def test_run_walk_forward_persists_lifecycle_baseline(tmp_path: Path) -> None:
@@ -125,3 +125,20 @@ def test_run_walk_forward_uses_real_panel_and_writes_return_matrix(
     assert payload["baseline"]["observation_count"] == 126
     assert {"as_of", "crypto_live_pool_rotation", "buy_hold_BTC"} <= set(return_matrix.columns)
     assert len(return_matrix) > payload["baseline"]["observation_count"]
+
+
+def test_baseline_uses_exact_tail_of_full_return_stream() -> None:
+    from quant_platform_kit.strategy_lifecycle.contracts import BacktestResult
+
+    index = pd.date_range("2024-01-01", periods=200, freq="D")
+    returns = pd.Series(range(200), index=index, dtype=float) / 100000
+    full_result = BacktestResult(
+        strategy_profile="crypto_live_pool_rotation", domain="crypto", param_set_id="", params={}
+    )
+
+    baseline = _baseline_from_return_tail(full_result, returns)
+
+    expected = returns.tail(126)
+    assert baseline.start_date == expected.index.min().date()
+    assert baseline.end_date == expected.index.max().date()
+    assert baseline.observation_count == len(expected)