feat(backtest): summarize a backtest result, or a decision log by its path

- callers no longer construct TradingMemoryLog to score a run; README and CLI pass the result

README.md

@@ -325,11 +325,10 @@ One run gives one decision, which cannot tell you whether the system decides wel
 
 ```python
 from tradingagents.backtest import iter_grid, run_backtest, summarize
-from tradingagents.agents.utils.memory import TradingMemoryLog
 
 dates = iter_grid("2026-06-01", "2026-08-01", every_n_days=7)
 result = run_backtest(["NVDA", "AAPL"], dates, config, selected_analysts=["market", "news"])
-print(summarize(TradingMemoryLog({"memory_log_path": str(result.log_path)})).render())
+print(summarize(result).render())
 ```
 
 From the CLI:

cli/main.py

@@ -1372,7 +1372,6 @@ def backtest(
     ),
 ):
     """Score past decisions over a grid of tickers and dates."""
-    from tradingagents.agents.utils.memory import TradingMemoryLog
 
     try:
         dates = iter_grid(start, end, every)
@@ -1395,7 +1394,7 @@ def backtest(
     except Exception as exc:  # a missing key or an unknown analyst is a setup error
         console.print(f"[red]{exc}[/red]")
         raise typer.Exit(code=1) from None
-    console.print(summarize(TradingMemoryLog({"memory_log_path": str(result.log_path)})).render())
+    console.print(summarize(result).render())
     console.print(f"\nRan {result.cells_run} cells, skipped {result.skipped}. Log: {result.log_path}")
     for ticker, date, reason in result.failures:
         console.print(f"[yellow]failed:[/yellow] {ticker} {date}: {reason}")

tests/test_backtest.py

@@ -120,7 +120,7 @@ def _log_with(tmp_path, rows):
         log.store_decision(ticker, date, decision)
         if outcome is not None:
             log.update_with_outcome(ticker, date, outcome[0], outcome[1], 5, "note", "2026-02-01")
-    return log
+    return tmp_path / "m.md"
 
 
 @pytest.mark.unit
@@ -245,4 +245,32 @@ def test_the_window_reported_is_the_one_the_outcomes_used(tmp_path):
     log.store_decision("NVDA", "2026-01-05", "**Rating**: Buy\n\nx")
     log.update_with_outcome("NVDA", "2026-01-05", 0.1, 0.04, 21, "note", "2026-02-01")
 
-    assert "21 trading days" in summarize(log).render()
+    assert "21 trading days" in summarize(tmp_path / "m.md").render()
+
+
+@pytest.mark.unit
+def test_a_backtest_result_is_summarized_directly(tmp_path):
+    """The result names its own log, so a caller never builds the log to score it."""
+    from tradingagents.backtest import BacktestResult
+
+    path = _log_with(tmp_path, [("NVDA", "2026-01-05", "Rating: Buy\n\nx", (0.10, 0.04))])
+
+    assert summarize(BacktestResult(run_id="r", log_path=path)).resolved == 1
+
+
+@pytest.mark.unit
+def test_a_log_path_that_does_not_exist_is_an_error_not_an_empty_summary(tmp_path):
+    missing = tmp_path / "no-such-dir" / "m.md"
+
+    with pytest.raises(FileNotFoundError):
+        summarize(missing)
+    assert not missing.parent.exists()
+
+
+@pytest.mark.unit
+def test_a_result_whose_cells_all_failed_summarizes_as_empty(tmp_path):
+    from tradingagents.backtest import BacktestResult
+
+    result = BacktestResult(run_id="r", log_path=tmp_path / "never-written.md")
+
+    assert summarize(result).resolved == 0

tests/test_rating_integrity.py

@@ -98,7 +98,7 @@ def test_an_unscored_decision_is_left_out_of_the_backtest_figures(tmp_path):
     log.store_decision("AAPL", "2026-01-05", REFUSAL)
     log.update_with_outcome("AAPL", "2026-01-05", 0.1, 0.04, 5, "note", "2026-02-01")
 
-    summary = summarize(log)
+    summary = summarize(tmp_path / "m.md")
     assert set(summary.by_rating) == {"Buy"}
 
 

tradingagents/backtest.py

@@ -175,9 +175,15 @@ def run_backtest(
     return result
 
 
-def summarize(memory_log: TradingMemoryLog) -> BacktestSummary:
-    """Score the settled decisions in a log, by rating."""
-    entries = memory_log.load_entries()
+def summarize(source: BacktestResult | str | Path) -> BacktestSummary:
+    """Score the settled decisions of a backtest, or of a decision log at a path, by rating."""
+    if isinstance(source, BacktestResult):
+        path = source.log_path      # a run whose cells all failed wrote no log: nothing to score
+    elif Path(source).is_file():
+        path = Path(source)
+    else:
+        raise FileNotFoundError(f"no decision log at {source}")
+    entries = TradingMemoryLog({"memory_log_path": str(path)}).load_entries()
     # A decision with no readable rating has no direction, so it can neither
     # count for nor against the system; it is reported as unscored instead.
     resolved = [(e, _alpha(e)) for e in entries