deepen OHLCV history + make the factor-IC pass honest about overlap/regime

Two changes so the cross-sectional signal results can actually be trusted.

(a) History depth — the binding constraint. Ingestion defaulted to 365 days, so
long-lookback factors (12-month momentum, 52-week high) were only computable on a
handful of weeks at the tail, and every IC reflected a single market regime.
- New `settings.ohlcv_history_days` (default 1825 ≈ 5y); new tickers backfill this
  far instead of 1 year.
- New manual "data_backfill" job (Admin → Jobs) re-fetches the full window for
  every ticker, ignoring incremental resume — run once to deepen existing
  1-year histories. Idempotent (upsert); resumes after rate limits.

(b) Factor-IC honesty. The IC was averaged over weekly rebalances whose 30-day
forward windows overlap, inflating the t-stat ~sqrt(6)x.
- IC now measured on NON-OVERLAPPING windows (weeks thinned to ~HORIZON apart).
- Each signal carries a `reliable` flag (>= 12 independent windows); BacktestPanel
  greys out and de-stars thin signals so a lucky 9-week IC of 0.3 can't masquerade
  as an edge.

332 backend tests pass; frontend build clean. No migration (config + job + an
added JSON field on the cached backtest report).

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

app/services/backtest_service.py

@@ -79,7 +79,8 @@ _CAL_BUCKETS = [(0, 20), (20, 40), (40, 60), (60, 80), (80, 100.01)]
 # ranking stocks by this signal sort tomorrow's winners from losers. This is the
 # test the per-setup hit-rate report can't do: it measures predictive power of a
 # signal, not the outcome of a target/stop structure built on top of one.
-MIN_CROSS_SECTION = 20   # min tickers present in a week to score that week
+MIN_CROSS_SECTION = 20    # min tickers present in a week to score that week
+MIN_RELIABLE_PERIODS = 12 # min non-overlapping windows before a signal's IC is trusted
 
 
 def _wrap_levels(level_dicts: list[dict]) -> list[Any]:
@@ -407,26 +408,53 @@ def _quintile_spread(pairs: list[tuple[float, float]]) -> float | None:
     return sum(p[1] for p in top) / k - sum(p[1] for p in bottom) / k
 
 
+def _week_ordinal(week_key: tuple[int, int]) -> int:
+    """Monotonic absolute week number from an (ISO year, ISO week) key."""
+    year, week = week_key
+    return year * 53 + week
+
+
+def _nonoverlapping_weeks(
+    week_keys: list[tuple[int, int]], stride: int
+) -> list[tuple[int, int]]:
+    """Thin to weeks at least ``stride`` apart so their forward windows don't
+    overlap — greedy earliest-first. Removes the autocorrelation that would
+    otherwise inflate the IC t-stat across adjacent weekly rebalances."""
+    kept: list[tuple[int, int]] = []
+    last: int | None = None
+    for wk in sorted(week_keys, key=_week_ordinal):
+        o = _week_ordinal(wk)
+        if last is None or o - last >= stride:
+            kept.append(wk)
+            last = o
+    return kept
+
+
 def _signal_evaluation(collected: dict) -> list[dict]:
     """Per-signal factor diagnostics, one row per candidate signal:
 
-      mean_ic               average weekly rank-IC (Spearman of signal vs fwd ret)
+      mean_ic               average rank-IC (Spearman of signal vs fwd ret)
       ic_t_stat             mean_ic / stderr — is the IC reliably non-zero?
-      ic_positive_pct       share of weeks the IC is positive (consistency)
+      ic_positive_pct       share of windows the IC is positive (consistency)
       mean_quintile_spread  avg top-minus-bottom-quintile forward return
+      reliable              True once there are >= MIN_RELIABLE_PERIODS windows
 
-    A signal with no edge lands near IC 0 and spread 0. Caveat: weekly rebalances
-    with a HORIZON-day forward window overlap, so the t-stat overstates
-    significance — read it as directional, alongside ic_positive_pct.
+    IC is measured on NON-OVERLAPPING forward windows (weeks thinned to ~HORIZON
+    apart) so the t-stat isn't inflated by autocorrelation. A signal with no edge
+    lands near IC 0 / spread 0; one with too few independent windows is flagged
+    unreliable rather than trusted on a lucky handful.
     """
+    stride = max(1, round(HORIZON / 5))  # ISO weeks spanned by the forward window
     rows: list[dict] = []
     for name in sorted(collected):
+        weeks_map = collected[name]
+        usable = [wk for wk, recs in weeks_map.items() if len(recs) >= MIN_CROSS_SECTION]
+        kept = _nonoverlapping_weeks(usable, stride)
         ics: list[float] = []
         spreads: list[float] = []
         sizes: list[int] = []
-        for recs in collected[name].values():
-            if len(recs) < MIN_CROSS_SECTION:
-                continue
+        for wk in kept:
+            recs = weeks_map[wk]
             ic = _spearman([r[0] for r in recs], [r[1] for r in recs])
             if ic is not None:
                 ics.append(ic)
@@ -450,6 +478,7 @@ def _signal_evaluation(collected: dict) -> list[dict]:
             "ic_t_stat": round(t_stat, 2) if t_stat is not None else None,
             "ic_positive_pct": round(sum(1 for x in ics if x > 0) / len(ics) * 100, 1),
             "mean_quintile_spread": round(sum(spreads) / len(spreads), 4) if spreads else None,
+            "reliable": len(ics) >= MIN_RELIABLE_PERIODS,
         })
     rows.sort(key=lambda r: r["mean_ic"], reverse=True)
     return rows
@@ -518,12 +547,13 @@ async def run_backtest(
         "signal_eval": _signal_evaluation(collected),
         "signal_eval_note": (
             "Cross-sectional rank-IC of price-only signals vs the forward "
-            f"{HORIZON}-day return (weekly rebalance, min {MIN_CROSS_SECTION} "
-            "names/week). |IC| ≳ 0.03 with a consistent sign is a real (if small) "
-            "edge; near 0 means ranking on it sorts nothing. Momentum factors and "
-            "high_52w are expected positive; reversal_1m and vol_6m are expected "
-            "negative (mean-reversion / low-vol anomaly). Overlapping windows inflate "
-            "the t-stat — read directionally."
+            f"{HORIZON}-day return (min {MIN_CROSS_SECTION} names/window). |IC| ≳ "
+            "0.03 with a consistent sign is a real (if small) edge; near 0 means "
+            "ranking on it sorts nothing. Momentum factors and high_52w are expected "
+            "positive; reversal_1m and vol_6m expected negative (mean-reversion / "
+            "low-vol anomaly). IC is measured on non-overlapping windows; signals "
+            f"with fewer than {MIN_RELIABLE_PERIODS} independent windows are flagged "
+            "unreliable (too few regimes — deepen history with the Data Backfill job)."
         ),
         "note": (
             "Sentiment & fundamentals held neutral (no point-in-time history). "