backtest: --record/--record-bars Parquet recorder + 2-year M5/M15 replay with indicator weights and calendar

Adds an opt-in recorder to the backtester so a replay keeps the evidence each
trade setup was decided on: every indicator's vote, value, confidence, base
weight, regime multiplier, effective weight and contribution on the primary and
both confirmation timeframes, penalties, the setup and outcome, and the Forex
Factory calendar events around the trade (4 h before entry to exit, all-day rows
on the New-York date). --record-bars adds one float32 row per analyzed bar with
the same votes plus 1 h/4 h/24 h forward moves in ATRs. Output is Parquet in a
folder per run with a manifest carrying the engine/gate fingerprints, the regime
weight table in force and the book MD5.

Nothing in the signal core changes: the trade book is byte-identical with
recording on or off (control-vs-recorded MD5 equal on both 2-year runs, M5
781 / +146.07R / PF 1.52, M15 793 / +117.62R / PF 1.41 at $0.65). Module 33
adds 12 tests (498 total): snapshot reproduces the engine's composite, absence
is no vote, calendar join edges/phases/holiday, write/load round-trip,
_replay_span parity with recording on, CLI flags, import and source guards.

Also: backtest/replay_scorecard.py (first-look scorecards, calendar buckets,
in-book and per-bar what-ifs, --check invariants), scripts/ff_calendar_scraper.py
restored from git history (the only way to regenerate the gitignored calendar
file), report in backtest/research/replay_recording_2026-09-09.md.

Claude-Session: https://claude.ai/code/session_01Nc9Ninzyi18QkWezrxeBvB

scripts/ff_calendar_scraper.py

@@ -0,0 +1,144 @@
+"""Forex Factory calendar scraper -> backtest/cache/calendar.parquet.
+
+Fetches weekly calendar pages and parses the event JSON embedded in
+`window.calendarComponentStates` (the `days:` array). Event `dateline`
+is a unix epoch, so timestamps carry no timezone ambiguity — verified
+2026-07-11 against known release times (NFP 12:30/13:30 UTC across DST,
+ECB 12:15/13:15) and against EUR/USD M15 volatility: the bar containing
+a high-impact event averages 3.5x normal range, and shifting event
+times by ±1h destroys that alignment.
+
+Usage:
+    python scripts/ff_calendar_scraper.py --start 2024-01-01 --end 2026-07-13
+    python scripts/ff_calendar_scraper.py            # refresh: last stored week -> today
+
+Merges into the existing parquet on event (id, dateline), so refreshing
+is idempotent and also picks up revised `actual` values for recent weeks.
+Be polite: ~1.6s between requests (a full year is ~52 requests).
+"""
+import argparse
+import json
+import time
+from datetime import date, timedelta
+from pathlib import Path
+from typing import List, Optional
+
+import pandas as pd
+import requests
+
+CACHE_DIR = Path(__file__).resolve().parents[1] / "backtest" / "cache"
+PARQUET = CACHE_DIR / "calendar.parquet"
+
+UA = ("Mozilla/5.0 (X11; Linux x86_64) AppleWebKit/537.36 "
+      "(KHTML, like Gecko) Chrome/126.0 Safari/537.36")
+
+FIELDS = ["id", "dateline", "name", "country", "currency", "impactName",
+          "timeLabel", "timeMasked", "actual", "forecast", "previous",
+          "revision", "actualBetterWorse"]
+
+
+def week_url(d: date) -> str:
+    return (f"https://www.forexfactory.com/calendar?week="
+            f"{d.strftime('%b').lower()}{d.day}.{d.year}")
+
+
+def extract_days(html: str) -> Optional[list]:
+    """Balanced-bracket extraction of the `days:` JSON array."""
+    i = html.find("days:")
+    if i < 0:
+        return None
+    j = html.find("[", i)
+    depth, in_str, esc = 0, False, False
+    for k in range(j, len(html)):
+        ch = html[k]
+        if in_str:
+            if esc:
+                esc = False
+            elif ch == "\\":
+                esc = True
+            elif ch == '"':
+                in_str = False
+        else:
+            if ch == '"':
+                in_str = True
+            elif ch == "[":
+                depth += 1
+            elif ch == "]":
+                depth -= 1
+                if depth == 0:
+                    return json.loads(html[j:k + 1])
+    return None
+
+
+def fetch_week(sess: requests.Session, d: date, retries: int = 4) -> Optional[List[dict]]:
+    for attempt in range(retries):
+        try:
+            r = sess.get(week_url(d), timeout=30)
+            if r.status_code == 200:
+                days = extract_days(r.text)
+                if days is not None:
+                    return [{k: e.get(k) for k in FIELDS}
+                            for day in days for e in day.get("events", [])]
+        except requests.RequestException:
+            pass
+        time.sleep(5 * (attempt + 1))
+    return None
+
+
+def scrape(start: date, end: date) -> pd.DataFrame:
+    sess = requests.Session()
+    sess.headers["User-Agent"] = UA
+    rows: List[dict] = []
+    failed: List[date] = []
+    d = start
+    while d <= end:
+        week_rows = fetch_week(sess, d)
+        if week_rows is None:
+            failed.append(d)
+            print(f"FAILED week {d}")
+        else:
+            rows.extend(week_rows)
+        d += timedelta(days=7)
+        time.sleep(1.6)
+    if failed:
+        print(f"WARNING: {len(failed)} weeks failed: {failed}")
+    return pd.DataFrame(rows)
+
+
+def merge_into_store(new: pd.DataFrame, path: Path = PARQUET) -> pd.DataFrame:
+    if path.exists():
+        old = pd.read_parquet(path)
+        combined = pd.concat([old, new])
+    else:
+        combined = new
+    # keep="last": a refresh overwrites stale rows (actuals fill in
+    # after release, revisions land weeks later).
+    combined = (combined.drop_duplicates(subset=["id", "dateline"], keep="last")
+                .sort_values("dateline").reset_index(drop=True))
+    combined.to_parquet(path, index=False)
+    return combined
+
+
+def main() -> None:
+    ap = argparse.ArgumentParser(description=__doc__.splitlines()[0])
+    ap.add_argument("--start", type=date.fromisoformat, default=None)
+    ap.add_argument("--end", type=date.fromisoformat, default=date.today())
+    args = ap.parse_args()
+
+    start = args.start
+    if start is None:
+        if PARQUET.exists():
+            last = pd.read_parquet(PARQUET).dateline.max()
+            start = pd.Timestamp(last, unit="s").date() - timedelta(days=7)
+        else:
+            start = date(2024, 1, 1)
+    print(f"Scraping {start} -> {args.end}")
+    new = scrape(start, args.end)
+    combined = merge_into_store(new)
+    print(f"Store: {len(combined)} events "
+          f"({pd.Timestamp(combined.dateline.min(), unit='s')} -> "
+          f"{pd.Timestamp(combined.dateline.max(), unit='s')}) -> {PARQUET}")
+
+
+if __name__ == "__main__":
+    main()