{
  "hypothesis": "H6 FADE THE PROVEN-WRONG: Telegram/WA signal channels whose calls had significantly NEGATIVE market-adjusted forward returns in a ranking window keep being wrong in a later window, so trading against them earns a positive market-neutral return net of costs. Proprietary data (our captured channel signals).",
  "registered_utc": "2026-11-04T23:30:59Z",
  "signals": "shadow_lab.db shadow_trades rows with variant_id LIKE '%_MIRROR' (one row per captured signal); provider = variant_id minus '_MIRROR'; dedup: first (provider, symbol, side) per rolling 24h",
  "prices": "Bybit linear 1h klines (public API). Entry = OPEN of the first 1h bar starting after the signal timestamp (knowable). Signals on symbols with no Bybit klines are dropped and counted",
  "outcome": "direction-adjusted market excess at 48h: d * [(coin exit/entry - 1) - mean(BTC, ETH same window)], d = +1 LONG / -1 SHORT",
  "ranking_window_A": "signals 2026-05-15 .. 2026-08-31",
  "fade_group": "providers with >= 30 deduped signals in A and day-clustered t of mean excess <= -2.0 in A",
  "follow_group_control": "providers with >= 30 in A and t >= +2.0 (persistence of skill, reported)",
  "trading_window_B": "signals 2026-09-01 .. 2026-11-01 (48h complete)",
  "fade_return": "-excess - 0.00214 (0.107% round trip on the coin AND on the BTC/ETH hedge)",
  "metric": "mean fade return per signal in B across the fade group",
  "confirm_bar": "B mean > 0 with day-clustered t >= 2.0 AND week-clustered t >= 1.5",
  "fail": "fade group empty, or B day-clustered t < 1.5 -> fading is not an edge",
  "secondary_reported": ["24h and 72h horizons", "per-provider B results", "Spearman of provider mean excess A vs B (persistence across all providers n>=30 in both)", "fade ALL providers pooled", "follow-group B result", "share of B fade return from the largest provider"],
  "no_tuning": "windows, horizon, threshold and costs fixed here"
}