📘 Trading Playbook

A. Intraday Hour-of-Day Bias

• Setup: Mode=Hour, H=1h, ColorBy=TStat, Lookback=3y, TZ=local.
• Plan: trade core setups only in top 3–5 green hours; in red hours reduce size/skip.
• Sizing: T≥2 → 1.0R; 1≤T<2 → 0.5R; T<1 → micro/skip.
• Risk: watch news/regime shifts.

B. Day-of-Week Swing Filter

• Setup: Mode=DOW, H=1d, ColorBy=Mean, Lookback=8y.
• Plan: add with trend on green days; use red days for partials.
• Risk: holidays/rollovers may lower n.

C. Monthly Tilt

• Setup: Mode=Month, H=1d, ColorBy=Sharpe, Lookback=10y.
• Plan: step up risk in historically strong months; be conservative in weak months.
• Risk: structural cycles can shift patterns.

D. Multi-Horizon Robustness

• Compare H=1h vs 4h and Lookback=2y vs 8y; overlap = stronger evidence.

E. Time-Sizing in Bots

Map normalized T-stat (0..1) to a size multiplier; enforce n ≥ MinObs.

What the tiles show (selected months)

• Oct — t=2.9, 64% | n=36 → one of the strongest months on this 8-day horizon; historically, forward 8-day shifts skew positive and statistically meaningful.
• Sep — t=0.9, 65% | n=34 → moderate positive tilt with solid sample size.
• Jan — t=1.1, 57% | n=35 → mild positive bias.
• Apr — t=1.1, 68% | n=34 → positive with a strong hit-rate cluster.
• May — t=-0.4, 46% | n=35 → slightly negative/neutral; treat with caution.
• Aug — t=-1.8, 43% | n=35 → the weakest month on the map; negative and relatively strong in magnitude.

How to trade this picture

• Lean in: If your system gives a long setup during October (or a nearby strong month like September), allow entries and consider a full size; monitor news/volatility as usual.
• Stand down / size down: In August and (to a lesser extent) May, be more selective with longs or favor quicker profit-taking; aggressive contrarian shorts still need proper technical triggers.
• Look for clusters: Here, Sep–Oct forms a two-month green cluster (consistency > single tile).
• Respect n: All months show n≈33–37, which is healthy for Month mode; still, validate on your symbol/timeframe.
• Align with horizon: The bias refers to +8 days from entry; plan targets/stops consistent with that holding period.

📎 Atlas vs Timeframe

🧭 First principles

Atlas computes forward return on the current chart timeframe. Switching TF (M15/H1/H4/D1…) changes:

• the bar stream and calculations,
• how Horizon is interpreted,
• the sample size n per bucket,
• the volatility context (ATR, winsorize impact).

⏱️ Horizon logic

• Bars — horizon is tied to TF: H=3 on H1 = 3 hours; on H4 = 12 hours.
• Hours/Days — fixed real time; internally converted to bars of the current TF.

Practice: want stable economic meaning across TFs → use Hours/Days.
Need “next N bars on this TF” precision → use Bars.

🗺️ Which mode fits which TF

• Hour-of-Day — use intraday TFs (M5–H1). On D1/W1 most bars fall into the same hour → the map loses meaning.
• Day-of-Week — best on H1–D1: enough observations and a clear weekly rhythm.
• Month — D1/W1 for calendar seasonality (swing windows).

🎛️ What changes when TF changes

• Sample size n: higher TF ⇒ fewer observations ⇒ more grey n/a (via Min Observations).
• Metrics (Mean/T-stat/Sharpe) are recomputed on the new bar series; color scale is re-normalized.
• ATR normalization uses ATR of the current TF — that’s OK: you’re analyzing the regime of this TF.

🚀 Plug-and-play presets by goal

🟩 Intraday timing (scalp/day)
TF: M15–H1 • Mode: HourOfDay • Horizon: 1–2 Hours • Color By: TStatistic • Lookback: 2–4y • MinObs: 30–60 • TZ: your session.
Use: filter best hours; size down/skip in red hours.

🟦 Session/weekly bias
TF: H1–H4–D1 • Mode: DayOfWeek • Horizon: 1 Day • Color By: Mean or Sharpe • Lookback: 5–10y • MinObs: 50–100.
Use: plan adds/partials by weekday.

🟨 Calendar seasonality (swing)
TF: D1/W1 • Mode: Month • Horizon: 1 Day (or 3–8 Days for swing) • Color By: TStat/Sharpe • Lookback: 8–10y.
Use: step up in strong months, be conservative in weak ones.

🧱 Robustness — when to tweak

• Winsorize % (0.5–2%) — taming fat tails (crypto, newsy weeks).
• ATR Normalization ON — compare regimes with different vol (XAU, BTC).
• Min Observations — raise on higher TF to hide noisy buckets.
• Timezone Offset — correct assignment by hours/days (critical for Hour/DOW).

🔎 Mini examples

A) XAUUSD intraday “strong hours”
TF=H1 • Mode=Hour • Horizon=1 Hour • ColorBy=TStat • Lookback=3y • TZ=+0.
Read: green hour clusters → allow/size longs; red → de-risk.

B) EURUSD weekly rhythm
TF=H1 • Mode=DOW • Horizon=1 Day • ColorBy=Mean • Lookback=8y • TZ=+3.
Read: Mon–Fri ranked; Sat/Sun = n/a (no trading).

C) BTCUSD monthly swing tilt
TF=D1 • Mode=Month • Horizon=8 Days • ColorBy=TStat • Lookback=8y • TZ=+0.
Read: Sep–Oct green cluster → favorable swing window; Aug red → caution.

✅ Do / ❌ Don’t

Do

• Match Horizon to your intended holding period.
• Prefer clusters + adequate n over single bright tiles.
• Cross-check multiple lookbacks (e.g., 3y vs 8y) and multiple horizons.

Don’t

• Use Hour mode on D1/W1.
• Compare TFs with Horizon=Bars expecting identical real-time length.
• Ignore TZ on intraday.

🧯 Quick troubleshooting

• Too many n/a → lower TF, extend Lookback, reduce MinObs.
• Map looks “totally different” after TF change → switch Horizon to Hours/Days to keep meaning stable.
• Colors feel “overheated” → increase Lookback, raise MinObs, set Winsorize 1–2%.
• Hour buckets misaligned → fix Timezone Offset.

🧩 One-liners

• Need exactly +1 day regardless of TF → Horizon = 1 Day.
• Need next 3 bars on this TF → Horizon = 3 Bars.
• Make crypto comparable across years → ATR Norm ON + Winsorize 1–2%.
• Hide weak buckets → MinObs 50–100 (DOW/Month) and 30–60 (Hour).