Algo TradingML: P(H)RL: Skip/1R/2R
QuarterlyBot combines an ML estimate of probability P(H) with RL policies deciding whether to
skip a setup or take 1R/2R at a fixed risk.
Stop is placed beyond quarterly High/Low; targets RR 1:1–1:3, with shift to BE after 1:1 for higher targets.
Two modes available: auto-trading or notifications. Join the waitlist →
How it works: ML P(H) + RL decisions
First, we compute P(H) — probability a setup will play out.
A supervised ML model uses historical features: structure (SMT/PSP),
timing (TPD confirmations: 5m→90m/1H, 15m→4H/6H, 1H→1D), and context (BPR, True Opens,
session filters). The model sees both "micro" signals (candle/shape) and "macro" backdrop
(position vs. quarterly range, median, imbalances).
Then an RL agent classifies the candidate as Skip, 1R, or 2R
based on expected return/drawdown at unchanged risk (SL at quarterly HL).
It's discipline over guessing: clear trade-management rules consistent across assets and timeframes.
True Opens: anchor analysis to the higher-timeframe open — a unified entry standard.
BPR: seek imbalance and return to prior quarter median/50%.
CISD: entry invalid without a close beyond the divergent bar.
The core is a fixed SL beyond quarterly High/Low and stepped targets 1:1, 1:2, 1:3.
This reduces execution variance and makes trades comparable. For targets >1:2
we move to BE at 1:1, stabilizing the right tail of outcomes.
We don't optimize a “perfect” exit; we use simple robust rules that are easier to test and maintain.
A single “risk language” for ML and RL improves cross-asset generalization.
Rules are clear, transparent, and emotion-free.
Methodology is easy to validate retro and live.
Markets and constraints
The system auto-tunes for CME (indices, metals, energy) and major FX pairs
(e.g., EURUSD, GBPUSD, DXY). We do not provide financial advice — this is a research tool
for discipline and risk control. Respect liquidity rules and avoid exotics without testing.
Who it's for & how to use
QuarterlyBot suits traders who want to shift from improvisation to rules and reproducibility.
If “risk first, then profit” resonates — fixed SL, strict RR, and RL decisions give a clear structure. Typical cases:
Swing/positional: fewer but cleaner signals; less operational noise, easier control.
Intraday: more candidates, more noise — strong filters and discipline required.
Notifications: bot sends signal and P(H); trader decides — a softer onboarding path.
Auto-trading: RL policy executes the plan; user controls risk parameters and limits.
What the bot does not do: “promise returns”, “predict the market”, or replace your risk management.
It automates a repeatable process and helps you stick to the rules.
Methodology & validation: from idea to deployment
Pipeline: data collection/cleaning → feature generation (SMT/TPD/BPR/True Opens, HL context) →
ML training/temporal validation → RL policies trained on fixed-risk simulated episodes →
deployment with monitoring. Principles:
Honest validation: train/validation windows separated in time; no future leakage.
Data drift: monitor feature/quality distributions; recalibrate on deviations.
Metrics: not only P/L; look at expectation vs. drawdown, stability, skip-rate.
Transparent rules: fixed SL/RR simplify version comparison and case analysis.
Data & privacy
For the waitlist we collect only e-mail and optional X/Twitter handle — to notify when access opens.
No personal trading data is processed without explicit consent. See Privacy
and Terms. Markets and risk parameters on the user's side — under full user control.
Roadmap & transparency
Pine bot: basic signals, checklists (done).
ML estimate: P(H) for candidates (active calibration).
Backtesting: public methodology and reports without “history painting”.
Public access: gradual quota expansion via waitlist.
How it differs from classic “bots”
Many solutions are indicator bundles with tuned parameters. QuarterlyBot flips it:
fixed risk as the base, ML for probability, RL for decision rules.
Instead of chasing a “perfect filter”, we build a robust process where results are a function of discipline, not a magic indicator.
FAQ
Is this a fully automated trading bot?
Two modes: auto-trading by RL policy and notifications where the trader decides. Switch in profile settings.
What is P(H) and how is it computed?
P(H) is the ML-estimated probability a setup works out. We use structure (SMT/PSP), timing (TPD), context (BPR/True Opens), temporal validation, and drift monitoring.
Why is the stop at quarterly High/Low?
A fixed SL makes trades comparable, reduces overfitting, and gives RL a unified “risk language”.
Which markets are supported?
CME indices/metals/energy and major FX pairs (EURUSD, GBPUSD, DXY, etc.). Exotics only after testing.
What RR and position-management rules?
Targets 1:1, 1:2, 1:3. For targets >1:2 we move to BE after reaching 1:1.
Can I tweak system parameters?
Only user risk/limit parameters are adjustable. System rules (SL, RR, CISD, sessions) are fixed for reproducibility.
Do you have backtests and reports?
Yes — public reports with methodology are in progress. Focus on honest validation, not “perfect” historical curves.
Is this financial advice?
No. QuarterlyBot is a tool and methodology. Risk and capital decisions are the user's responsibility. See Risk Disclosure.
How do I get early access?
Leave your e-mail and X/Twitter handle on the waitlist. Invitations are sent in batches.
What if signal quality drops?
We monitor drift and calibration. If you see divergence — let us know; we'll inspect data segments and update the model/policy as needed.
QuarterlyBot