LIVE PRODUCTION · 90+ ENDPOINTS · 14 BACKGROUND THREADS

TradeOnly is discipline as software. built for the 0.1%. a self-improving cage. your unfair advantage.

A self-improving, multi-path trading system that prevents the silent ₹1.3M loss pattern retail traders unknowingly reproduce. Built on four years of structured backtest data and live Dhan integration. Every order is checked by 11 cage rules and 9 phases of intelligence before it reaches the broker.

₹0M
Loss Prevented
Historical bypass losses now blocked at order-router level
0 yrs
Backtest Moat
1,086 structured pattern outcomes since 2022 — live-matched in real time
0%
Cage Open-Rate
Multi-path cage opens daily — up from 1.5% in single-gate v1
↓ scroll · the data tells the story
₹1.3M LOSS PREVENTED 4 PATHS TO A GOOD TRADE 11 WALLS AGAINST A BAD ONE 1,086 BACKTEST PATTERNS SELF-IMPROVING NIGHTLY 90+ LIVE ENDPOINTS 14 BACKGROUND THREADS DISCIPLINE AS SOFTWARE ₹1.3M LOSS PREVENTED 4 PATHS TO A GOOD TRADE 11 WALLS AGAINST A BAD ONE 1,086 BACKTEST PATTERNS SELF-IMPROVING NIGHTLY 90+ LIVE ENDPOINTS 14 BACKGROUND THREADS DISCIPLINE AS SOFTWARE
The Problem · Indian Retail Options Market

Retail traders lose discipline, not direction.

According to SEBI's 2023 study, 90% of retail F&O traders lose money over any given financial year. Not because they pick wrong directions — they pick correctly 43% of the time. They lose because they trade more when they should trade less, size up after losses, and override their own rules at exactly the wrong moments.

90%
of Indian retail F&O traders lose money. Average loss: ₹1.1L per trader per year. (SEBI Jan 2023)
36×
overtrading on bad days. Our reference trader took 36 trades in one session — daily cap was 6.
5×
EXIT_NOW signals manually overridden in a single day — every override compounded losses.
24×
worse outcomes per trade when traders bypass their own rules. Our data: -₹212/trade with rules · -₹5,107/trade without.
The Insight That Changed Everything
The discipline already works. The bypass is the bug.
We analyzed 301 trades across our reference operator's history. When trades passed through the system's rules, win rate was 50% and average outcome was essentially break-even. When the same trader bypassed the rules — via TradingView webhooks, manual Dhan, or one-click overrides — the same trader's win rate dropped to 38% with a ₹5,107 average loss per trade.

The traditional advice — "trade better signals" — solves the wrong problem. The real fix: make the system structurally unbypassable while opening the door for more good trades. That's what we built.
The Solution · Cage v2 Multi-Path

Four paths to a good trade. Eleven walls against a bad one.

Earlier discipline systems either traded everything (defeats the point) or traded nothing (no opportunity). Cage v2 introduces a multi-path approval system: any of four independent signal pathways can open the cage. Behind that, the bridge enforces eleven structural refusals — at the order-router level — so even a moment of weakness can't fire a bad trade.

Path 1 · Standard
Classic Consensus
Decide ≥ 70% AND
Predict ≥ 70% AND
Regime aligned
The textbook entry: macro signal + strike-level signal + trending regime all agree.
Path 2 · Big Trade Lane
Single-Signal Conviction
BigTrade STRONG GO
(score ≥ 80) +
safety rails intact
When a single signal is screaming — institutional-grade conviction overrides need for consensus.
Path 3 · Consensus
Three-Signal Quorum
Decide ≥ 60% AND
Predict ≥ 60% AND
BigTrade ≥ 70
Looser bar but requires three independent signals to all vote moderately positive.
Path 4 · Breakout
Momentum-In-Motion
Breakout score ≥ 85
+ valid strike/side
+ regime not CHOP
Catches gamma-squeeze moves where premium is already exploding and waiting costs you the trade.
The 11 Order-Router Refusals (Phase 1 Defense)
The system refuses orders, not warns about them.
① Daily loss cap breached
② Daily trade cap breached
③ Tilt cool-down active (15-30 min after big loss)
④ Auto-flat hard-lock (3+ system overrides today)
⑤ Sizing tier locked (>10 lots needs 3 greens)
⑥ Breakout daily cap (max 2/day)
⑦ Red-day un-confirmed
⑧ CHOP regime hard-refuse
⑨ Coach EXIT_NOW lockout (60s)
⑩ Re-entry cooldown (5 min same direction)
⑪ Pre-entry quality score < 50
DAILY LOSS CAP · TILT COOL-DOWN · CHOP REFUSE · EXIT_NOW LOCKOUT · RE-ENTRY COOLDOWN · QUALITY FLOOR · SIZING TIER · BREAKOUT CAP · RED-DAY CONFIRM · AUTO-FLAT RATCHET · HARD-LOCK DAILY LOSS CAP · TILT COOL-DOWN · CHOP REFUSE · EXIT_NOW LOCKOUT · RE-ENTRY COOLDOWN · QUALITY FLOOR · SIZING TIER · BREAKOUT CAP · RED-DAY CONFIRM · AUTO-FLAT RATCHET · HARD-LOCK
The Moat · Why This Compounds Over Time

The system gets smarter every night the user sleeps.

Three asymmetric assets create an unfair, defensible position that no competitor can replicate without years of accumulated data and engineering. Every trade feeds the moat; every night the moat deepens.

◆ Asset 1 · Pattern Outcomes
1,086
structured backtest patterns since 2022
Four years of gap_fade, expiry_pin, and 5+ other intraday setups stored with full entry/target/SL/exit history. Live-matched against current state every poll. A new entrant would need 4 calendar years of accumulated data plus the schema-design work to even begin replicating this. Live endpoint: GET /patterns/live.
◆ Asset 2 · Self-Tuning Classifier
2,388
labeled gate-state observations, growing daily
Every cage decision is logged with full signal context, joined with the trade outcome that followed. A heuristic ML classifier retrains hourly and adjusts entry-quality scores by ±10 points based on which signal combinations are working for this specific user, in this specific market regime. The system that sleeps with you wakes up smarter than you.
◆ Asset 3 · Daily Auto-Calibration
16:00
IST · automated nightly retrain (weekdays)
After every market close, the bridge retrains the classifier and recomputes per-hour win-rate multipliers from the last 30 days. Tomorrow's cage carries yesterday's lessons — without the user lifting a finger. The classifier has already learned that 13:00 IST has a 75% historical win-rate and 09:00 IST has 25%. Tomorrow's cage acts on that knowledge.
◆ Asset 4 · Adaptive Cage Itself
±15pts
dynamic threshold range based on context
Cage thresholds adjust live based on: 30-day source win rates (winning sources get easier thresholds), VIX regime (high vol → lower bar), auto-flat ratchet (past mistakes today raise tomorrow's bar), and per-hour performance. The cage is never the same twice. Competitors with static rules can't keep up.
How It Works · 4 Phases of Intelligence

Defense + Offense + Self-Improvement + Microstructure.

Every order is evaluated through four progressive layers of intelligence. The first layer blocks structural mistakes. The next three layers evaluate quality, learn from outcomes, and read real-time microstructure for edge.

01
Phase 1 · Defense
Make the existing system unbypassable.
Order-level enforcement at the bridge layer. TradingView webhooks, manual Dhan, one-click fires — all subject to identical refusal rules. No trade that violates discipline reaches the broker.
TV-webhook cage Re-entry cooldown CHOP refuse EXIT_NOW lockout Quality floor 50
02
Phase 2 · Signal Generation
Turn unused data into decision-grade signals.
The 1,086-row pattern table comes online. Open Interest delta tracking surfaces institutional accumulation vs short squeezes. Greeks-aware sizing scales position based on Gamma/Theta/Delta — not gut feel.
Pattern matcher OI delta Greeks intel
03
Phase 3 · Self-Improvement
The system learns from every outcome.
Multi-timeframe confluence scores 1m + 5m + 15m signal agreement. The pattern classifier trains on labeled gate-states. Daily auto-calibration retrains weights at 16:00 IST. The next morning's cage carries the lessons of yesterday's trades.
Multi-TF confluence ML classifier Daily auto-cal
04
Phase 4 · Microstructure
Real-time tape, pre-market context, position scaling.
Tape filter approximates aggressor flow from spot velocity. Pre-market routine fires at 09:00 IST and computes max-pain, gamma flip strikes, and vol regime classification. Position scaling suggests pyramid-adds on winners that still have momentum.
Tape filter Pre-market engine Pyramid scaling
By The Numbers · Engineering Rigor

A real production system, not a notebook.

This isn't a strategy in a Jupyter notebook. It's a production-grade observability + execution platform with full ops infrastructure, automated tests, reboot-survival, and self-healing background services. Every commit is gated by 116 verification markers.

0
REST endpoints
authenticated · rate-limited · audited
0
Background threads
market-aware · weekday-only
0
Backend lines
Python · Flask · single-file bridge
0
Frontend lines
React · Vite · 22 production tabs
0
SQLite tables
trades · gate_history · patterns · snapshots
0
Refusal rules
order-router level · zero bypass
0
Verification markers
36 frontend + 80 backend · all green
0
Auto-recovery levels
L1 cache-clear · L2 sentinel · L3 self-restart
Tech Stack · Production Infrastructure

Built like a hedge-fund's trade desk, runs on a Mac.

Boring, reliable, well-understood components — composed thoughtfully. No ML frameworks, no GPU dependencies, no exotic infrastructure. The complexity is in the rules, not the stack.

🐍 Python 3.12 · Flask
⚛ React 18 · Vite
🗄 SQLite (8 tables)
📡 Dhan REST API + websockets
☁ Cloudflare Tunnel + Access
🔐 AWS EC2 SSH (static-IP egress)
🎙 Web Speech API (Sudha voice)
⚙ launchd KeepAlive (reboot survival)
📊 Bloomberg-inspired UI
🛡 Per-tab rate limiting

Discipline at the moment of action is a software problem.
We solved it.

— TradeOnly · Investor Brief · 2026
PYTHON FLASK REACT VITE SQLITE 8 TABLES DHAN REST API CLOUDFLARE TUNNEL AWS EC2 STATIC IP LAUNCHD KEEPALIVE 116 VERIFICATION MARKERS PYTHON FLASK REACT VITE SQLITE 8 TABLES DHAN REST API CLOUDFLARE TUNNEL AWS EC2 STATIC IP LAUNCHD KEEPALIVE 116 VERIFICATION MARKERS
Roadmap · From Personal System to Platform

Built for one. Built to scale to many.

Today the system runs as a single-operator platform — battle-tested through real losses and real wins. The path to scale is clear and incremental: each phase compounds the previous, each step is already de-risked by the live deployment.

Live
Personal trading desk
Single-operator deployment. Cage v2 with 4 paths + 11 refusal rules. 9 phases of intelligence operational. Self-improving daily. Hosted on Mac mini with Cloudflare Tunnel + AWS static-IP egress.
Phase Next
Multi-broker · multi-asset
Generalize the cage layer beyond Dhan. Plug in Zerodha, ICICI Direct, Upstox via the broker abstraction. Add NSE equity + commodity options. Same discipline framework, same moat.
Phase Future
B2C SaaS · sub-account model
Multi-tenant deployment. Each user gets their own cage instance, their own classifier weights, their own pattern library. Subscription tier model. The system that knows YOUR trading personality.
Phase Future
Institutional licensing
White-label the cage engine for prop desks, family offices, RIAs. The discipline layer is broker-agnostic, asset-agnostic, and strategy-agnostic. Sells into anyone with execution authority.
Phase Future
Pattern marketplace
The 1,086-pattern library becomes a community-contributed asset. Vetted setups + win-rates published transparently. Network effects compound the moat: more users → more patterns → better cage for everyone.
Phase Future
Trader behavior analytics
Aggregate (anonymized) bypass-attempt patterns, override frequencies, regime-specific behaviors. Sell behavioral analytics to brokers + regulators concerned about retail F&O losses. Mission-aligned monetization.

The discipline crisis is a software problem.
We've already solved it.

A live, production-grade trading discipline platform, built and battle-tested by a working options trader. The moat is real. The technology is shipped. We're ready to scale.

📧 Request a demo ↑ Re-read the data