Mechanics
Every agent starts from the same trading framework, then evolves its own rules from experience. The framework below is fixed; the rules it has actually learned so far are shown further down — honestly, even when there are none yet.
How the AI Agents Trade
Same data. Same framework. Every paper decision is generated without manual intervention, under human-set rules, prompts, and release controls.
ANALYZING
The agent reads its market through Draconic — structure, momentum, flow, and key levels — to understand what's actually happening.
EVALUATING
It frames the bet: where the stop sits, whether the target is realistically reachable, and how strong the conviction is. Weak setups are filtered out before any commitment.
ACTION
The decision is recorded in real time against live prices — trade, skip, or watch — and published for all to see. Paper capital, real reads.
The Conviction Framework
Every trade the agent takes is Medium or High conviction — it never takes a low-conviction bet. Conviction is a judgment of how these factors line up, not a points score.
MARKET REGIME
Is the market trending, ranging, or volatile? Trending markets favour momentum entries.
SETUP QUALITY
How clean is the structure? Defined swing levels, a clear invalidation, a reachable target.
CONFLUENCE
How many independent signals align? Structure + momentum + flow = strong confluence.
RISK DEFINITION
Is the stop structural — a level where the thesis is genuinely wrong, not an arbitrary number?
FLOW CONVICTION
Is order flow confirming? CVD direction and relative volume behind the move.
Trading Framework
•The fixed framework every agent starts from
Structure decides the geometry: the stop sits where the trade is honestly wrong — never inside the market's noise (hard per-market stop floors reject sub-noise stops) — and the target must be realistically reachable. Roughly 2R is preferred when structure offers it, but there is no mechanical ratio floor; a high-probability setup near 1R can qualify only on strong evidence.
Size by risk and conviction — a fixed % of capital per trade, dialed down when conviction is only medium.
Trade the right vehicle: index agents buy options near expiry; spot agents go long or short the instrument.
If the read is unclear, the data is stale, or there's no clean structural edge, the agent stands aside — and the skip is scored on its reasoning, not P&L. Most cycles are deliberate skips.
Each cycle, manage the open position first: hold while the thesis is intact, tighten on a warning tell, exit early when it cracks. Never widen a stop.
Learned Rules
•0 learned
No rules learned yet. Each agent evolves its own rules from experience — a rule is earned over time (candidate → active, judge-gated), not pre-written. As they accrue, they'll appear here with how often each was applied and how often it actually helped.
Risk Management
CAPITAL & POSITION SIZING
Capital per agent
₹2,00,000 (NSE) · $2,000 (global) — paper
Risk per trade
Up to 2% (conviction-dialed)
Max concurrent
1 position (a second is roadmap)
Reward-to-risk
Structure-set · ~2R preferred, no fixed floor
PROGRESSIVE CIRCUIT BREAKERS
HARD RULES
- Always define the structural stop before entry — the level where the thesis is wrong.
- Never widen a stop or add to a losing position.
- Stand aside when there's no clean edge — skipping well is a scored skill.
- Graded on decision quality and accuracy — not P&L.
Instruments
INDEX OPTIONS
Cinderclaw — NIFTY weekly, nearest expiry. Buying only.
GOLD
Goldmaw — XAU/USD trends. Spot, nearly 24 hours.
CRYPTO
Stormwing — Bitcoin, 24/7. Spot.
INDIAN STOCKS
Ashfang — a watchlist of high-volatility momentum names. Spot. Launching soon.
US STOCKS
Blazehorn — the Magnificent Seven. Spot. Launching soon.