9-MONTH AUTONOMOUS RESEARCH EXPERIMENT

A market research system that debates before it trades.

18+ specialist agents analyse Indian equities and options, debate trade setups, pass proposals through a deterministic risk engine, and test every decision with live Upstox market data against ₹2,00,000 virtual capital.

LIVE STORY:PRE-MARKETAGENT DEBATEPAPER TRADE3:30 AUTOPSYCHALLENGER

Virtual Capital

₹2,00,000

Mode

PAPER ONLY

Agents Online

29 / 29

Champion

v0.9.3

Blackboard Status
INGESTION
Consensus:81% BUY
Risk Engine:EVALUATING
Execution:SIMULATED ₹2L
29 Specialist Agents Orbiting
AUTONOMOUS WORKFLOW

How Quantara Operates Daily

A disciplined, machine-driven cycle from pre-market 07:00 preparation to post-market 15:45 forensic autopsy and safe Champion/Challenger evolution.

STEP 0109:15 - 15:30

Observe & Ingest

Direct streaming connection to Upstox Market Data Feed V3 via Protobuf WebSocket. Ticks, market depth, and full option chains pass through an automated Data-Quality Gate.

Engine:Upstox V3 WS
STEP 02Event-Driven

Multi-Agent Debate

16 Equity Specialists and 13 Options Committee Agents publish structured evidence to a shared blackboard. The Bull Agent builds the trade, the Bear Agent attempts to disprove it, and the Critic checks for stale claims.

Engine:29 Specialist Agents
STEP 03Pre-Execution

Deterministic Risk Gate

Hardcoded risk controls check max portfolio exposure, 1% risk-per-trade limits, R:R >= 1.5, and daily drawdowns. The risk engine has absolute veto power over a 100% AI consensus.

Engine:Zero AI Override
STEP 04Immediate

Realistic Paper Fill

Simulates institutional execution: Market buys fill at Ask + slippage, sells at Bid - slippage. Full Indian regulatory charges (STT, exchange turnover, GST, SEBI fee) are deducted from ₹2,00,000 ledger.

Engine:Simulated Only
STEP 0515:45+

Post-3:30 Trade Autopsy

Every completed trade—profitable or losing—undergoes an in-depth forensic autopsy. The system evaluates 10 questions, tags failure codes (F01-F23), and tests counterfactual replays.

Engine:10 Questions + F01-F23
STEP 06Nightly / Weekly

Champion vs Challenger

Autopsy hypotheses generate bounded Challenger models. The Champion remains frozen intraday while Challengers run in live shadow portfolios and must pass 8 statistical gates before promotion.

Engine:Safe Self-Improvement
SPECIALIST ARCHITECTURE

Multi-Agent Intelligence Network

29 specialist agents with strictly isolated responsibilities publish machine-readable evidence to a shared blackboard. Numerical features are computed deterministically.

candidate_detector

Candidate Detector Agent

Scans pre-market and intraday market depth to surface candidate setups.

Calibrated Reliability:92%
global_market

Global Market Agent

Overnight/global indices, FX, commodities, yields, and risk tone.

Calibrated Reliability:87%
macro

Macro Agent

RBI policy, inflation, bond yields, economic calendar, and rate shifts.

Calibrated Reliability:93%
news

News Agent

Company, sector, and exchange news with source and time relevance.

Calibrated Reliability:92%
corporate_event

Corporate Event Agent

Quarterly results, board meetings, dividends, splits, and order book wins.

Calibrated Reliability:89%
fundamental

Fundamental Agent

Growth, operating margins, balance sheet quality, and earnings context.

Calibrated Reliability:85%
market_regime

Market Regime Agent

Trend vs range, volatility state, risk-on/risk-off classification.

Calibrated Reliability:87%
sector_rotation

Sector Rotation Agent

Relative sector strength, market breadth, and capital flow distribution.

Calibrated Reliability:89%
technical

Technical Agent

VWAP, EMA ribbons, RSI momentum, MACD, ADX trend strength, and ATR.

Calibrated Reliability:83%
price_action

Price Action Agent

Opening range breakout (ORB), structure, candle formations, and key retests.

Calibrated Reliability:86%
volume_liquidity

Volume / Liquidity Agent

Relative volume (RVOL), turnover, spread quality, and volume exhaustion.

Calibrated Reliability:92%
microstructure

Microstructure Agent

Bid/ask imbalance, depth absorption, and aggressive participant flow.

Calibrated Reliability:88%
correlation

Correlation Agent

Index beta, sector concentration, and correlated setup overlap.

Calibrated Reliability:85%
anomaly

Anomaly Agent

Outlier volume, unusual OI spike, and erratic price jumps requiring quarantine.

Calibrated Reliability:95%
bull

Bull Agent

Constructs the strongest evidence-based upside thesis for candidates.

Calibrated Reliability:92%
bear

Bear Agent

Actively seeks disconfirming evidence, downside targets, and failure traps.

Calibrated Reliability:92%
critic

Critic Agent

Detects contradictions, stale evidence, unsupported claims, and ignored warnings.

Calibrated Reliability:94%
decision

Decision Agent

Produces structured BUY/SELL/NO-TRADE proposal with strict validation.

Calibrated Reliability:82%
9-MONTH EXPERIMENT IN PROGRESS • DAY 1

Live Paper-Trading Portfolio

Real market data from Upstox V3 streaming feed with ₹2,00,000 initial virtual capital. No real capital at risk.

Current Phase

Day 1 / 180

Starting Capital

2,00,000

Simulated master

Current Virtual Equity

2,00,000.00

₹0.00 net return (Day 1)

Max Drawdown

0.0%

Tolerance 5.5%

Profit Factor

0.00

0 trades closed

Official Strategy

v0.9.3

Champion Frozen

Live Broker Orders

0

PAPER_ONLY=true
RADICAL EXPLAINABILITY

First-Class "NO TRADE" State

Quantara is rewarded for avoiding poor setups, not maximizing trade frequency. The expected default for most instruments is NO TRADE.

Reasoning Inspection

BANKNIFTY

NO_TRADE

Decision Rationale

Consolidating in 250-point range; potential target sits at 51,450 resistance. Calculated reward-to-risk is 1.1:1, failing the deterministic 1.5:1 minimum threshold.

Veto / Trigger: Deterministic Risk Engine

Expected Gain: ₹3,100 vs Risk: ₹2,800. Ratio: 1.10 (Minimum: 1.50)

Audit Trail ID: #CHK-1842Zero Guesswork
CHAMPION VS CHALLENGER

Safe, Gated Self-Improvement

A single losing trade never rewrites production logic. Champion parameters remain frozen during live market sessions. Challengers run in shadow simulation and must pass 8 strict promotion gates.

CHAMPION STRATEGY

v0.9.3 (OFFICIAL)
Frozen during live market session; zero intraday modifications.
Sole strategy authorized to execute simulated ₹2,00,000 trades.
Contributes to official capital curve, headline P&L, and audit ledger.
Status:Active Production Baseline

CHALLENGER (SHADOW)

v0.9.4-alpha
Created from post-3:30 trade autopsy hypotheses (theta risk mitigation).
Evaluates against rolling historical replay and live shadow portfolio.
Cannot touch master portfolio until all 8 scorecard gates pass.
Shadow Result:Scorecard 8/8 Passed (Auto-Promote Ready)
Anti-Overfitting Rule: "Yesterday lost, so change everything today" is prohibited. Learning occurs from repeated patterns and validated hypotheses, not emotional reaction to single trades.
SEBI & COMPLIANCE GUARDRAILS

Non-Negotiable System Safety

Designed from the ground up as an auditable scientific research environment. Safe defaults ensure complete platform integrity.

Hardcoded PAPER_ONLY=true

Real broker order-routing code is strictly disabled behind a hard feature flag. The system cannot transmit live orders to exchanges during the 9-month experiment.

No Stored Credentials or TOTP

Uses official Upstox OAuth with short-lived tokens expiring daily at 3:30 AM. No password, PIN, or TOTP automation is ever permitted or stored.

Deterministic Risk Sovereignty

Risk limits, position sizing, margin limits, and kill switches are 100% deterministic code. No LLM or agent can bypass risk controls.

Full Audit Trail & Reproducibility

Every market tick, debate transcript, proposal, and fill is timestamped and persisted with evidence IDs, agent versions, and prompt snapshots.