ARTHASTRA — Investor Briefing

The Opportunity

Why Arthastra exists

◆ Market Opportunity

$2.4T Retail Investment Market

Bloomberg Terminal costs $24,000/yr. We deliver institutional-grade signal at 1% of the price — research analytics and defense intelligence purpose-built for the retail investor.

TAM: $24B addressable

⚡ Velocity Signal

Built in 30 Days. In Production.

Not because we had to — because the architecture was already clear. Speed is evidence of clarity, not haste. The system is live with real users and real picks being logged daily.

30 daysProductionLive

● Comparables

Bloomberg + Palantir + Koyfin

Bloomberg built market data infrastructure. Palantir built defense analytics. Koyfin built beautiful data. Arthastra merges all three — for the retail investor, at retail price.

BloombergPalantirKoyfin

Core Platform Architecture

Three-layer intelligence system

Arthastra operates as a layered intelligence system composed of three core engines. Each layer has a defined responsibility, a governance model, and a measurable output.

■ QUANT_LAB

Quantitative Research Engine

Processes market data across equities, crypto, commodities, and FX to generate structured research signals through feature engineering, regime detection, signal scoring, and statistical validation. Every research run produces deterministic manifests and experiment logs, ensuring all signals are reproducible and auditable.

Statistical Models Regime Detection Drift Monitoring Audit Trail

◆ YUNI

Platform Intelligence Layer

The operational governance layer that runs Arthastra behind the scenes. Continuously monitors platform health, signal engine status, and security events. All platform actions pass through a structured governance pipeline ensuring the system operates within defined boundaries. Maintains persistent operational memory of system activity.

Governance Pipeline Operational Memory Security Monitoring Anomaly Detection

◉ ASTRA

Research Interface

The analytical interface that surfaces market signals produced by QUANT_LAB and presents them through structured research summaries, pattern observations, and market context. Rather than executing trades automatically, ASTRA provides signal explanations and research context — every investment decision remains fully controlled by the investor.

Signal Explanations Research Summaries Pattern Observations Investor-Controlled

Governance Pipeline

INPUT → INTERPRETATION → POLICY REVIEW → CAPABILITY GATE → EXECUTION → AUDIT LOG

Every action inside the platform passes through this pipeline before execution. Unregistered actions are refused. This architecture prevents uncontrolled behavior and keeps the system operating within defined boundaries.

ASTRA Picks Performance

Live trading signals

Win Rate —

Picks —

Avg Return —

Ticker	Entry Price	Date	Confidence	Outcome	Return
Connecting to data feed…

Monthly Win Rate %

Loading monthly data…

Confidence Score vs Actual Return

System Changelog

How we improve ASTRA's pick quality — documented

✓ Deployed 2026-03-09

Confidence Floor Added

PROBLEM

29 picks below 60% confidence had a 0% win rate and -4.59% average return. ASTRA was executing low-conviction signals with no edge.

0% win rate on 29 low-confidence picks

FIX

All BUY/SELL picks below 70% confidence are now blocked before reaching the trade log.

Only 70%+ confidence picks accepted

Impact: Eliminates the worst-performing tier. High confidence (80+) picks historically show 97% win rate.

✓ Deployed 2026-03-09

Duplicate Pick Guard Added

PROBLEM

ASTRA's signal loop was emitting the same pick on every cycle. XOM BUY appeared 17 times in one day, GDXJ BUY 11 times — inflating pick count and distorting the track record.

Up to 17 identical picks per ticker per day

FIX

One BUY or SELL per ticker per day. Any duplicate is blocked and logged with the reason.

Maximum 1 pick per ticker per day

Impact: Pick count now reflects unique decisions, not loop noise.

✓ Deployed 2026-03-09

Direction Conflict Lock Added

PROBLEM

ASTRA generated both BUY and SELL signals on XLK the same day. The SELL was correct (100% win rate). The BUY was wrong (0% win rate). Conflicting signals cancelled each other out.

BUY and SELL on same ticker same day was possible

FIX

If a BUY exists for a ticker today, a SELL is blocked, and vice versa. First signal wins.

Direction locked per ticker per day

Impact: Eliminates contradictory positions on the same ticker.

Live Signal Gate Activity

Accepted —

Rejected —

Low Confidence Blocked —

Duplicates Blocked —

Conflicts Blocked —

Time	Ticker	Action	Confidence	Status	Reason
Loading gate activity…

Quant Model

Multi-asset performance breakdown

🗓 Live Soon

Full Quant Breakdown — In Development

The quant model is actively processing picks across equities, crypto, commodities, and FX. Asset-class accuracy breakdowns, rolling accuracy trend charts, and pick volume analytics will populate here automatically as sufficient data accumulates. We show nothing until the numbers are real.

Now Live

Equities picks

Now Live

Crypto signals (ARBI)

Q2 2026

Commodities

Q3 2026

FX coverage

ARBI — Live Crypto Signal Engine

Running 24/7 · Binance.US live data · Paper execution

Live — Training Continuously

What ARBI Is Building

ARBI is Arthastra's autonomous crypto signal engine. It runs 24/7 against real Binance.US market data — live WebSocket prices across 8 symbols — executing every trade in paper mode. The result is a continuously growing knowledge base of real signal quality data: which strategies work in which regimes, how long positions should be held, and when not to trade. No money is at risk. The learning is genuine.

Live Now

Binance.US WS feed

Live Now

4 strategy stack

Live Now

Regime detection

Live Now

Railway 24/7 deploy

Q2 2026

Live execution

Community Trades

Collective Win Rate

Recent Signals

Engine Status

Strategy Breakdown

Loading strategy data…

Portfolio Exposure

Loading portfolio…

Ticker	Strategy	Direction	Confidence	Time
Connecting to ARBI engine…

ARBI Engineering Log

Data integrity fixes — documented

✓ Deployed 2026-03-16

Synthetic Price Guard — Entry & Exit

PROBLEM

BNB/USDT has no live feed on Binance.US. The engine was trading it using freshly re-seeded synthetic prices each tick — producing fabricated PnL like +21.24% / $4.19 / 0-minute holds. When live data failed for any symbol, open positions were also being exited using synthetic prices, producing fake returns of +15%, +24%, or -60%.

Fabricated PnL corrupting session data

FIX

Both entry and exit evaluation now check source == "synthetic" and skip entirely. Positions stay open until real WebSocket or live REST data returns. Synthetic fallback is now inert.

Only live-priced trades enter the knowledge base

Impact: All PnL in the knowledge base now reflects real market conditions. BNB remains in the symbol list and will activate automatically when a live feed is available.

✓ Deployed 2026-03-16

Session Win Rate — Metric Isolation Fix

PROBLEM

A live Binance.US session showed "Trades this session: 0" alongside "Win rate (this session): 41.4%" — a statistical impossibility. The session win rate was reading all-time historical data from the DB instead of in-memory session counters.

Historical metrics were mislabeled as session metrics

FIX

Added a dedicated session_wins in-memory counter, incremented only on real closed exits. Session win rate now derives exclusively from this counter, with an N/A guard when no trades have closed.

Session and all-time metrics fully isolated

Impact: Session reports now show statistically valid numbers. Historical all-time metrics are still tracked separately in the knowledge DB.

✓ Deployed 2026-03-16

Shutdown Report — Open Position Accounting

PROBLEM

Pressing Ctrl+C printed no session summary at all — the shutdown report was swallowed by a pipe race condition between Python and tee. Even when visible, all positions showed "Trades: 0" because stop() wrote directly to the DB, bypassing the trade counter.

Session report was empty or invisible on shutdown

FIX

stop() now routes all open positions through the standard _paper_exit() path at entry price (pnl=0, reason=session_end), ensuring they appear in the trade count. Stdout is explicitly flushed before the process exits.

Full session report printed reliably on Ctrl+C

Impact: Every session now produces a complete summary: trades, win rate, balance, exit breakdown, and regime breakdown — even when interrupted mid-session.

War Room Alpha — Defense Sector Intelligence

Geopolitical signal mapping

🗓 Live Soon

Geopolitical → Market Correlation Engine — In Development

War Room Alpha maps active geopolitical conflicts to defense sector equities, tracking the lag between conflict escalation and stock price movement. This is Arthastra's core differentiator — institutional-grade defense intelligence made accessible. Signal data will populate here as the engine goes live.

Q2 2026

Conflict mapping

Q2 2026

Defense correlations

Q3 2026

Live signal feed

Q3 2026

Alert system

Platform Stats — Live Metrics

As of today

Total Picks Logged

Overall Win Rate

Days Live

Waitlist Users

Cumulative Performance — Equity Curve (Normalized 100)

Based on closed picks · Normalized basis

Intelligence at the intersection of markets, data, and geopolitics.

The Opportunity

Core Platform Architecture

ASTRA Picks Performance

System Changelog

Quant Model

ARBI — Live Crypto Signal Engine

ARBI Engineering Log

War Room Alpha — Defense Sector Intelligence

Platform Stats — Live Metrics

Ready to discuss an investment in Arthastra?

Intelligence at the intersection of markets, data, and geopolitics.

The Opportunity

Core Platform Architecture

ASTRA Picks Performance

System Changelog

Quant Model

ARBI — Live Crypto Signal Engine

ARBI Engineering Log

War Room Alpha — Defense Sector Intelligence

Platform Stats — Live Metrics

Ready to discuss an investment in Arthastra?

Express Your Interest