Starting paper capital
Workspace / Overview
PAPER ONLY
K
NIFTY · BANK NIFTY
Research. With a record.
A private workspace for testing options ideas with Jev, Codex and your Brain.
◈
Paper workspace is ready
Zerodha selected. Connect your Kite app in Broker & data to record market quotes.
Continuous paper account
Checking status…
Latest experiment P&L
—
Run an experiment to measureMaximum drawdown
—
Latest baseline · after estimated costsCompleted experiments
0
Saved with data & configuration hashesEXPERIMENT PERFORMANCE
Paper equity
⌁
Your evidence starts here.
Import historical option snapshots or run a synthetic fixture to test the software.
Rule baseline Baseline + Jev filterNo-trade reference: ₹1,00,000
SYSTEM CONNECTIONS
Your research stack
AI scores are research inputs. Position sizing and entry limits are enforced by code.
THE RESEARCH LOOP
Each step can be inspected.
01ObserveTime-stamped snapshots
02CompareBaseline + Jev
03ConstrainRisk & execution rules
04ReviewCodex + human labels
FORWARD OBSERVATIONS · SEPARATE PAPER ACCOUNTS
Continuous paper
NIFTY & BANKNIFTY. Each account has its own balance, trades and risk limits; results are never combined.
Original account: cash, trades and loss locks are retained.
Waiting for worker status.
Runs on the server while your browser is closed. Regular trading days, 09:15–15:30 IST; new entries before 15:00, session exit from 15:15. Reconnect Zerodha each trading day in Broker & data when its session expires.
Paper equity
—
Cash + executable exit estimateAvailable virtual cash
—
Preserved across restartsCumulative paper P&L
—
After estimated costsClosed paper trades
0
One open position maximumLuna · Financial Brain
View all 13 method statuses
Python supplies calculations and enforces risk. Luna interprets the available evidence for entry and HOLD / EXIT. Estimated or missing inputs stay labeled; attached methods do not establish a profitable strategy.
Trade-profit probability
Not yet validated
Model confidence describes an answer, not the chance of profit. Rule confidence means only a rule matched. New Luna decisions use their own frozen calibration cohort. Jev distributions, Luna choices and historical trades are never pooled. Trade-profit probability needs separate evidence from closed paper trades after modeled fees and slippage. It does not imply positive expected profit. AI and hosting costs are excluded. Research only; this layer cannot change entries, exits or risk limits.
Validation evidence
Jev / OpenAI / rules comparison
Checking scheduled paper comparison…
Open position
Loading current paper-account limits.
Loading exit policy.
Market context
Twenty fresh price snapshots are needed; these are not OHLC candles. When free to enter, scans up to 21 nearby strikes per call/put type within 2% of each index, at the nearest future expiry. Selects the nearest contract passing budget, risk, spread and depth checks. A Jev decision can fill only on a later quote.
Continuous paper ledger
LATEST 100 EXITSSnapshot simulation with estimated fees and slippage. Top-of-book quantity must cover a simulated fill; missing prices never become invented fills. Full records remain stored on the server.
Shadow ledgers
FROZEN · SAME QUOTESCounterfactual ledgers on this account’s own quotes: no trade, the deterministic rule with fixed exits, the same rule with a trailing stop, and the live entries with fixed exits. Rules and limits are frozen when the ledgers start; daily loss is observed, not enforced. Each difference isolates one component; none is validated profitability.
Observation & decision journal
Latest 80 events, including warmup and no-trade reasons. All events are stored.
Daily financial review
MEASURE BEFORE YOU SCALE
Experiments
Every run starts with its own virtual balance. Results are never compounded across experiments.
1. Choose your evidence
Synthetic quotes and lot sizes are invented software fixtures. They do not represent exchange prices or historical performance.
Download CSV format example ↗One contract per file, 60–3,000 bid/ask snapshots. Lot size and expiry come from your file. Imported provenance remains unverified.
2. Run a controlled comparison
₹20 fee per side plus estimated levies. Loss-based entry lock: 2% per day; 10 entries per day; premium capped at 10% of initial capital. Gaps may exceed planned stop losses.
70% context / development30% evaluation
Signals use only prior observations; fills use the next quote. Threshold 0.8 is provisional, not calibrated accuracy. Reusing the evaluation slice makes it development data.
Jev budget: 250 decisions per run, 10 comparisons/day. No live order endpoint exists.
Experiment history
TRACE → LABEL → IMPROVE
Decision journal
Inspect inputs and the earliest failure before changing a prompt or strategy.
Run an experiment to collect decision traces.
BOUNDED AI REVIEW
AI research
Codex reviews recorded experiments with methods retrieved from your Brain.
Review only: no orders, shell tools or risk-policy changes. Reports are AI judgments, not independent financial validation. Up to 5 reviews/day.
ZERODHA · KITE CONNECT
Broker & data
Connect your account, record real option quotes, then test with ₹1,00,000 virtual capital.
1. Set up your Kite app
In Kite developer console ↗, choose Connect with market data and use these settings:
- App name
- Keyoor Options Lab
- Redirect URL
https://bot.keyoorai.com/broker/zerodha/callback- Postback URL
- Leave blank
Connect data plan is listed at ₹500 per app/month. Confirm the final amount in Zerodha before subscribing. The free Personal plan does not include live or historical data.
Save or replace app credentials
2. Connect your account
Login and OTP happen on Zerodha. Kite sessions expire at 06:00 IST the next day or earlier if revoked; reconnect when needed.
This app uses broker access only for market data. It cannot place, modify or cancel orders. Kite credentials themselves may permit trading, so keep them private.
Disconnect removes this workspace's saved session and stops recording. It does not revoke access on Zerodha.
3. Record option quotes
Regular session: 09:15–15:30 IST on trading days. One contract at a time, roughly one snapshot per minute. Quotes older than 60 seconds or with missing depth are rejected. Recording stops at session end, on pause, or after three consecutive errors. Holidays are not independently verified.
This version records bid/ask depth, spot, volume and OI. Greeks and IV are not supplied here. At least 60 accepted snapshots are needed; save the recording as a dataset to run the existing paper simulator. No automatic trading runs.
Recordings
SOURCE-GROUNDED ENGINEERING
Brain & methods
Saved knowledge becomes explicit prompts, checks and measurable experiments.
Trading skills
13 research methods now feed Luna entry/exit context and Codex reviews. Each card shows what the paper engine currently supports and what still needs data or code.
AI engineering methods
These sources inform prompts and system design. Financial research and recorded trades provide trading evidence.
Prompt contracts
The exact prompts used by this version. Dataset, configuration and model versions are saved with each run.
Luna · financial Brain, entry and exit
Jev · observed regime classification
Codex · experiment review
These are our applications of engineering sources. The named authors have not endorsed this strategy. Options strategy and market-microstructure expertise still need specialist review.