JEV INVESTMENT LAB

Real market data. Real decisions. Historical experiment.

DECISION-MAKING EXPERIMENT

JEV-20260922-V2

A historical simulation that asks whether a structured decision model can repeatedly choose under uncertainty — using stock investing as the test bed, not as a product.

WHY THIS EXPERIMENT?

Jev is a new class of AI model designed to make structured decisions rather than generate traditional text.

I wanted to test a simple question:

“How good is Jev at making difficult decisions when given a large amount of context, a defined set of choices, and real constraints?”

Stock investing is a useful test case because decisions are:

  • uncertain
  • sequential
  • context dependent
  • constrained by capital
  • affected by previous decisions
  • evaluated over time rather than on a single question

This experiment uses stock investing as a decision-making test bed. It is not intended to provide financial advice or demonstrate an automated trading strategy.

EXPERIMENTAL — NOT INVESTMENT ADVICE

This is a historical simulation designed to explore AI decision-making. It is not a recommendation to buy or sell securities and is not an automated trading system.

WHAT IS THIS EXPERIMENT TESTING?

INPUT

  • Market data
  • Technical context
  • Portfolio state
  • Available capital
  • Decision constraints

JEV

  • Structured decision model
  • typesafe-ai/jev

DECISION

  • BUY
  • HOLD
  • SELL
  • NO ACTION

PORTFOLIO

  • Holdings change
  • Cash changes
  • Value marked at close

1.Can Jev make decisions repeatedly rather than answer a single question?

2.Can it make decisions while respecting portfolio constraints?

3.Can it incorporate changing market and portfolio context?

4.Can it decide when to enter and exit positions?

5.How does its decision-making behave over a sustained historical replay?

The experiment evaluates the quality and behavior of the decisions produced by Jev. Portfolio return is an outcome of the simulation, not the sole definition of decision quality.

EXPERIMENT SCOPE

CAPITAL

₹10,00,000

UNIVERSE

NIFTY 100 (current)

PERIOD

2026-03-10 → 2026-08-31

TRADING SESSIONS

118

MODEL

typesafe-ai/jev

DECISION FREQUENCY

End of each trading session

EXECUTION

Next trading session OPEN

DECISION OPTIONS

BUY / HOLD / SELL / NO ACTION

MAX POSITIONS

5

MAX INITIAL POSITION ALLOCATION

20%

LEVERAGE

None

SHORTING

None

TRANSACTION COST

10 bps

SLIPPAGE

5 bps

BENCHMARK

NIFTY 100 buy-and-hold

EXPERIMENT GUIDELINES

These rules were frozen before the replay and were not changed based on observed results.

  • Jev evaluates eligible stocks at EOD.
  • BUY on an unheld stock can initiate a position.
  • BUY on an already-held stock does not add shares.
  • HOLD maintains the existing state.
  • SELL exits the entire existing position.
  • SELL on an unheld stock does nothing.
  • NO ACTION leaves the portfolio unchanged.
  • Maximum 5 positions.
  • Sell decisions are processed before new buys.
  • A position slot is freed only after a sell actually executes.
  • New buys are executed at the next trading session OPEN.
  • Portfolio is marked using CLOSE.
  • No leverage.
  • No shorting.
  • Transaction costs and slippage are included.
  • Remaining capital stays in cash.

No future market information is supplied to Jev for a decision.

WHAT INFORMATION DID JEV RECEIVE?

For each eligible stock on each decision date, Jev received structured market context and the live portfolio state — only information available as of that session’s close.

MARKET CONTEXT

  • 1 / 5 / 20 / 60 day returns
  • SMA 20 / 50 / 200
  • RSI
  • Volatility
  • Volume / volume ratio
  • Distance from 52-week high
  • Drawdown
  • NIFTY 100 market returns
  • Relative performance

PORTFOLIO CONTEXT

  • Current holdings
  • Available cash
  • Portfolio constraints
  • Current position state

JEV DECISION

  • BUY
  • HOLD
  • SELL
  • NO ACTION

RESULT

₹10,00,000₹11,22,524.61

Jev simulated return

+12.25%

NIFTY 100

+1.11%

Trades

19

Successful decisions

10,299

Max positions

5

The result surprised me — but the return is not the experiment’s only output. The more interesting question is what Jev actually decided, when it decided it, and why those decisions produced this portfolio path.

CAN WE TRUST THE REPLAY?

The V2 experiment was independently audited after completion.

  • Trade-level accounting✓ PASS
  • Portfolio reconstruction✓ PASS
  • Constraint validation✓ PASS
  • Yahoo raw OPEN validation✓ PASS
  • Benchmark reconciliation✓ PASS

EXPERIMENT PORTFOLIO

₹11,22,524.610143

INDEPENDENT RECONSTRUCTION

₹11,22,524.610143

DIFFERENCE

₹0

The experiment database and methodology were not modified during the audit.

External NSE reference data was requested for all 19 trades, but NSE returned HTTP 503 for those requests. Those reference fields are therefore recorded as unavailable; the experiment’s Yahoo Finance data was not replaced.

Open the audit explorer →

IMPORTANT LIMITATIONS

This is an experiment, not evidence that Jev can reliably outperform the market.

  • Historical simulation only.
  • 118 trading sessions is a short evaluation window.
  • The universe uses current NIFTY 100 membership, introducing survivorship bias.
  • Yahoo Finance data is an experimental data source and is not an authoritative commercial market-data feed.
  • This is not live trading.
  • Results do not establish future investment performance.
  • The experiment was not designed as financial advice or a deployable trading strategy.

EXPLORE THE EXPERIMENT

The Overview tells you WHAT happened.

Replay shows you HOW it happened.

Portfolio shows you WHERE the money moved.

Trades shows you WHAT Jev actually did.

Analysis helps understand WHY the decisions matter.

Audit checks WHETHER the result is internally consistent.

DATA & REPRODUCIBILITY

Technical cache details for readers who want the underlying market-data footprint. This section is secondary to the decision-making story above.

MARKET DATA CACHE

Local Yahoo Finance cache behind JEV-20260922-V2. Experiment window uses sessions from 2026-03-10 through 2026-08-31.

Stocks cached

100 / 100

Trading sessions in cache

384

Failed downloads

0

Cache as of

2026-09-22

SOURCE

VendorYahoo Finance · yahoo-finance2 4.0.2

UniverseFixed current NIFTY 100 universe

Constituent filedata/universe/ind_nifty100list.csv

BenchmarkNIFTY 100 (^CNX100)

Session clockAsia/Kolkata

Report time22/9/2026, 20:56:58 IST

CACHE WINDOW

First session requested2025-03-03

First trading session2025-03-03

Last trading session2026-09-22

Experiment start2026-03-10

Experiment end2026-08-31

Placeholder sessions excluded2026-01-15, 2026-05-01, 2026-05-28, 2026-06-26, 2026-09-14

ADJUSTMENT

Yahoo reported 9 split events. Overnight close ratios classify 9 as already split-adjusted and 0 as unadjusted traded prices. Dividend events: 259. Symbols where adj_close differs from close: 95. A split-adjusted history downloaded on 2026-09-22 embeds split factors whose ex-date is after a simulated decision date. That changes the rupee level, not the split-adjusted return. `adj_close` also embeds later dividends and is not a point-in-time price.

LATER LISTINGS

SymbolFirst real sessionSessions before listing
ENRIN2025-06-1972
TATACAP2025-10-13151
TMCV2025-11-12171

SAMPLE BARS

Reliance Industries Ltd. RELIANCE.NS

DateOpenHighLowCloseAdj closeVolume
2025-03-031,2041,206.451,1561,171.251,161.221,79,44,938
2025-03-041,162.21,1741,159.551,161.91,151.951,13,77,373
2025-03-051,1611,1831,1571,175.61,165.5386,64,095
2026-09-181,2451,247.31,226.41,226.41,226.41,51,22,715
2026-09-211,234.11,249.11,232.51,247.41,247.41,00,07,218
2026-09-221,247.61,251.91,237.41,240.41,240.41,06,81,733

HDFC Bank Ltd. HDFCBANK.NS

DateOpenHighLowCloseAdj closeVolume
2025-03-03869.9871.53847.05850.78825.772,12,05,364
2025-03-04849857.15846.63855829.871,99,20,958
2025-03-05850.98855.4844.13845820.172,15,83,872
2026-09-18715.25733.8715.257317313,94,00,710
2026-09-21731742.65729.05739.5739.53,54,82,683
2026-09-22738.85749.3738.6738.6738.63,54,73,028

Infosys Ltd. INFY.NS

DateOpenHighLowCloseAdj closeVolume
2025-03-031,692.31,728.61,692.31,708.61,624.3475,04,969
2025-03-041,6951,6991,6701,688.31,605.0467,59,673
2025-03-051,692.451,732.951,692.451,711.51,627.181,80,782
2026-09-181,061.91,061.91,0381,051.41,051.42,21,24,953
2026-09-211,041.51,044.51,030.31,038.51,038.560,74,358
2026-09-221,037.71,0411,019.21,029.41,029.493,38,840

INDUSTRIES IN THE FILE

  • Financial Services23
  • Automobile and Auto Components9
  • Capital Goods9
  • Fast Moving Consumer Goods8
  • Healthcare8
  • Metals & Mining7
  • Information Technology6
  • Oil Gas & Consumable Fuels6
  • Power6
  • Construction Materials4
  • Consumer Services4
  • Chemicals2
  • Consumer Durables2
  • Realty2
  • Services2
  • Construction1
  • Telecommunication1