Question 1

Has any of this been backtested — can I trust it as a signal?

Accepted Answer

No backtesting exists, and the README says so directly: the judgment rules have not been validated against returns, and Jev's output probabilities are not win rates. Position management, automatic order placement, and buy, stop-loss and target prices are explicitly outside the project's scope. It is built for reading market structure at the latest close, not for deciding entries.

Question 2

What does Jev actually judge, and what does the code decide?

Accepted Answer

Python computes the indicators, identifies the candlestick patterns and builds the technical facts, then Jev scores direction per dimension with Choice and volatility risk with Noul. Aggregating direction, disagreement and volatility risk back into one buy, hold or sell is done in code, so the summary rules are inspectable rather than model prose. The applicable horizon is picked from the two the model evaluated.

Question 3

What happens when the market data or the model call fails?

Accepted Answer

You get no verdict rather than a default one. If the model or the required quotes fail, no buy, hold or sell is output, and with no Jev key configured the interface shows that no verdict has been formed. During a scan the job pauses automatically on a model failure or three consecutive data-processing failures, and can be resumed later with --resume and --retry.

Question 4

What does it cost to run, and what do I need installed?

Accepted Answer

Python 3.11, plus Node 22.12 for the web interface; Windows goes through WSL. Tushare Pro needs your own token with access to the endpoints in use, while AKShare needs no key, and every model call is billed to your own TypeSafe account. Because there is no technical pre-filter, model calls scale roughly with the size of your scan list, and batches are capped at 50 codes.

Question 5

Which indicators does it actually compute?

Accepted Answer

Eight dimensions, listed rather than tallied. Trend covers SMA and EMA at 5, 10, 20, 60, 120 and 250 plus WMA, HMA, KAMA, DMI and ADX, Aroon, SAR and Supertrend. Momentum has MACD, RSI at 6, 14 and 24, KDJ, Stochastic, CCI and divergence candidates on confirmed turning points. Volume-price, volatility with ATR and Bollinger, price structure, 14 candlestick patterns, multi-timeframe resonance and relative strength versus CSI 300 and the sector complete it.

Question 6

What leaves my machine, and who pays for it?

Accepted Answer

The workbench and its records stay local, but technical state is sent to TypeSafe, and fetching quotes needs the network. Quotes come from your own AKShare or Tushare account; the project neither provides, hosts nor resells market data. Data-source permissions and model-call charges fall to your own accounts. It is a local single-user service with no public multi-user authentication, MIT licensed, and developed independently with no affiliation to TypeSafe AI.

Question 7

Can I scan a whole watchlist, and what gets filtered out?

Accepted Answer

Paste codes or use your watchlist and scan in batches of at most 50, deduplicated and in order, with progress, pause, resume and retry; Ctrl+C pauses. By default it filters ST names, delisting consolidations, no-volume names and those with too little history, and each stock enters the model as soon as its quotes arrive instead of waiting for the list. The job auto-pauses on a model failure or three consecutive data failures, then resumes with --resume and --retry.

jev-A-share-trader

Highlights

Questions about this entry

More Jev resources like this

From the Jev guides

From the Jev community