600+ passing tests behind every number on screen
EdgeTest replays historical crypto markets one bar at a time. You take entries and exits as the tape moves, pay spread, slippage and commission on every fill, and end up with a permanent, filterable record of what you actually did — plus a prop-firm evaluation simulator that enforces the rules that end most funded runs.
Why it exists
Scroll a chart back and you can already see where price went. Fill at the printed close and your costs vanish. Re-enter a symbol you are already long and a naive simulator happily opens a second, imaginary position. Each shortcut makes the equity curve prettier and the practice worthless.
EdgeTest was built to remove those shortcuts one at a time. Bars are revealed in sequence. Fills pay a half-spread offset, slip by a configurable number of ticks or percent, and are charged commission, optionally per side. A second order on a symbol you already hold nets against the existing position — adding and re-averaging, or reducing, closing, or flipping it — with realized PnL booked only on the part you actually closed.
It grew incrementally over many sessions, one milestone at a time: the replay engine, then the order and position accounting, then drawings and indicators, then prop-firm rules, then the journal, then the deep analytics. Each pass left its reasoning in the code — the conventions chosen, the alternatives rejected, and the things deliberately not built.
A human cannot look at a chart and tell whether a Sortino ratio, a trailing drawdown floor, or a netted average entry price is correct.
Every new money-math metric gets a hand-derived fixture. Not a snapshot of whatever the code happened to produce — an independently worked-out number the code then has to match. Snapshot tests only prove behaviour did not change. They cannot prove it was ever right.
How it’s tested
A backtester is only worth the arithmetic underneath it. This is how that arithmetic is held in place.
Fill costs, realized PnL, R-multiples, MAE/MFE, drawdown floors, payout dollars and netting splits are each asserted against numbers worked out by hand first and written into the test file. Nobody can eyeball whether a Sortino ratio is right by looking at a chart.
The engine is pure math with no React and no fetch, precisely so it can be tested this way. The replay scheduler takes its clock by injection, and the Monte Carlo resampler takes its random source by injection, so both are deterministic under test.
Session state, drawings, chart settings and templates are not assumed durable. They were verified by shutting the app down and bringing it back up, then checking the restored state matched.
Where a tool is not built, the interface says so and gives the reason, instead of rendering a button that quietly does nothing. Twenty-four toolbar entries are labelled that way today.
$ npx vitest run
Test Files 44 passed (44)
Tests 686 passed (686)Output from a real run of the suite in this repository.
What’s in it
Everything below is implemented and shipped in the app today.
Historical crypto candles, revealed one bar at a time, so you cannot see the future you are about to trade.
A paper fill at the printed close is a lie. Every entry and exit here pays the spread, slips, and is charged commission.
Trading the same symbol again does not open a second ticket. The engine nets against what you already hold.
Evaluations fail on rules, not on returns. This models the rules, including the ones that quietly end runs.
The numbers you would actually change your behaviour over — each one covered by its own fixture.
Every closed trade is a permanent record you can filter, annotate, and replay in its original context.
Seventy-one working drawing tools, thirty-two indicators, fifteen chart types. Nothing is a screenshot.
Cursor, balance, open positions, drawings, and chart settings are durable — proven by real cold restarts, not by hope.
BTC, ETH, SOL, BNB, XRP, ADA, AVAX and DOGE, each with its own tick size feeding the fill-cost model.
One session can hold several symbols with one balance across them, arranged in up to three panes.
Scrub back to any earlier bar and read the state as of that instant. Reviewing history cannot alter it.
Pricing
Accounts are free to create. Trial gives you one real session — up to 249 trades or 7 days from signup — to actually put the engine through its paces, not a crippled demo. Upgrading happens from the Account page after you're in — sales tax for your own country is calculated automatically at checkout.
One real session — a genuine trial, not a nagware demo.
Freeforever
Full advanced analysis and Prop Sim, on a lifetime cap.
$5per month
Everything, fully unlimited.
$10per month