Pick an instrument and a date. The chart plays forward one candle at a time and you place simulated trades seeing exactly what you would have seen at the time — no more. It is a practice ground for discretionary traders.
No account. No API key. No server. Market data goes straight from the provider to your machine. Nothing this project operates ever touches it.
Free and open source under the AGPL-3.0. There is no paid tier and never will be.
Pre-release. Everything below works and is covered by 121 tests running on Windows and Linux, but this has not been through real-world use yet. Treat the numbers as a practice tool, not as evidence for risking money.
Grab the installer for your system from the latest release, or build it yourself.
Installers are unsigned — the project has no code-signing certificate — so the first launch shows a warning:
| What you will see | What to click | |
|---|---|---|
| Windows | "Windows protected your PC" | More info → Run anyway |
| macOS | "cannot be opened because the developer cannot be verified" | System Settings → Privacy & Security → Open Anyway |
| Linux | nothing | — |
- Open the app. The form is already filled in with Bitcoin and a recent month.
- Press "Download this range." Crypto arrives in about five seconds.
- Press "Start replay."
- Press play, or tap → to step one candle at a time.
- Buy or Sell when you see something you like.
That is the whole product. Everything below is detail for later.
| Control | |
|---|---|
| Space | play / pause |
| → ← | one candle forward / back |
| Shift + → ← | ten candles |
| 1m … 1w | change timeframe; your place in the replay does not move |
| Jump to | skip to a date |
| Review | stats, equity curve, journal, export |
Stepping backwards really undoes things. Step back past the moment a trade filled and that trade is gone — not hidden, gone. Step forward and it fills exactly as it did before. Your position can never be a leftover from a future that has been rewound.
A backtester that quietly flatters you is worse than none at all, because it manufactures confidence you have not earned. Three places where this one refuses to:
⚠ assumed trades. When one candle touches both your stop loss and your take profit, there is genuinely no way to know from OHLC which came first. This app assumes the stop hit first — the pessimistic answer — flags the trade, and counts it on the Review screen. It will not pretend to know.
Real vs synthetic spread. The badge in the top right always states which you are getting. A spread you typed in is your assumption, not history, and the app says so rather than letting a number look like a fact.
Gaps stay gaps. Missing data is never interpolated or forward-filled to make a chart look tidy. A closed market and a hole in the provider's data are drawn differently, because reading price action across a data hole is a mistake.
The full contract is in docs/invariants.md — worth reading if you intend to trust the output.
| Provider | Covers | Key needed? | Speed |
|---|---|---|---|
| Binance | Crypto, full minute history | No | ~5s per month |
| Dukascopy | Currencies and gold | No | slow; start with a week |
| CSV | Your own file | No | instant |
| Databento | CME futures, equities | Yes — your own | untested, see below |
Both free providers work on first launch with nothing configured.
Your API key is yours. It lives in your OS keychain — Windows Credential Manager, macOS Keychain, Linux secret service. Never in a config file, never in a log, never sent anywhere but that provider's own API.
Downloaded candles are cached so you never fetch twice, and nothing is deleted automatically. Each session also keeps its own copy of the candles it started with, so a provider quietly revising history later cannot change trades you already took.
Close the app whenever you like, including with a position open. It is never force-closed — the position is saved exactly as it stood and resumes at the same candle.
Sessions are stored as a log of what you did, which is also why replaying one always produces the same trade log.
Review → Export writes trades.csv, round-trips.csv, journal.csv and
session.json next to the session. Local files. Nothing is uploaded.
Note what the review screen refuses to invent: profit factor shows — rather than infinity when nothing has been lost yet, and average R shows — when a trade was taken without a stop, because R is undefined without one.
Stated plainly, because a trading tool that hides its gaps is the problem it claims to solve:
- Tick-derived spread is not wired up. Dukascopy publishes real bid/ask, but the app still charges the spread you typed. The badge tells you the data supports better, not that you are getting it.
- The Databento adapter has never run against the live API. It needs a paid key. Its symbol handling is very likely wrong as shipped.
- Holiday detection is a heuristic, not a real calendar, so an unusual outage can be mislabelled as a holiday.
- Installers are unsigned (see Install).
- The UI has no automated tests. The Rust core has 121.
No automated strategies or optimisers. No live broker connections. No accounts, no cloud sync, no sharing. No AI features. No server component of any kind.
Where more features and less friction were in conflict, friction won.
git clone https://github.com/TimFarmr/bar-replay
cd bar-replay/ui && npm install && cd ..
# Build an installer for your platform (output in target/release/bundle/)
cd crates/app && ../../ui/node_modules/.bin/tauri buildFor development, run npm run dev in ui/, then cargo run -p bar-replay-app
in another terminal.
Use
tauri build, notcargo build --release, for anything you intend to run standalone. A plain cargo build leaves the app pointing at the dev server instead of embedding the front end.
There is also a CLI for inspecting the data cache without the UI:
cargo run -p replay-cli -- instruments
cargo run -p replay-cli -- fetch binance BTCUSDT --from 2024-01-01 --to 2024-02-01
cargo run -p replay-cli -- candles binance BTCUSDT --tf 1hTests: cargo test --workspace. The no-lookahead property test takes about two
minutes on its own — it runs eight thousand queries.
CONTRIBUTING.md covers the layout and house rules. Read docs/invariants.md first; several decisions in docs/adr are deliberate and load-bearing.
AGPL-3.0-only. See LICENSE.
Chosen to prevent a closed-source hosted fork. Market data must never be routed through infrastructure this project operates — that constraint is legal, not stylistic.