Backcast Labs against a typical subscription platform

That is the plan, and how much work it is depends on what it is written in. Anything already in Python — your own script or a strategy from a Python framework — goes straight into Strategies → Import and comes out as a notebook. Anything written in another platform’s own scripting language has to be converted to Python first, and that step is not ours. Backcast Labs reads Python and nothing else, and the site says so here rather than after launch. That conversion is where the losses happen, and they are not losses we can report on because we never saw the original. Step two is ours: the importer maps entries and exits onto engine orders, promotes every tunable constant to a sweepable param(), and prints a repainting audit — every read of data that was not available at that bar, with its lookahead flag — before the first run. Behaviour with no engine equivalent, an intrabar risk rule for instance, comes back as a warning, not a silent substitution.

No, and that is the honest answer rather than a comfortable one. The guest notebook has no importer and no repainting audit. It ships the output of an import — the Imported signal stack notebook, already loaded — so you can see what a strategy from elsewhere looks like once it is in and what the validation does to it. The importer itself, the lookahead flags and the drift monitor are in the installed engine, on your own files. If rehearsing the import on your own strategy is what would decide it, join early access and we will tell you when the installed engine is available.

The community algorithm library and the pre-aligned cloud data. Also, and less often admitted: your past runs, your saved reports and your research history do not come across. There is no importer for them, so the research log starts on your first run here. And Backcast Labs does not trade — it validates and writes a signed configuration for whatever executes for you, so if your cloud backtester was also your execution venue, you are replacing one product with two. What you gain is unlimited bars, your strategies on your own disk, validation methods that would otherwise be homework, and no monthly bill for your own CPU time.

Only if its timestamps carry a zone. The importer refuses naive timestamps outright — no guessing, no "assume UTC" — because ambiguous sessions are the single largest source of accidental lookahead, and a header that merely says “UTC” is not the same as a zoned timestamp. If your export from another tool is naive, you re-stamp it before it loads. Beyond that: CSV and Parquet only, so an XLSX or PDF statement has to be converted first, and exchange history pulled through your own keys is subject to that venue's rate limits, which for a multi-year backfill is measured in hours rather than minutes.

No for the studio — templates, parameters and every validation panel are graphical. Yes if you want to write a strategy from scratch: that is plain Python against a small engine API, about thirty lines for a trend follower.

The axes are the ones Backcast Labs is built around: one-time pricing, local execution, honest defaults, built-in validation, data breadth and no lock-in. A platform picked for its community library or its fundamental screener will win on axes that are not in this table.

Most people do. Chart wherever you like, research in a notebook, then run the honest test here before any capital moves. The engine exposes a Python API precisely so it can sit inside an existing workflow.