The old screener had a beautiful filter panel and a fatal flaw: filters composed with menus can only express what the menu designer anticipated. Every genuinely interesting question our users asked required a combination we had not built a control for, and the panel grew until finding the filter took longer than running the screen.
So we took the panel apart and replaced it with a query language of about a dozen keywords. You type conditions, they compose with and, or and parentheses, and the whole thing is a string you can save, share in a message, or paste into a backtest as a universe definition. `mcap > 2b and rvol(5) > 2 and close > sma(50) and sector != energy` is a screen, a bookmark and a piece of documentation all at once.
Everything is typed, so mistakes surface as you write rather than as an empty result set. Comparing a percentage to a dollar amount is a syntax error with a clear message. Field names autocomplete with their units and their update frequency inline, which turned out to matter more than we expected — a surprising share of confusing screens came from people combining a quarterly fundamental with an intraday technical without noticing the mismatch.
The new filter set covers thirty-one additions we could not previously express: short interest and days to cover, options skew and term structure, insider transaction windows, index membership changes, and a set of relative-strength measures computed against sector rather than the broad index. All of them work identically across equities, ETFs and crypto where the concept applies, and return an explicit "not applicable" rather than a silent zero where it does not.
Results stream progressively. The old screener assembled the full result set before painting anything, which felt slow even when it was not; the new one paints the first matches within about 200 ms and fills in as ranking completes. Median perceived latency on a broad-universe screen dropped by roughly half without the underlying query getting meaningfully faster.
The part we are proudest of is the smallest: any preset promotes to an alert or a backtest universe with one click, carrying its exact definition. The gap between "this screen is interesting" and "notify me when it triggers" or "how would this have performed" used to be twenty minutes of rebuilding. Now it is a button, and the definition can never drift between the three surfaces.
Lena Kowalczyk
Product Lead, Research Tools
Writes for AlgoBeam on product. Every figure quoted above can be reproduced in the backtester with the same settings.