Quantery 1.5.0: every number can now show its work
Quantery 1.5.0 is out. One theme runs through the whole release: a number you can't trace is a claim, and a research tool shouldn't ask you to take claims on faith. So in 1.5.0, every value a scan produces can explain itself, down to the SEC filing it came from.
Click any number, see the filing behind it
Your thesis's feature values now appear as columns in the survivors table, and every one of them is a button. Click it and an explain panel opens with the full story of that number:
- The criterion's rules with your thresholds filled in, and the rule that actually matched. "Scored 2 because fcf_yield >= 0.06", with the rules that didn't match shown dimmed.
- The value the scan recorded next to today's live value, labeled whenever they differ. They can differ for identifiable reasons: newer filings, backfilled fields, market movement, or different scan settings. Both are shown.
- Every input's fiscal period, filing date, and source, with a link to the filing on SEC EDGAR where the specific filing is stored.
Values built from several inputs (trailing-twelve-month sums, ratios, growth rates) appear in no single filing, so the panel labels them as computed and expands them: a free-cash-flow yield unfolds into the quarterly cash-flow reads underneath it, each with its own filed date and link.
The company page gets the same treatment. Score pills, the quant gate, and the thesis metric cards all open the panel, and a Data and filings link opens your stored fundamentals with identical provenance.
One limit worth knowing: rows ingested before 1.5.0 don't carry a stored filing accession, so their links go to the company's EDGAR filing index rather than the specific document. The panel says so instead of guessing. As fresh data comes in, the deep links fill in on their own.
Open any past run, and diff two of them
The Run history card is now interactive. Click any recorded scan and you're viewing that run's survivors, with a banner naming the run and a Back to latest button so you always know which day you're looking at.
With two or more recorded runs, a Compare button diffs any pair into Added, Dropped, and Still passing, with each symbol's score change and rank movement. The panel tells you when very large runs are compared on their top 1,000 rows and when the two runs used different (or unknown) thesis versions, so a diff never quietly mixes rulesets.
Know what your local data actually holds
Every company page now carries a Data coverage card: per dataset, the date range, row count, and freshness your local data holds, with sources for prices and fundamentals, plus missing calendar quarters and price gaps.
The wording maps strictly to recorded evidence. "Last collected" means a pull succeeded. "No 10-K or 10-Q data found" means SEC EDGAR was asked and had nothing. "Last attempt failed" means the pull errored. "Not collected yet" means exactly that. The card never claims a negative the data didn't record.
It also states the rule that quietly shapes every backtest: a company is included only on dates with visible fundamentals and fresh prices. Short or gappy history means it drops out of parts of a run, and now you can see that before you wonder why a name is missing.
The same audit trail for your AI assistant
If you've connected Claude or another assistant over MCP, it gets the same transparency you do. Three new read-only tools:
get_fundamentals: raw point-in-time fundamentals with filing identity and EDGAR references.explain_survivor: any survivor's numbers down to rules and filings, with unrecoverable provenance reported as unavailable so an agent can't fabricate attribution.get_symbol_coverage: the Data coverage card's per-symbol view.
get_survivors and get_scan also gain an opt-in include_features parameter for the full feature map. Setup lives in the docs.
Smaller changes and fixes
- Symbols where SEC EDGAR reports no 10-K or 10-Q data are rechecked every 30 days instead of every night. A confirmed empty answer is remembered; errors are always retried, and a manual full refresh still checks everything. On universe-wide nightly pulls this removes real repeated work.
- Re-running a scan on the same day no longer leaves results from the earlier run behind. Symbols that dropped out of the newer run are removed, and the job log notes when a re-run replaced a larger earlier result set.
- Data quality flags recorded during evaluation (stale filings, missing fundamentals) were being dropped before display. Each survivor now shows them as labeled chips, separate from the AI risk flag.
Getting 1.5.0
Windows and Linux update in-app. On macOS, grab the new build from the download page; in-app updates there arrive once signed builds ship. If you hit anything odd, the Bug Report button in the app's Settings pre-fills an email with your version and platform.
Research tooling, not investment advice. Nothing here is a recommendation to buy, sell, or hold any security. Screens, scores, and backtests are informational only; backtested results are hypothetical, exclude costs such as commissions and slippage, and do not guarantee future results. Verify against primary filings and make your own decisions.
Want to try this on your own rules? Quantery is free for 14 days: the full app, no card required.
← All articles