# Biotech desk

> An open-source, unattended nightly research desk for small-cap biotech stocks, written in Python with no third-party dependencies and released under the MIT licence. Built by Svetoslav Rankov, Senior Director of Software Engineering at EDITED (https://sved.net), who specifies, directs and reviews it while an AI agent writes most of the code. Source: https://github.com/svedbg/pharma

Every weeknight it reads SEC EDGAR filings, prices, Form 4 insider trades, FINRA short volume and ClinicalTrials.gov records for a watchlist, computes deterministic signals, and notifies only when a name changes tier. A veto layer blocks any buy signal that sits on a priced share offering, a delisting notice, an auditor change, a recent collapse, or under 1.5 quarters of cash, until that veto is refuted with evidence from the filing.

Key measured facts (against the XBI biotech ETF, three-month horizon):
- "Buy near the yearly low" underperformed by 2.68%.
- "More oversold is better" underperformed by 9.59%.
- Oversold confirmed by heavy selling volume outperformed by 1.72%; this is the only rule that survived and the actionable tier requires it.
- Oversold without volume confirmation underperformed XBI by 2.7 percentage points median over 60 sessions.

Operating facts:
- About 59 companies scanned per weeknight in roughly four minutes.
- Data cost is zero; every source is free and keyless.
- 273 automated tests (counted at build time); one fails the build if the runtime gains a third-party import.
- Signal tiers: none, watch, setup, act. Only act is actionable.
- A separate heartbeat process alarms if no report appears for three weekdays, because silence is the system's normal output.

It is research support for the user's own decisions. It is not financial advice, places no orders, and is not a product.

## Pages
- [Landing page](https://desk.sved.net/): overview, veto layer, tier ladder, measured results, setup, FAQ
- [The 91% question](https://desk.sved.net/brief.html): long-form capability brief on why a 91% one-day fall is usually not a buying opportunity
- [Source repository](https://github.com/svedbg/pharma): code, README, CLAUDE.md architecture notes