Research

A living equity research dashboard.

An autonomous system tracks 20 stocks, reads their news and SEC filings every night, scores each informational event, and flips ratings by formula. No human in the loop. Every rating change is logged with the exact events that caused it.

Updated nightly. Last run: July 24, 2026.

This is an automated research experiment, not investment advice. An AI model scores public information and a formula flips ratings. It can be wrong, late, or stale. Nothing here is a recommendation to buy or sell anything. I may hold positions in stocks shown. Do your own research.

Ratings

7 Strong Buy · 2 Buy · 9 Hold · 2 Sell

Where the model currently stands across the universe.

Latest flips

TSLA Hold → Sell (Jul 19) · AAPL Hold → Strong Buy (Jul 19) · UNH Buy → Strong Buy (Jul 17) · V Hold → Buy (Jul 17) · CRM Buy → Hold (Jul 17)

Evidence base

3,210 events

News stories and SEC filings scored so far. Every one is on the record.

The board

Click a column to sort. Click a ticker for its full event history.

Ticker Company Rating Last event
JPM JPMorgan Chase Strong Buy Jul 21 Thesis →
GS Goldman Sachs Strong Buy Jul 21 Thesis →
UNH UnitedHealth Strong Buy Jul 21 Thesis →
XOM Exxon Mobil Strong Buy Jul 21 Thesis →
AVGO Broadcom Strong Buy Jul 21 Thesis →
PLTR Palantir Strong Buy Jul 21 Thesis →
AAPL Apple Strong Buy Jul 21 Thesis →
TSLA Tesla Sell Jul 21 Thesis →
MSFT Microsoft Sell Jul 21 Thesis →
AMD Advanced Micro Devices Buy Jul 21 Thesis →
V Visa Buy Jul 21 Thesis →
WMT Walmart Hold Jul 21 Thesis →
LLY Eli Lilly Hold Jul 21 Thesis →
META Meta Platforms Hold Jul 21 Thesis →
ORCL Oracle Hold Jul 21 Thesis →
CRM Salesforce Hold Jul 21 Thesis →
NVDA NVIDIA Hold Jul 21 Thesis →
GOOGL Alphabet Hold Jul 21 Thesis →
COST Costco Hold Jul 21 Thesis →
AMZN Amazon Hold Jul 21 Thesis →

How it works

Every night after the US close, a pipeline on my machine pulls each stock's company news and fresh SEC filings (8-K, 10-Q, 10-K, 13D/G). New events are deduplicated against everything already seen, then scored in a single AI pass against a written rubric: an integer from -5 (thesis-changing bad) to +5 (thesis-changing good), measuring informational surprise relative to what a reasonable holder already expected. Most news scores zero. That is the point.

Each event's score then decays exponentially with a half-life that depends on what kind of information it is: earnings and guidance fade over ~45 days, legal and regulatory news over ~60, product news over ~14, macro noise over ~10. A stock's cumulative score is the sum of its still-active contributions, recomputed fresh every night.

Ratings flip by formula with hysteresis: the model enters Buy above +5.0 and Strong Buy above +7.5, but does not exit a tier until the score falls well below where it entered (+3.5 and +6.0 respectively, mirrored for Sell at -5.0/-3.5). Those gaps keep ratings from flapping on borderline scores. Every flip is recorded with the exact events that triggered it, which you can inspect on each stock's page.

News data via Finnhub. Filings via SEC EDGAR. Scores and ratings are model states, not recommendations. This dashboard exists to test whether informational events can be scored systematically, and to keep the system honest by making its full history public.