What is this page?

The portfolio overview for 14 tracked companies across four sectors: AI Infrastructure, Defense & Security, Quantum Computing, and Energy & Power. Every number on this page updates automatically through a pipeline of scheduled tasks, API calls, and analyst vetting. Click any ticker to open a deep dive with financials, analyst details, and health scores.

Stat boxes

Upgrades / Downgrades (30d) — how many of your 14 tickers received an upgrade or downgrade from a vetted analyst in the last 30 days. Only actions from analysts who passed the composite scoring system are counted (see "How analysts are chosen" below). A ticker with multiple actions still counts once. Earnings Reported — companies that reported earnings in the last 14 days. A "beat" means EPS actual ≥ EPS estimate. Source: FMP earnings-company endpoint, pulled daily. Thesis Alerts — count of tickers with active concerns. Combines two sources: (1) client-side threshold checks run on every page load (price above PT consensus, consensus at Hold or Sell), and (2) AI-evaluated alerts from the daily Claude Opus analysis routine (PT declining, revenue decelerating, cash runway short, filing red flags, etc.).

Sector pills

One per sector. Company count, total Buy ratings across all companies in that sector, and average price target upside. Upside formula: (PT consensus - current price) / current price × 100.

Table columns explained

Price — last close from FMP quote data, updated daily by the scheduled task and on-demand via the Refresh button. P/E (Price-to-Earnings) — how much investors pay per dollar of earnings. Lower means cheaper relative to earnings. The dashboard uses FMP's trailing-twelve-month (TTM) P/E when available. If missing, it falls back to the most recent quarterly EPS × 4, with a guard: if net income is >80% of revenue, it's likely a one-time gain (e.g., IONQ's $805M), so P/E is suppressed rather than showing a misleading number. Pre-revenue companies (IONQ, RGTI, QBTS, OKLO) always show "—" because P/E is meaningless without earnings. Consensus — the overall analyst recommendation: Buy, Hold, or Sell. Determined by the majority of rated analysts. Only vetted analysts contribute — see below. Buy / Hold / Sell — raw count of analyst ratings in each category. "Strong Buy" is grouped with Buy; "Strong Sell" with Sell. PT Consensus — the average price target from vetted analysts who issued one. Not all analysts provide price targets, so the count may differ from the Buy/Hold/Sell total. Upside — percentage difference between current price and PT consensus. Green = stock is below the target (analysts expect it to go up). Red = stock is above the target. Mkt Cap — market capitalization (share price × total shares outstanding).

How analysts are chosen

Not all analyst opinions are equal. The dashboard uses a custom vetting pipeline that runs on the host machine via Windows Task Scheduler: Step 1 — Scraping: Every night at 10 PM ET, benzinga_scraper.py pulls that day's analyst ratings from Benzinga. New analysts with ≥70% Benzinga accuracy get their full history scraped by benzinga_discover.py. Step 2 — Scoring: Every two weeks, analyst_rankings.py evaluates every analyst's track record using historical price data. A "Buy" is correct if the stock rose ≥10% within 12 months; a "Sell" is correct if it dropped ≥10%. Recent calls (last 12 months) are weighted 2× more than older ones. Each analyst gets a composite score (0–100) from four components: buy accuracy (40% of score — how often their Buy calls are right, recency-weighted), directional accuracy (25% — all calls), portfolio ticker coverage (20% — how many of your 14 tickers they cover), and statistical confidence (15% — penalizes thin sample sizes, scales linearly to 100 rated calls). Step 3 — Qualification: Analysts with a composite score ≥ 45 enter the qualified pool (adaptive size, 20–60 analysts). They must stay above the quality floor for 2 consecutive biweekly evaluations to reach "qualified" tier, and 5 evaluations with a buy accuracy ≥ 50% to reach "veteran." Demotion is slow — qualified analysts get a 2-evaluation watch-list buffer before dropping. The daily collector (a scheduled GitHub Actions workflow) reads the latest vetted analyst data and writes it into daily.json. This is why you see specific analyst names, ranks, and accuracy scores rather than anonymous aggregates.

Refresh button

Pulls live quotes from FMP via the Cloudflare proxy, updates prices in the browser, then auto-saves to GitHub via save.js. Analyst ratings and price targets are NOT re-fetched — those come from the Benzinga pipeline and only update nightly. 60-second cooldown between refreshes.

How data gets here

Each morning a scheduled GitHub Actions workflow (daily.yml) runs the daily collector: it reads companies.json for the ticker list, pulls quotes, earnings, and insider trades from FMP plus vetted Benzinga analyst data, builds pipeline health, and commits public/daily.json to the repository. Separate scheduled workflows refresh macro data, SEC filings, and the weekly financials. On page load the browser fetches daily.json, weekly.json, and analysis.json from the Cloudflare Pages site and renders everything. Persistent state (the SQLite databases) lives in a Cloudflare R2 bucket that the workflows sync at the start and end of each run.