How it works

1. Plan. Given a ticker, the engine writes a short research plan tailored to that name before it looks anything up.
2. Research with tools. It then calls live tools in the order its plan needs: price action and market context, company fundamentals, earnings history and estimates, a quant comparison against peers and benchmarks, the macro backdrop, and web research over company news, filings and the industry. For crypto it swaps in on-chain fundamentals, flows and catalysts. It decides what to call and when it has enough, under a hard budget of tool calls.
3. Commit to a view. A synthesis pass reads only the collected evidence and must pick one view, from strongly bearish to strongly bullish, plus "no clear edge" when that is the honest answer. It gives two separate scores. Edge (0–10) is how attractive the opportunity is at the current price on a fixed scale: 4 and below means no edge, 5 to 6 a lean, 7 to 8 a call worth sizing, 9 and up a rare, exceptional mispricing. Evidence (0–10) is how complete and well-sourced the research is. Neither is a confidence rating: strong evidence and no edge is a perfectly normal result, and it is the reason no-edge calls are published without an edge number. The track record tests whether higher edge scores actually beat the market more often.
4. Audit. A second, adversarial pass checks every number, source and claim in the opinion against the evidence. If it flags something, the opinion gets exactly one correction round. The verifier's notes are published with the call, including when it changed the view.
5. Chart and publish. One chart is chosen to support the final thesis, and the call publishes to the feed as-is. Nothing is edited afterwards.

A run usually takes about 5 minutes and up to 20. While it runs, the request form shows the real stage and the real tool calls as they happen.

• Entry is the closing price on the day the call was published. It is written once and never recomputed.
• Benchmark is SPY for stocks and ETFs, BTC for crypto. Every result is the call's return minus the benchmark's over the same window.
• Hit or miss applies to directional calls only. A bullish-family call hits when it beats the benchmark, a bearish one when it lags it. Neutral and "no clear edge" calls make no claim, so they are shown but not scored.
• Horizons. One week and one month are shown so nothing hides in the meantime; three months is the engine's stated horizon and the final verdict. Closed horizons are frozen.
• Nothing is curated. Every completed run is published automatically. The only operator action is hiding a broken or duplicate run, and the track record states how many times that has happened.

Anyone can name a ticker on the calls page, a ticker page or a call page. The symbol is checked on the spot against our universes and a live lookup; unknown symbols are refused with a message. Accepted requests run automatically, one at a time, up to 6 runs per day. A ticker with a call newer than 1 day is served that call instead of a new run. Leave an email and we send one message when the call publishes. That is the only email the request form ever sends.

The portfolio is separate from the research calls. It is traded by rules from the same nightly technical scan the radar shows: long only, equal-weight slots, no leverage, every fill published with the reason the scan gave at the time, and marked against the same money in SPY. Research calls are opinions and are never traded there.

• Prices and history. End-of-day and intraday market data from EODHD, with Finnhub and Alpaca as fallbacks. Crypto prices from CoinGecko. The numbers on this site are daily closes and may differ from live quotes.
• Fundamentals and filings. SEC company facts (XBRL) straight from the regulator, Financial Modeling Prep for statements and estimates, Finnhub for earnings history and the calendar.
• News and research. Web research through Perplexity and Linkup, with every source tiered: primary filings and company IR first, major wires second, recognized research third, general finance sites fourth, blogs and SEO content last. Sources are listed on each call.
• Crypto. DefiLlama for protocol revenue, fees and TVL, plus the on-chain and flow data behind the crypto tools.
• Models. The planner, synthesis and verifier are large language models run with tool access; which model made each call is recorded with the run. Models are a means to read evidence faster, not a source of facts: every claim should trace to a listed source.

• No cookies, no trackers. Page views are counted on our own server. The only per-visitor value is a hash of your IP and browser that is salted with a secret that rotates daily, so we can count daily uniques but cannot follow one person across days. Raw IP addresses are never stored for page views.
• Request form. The ticker, the time, and a short hash of your IP are kept to rate-limit abuse and to avoid queueing the same ticker twice from one network. If you leave an email, it is used for exactly one notification when the call publishes.
• Waitlist. Your email is stored with the signup time and used only to tell you when early access opens.
• Removal. Message @AlphaMrktPulse on X from the account you want removed, or reply to any email we sent you, and we delete it.
• Third parties. Fonts load from Google Fonts and the site is served through Cloudflare; both see the standard request metadata any web host sees. Nothing you type is sent to them.