The One SEC Form That Reveals Dilution Before It Hits Your Position — and How I Auto-Scan It

Every retail trader learns the hard way that some 10-15% overnight gaps come out of nowhere, unrelated to news. It's shelf-registered dilution. The company filed an S-3 months ago, then quietly executed on it via a 424B5 or 424B3 the night before. By the time the volume shows up, the price is already gone.

Here's the mechanic and the thing I built to catch it before you're in the trade.

The two-filing dance

An S-3 registration is a company saying "I might sell more stock later." No shares issued yet, no immediate dilution — but the shelf is now loaded.

The 424B5 (and 424B3) is a company saying "I am selling shares now, off the shelf I registered earlier." That's the actual dilution event. It's technically a prospectus supplement, but the market reads it as "the ATM machine just fired."

The gap between the S-3 and the 424B5 can be days or years. That's the window where you can front-run the dilution if you're watching the filings.

You can pull all of it from SEC EDGAR for free. But EDGAR gives you a firehose of 10,000+ filings a day and no way to say "just show me the 424B5s that landed on tickers I'm holding, and only if the raw amount is more than 5% of float."

What I built

A polling service that:

1. Watches EDGAR's atom feed every 60 seconds for new 8-K, S-3, 424B5, and 424B3 filings.
2. Parses the 424B5 body for the offering size and cross-references it against float from EDGAR's XBRL structured facts.
3. Scores each ticker on a proprietary red-flag taxonomy (dilution proximity, listing-standard notices, officer clustering, cyber Item 1.05, bankruptcy language). LOW/MODERATE/ELEVATED/HIGH bands.
4. Ships a $49 forensic report per ticker with every flag cited to the exact EDGAR accession number.

The scoring engine is the interesting part. Half of what looks like "sudden news" is actually a signal that was sitting in a prior 8-K three weeks earlier — an Item 5.02 officer departure, a buried 8.01 disclosure, an S-3 shelf registration. The tool assembles the narrative before the market does.

What one looks like

Ten pre-generated sample reports on tickers most retail traders already watch — AAPL, NVDA, TSLA, COIN, GME, PLTR, MSFT, SOFI, JPM, AMC — are here: https://filingfirehose.com/reports/samples. No paywall. They're identical in format to what a paying customer receives on any ticker.

If you want one on a specific ticker you're watching, that's the $49 report. One-time, no subscription, no-signals refund guarantee — if the engine finds zero red flags on your ticker, you get the $49 back.

The dev side

A few implementation notes for anyone building similar:

• Never do parquet scans inside a request handler. I tried "wrap the scan in a threaded timeout" and took production down for 45 minutes when the polars binary held a futex lock across the timeout. Now every heavy scan is a nightly precompute → JSON blob → request handler just reads the blob. Response times dropped from 3s p99 to 8ms.

• Cite everything. An LLM-written brief that says "we found dilution risk" is useless. One that says "see accession 0001193125-25-107824, an S-3 shelf registration for $500M filed on 2025-04-12, followed by a 424B5 on 2025-06-03 for $60M" is forensically defensible. Every claim in the paid report links to the source filing.

• Trust the LLM for synthesis, not for extraction. Accession numbers, item codes, and dates come from a deterministic parser. The LLM only assembles the narrative. This avoids the "the AI made up a filing" credibility death spiral.

• The classifier is open-source. github.com/jaablon/buried-events-parser — MIT-licensed, pure Python, zero LLM. You can audit exactly how the 8-K item classifier works before trusting any output from the paid tool.

The honest limitations

• US SEC filings only. No international.
• Forensic disclosure signals, not fundamentals. If you want DCF models and 15-year historical financials, use TIKR or StockAnalysis.com.
• No live chat / community. It's a report, not a subscription.

If you're deep in SEC filing analysis and think there's a signal class I'm missing from the taxonomy — reply here or hit me at [e-mail]. Genuinely want feedback from other people who spend time in EDGAR.

---

Nothing in this post is investment advice. Dilution-risk detection is a disclosure-language read, not a recommendation to buy or sell any stock mentioned.