How Catalyst works

Data sources

Every signal on Catalyst originates from a public government dataset. We do not buy data, scrape paywalled feeds, or accept tips.

• SEC Form 4 — corporate insider transactions filed under Section 16 with the SEC. Pulled from EDGAR every 15 minutes.
• STOCK Act PTRs — Periodic Transaction Reports filed by members of Congress within 45 days of a trade, sourced from House Clerk and Senate eFD.
• Form 8-K — material-events filings used as a secondary signal-quality check on the issuing company.
• Bioguide.congress.gov — public-domain member photos and chamber metadata.

How a Form 4 becomes a signal

Filings pass through a deterministic filter before any AI is involved. Three importance tiers decide whether a trade is stored quietly, summarized, or escalated.

• NOISE (< $5,000) — stored in the database but not surfaced or analyzed. Reserved for inadvertent reporting, fractional grants, and de-minimis activity.
• LOW ($5K–$100K) — surfaced in the live feed. Eligible for AI analysis only on the buy side, and only when three or more separate insiders buy the same ticker within 60 days (a cluster signal). LOW sells are stored and shown, but not AI-analysed individually.
• MEDIUM ($100K–$500K Code-P, or $100K–$1M Code-S) — routed to a Haiku model for a 3-sentence pattern note.
• HIGH ($500K+ Code-P, $1M+ Code-S, or senior-title threshold) — routed to a Sonnet model for a deeper 4-sentence write-up with bull/bear framing.

The thresholds above apply to officers and directors — people with day-to-day visibility into the company. For 10%+ beneficial owners (PE firms, activist funds, large institutional holders), we use higher thresholds because a $250K trade by a billion-dollar fund is rarely informational. Beneficial-owner buys reach Tier 2 at $1M and Tier 3 at $5M; sells reach Tier 2 at $10M and Tier 3 at $50M.

Every AI analysis is required to cite a specific number from the filing. We forbid hedge words like "may," "could," and "suggests" from the model output, so the published narrative reflects what the data actually shows.

Conviction score

Each signal carries a 1–10 conviction score. The score is the model's own assessment of how strong the setup looks given the inputs it sees — insider role, dollar value, holdings delta, cluster context, price timing, and any prior Catalyst signals on the same ticker. We also track 30-day and 90-day actual returns vs SPY for every published signal so conviction calibration can be reviewed honestly over time, but those outcomes are not currently used to retrain the model.

• 1–3 — noise tier, almost never published
• 4–6 — mildly interesting; worth knowing about
• 7–8 — strong setup; cluster, first-time buyer, or senior officer
• 9–10 — flagship trade; reserved for rare combinations of size, role, and timing

What we do not do

• We do not give personalized investment advice.
• We do not predict prices.
• We do not infer non-public information.
• We do not accept paid placement of signals or stocks.

See the full investment disclaimer.

Updates and corrections

If you spot a misclassified filing, a missing trade, or a model output that misreads the data, email corrections@catalystview.xyz with the SEC accession number. We fix issues quickly and publicly.