Never trade on evidence you don't trust

01 · Sources

Primary sources first, named on every row

The desk scans official and institutional feeds — SEC EDGAR daily indexes, central-bank and regulator releases, government data like USGS and GDACS — plus curated secondary coverage for breadth. Every cached item keeps its source name and URL, and every event card lists the evidence behind it, so you can always click through to the original document.

• Insider data is parsed, not summarised. Form 4 filings are decomposed transaction-by-transaction; machine-readable congressional PTR PDFs are parsed line-item by line-item, across all asset types.
• Nothing is deleted. Rows that age out of the live window move to an append-only archive and stay browsable, flagged archived.

02 · Scoring

Deterministic formulas, published here

The rankings that order every page are deterministic — the same inputs always produce the same score, with no model in the loop.

Corroboration counts distinct sources, so ten copies of one wire story score like one. Stories with no market transmission path — celebrity, entertainment, lifestyle — are floored deterministically so financial signal stays on top. Filings rank by a published importance formula (form type, filer size, timing); rows later refined by the AI ranking pass are labelled AI-refined so you can tell the two apart. Insider flow's balanced sort weighs trade size, congressional provenance and filer seniority against a 14-day recency half-life.

03 · AI & spend

What AI does — and what it never touches

AI on Silecat has exactly two jobs: light ranking passes that tag and re-rank what the deterministic pipeline already collected, and Premium Analytics — iterative research runs over a single target's primary sources that produce a three-horizon market read with its uncertainty stated.

• AI never fabricates rows. Every event, filing and trade on the site exists in a source document first. Analyses cite the evidence they read and list their open questions in evidence gaps.
• Multiple models check the work. Premium Analytics research runs over several iterations across more than one AI model, cross-checking each pass against the source documents to combat hallucinations before anything is published.
• Spend is governed. Every AI job runs inside per-run, per-cycle and daily budget caps, and the operator's console meters every token spent — per refresh cycle, per analysis, per member run.
• Member runs are private and metered. Your analyses spend your own monthly tokens at flat, published prices — the exact count is always on your desk — and their results are visible only to you.