Methodology
What we scan, how we score, where it stops.
Risk Radar is deliberately simple and auditable. We read public SEC 8-K filings, flag distress language, score it, and hand-curate the result. Every number traces back to a clause you can open yourself.
Where the data comes from
Source is SEC EDGAR full-text search — U.S. government, public-domain data. No account, no API key, no paid feed. We query 8-K filings from the last several days for distress-associated phrases, fetch the matched document, strip the HTML, and score the plain text. Requests identify themselves per SEC policy and are rate-limited well under the published limit.
Signals we scan for
The scorer matches weighted phrase categories. Higher weight means a stronger distress signal.
| Category | Weight | Example language |
|---|---|---|
| Debt default / covenant | 4 | event of default, covenant breach, accelerate obligations, immediately due and payable, forbearance |
| Liquidity / going concern | 3 | going concern, substantial doubt, minimum liquidity, cash and equivalents |
| Financial reporting | 3 | non-reliance, should no longer be relied upon, material weakness |
| Control change / credit amendment | 2 | change of control, waiver, refinancing, amendment to credit agreement |
| Exchange listing deficiency | 2 | notice of delisting, listing rule, minimum bid price, continued listing |
Severity scoring
Each matched phrase adds its category weight to a filing's total score. We bucket the total:
- LOW — score under 4. A single weak signal; usually routine.
- MEDIUM — score 4 to 8. Worth a look; context decides.
- HIGH — score 9 or more. Multiple or strong signals stacked in one filing.
Human curation
The raw score is triage, not a verdict. Its biggest known weakness is that large, healthy investment-grade issuers file routine indentures that define "event of default" — pure boilerplate that inflates a score without indicating any distress. So every issue is curated by hand: we separate genuine microcap / going-concern / punitive-default situations from large-cap debt mechanics, and we flag the boilerplate openly instead of hiding it. That triage is the product.
Confidence labels
Each curated item carries a confidence note explaining how strong the read is — for example, multiple independent distress signals in one filing scores higher confidence than a single going-concern line that is standard for the issuer's stage (such as a late-stage SPAC).
Limitations
- Keyword scoring can miss distress phrased in language we don't yet match, and can over-fire on boilerplate (which is why we curate).
- We scan a rolling recent window of 8-K filings, not every filing type. Insider sales, debt maturities, lawsuits, and shelf offerings are roadmap, not current coverage.
- A flag is a prompt to read the source filing yourself, nothing more.
- Scores are deterministic and reproducible, but they are a heuristic, not a credit rating.