How DredgeCap Generates Risk Profiles

1. Source Data

Every claim on a DredgeCap ticker page is sourced from public SEC filings — primarily 10-K (annual report), 10-Q (quarterly report), 8-K (material event), 20-F (foreign private issuer annual), and S-1 / S-3 (registration statements). DredgeCap pulls these directly from SEC EDGAR's public XBRL/HTML feeds. There is no proprietary data source, no insider information, no analyst consensus, and no paywalled-database reliance. If a fact appears on a ticker page, you can in principle verify it by reading the same filing.

Filings are cached locally so re-analysis doesn't depend on a live EDGAR connection, and the pipeline tracks each ticker's most recent "annual anchor" (10-K or 20-F) along with all subsequent quarterly and material-event updates.

2. Applicability Gate

Before a profile is generated, DredgeCap checks whether the ticker is actually analyzable. Common reasons a ticker fails the applicability gate:

• No annual anchor.A ticker with only old quarterly filings and no recent 10-K can't be evaluated against a stable financial baseline.
• Stale anchor. If the most recent annual filing is more than 24 months old, the disclosure base is too dated for a current risk read.
• Insufficient filing basis. Some delisted / suspended tickers have technical EDGAR presence but no substantive disclosures to analyze.
• Non-corporate instruments.Trust units, ETFs, warrants, and other non-operating-company instruments don't fit the company-profile shape.

When the gate fires, the ticker page renders an explanation of why no risk profile is available — rather than fabricating one from thin data.

3. Structured Analysis

For tickers that pass the applicability gate, the cleaned filing text is run through DredgeCap's analysis engine — a large language model under a structured prompt that requires:

• Citations on every factual claim. Each number or quote must be tagged with the form type and filing date it came from. Uncitable claims are rejected.
• No forward-looking speculation.The model is instructed to summarize what filings disclose, not predict what will happen. Phrases like "could" / "might" / "projected to" are restricted to direct quotes from filings.
• No restricted terminology.Specific finance- loaded terms (e.g., "toxic financing" for non-toxic instruments, "death spiral" for ordinary convertibles) are gated to require explicit conversion-mechanic evidence in filings.
• Going-concern conservatism.The auditor's going-concern flag is reported separately from boilerplate risk-factor language. A profile only flags going concern when the auditor has actually issued a qualified opinion.

4. Schema Validation

LLM output passes through a separate validator before it's written to the public profile database. The validator catches:

• Hallucinated dates (e.g., "filed 2024-13-31")
• Fabricated numbers (values not present in source filings)
• Off-topic narratives (text that doesn't reference the specific ticker)
• Format drift (e.g., date formats other than MM/DD/YYYY when the prompt requires it)
• Empty extraction fields where filing text clearly contains the data

Profiles that fail validation are not published; they queue for re-generation under updated prompt versions.

5. Risk Score

The headline 0–10 risk score is a composite of four sub-scores that each measure a different disclosure-driven dimension:

• Dilution risk — historical share-count growth + outstanding convertible / warrant overhang.
• Debt toxicity — total debt to equity, plus specific concerning instruments (default-triggering covenants, floor-less convertibles).
• Liquidity — cash on hand vs. operating burn rate, runway in months.
• Profitability — operating margin, revenue trajectory, gross margin trend.

A separate going-concern flag and primary risk driverfield surface the dominant single risk for the ticker (if any), so the headline 0–10 doesn't collapse signals that deserve attention into a single number.

Note: the score reflects what the filings disclose. It is not a prediction of the stock's future returns. A 9/10 risk score doesn't mean the stock will fall; it means the disclosed financial profile is unusually fragile relative to typical public companies. Stocks with 9/10 risk profiles can still outperform — the score measures the floor under existing shareholders, not the ceiling.

What We Don't Do

• No buy/sell recommendations. DredgeCap produces risk descriptions, not investment calls.
• No price-target forecasts. Models that try to forecast price are doing a different job than this site.
• No insider data.Everything is from public SEC filings. We don't have a Bloomberg terminal, a Refinitiv subscription, or any non-public source.
• No social-media sentiment scoring. Reddit / Twitter sentiment is omitted; we want signal from filings, not from chatter.
• No paid-coverage incentives. Companies cannot buy coverage or influence their risk score. The score is a function of what their filings say.

How to Verify Any Specific Claim

Every fact on a ticker page is tagged with its source filing and date. To verify, follow the citation to SEC EDGAR and read the referenced filing yourself. If you find a discrepancy between what DredgeCap says and what the filing says, it's a bug — please report it via X / Twitter. Bug reports drive the validator's rule additions.