Methodology
How claims are identified, checked, and scored
Elonalysis monitors a defined set of public channels for falsifiable claims made by Elon Musk. Sources include posts on X (formerly Twitter), earnings calls, SEC filings, press releases, interviews, and court documents.
An automated ingestion pipeline scans these channels continuously. Natural language processing identifies statements that contain verifiable factual assertions, forward-looking predictions, or commitments that can be tracked over time. Opinions, jokes, and purely subjective statements are excluded unless they contain embedded factual claims.
Each candidate claim is deduplicated against the existing corpus. Restated or rephrased versions of previously tracked claims are linked rather than duplicated. The system prioritizes claims with high engagement or public significance.
Each claim passes through a multi-stage AI pipeline. The system first decomposes compound claims into individually verifiable sub-claims. For each sub-claim, the evidence retrieval module searches a curated index of authoritative sources: government databases, regulatory filings, peer-reviewed research, and established news organizations.
Retrieved evidence is evaluated for relevance, recency, and source credibility. The verdict generation module weighs confirming and contradicting evidence to produce a structured verdict, a plain-language summary, and a confidence score.
Verdicts below a configurable confidence threshold are flagged for manual review. All verdicts include source citations and archived links so readers can verify the reasoning independently.
| Verdict | Definition |
|---|---|
| ✓True | The claim is accurate and supported by reliable evidence. No material omissions or distortions were found. |
| ~✓Mostly True | The core claim is accurate but contains minor inaccuracies, omissions, or imprecise language that could be misleading without context. |
| ⚠Misleading | The claim contains factual elements but presents them in a way that creates a false impression. Key context is missing or selectively omitted. |
| ~✕Mostly False | The claim contains some factual basis but is predominantly inaccurate. The overall impression conveyed does not reflect reality. |
| ✕False | The claim is demonstrably incorrect based on available evidence. No credible source supports the assertion as stated. |
| ?Unverifiable | The claim cannot be confirmed or denied with currently available evidence. This includes future predictions that have not yet reached their target date. |
| ↑Exaggerated | The claim is directionally correct but significantly overstates the magnitude, scope, or certainty of the underlying facts. |
| →Moved Goalposts | A previously made commitment or prediction has been restated with altered criteria, timelines, or definitions of success. |
| ⊘Broken Promise | A specific commitment with a defined timeline or deliverable has passed its deadline without fulfillment and without an updated timeline. |
| …Pending | The claim has been identified but has not yet completed the fact-checking pipeline. A verdict is forthcoming. |
A claim requires a minimum of one independent, verifiable source before a non-pending verdict is issued. For claims rated with high confidence (≥90%), at least two corroborating sources from distinct organizations are required.
Acceptable sources include government agencies (SEC, NHTSA, GAO, FDA), peer-reviewed publications, established wire services (Reuters, AP), major newspapers of record, and official corporate filings. Social media posts, anonymous sources, and opinion columns are not counted as primary evidence, though they may provide supporting context.
All cited sources are archived via the Wayback Machine or Archive.today at the time of verdict issuance. If an original source becomes unavailable, the archived version is linked instead.
This system has known limitations. It cannot fact-check claims about private conversations, internal company decisions, or subjective experiences. Forward-looking predictions are marked as “unverifiable” until their target date passes.
The AI pipeline may occasionally misinterpret sarcasm, irony, or context-dependent statements. Confidence scores reflect the system’s certainty in its analysis, not the absolute truth of the underlying claim.
Coverage is limited to English-language public statements. Claims made in private settings, encrypted channels, or non-English media are not tracked. The platform does not have access to proprietary internal data from any of the tracked companies.
Errors are corrected promptly. When a verdict is found to be incorrect due to new evidence, source errors, or analytical mistakes, the claim is updated with a correction notice that details what changed and why.
Original verdicts are preserved in the revision history. The system does not silently alter past assessments. All corrections include a timestamp and an explanation of the error.
To report an error or submit a correction request, contact tips@elonalysis.com. Include the claim ID, the specific error, and supporting evidence.