How It Works

Arc-Hives verifies news claims against primary sources — official records, government data, SEC filings, and original transcripts — not other news articles.

The verification pipeline

  1. 1

    Paste any article URL or claim text

    Submit a full article or a single sentence. The pipeline accepts both. For articles, it fetches and parses the content directly — you don't need to copy anything manually.

  2. 2

    The pipeline extracts and decomposes every factual claim

    Each verifiable assertion in the article is pulled out individually. For complex claims, the system breaks them down into sub-questions that must each be satisfied: Who said it? What does the data actually show? Does the timeframe match? This first-principles decomposition is what separates verification from search.

  3. 3

    Each claim is checked against a strict evidence hierarchy

    The system queries primary sources first — government databases, official statistics APIs (BLS, World Bank, SEC EDGAR, Federal Reserve), court records, and official press releases. Only if no structured primary source is found does it fall back to web search, and even then it requires independent external corroboration.

    Sources are ranked by evidential weight. An official BLS employment report outranks a Reuters article citing that same report. A press release outranks a paraphrase of it. The reasoning reflects where each piece of evidence sits in the hierarchy.

  4. 4

    You get a verdict, a confidence score, and the full evidence trail

    The final output shows you exactly which sources were consulted, what each source said, and how the verdict was reached. Every step is visible — you can follow the reasoning yourself and decide whether you agree with it.

    The record is permanent. If new evidence emerges, a new analysis run is added alongside the original — nothing is overwritten. You can see how the evidentiary picture has changed over time.

A claim, step by step

Here's how the pipeline handles a real statistical claim.

Input claim

“Germany's exports of goods and services accounted for 43.4% of its GDP in 2023.”

Sub-questions generated

1.What was Germany's export value as a percentage of GDP in 2023?
2.Does the World Bank or Eurostat publish this indicator for 2023?
3.Is the claimed value (43.4%) within the reported range?

Primary source located

api.worldbank.org — Indicator NE.EXP.GNFS.ZS

Germany Exports of goods and services (% of GDP) — 2023: 42.97%

Verdict

Mostly Confirmed90% confidence

World Bank data confirms 42.97% — 0.43 percentage points below the claimed 43.4%. The direction and magnitude are correct. The gap falls within normal data-revision tolerance between national statistical offices. Claim is substantially accurate.

Why not just ask an AI?

AI assistants synthesize and recall — they generate plausible-sounding answers from patterns in training data. They do not verify. A large language model answering “was this statistic accurate?” is drawing on whatever it was trained on, which may be months or years out of date, and cannot cite a specific document it actually retrieved.

Arc-Hives does something different: it queries live primary sources — government APIs, official databases, SEC filings — retrieves the actual data, and produces a citable record that references specific documents. The reasoning is visible, the sources are linked, and the record is permanent.

AI assistant

  • Recalls from training data
  • May be months out of date
  • Cannot cite a retrieved document
  • No permanent record
  • Confident tone regardless of accuracy

Arc-Hives

  • Queries live primary sources
  • Current data at time of run
  • Links to the actual document
  • Permanent, timestamped record
  • Explicit confidence score with reasoning
Try It Free