TruthForge

TruthForge reads the article on the page you're viewing and critiques how it frames the truth — not whether its facts check out in isolation, but whether the piece is built to mislead: what it leaves out, what it insinuates without asserting, what belief it plants that a literal fact-check would miss. Truth and candor are kept separate, always — a statement can be fully accurate and still built to mislead, and this tool never collapses that into one true/false verdict. For every finding that plants a belief, it names the exact phrase that plants it and gives the grounded correction: "Plants: … / Actually: …" Each finding carries its own calibrated confidence — there is no article-level score, badge, or percentage anywhere. It reads across eight lenses — framing & tone, sourcing balance, missing context, factual/verification gaps, causal/motive insinuation, omission/selection, rhetorical technique, and fabrication/manipulation — and shows only what's actually there; a category with nothing to critique is simply omitted. On request, it widens perspective coverage (real sources from the missing political direction, or an honest "nothing found" — never a fabricated balance), audits an institutional source's track record against specific past claims independent of lean, and researches the individual byline author the same way. It runs entirely from your own API key (Anthropic Claude, Google Gemini, or xAI Grok — your choice) called directly from your browser — there is no backend server, and the developer never sees your article text, your key, or your results. Not a fact-check oracle. Not a persuasion tool. Not confidently wrong.