For years we have insisted that how we verify matters as much as what we conclude. Today we are putting that claim on the record.
We just shipped the full Methodology page, a deep, unredacted technical brief on how Veremet turns Information Chaos into a transparent, tamper-proof record of truth. If you have ever asked us how a Consensus Graph entry is actually produced, this is your answer.
This post is the executive summary. The page itself is the receipts.
Why opacity is the disease, not the cure
The dominant fact-checking model has two failure modes. The first is the editorial board: a small group of professionals making private judgments and publishing verdicts as edicts. The second is the algorithm: a black-box scoring system that ranks claims without explaining itself. Both ask you to trust authority over evidence.
Neither survives contact with a 21st-century information environment. Verdicts arrive too late and at too low a volume to matter. Algorithms cannot exercise ethical judgment, and editors cannot match the velocity of synthetic media.
The architecture we describe in the methodology rejects both. It is not a scoring engine. It is a public utility for evidence and a public network of accountable judgment, working in deliberate symbiosis.
Two engines, one truth
The first thing to understand about the platform is that it has two distinct layers, and they are non-substitutable.
Engine 1: The Public Intelligence Utility (the "Hard" Layer). This is the raw machine. It runs continuous scans across more than ten global sources—GDELT, SEC EDGAR, Congress.gov, and others—and produces between 96 and 144 investigations every day, sourcing more than 1,400 pieces of evidence in the process. We measured that throughput against published benchmarks for traditional newsrooms: it is 50 to 300 times faster. This is what makes per-investigation costs collapse and why real-time verification is even possible.
Engine 2: The Public Intelligence Network (the "Soft" Layer). This is the human element, and it is the part that machines cannot replace. AI models do not have ethical judgment. They cannot weigh dignity, context, or moral weight. Our decentralized community of Verifiers does that work, finalizing the Consensus Graph through weighted voting and structured evidence submission. The hard layer can find ten thousand sources in a minute. The soft layer is what tells you which one matters.
If you remove the Verifiers, you get a fast but soulless oracle. If you remove the utility, you get a thoughtful but toothless committee. Veremet is the architecture that refuses to choose.
The Clarity Protocols: a five-stage pipeline
Every investigation flows through the same five deterministic stages. We call them the Clarity Protocols, and they are designed so that any third party—academic, regulator, or rival—could audit our work end to end.
1. Ingestion and the Discovery Seeder. The pipeline starts with 7+ autonomous AI workflows orchestrating twelve+ domain verification tools. They surface candidate claims continuously, without waiting for a topic to go viral.
2. The Provenance Firewall. Before a claim enters analysis it has to be traced back to its primary source. If the trail dead-ends at a recycled tweet, it never enters the graph as evidence; it is logged as a vector of propagation, not as fact.
3. The Reality Gap Engine. This stage compares the viral narrative against what the primary evidence actually says. The "gap" between the two is what tells us we are dealing with misinformation, hyperbole, or honest dispute. It is also what produces the most surprising findings on the platform.
4. Polarization and Consensus Scoring. We use Shannon entropy to mathematically distinguish genuine consensus from manufactured agreement. A topic where every published source says the same thing and a topic where genuinely independent sources converge look identical at a glance—and very different under information theory.
5. The Chronicle and AI Dossiers. The final stage synthesizes all of the above into a readable, source-linked briefing. Every claim in a dossier is one click away from the underlying primary evidence. There is no "trust us"; there is only the citation.
These five stages are not aspirational. They run on every investigation, every day.
Safety as a non-negotiable
Speed and openness do not earn the right to publish unsafe content. Before any investigation reaches our database, it passes through a strict pre-publish safety gate that screens for hate speech, harassment, and dangerous content using Gemini 2.5 Flash.
We frame the philosophy as "Invitation over Judgment." The point is not to suppress views; it is to keep the platform usable as enterprise-grade infrastructure. If a financial institution, a regulator, or a journalism outlet is going to consume the API, the stream has to be brand-safe by construction, not by case-by-case moderation after the fact.
That gate is one of the reasons we can credibly offer the Consensus Graph as a B2B utility rather than as another opinion publisher.
Why "fact-checker" is the wrong word for what we do
The methodology page includes a comparison table we have been circulating internally for some time. The numbers are worth quoting because they reframe what category the company is in.
A traditional fact-checking organization typically issues a verdict in one to seven days. Veremet issues one in fifteen to thirty minutes. Traditional outlets publish two or three investigations daily. We publish over one hundred. The marginal cost of a traditional fact-check is roughly $500; ours is about $0.10. Their methodology is opaque and editorially curated; ours is transparent and community-audited.
These differences are not incremental. They are categorical. A 5,000x cost reduction does not produce a better fact-checker—it produces a different kind of institution. We do not see ourselves as the next PolitiFact. We see ourselves as the public infrastructure that makes verification a commodity service rather than an artisanal verdict.
That is the meaning of the "Anti-Commodity Manifesto" framing in the methodology: by industrializing the boring parts of verification, we free human judgment to do what only humans can do—exercise principled, accountable judgment on the cases that actually matter.
What this means for you
If you are an analyst, journalist, or curious citizen, the most direct way to engage is to become a Verifier. Every investigation gets stronger when more accountable humans weigh in, and the network's reputation system is designed to reward rigor, not volume.
If you are an enterprise—a trading desk, a newsroom, a platform integrity team, a regulator—the Consensus Graph is available as an API. We brand-safe it at the source so you do not have to. Talk to us about access, SLAs, and custom dossiers.
If you are skeptical, that is the appropriate response. Read the methodology in full and judge it on its merits.