Common questions and worries — answered honestly.

If you read the simple version and thought "but what about…", this is for you. The hard questions don't get easier ones.

Honest answer: partially yes, and we're designing around it rather than pretending it can't happen.

Russia, China, Iran, the US — every state has narratives it wants to legitimise. If Veritas is successful, state actors will run validators. We can't realistically stop them by gatekeeping; gatekeeping would defeat the whole "different communities can participate" idea, and they'd just route around it.

What we can do — and what the design tries to do:

• Make state affiliation visible. If a validator is state-aligned, that has to be disclosed in the credential. Your reader software shows you. You can choose to ignore them.
• Don't let state validators dominate by reputation. The system requires multiple jurisdictions for a domain to be "recognised" by the reference servers. State X can't single-handedly make a consensus domain look mainstream.
• The user is in charge. Your consensus profile decides which validators count for you. Default profiles don't include state-narrative validators.

Will it work perfectly? No. The critical review calls out that a state actor with about $2 million per year could buy a serious presence in the system. We're working on that — it's a known weakness, not a hidden one.

Community Notes is great. It works in one specific environment (X), with one specific algorithm, on one specific kind of content (tweets). It produces one verdict per note — the bridging algorithm does well, but it's a single-frame system inside one platform.

Veritas is different in three specific ways:

• Cross-platform. A claim made on a news site, a Reddit post, and a YouTube transcript can share the same identifier and the same checks. Community Notes only works inside X.
• Multiple verdicts coexist. Scientists and historians might come to different conclusions. Veritas keeps both. Community Notes picks one.
• Cascading retraction. If a source gets retracted, every claim that used it gets flagged automatically. Community Notes doesn't propagate that signal.

Honestly: Community Notes is the closest thing that works at scale. We respect it. Veritas is trying to do something broader, but it's not yet proven; Community Notes is.

Wikipedia tries to produce one neutral article per topic. The community works hard on it; the policy is "neutral point of view." It works well for many subjects and has documented struggles in politically-fraught topics (Eastern European history, Israel-Palestine, climate, anything Indian-political).

Veritas takes a different bet: some questions don't have one neutral answer. Different communities legitimately come to different conclusions. Instead of forcing one verdict, Veritas records all the verdicts with attribution — and lets the reader pick the frame they want to see.

Wikipedia is great for "what is the boiling point of water." Veritas is for "did event X cause outcome Y in conflict Z" — where reasonable scholars disagree.

They're complements, not competitors. Wikidata claims could plug into Veritas as one validator's contribution.

Probably yes for some claims. Disagreement doesn't go away because we have a protocol. The difference is what disagreement looks like:

• On Wikipedia, disagreement happens in talk-page edit wars; the article displays one verdict.
• On Veritas, disagreement is recorded openly — "scientific-default says X, revisionist-academic-default says Y, both with citations." The reader sees both.

This doesn't end the argument. It makes the argument visible and structured. That's the goal.

Nobody decides "what's true" centrally. But several editorial choices are made by the foundation, and we should be honest:

• Which validators get foundation-endorsed badges? Foundation decides, with appeal process.
• Which "default consensus profiles" ship in browsers / apps? Foundation curates ~5, others are community-published.
• What gets refused at the protocol layer? Foundation publishes a narrow list (CSAM-verification etc.); a 9-seat panel revises it.
• Which countries get chapters? Foundation negotiates affiliation agreements.

So: the foundation has more editorial power than the marketing sometimes suggests. The critical review pointed this out, and v0.3 will reduce these surfaces and label the remaining ones honestly. Right now we count five. Target: three or fewer.

Then the main funding source dries up and the validator-compensation model has to shrink.

This is the biggest single risk in the project. If we can't get one major AI lab to do a pilot integration with measurable hallucination-reduction results, the economic model is theoretical.

Backup plans:

• Open-weight models (Mistral, Llama, etc.) might integrate without commercial pressure.
• Foundation grants alone can fund a smaller-scale pilot operation.
• Worst case: the protocol exists as a public-good standard with limited operational scale.

But we shouldn't pretend: if AI labs ignore this, the ambitious version of the project doesn't happen.

Fair worry. Most projects that mix "fact-checking" with "blockchain" have failed (Civil, Po.et, Factmata, Bitpress — none reached useful scale). We've read all those postmortems.

The blockchain in Veritas does one specific job: it lets validators who hate each other still operate on the same ledger. State-aligned validators, religious validators, dissident validators — there is no neutral institution that could credentialise them all. A blockchain doesn't need a neutral credentialing institution; it just records signed events.

The blockchain is not in the hot path. AI grounding queries don't touch the chain — they hit a fast cache. Daily reading hits the cache. The chain only handles the slow stuff: settlement, payment, permanent record, retraction quorum.

Token economics follow the Chainlink pattern (service-payment utility, no mandatory burn). Not the speculation-token pattern of the failed predecessors.

If after looking we conclude blockchain doesn't earn its keep, the design can fall back to a federation-only architecture. It would lose the "mutually-hostile validators" property — which is the whole point — but it would still work.

The chain is universal — anyone in any country can post a signed attestation. The chain doesn't care about jurisdiction.

The display layer respects local law. A German-chapter aggregator complies with German law (no Holocaust denial); a US-chapter aggregator complies with the First Amendment. Users connect to the chapter for their jurisdiction, with disclosure of what's filtered.

For users in countries where the local chapter is captured by the state, two paths:

• Use a different chapter's aggregator (with VPN-style implications they must understand).
• Run a personal aggregator with no chapter filter at all (technical but possible).

This isn't a censorship-circumvention tool. It's an honest acknowledgement that one filter doesn't work for the whole world.

Those organisations are the validators in our model. They already produce careful checks. The difference is:

• Their checks today are stuck on their own websites. Hard to consume programmatically.
• There's no system that lets a 2032 reader of a 2031 article know that the source the 2031 article cited was retracted in 2033.
• Their reach is limited; the original misinformation spreads further than the correction.

Veritas wants to be the infrastructure on top of which fact-check organisations work. Not a replacement. The IFCN signatories could be the founding cohort of validators.

The chain side scales fine — modern Layer-2 blockchains handle the volumes we need at very low cost (fractions of a cent per attestation).

The human side is the bottleneck. Verification still takes expert hours. We can never check every claim on the internet. The realistic goal is to cover claims where checking matters most: scientific publications, government statements, major news, contested historical questions, anything cited at scale.

That's still millions of claims. It's a lot of work. The investigation market (parties pay for claims they care about being checked) is meant to scale verification effort to where it matters most economically — but it can't make the labour problem disappear.

A few defences, layered:

• Posting costs a small fee. Per-attestation cost discourages spam.
• Rate limits per validator credential. Even with money, you can't flood.
• Reputation math. Validators who post low-quality attestations have their reputation drop; future attestations carry less weight.
• Aggregator filtering. The reference aggregators don't surface attestations from low-reputation or unknown validators by default.
• User filtering. Your consensus profile can blocklist specific validators.

The critical review pointed out that current per-attestation cost (~$0.05) is too low to deter a determined spammer with $2,000. We're looking at scaling the fee.

Your consensus profile lives on your device. It's never uploaded unless you explicitly choose to share it.

The aggregators that you query don't need to know who you are. You can connect anonymously. The query itself reveals what claim you're looking at — that's a privacy trade-off, but it's the same as querying any website.

Personal data never goes on the blockchain. The chain stores claim hashes and signatures only. Evidence files stay in regular off-chain storage, and can be deleted when their subjects request it (GDPR-compliant).

The architecture is against suppression — anyone can post; anyone can consume any view. But we should be honest about the failure modes:

• The default profiles matter enormously. If 99% of users use the foundation's default consensus profile, what gets included becomes a power decision. We're working to make the default deliberately diverse and to surface "opposite views" daily so users see frames they wouldn't pick.
• Aggregator filters can be captured. A national chapter under government pressure could filter dissident views. The chain still has them — but the average reader doesn't.
• The "narrow refusal list" is a chokepoint. Today it's narrow (5 items, all genuinely awful). The danger is mission creep. We've committed to a 7-of-9 panel supermajority + 60-day public comment period for any expansion. It's a structural friction, not a guarantee.

This is one of the hardest open questions. We don't claim to have solved it.

Today: nothing is live. This is a working paper.

If funded:

• Months 0–6: Build the consumer "consensus quiz" (a fun product on its own that helps users discover their consensus profile). Build reference servers on a test chain.
• Months 6–18: Pilot with 5–10 institutional validators across multiple countries. One AI-lab integration with measurable hallucination-reduction results.
• Months 18+: Scale up. Open country chapters. Real users.

We're not promising any of this. We're saying: if the funding lands and the pilots work, this is the timeline. If they don't, we publish what we learned.

The foundation is non-profit. The validators (universities, libraries, newsrooms) get paid for their verification work — about 60–70% of inbound revenue goes to them.

The remaining 20–30% covers operations: legal, security audits, software development, foundation staff. 10–15% is held in reserve for stability.

No equity. No founder windfall. No private ownership. All flows are published. The treasury is auditable annually.

Validators getting paid isn't a problem — they should be. Verification is real work.