Working paper · v0.2 · Deck Collaborative Fact-Checking Working Group · April 2026
§ A proposal

Veritas Protocol

A hybrid chain-plus-federation substrate for plural, verifiable factual claims on the open web — where mutually-hostile frames can all participate, users compose their own verdicts via a portable consensus profile, and AI systems ground their outputs against a signed, open corpus.
veritas-working-paper.pages.dev
fig. 01 / 16
Veritas Protocol · v0.2Thesis02 / 16
Thesis

The open web has no trust layer for factual claims. v0.2 proposes one that is genuinely plural.

§ 01

The substrate is permissionless to write, so mutually-hostile validator cohorts — state, religious, dissident, sub-cultural — can all post. No single foundation credentialises enemies.

§ 02

Readers compose verdicts via CPML — a user-owned Consensus Profile Markup Language file. The protocol does not pick a verdict; the user's CPML does.

§ 03

Hybrid architecture: blockchain (Base L2) for settlement and permanent record; federation of aggregators for fast, edge-cached, per-CPML reads.

§ 04

Economics close through AI-lab grounding fees, certificate subscriptions and investigation commissions. Validators earn 60–70%; no advertising; no equity.

Veritas Protocol · v0.2Problem03 / 16
Problem

Six gaps the web has, that v0.2 closes together.

Provenance gap
Claims travel without cryptographic lineage.
§ 3.1
Retraction-propagation gap
Retraction Watch records thousands/year; readers almost never learn.
§ 3.2
Pluralism gap
Single-verdict fact-checking picks a frame or abstains.
§ 3.3
AI-grounding gap
No open, canonical, verdict-carrying corpus for generative models.
§ 3.4
Adversarial-frame gap
Mutually-hostile validator cohorts cannot participate in existing systems.
§ 3.5
Labour-sustainability gap
No market prices contested-claim verification; sector is chronically under-resourced.
§ 3.6
Working paper · §3
Veritas Protocol · v0.2Part I — Architecture04 / 16
Part I — Architecture

The tool answers the question it is asked.

Blockchain coordinates mutually-distrustful parties. Federation composes known actors at speed. The protocol needs both — chain in the cold path, federation in the hot path.

Veritas Protocol · v0.2Architecture05 / 16
Fig. 01 — Three layers

Chain as settlement and permissionless write; federation as cache and policy; consumer as CPML.

PROTOCOL LAYER · chain-anchored · permissionless · Base L2 (Phase II) - claim records (content-addressed) - attestation records (signed per-domain verdicts) - validator credentials (self-issued / third-party / foundation-issued; all first-class) - cascade events; payment settlement; refusal registry ↕ reconciliation on TTL + cascade signal FEDERATION LAYER · aggregators + caches + country chapters - indexes chain events; composes per-CPML views - applies editorial filters (aggregator's own policy) - country chapters: jurisdictional filter sovereignty - edge-cached materialised snapshots per domain - libp2p gossip for cascade propagation ↕ fast cached reads; local CPML application CONSUMER LAYER · CPML + quiz + apps + AI systems - consumer holds a CPML (local; optionally shared) - consumer picks aggregator(s); foundation offers several, not one - AI grounding calls hit cached per-domain snapshots (MCP-native) - quiz MVP produces starter CPML; daily refinement - payment flow closes back to chain on billing cycle
Working paper · §5
Veritas Protocol · v0.2Permissionless write06 / 16
The distinctive choice

Mutually-hostile validators as the design requirement, not the edge case.

Plural verdicts are meaningful only if they admit actors who believe each other to be evil and lying. A protocol that credentialises only validators sharing foundation norms is single-frame with inclusive marketing.

No foundation can credentialise mutual enemies neutrally. The write layer must be permissionless. Gate-keeping moves down the stack, to aggregators and consumer-side CPML filters.

What this unlocks
Frames the protocol now admits.
State-aligned narratives · ideologically-opposed religious traditions · dissident academic communities · sub-cultural epistemic groups · any self-declared rapporteur body with a published editorial standard.
What this refuses
Narrow operational refusals only.
CSAM-verification · non-consensual imagery · credible threats · mass-casualty-weapon synthesis · active illegal markets. Topic-based refusals never.
Ideas · 01
Veritas Protocol · v0.2CPML07 / 16
Consumer layer

CPML — the user-owned consensus profile.

A portable, signed, shareable file describing which consensus domains a reader trusts for which topics, with what weights, under what conflict-resolution rules.

Formal model: Value-based Argumentation Framework audiences (Bench-Capon 2003). The reader's signing key is the sole authority on which profile applies — an explicit inversion of the 1996 W3C PICS failure pattern.

Four to six starter CPMLs with published editorial rationale; third-party CPMLs (conspiracy-research, religious-tradition, sub-cultural) are architecturally first-class.

{ "@context": "https://veritas-protocol.example/cpml/v0.1", "@type": "ConsensusProfile", "domains": [ { "name": "scientific-default", "weight": 1.0, "topics": ["science"] }, { "name": "historical-academic", "weight": 1.0, "topics": ["history"] } ], "validators": { "trust": ["did:web:royalsociety.org"], "distrust": ["did:web:known-badactor.example"] }, "conflict": { "rule": "show-disagreement", "min_validators_per_verdict": 2 }, "display": { "show_falsified": "suppress", "show_contested": "surface-honestly", "surprising_opposite_view": "daily" } }
Ideas · 04
Veritas Protocol · v0.2Cascading falsification08 / 16
Primitive

Cascading falsification — retractions reach the reader.

Trigger
Quorum-signed retraction.
New evidence, successful dispute, or upstream-source retraction. Single-validator retractions mark as DISPUTED-PENDING, not falsified.
Propagate
Gossip + chain.
libp2p GossipSub reaches subscribed consumers in seconds; chain confirmation arrives minutes later. Classical Doyle JTMS algorithm.
Notify
Every cache, every client.
Aggregator caches invalidate on high-severity cascades; AI systems suppress on grounding call; publisher sees event and can update.

Worked example. A 2030 study cited by a 2031 brief cited by a 2032 news article is retracted in 2033. Under Veritas, every dependent claim is marked CASCADE_PENDING within seconds; AI grounding calls suppress; readers see retraction-pending indicators. None of this happens on the web today.

Working paper · §5.5
Veritas Protocol · v0.2AI-read surface09 / 16
Integration surface

AI-read API — grounding at inference, per user's CPML.

// pre-emit grounding check const report = await veritas.check({ claims: extractedClaims, cpml: userCpmlId, // composed at edge minValidators: 3, }); // report.falsified · remove from output // report.contested · surface honestly // report.unsupported · flag // report.retraction_pending · warn user // sub-50ms at edge; verifies Merkle // proofs + validator signatures client-side.

Cacheable REST API. Per-domain daily snapshots served from edge CDN. MCP-native integration for Claude / Claude-compatible agents.

Public evidence on retrieval-augmented grounding shows measurable reductions in confidently-wrong outputs. The protocol makes that corpus open, signed and canonical — so every AI system can use it, not just those that built their own.

Phase II pilot: one laboratory, one benchmark, published hallucination-reduction delta. Falsification test for the whole AI-grounding thesis.

Working paper · §5.6
Veritas Protocol · v0.2Investigation market10 / 16
Economic flow

Investigation market — real money to contested claims.

Parties commission investigations of contested claims; escrow funds on-chain; qualified validators bid or are auto-assigned; validators execute and sign a verdict; escrow releases on completion plus quorum.

Adversaries pay for each other's claims to be investigated. Israeli-aligned and Palestinian-aligned parties, both believing the other false, both fund the mechanism that produces evidence. The protocol records both verdicts honestly.

Investigative journalism finds a new revenue stream: per-case payment from parties with stake.

Starting fees
Phase II schedule.
$300 quick · $1K standard · $3K deep · $10K+ extended · $2.5K adversarial cross-investigation. Flat-fee tiers prevent bidding wars.
Safeguards
Structural.
Mandatory 3+ investigator quorum · loser-pays-all · public-interest fund for under-resourced claims · commissioner pattern detection · published audit.
Ideas · 06
Veritas Protocol · v0.2Economics11 / 16
Economics

Chainlink survivor pattern — service-fee utility, no mandatory burn at launch.

6
Revenue streams

AI-lab grounding · certificates · user subs · content-producer priority · investigation commissions · donations.

60–70%
To validators

Treasury policy: validator compensation 60–70% · operations 20–30% · reserve 10–15%. Published and auditable.

CH / EU
Legal structure

Swiss Stiftung + Swiss Verein + EU GmbH. MiCA utility-token classification. US 501(c)(3) + Wyoming DAO LLC deferred.

Token drops the burn-for-USDT mechanism at launch (tokenomics recommendation — the single weakest Howey-test link). Pure service-flow utility. Validators compensated in USDT from treasury. If regulatory posture allows, burn may be activated later by governance vote.

Ideas · 03
Veritas Protocol · v0.2Governance12 / 16
Governance

Narrow at protocol; plural at consumer; jurisdictional at chapters.

Protocol layer
5 operational refusals.
CSAM · non-consensual imagery · imminent threats · mass-casualty-weapon synthesis · illegal markets. Operation-based; never topic-based. Revisable only by 7-of-9 panel supermajority + 60-day comment.
Consumer layer
CPML composition.
No foundation arbitrates which frames count. Users pick starter CPMLs + edit + share + fork. Plurality emerges from read-layer composition, not write-layer filtering.
Chapter layer
Country sovereignty.
EU, US day-one; UK, Switzerland next. Each chapter's filter rules are published and apply to display, not to write. Chain log is universal.

9-seat multi-stakeholder dispute panel: 3 legal · 3 editorial · 2 civil-society · 1 rotating user-community. 3-year terms, geographic distribution, published decisions with dissenting opinions. Panel-independent removal procedure.

Ideas · 09 · Refusals & panel
Veritas Protocol · v0.2Part II — Plan13 / 16
Part II — Plan

Three phases. Explicit gates. Exits are cheap.

Consumer quiz ships first. Federated validator network and AI-lab pilot second. Country chapters and own-L1 evaluation third. The protocol is a standard, not an operating platform — exit at any phase costs near-zero.

Veritas Protocol · v0.2Roadmap14 / 16
Roadmap

Three phases, built in the open.

Phase I · 0–6 mo
Quiz MVP + testnet contracts.
Deliverables: 10-question consensus quiz (standalone-viable); reference chain contracts on Base testnet (EAS-based); /factcheck v0.2 spec draft; reference aggregator v0.1; standards-body submissions (SCITT, W3C, schema.org). Gate: 20+ sites publishing v0.2; 5+ third-party-attesting orgs; 1+ AI-lab letter of intent; ~10K quiz users.
Phase II · 6–18 mo
Federation + AI-lab pilot.
Deliverables: foundation in place; 5–10 institutional validators across 3+ jurisdictions; 3–5 chartered consensus domains; first cascade event; AI-lab integration with published benchmark; investigation market live; EU + US chapters; dispute panel seated. Gate: measurable benchmark impact; federated network; peer-reviewed publication; 3+ revenue streams.
Phase III · 18+ mo
Scale + chapters.
Deliverables: UK, Japan-or-Brazil chapters; consensus-domain expansion; own-L1 evaluation against criteria (baseline: remain on L2); IFCN interop; 100K+ quiz users. Indicative cost: ~US$ 300–500K Phase I · ~US$ 600–900K Phase II · steady state Phase III.
Working paper · §13
Veritas Protocol · v0.2Build elements15 / 16
What we have to build

Every component has a name, a specification target, and an operational owner.

Protocol
·veritas-contracts (Solidity suite)
·veritas-l2-deploy
·veritas-scitt-adapter
·veritas-token + veritas-treasury
Federation
·veritas-aggregator (fork Ozone)
·veritas-cache + veritas-mcp-adapter
·veritas-sdk (TS / Py / Go)
Consumer
·cpml-spec (W3C CG input)
·cpml-resolver
·veritas-quiz (Frame / Compass)
·veritas-browser-ext
Publisher
·veritas-publisher (CLI + CMS plugins)
·veritas-validator-stub
·veritas-cert-cli
Governance
·foundation-charter
·dispute-panel-charter
·hard-list-refusal-spec
·chapter-framework
·starter-cpmls (4–6)
Integration
·ai-lab-pilot
·retraction-watch-integration
·claimreview-bridge
·investigation-market-contract
Working paper · §14
Veritas Protocol · v0.2Call for research partners16 / 16
Invitation

We are looking for research partners to build this with us.

Universities, libraries, standards bodies, journalism organisations, AI laboratories, foundations, policy institutes, sub-cultural and ideologically-diverse epistemic communities — if any of this resonates with your agenda, we would very much like to talk.
Co-author
the specification
Operate
a pilot validator
Audit
the protocol in the open
Charter
a consensus domain
veritas-working-paper.pages.dev · Draft v0.2 · April 2026
Collaborative Fact-Checking Working Group
← → · Space · Home/End · R reset · P print