# 11 — The Blockchain Debate

> *A record of the argument that moved the architecture from federation-only (v0.1) to hybrid (v0.2). Kept as a design artefact because the reasoning matters more than the conclusion.*

## Why this document exists

Design decisions that reverse earlier decisions are often forgotten — the final architecture is adopted, the earlier version is archived, and the reasoning that connected them disappears. Future contributors deserve to read the argument that produced the architecture, not just the architecture itself. If the argument turns out to be wrong, they need the premises to attack. If it turns out to be right, they need the premises to defend.

This document reconstructs, in neutral voice, the argument that led from v0.1 (federation, no chain, no token) to v0.2 (hybrid chain + federation with utility-token economics). It is not a narrative of who said what, but a distillation of the propositions and their cumulative weight.

## The starting position (v0.1)

The v0.1 working paper argued for federation over blockchain on five grounds:

1. **Empirical record.** Token-incentivised fact-checking projects (Civil Media, Po.et, Factmata, Bitpress, Fact Protocol, Swarm Network, DEFC, ProBlock) have no record of sustained adoption in roughly eight years of attempts. Meanwhile, the operationally-successful systems in adjacent space (schema.org/ClaimReview, C2PA, Sigstore, Certificate Transparency, X Community Notes' bridging algorithm) are all non-chain.

2. **Cost.** A Certificate-Transparency-style federated log with cosigned witnesses offers tamper-evidence with a security model close to an operated chain, at approximately three orders of magnitude lower cost per event.

3. **Regulatory exposure.** Tokens are securities-adjacent in most jurisdictions. On-chain data conflicts with GDPR Article 17. Federation avoids both.

4. **Institutional validator mismatch.** Universities, libraries, newsrooms — the actual validator pool — are not set up to manage wallets, pay gas, or hold positions in volatile assets.

5. **Tool-to-problem fit.** Blockchain answers *"how do mutually-distrustful parties coordinate without a trusted fiduciary?"*. Veritas v0.1 had a trusted fiduciary (the foundation) and validators who could trust each other's credentials. The chain was answering a question v0.1 wasn't asking.

## The four propositions that changed the position

### Proposition A — Mutual hostility is the design requirement, not the edge case

**Claim.** The value of plural verdicts comes from admitting actors who believe each other to be evil and lying. If the protocol admits only validators who share institutional norms, the "plurality" is single-frame with decoration. The design intent must include state-versus-state, religion-versus-religion, dissident-versus-official, and sub-cultural-versus-mainstream frames explicitly.

**Consequence.** No foundation can credentialise mutual enemies neutrally. Any foundation that tries will be accused of capture by each side, and both accusations will contain truth because the foundation must apply *some* editorial standard, and that standard will not be neutral across the full frame space.

**Implication.** The write layer must be permissionless. Credentialing moves down the stack. The chain's distinctive property — coordination without a mutually-acceptable fiduciary — becomes load-bearing.

### Proposition B — AI-hallucination economics create a paying customer that did not exist in 2018

**Claim.** Earlier token-incentivised fact-check projects failed economically because there was no paying customer at scale. In 2026, the cost of AI hallucinations (training compute wasted on self-correction, inference compute wasted on retrieval filtering, enterprise compute wasted on post-hoc safety layers) is an addressable market in the tens of billions of dollars. Even a fraction of a percent of that flow redirected to validator compensation via grounding-fee economics is substantially more than the fact-check sector has ever received.

**Consequence.** The failed-tokenomics track record is partially explained by a missing customer. A utility-token design with treasury buyback, service-fee flow from AI-grounding calls, and auxiliary revenue streams (site certificates, content-producer priority, investigation commissions) can close an economic loop that was impossible in 2018.

**Implication.** The tokenomics objection from v0.1 (`53-tokenomics-hard-analysis.md`) needs re-examination. The `-99.5%` validator ROI calculated under 2018-style assumptions does not hold under 2026 market conditions.

### Proposition C — Hybrid architecture resolves most of the v0.1 cost and latency objections

**Claim.** Chain-as-settlement + federation-as-cache preserves chain security guarantees (immutability, neutrality, permissionless write) for the operations that need them (settlement, canonical log, cascade confirmation) without putting the chain in the hot path for AI-grounding reads.

**Consequence.** Grounding-call latency is bounded by cache-layer performance (tens of milliseconds at edge), not by chain finality. Chain cost is bounded by settlement-rate volumes, not by read-volume volumes. GDPR compliance is handled at the evidence-storage layer (off-chain, crypto-shreddable), not at the attestation-log layer (which stores only hashes and signatures).

**Implication.** Three of the five v0.1 objections (cost, regulatory exposure, institutional mismatch) are partially or substantially mitigated. The remaining objections (empirical record, tool-to-problem fit) need independent re-evaluation in light of Propositions A and B.

### Proposition D — Permissionless write + CPML-based read is a different architecture from token-incentivised-fact-check

**Claim.** The failed predecessors (Civil, Po.et, etc.) all coupled token incentives to the *validator credentialing* question — validators had to hold tokens, stake for slashing, and be economically pre-committed. Veritas v0.2 decouples these. Anyone can post to the write layer. Validators can be institutional, individual, self-declared, or third-party-endorsed. Token economics run at the *service-flow* level (users pay for grounding calls) and *compensation* level (validators are paid in utility tokens or directly in fiat from treasury), not at the *credentialing* level.

**Consequence.** The failed-tokenomics track record does not apply directly to this design because the failing element in the prior designs (forcing validators into a token economy as a gate to participation) is not present in v0.2.

**Implication.** Proposition B's AI-economics argument is not inherited from a graveyard of failed attempts; it is a new economic flow that previous designs did not attempt.

## The cumulative weight

None of the four propositions, on its own, reverses the v0.1 argument. All four together do.

Proposition A alone would suggest adding a permissionless-write layer on top of a federated read layer, possibly without a chain at all (e.g., a Nostr-relay-style architecture). But Proposition A + Proposition B also requires economic flow, and economic flow tied to decentralised identity without a shared ledger is fragile (who settles payments? Who records validator compensation? Who enforces the burn mechanism against treasury reserves?). Chain-anchored settlement resolves these cleanly.

Proposition C demonstrates that the cost / latency objections do not apply at the hybrid level; they applied to the v0.1 strawman of "put everything on-chain."

Proposition D shows that the empirical-failure argument does not apply at the specific mechanism level; it applied to a different class of designs.

## What the debate did *not* settle

1. **Choice of chain.** The debate concluded that "some chain" should be the settlement layer. It did not conclude which. Current recommendation: prototype on Base / Optimism L2; decide own-L1 vs ride-existing at Phase III gate based on empirical data and investor signals. (`10-chain-selection.md`)

2. **Token regulatory structure.** The debate concluded that a utility token with treasury buyback is the target design. It did not settle the jurisdiction or whether to launch with burn-for-USDT enabled from day one or ship with pure-utility and add burn later. (`03-tokenomics-design.md`)

3. **Exact foundation role.** The debate concluded that the foundation is not the editorial arbiter. It did not specify the minimum role the foundation retains (spec stewardship? Dispute-panel operation? Reference-aggregator operation?). Current sketch: all three, but transparently and contestably.

4. **Default CPML policy.** The debate concluded that consumer-side composition via CPML is the mechanism for plural verdicts. It did not settle what the foundation-supplied starter CPMLs should look like or who curates them. (`04-cpml.md`)

## Falsification conditions

The v0.2 architecture is accepted on the arguments above. It can be un-accepted if any of the following turn out to be true:

1. **AI laboratories refuse to adopt grounding calls into their inference stack.** If no laboratory integrates during Phase II of the roadmap despite a measurable hallucination-reduction benchmark, Proposition B fails. Falls back to v0.1 federation with grant-only funding.

2. **The mutual-hostility requirement proves illusory in practice.** If, across multiple rounds of recruitment, no credible mutually-hostile validator cohort materialises — if everyone who actually signs up is drawn from overlapping institutional frames — then Proposition A was a theoretical argument without real-world carriers. Falls back to v0.1.

3. **Chain-settlement costs scale non-linearly with attestation volume.** If transaction costs at the target scale make per-attestation economics negative even after AI-lab revenue, Proposition C fails on cost grounds. Consider sidechain, off-chain settlement with periodic anchoring, or return to federation.

4. **Regulatory closure on tokens.** If SEC enforcement, MiCA implementation, or equivalent action makes utility-token-with-burn economics legally impossible in all founding jurisdictions, Proposition B's specific flow fails. Fall back to USDT-only payment flow (no utility token) or to grant-funded federation.

Any of these conditions, if observed, must trigger a public re-examination of the architecture. The debate is not closed; the current conclusion is contingent.

## Reading order for someone picking up this design cold

1. `01-mutually-hostile-validators.md` — the design requirement.
2. `02-hybrid-architecture.md` — the resulting architecture.
3. `03-tokenomics-design.md` — the economic layer.
4. `04-cpml.md` — the consumer composition layer.
5. `07-quiz-mvp.md` — the initial consumer product.
6. `10-chain-selection.md` — the open tech question.
7. (`12`–`N` future editions) — how the architecture evolved.

And, for the counterposition, the archived v0.1 whitepaper at `/v0.1/paper/`. If the falsification conditions above become observed fact, v0.1 is the fallback architecture, not a discarded one.