What we're going to fix — the v0.3 plan in plain English.

Stop pretending one revenue source is six revenue sources.

The biggest finding from the reviewers: the project depends almost entirely on AI laboratories paying for grounding access. The paper currently lists this as one of six equal income streams. It isn't.

v0.3 will say so plainly. Three revenue scenarios — pessimistic ($5–15M/year), base case ($50–150M), optimistic ($500M+). Plus a fallback model: what if AI labs don't sign up at all? Smaller operation, slower validator network, no investigation market at scale, foundation grant survival mode.

Single reconciled cost table: phase costs, validator counts, fee tiers — same numbers across paper, brief, deck, ideas. No drift.

Resolve the contradiction we pretended wasn't there.

The protocol says "anyone can write" and also "we refuse five operations." Both can't be true in the strong sense. v0.3 will specify, layer by layer, where each refusal class lives.

At the protocol level: only one true refusal — the operation of "verifying CSAM" is incoherent (you can't verify what you can't possess legally). One item maximum on this list.

At the aggregator level: each aggregator publishes its own filter policy. Foundation reference aggregators filter certain categories; third-party aggregators may or may not.

At the user level: the consensus profile decides what gets surfaced for that user.

Items 3 and 4 of the current refusal list (credible threats, mass-casualty-weapon synthesis) move down to aggregator-level filtering — which is honest about what they actually are.

"No single authority" was never quite true. Say so.

The foundation operates five editorial surfaces: which reference aggregators get the foundation badge, which starter consensus profiles ship, what's on the refusal list, which countries get chapters, who's on the dispute panel.

v0.3 acknowledges this openly. The honest framing is "distributed authority with five disclosed editorial surfaces with appeal paths."

Concrete reductions: only one minimal reference aggregator (others are third-party); only one minimal new-user consensus profile (others are community-published); chapters and panels stay (justified by legal accountability, not editorial preference).

Goal: from 5 surfaces to 3. Long-term goal: even fewer.

The reviewer red-team listed 14 specific holes. Close them.

Sentinel found that adversarial-cost floors are too low. Costs don't scale with what's at stake. v0.3 implements 14 specific defences:

• The number of validators required to confirm a cascade scales with the value at stake. Below $10K, three validators. Above $1M, fifteen + a 72-hour confirmation window.
• Retractions need a signed credential from the original source — not just any validator claiming a retraction happened.
• State-aligned validators must disclose their affiliation in their credential.
• Foundation-published consensus profiles are cryptographically signed; readers refuse to load unsigned profiles by default. Stops the "trojan profile via SEO" attack.
• Validator signing keys rotate every 90 days.
• Posting a fake attestation costs five cents minimum, with rate limits per credential.
• For the first 12 months a "cold-start validator pool" of ≥20 institutions anchors reputation math. Closes the early-stage attack window.
• To recognise a consensus domain, three validators in three jurisdictions are required. No single state can dominate.
• Plus six more specific defences against oracle-manipulation, withdrawal attacks, investigation-market gaming, etc.

The hardest piece to write. We have to write it.

The paper invokes "epistemic pluralism" — the idea that different communities legitimately use different evidence standards. Sophisticated readers immediately ask: doesn't this slide into relativism (everything is just opinion)?

The reviewer pointed out: yes, if you don't carefully address it, that's exactly what readers will conclude. The paper doesn't address Boghossian's standard philosophical objection.

v0.3 adds a 1,500-word section that:

• States explicitly that we hold the Lynchian position (frame-scoped evidence standards, world is not divided into incommensurable truths).
• Explicitly disavows the Rortyan and Foucauldian framings.
• Engages directly with Boghossian's incoherence objection.
• Names the architectural universals (bare empirical claims, validator attribution, retraction events) that compose the same way regardless of what consensus profile you use.
• Provides specific test cases — Holocaust denial, flat-earth, religious-scientific conflicts — and shows how the system handles each.

Plus six new internal failure conditions. The current paper lists only external ones (e.g., "AI labs don't integrate"). Internal ones include: "the foundation editorially drifts"; "user consensus profiles converge into a single default profile"; "investigation market gets captured"; "Boghossian engagement fails public test."

Stop citing a formal framework we haven't actually implemented.

The CPML format (the file that holds your consensus profile) currently cites a formal framework called Value-based Argumentation Frameworks (Bench-Capon 2003, Amgoud-Vesic 2011). The reviewer pointed out that the JSON sketch isn't actually an instance of that framework. We're using the citation's authority without doing the formal work.

v0.3 picks one of three paths:

• Path A: Build a real Value-based Argumentation Framework implementation. Substantial formal work; a real reference resolver in code.
• Path B: Switch to a different formal basis (truth-maintenance systems, trust-based recommendation, probabilistic logic — there are options).
• Path C: Drop the formal-framework citation entirely. Present CPML as an engineering artefact, no theoretical claim.

Plus: define behaviour for empty profiles, contradictory profiles, profiles referencing non-existent consensus domains.

Single source of truth for every number. Glossary. Reference cleanup.

One spreadsheet containing every numerical claim across all v0.3 documents. Phase costs, validator counts, chapter counts, fee tiers, latency targets, revenue tiers. Every document draws from the same row of the spreadsheet. No more drift between paper, brief, deck, supporting documents.

A glossary at /glossary/ for the load-bearing terms (consensus domain, validator, CPML, etc.) — pick one term per concept and use it consistently.

Every reference in the paper's bibliography is either cited in the body or removed. Every citation marker in the body resolves to a bibliography entry. (The reviewer found seven references listed but never cited.)

The five claims currently missing the "[unverified]" hedge get either restored hedges or properly sourced numbers.

Replace the placeholder nousaeternos@gmail.com contact destination with a working-group address (or accept form-only contact).