# Veritas Protocol v0.3 — Plan

*Response to the 2026-04-21 critical review.
Each reviewer finding → classified → mapped to a v0.3 workstream → sequenced.*

---

## 1 · Summary of v0.2 problems

Four independent reviewers (super-reviewer, athena, sentinel, aegis) produced:

| Reviewer | Critical | High | Medium | Low |
|---|---:|---:|---:|---:|
| super-reviewer (whitepaper) | 7 | 11 | 9 | 6 |
| athena (logical / philosophical) | 3 | 4 | — | — |
| sentinel (adversarial / threat-model) | 5 CRIT attacks | multiple | — | — |
| aegis (consistency + QA) | 5 | 14 | 21 | 9 |
| **Totals** | **~20** | **~30+** | **~30** | **~15** |

The four reviewers independently converged on the same four structural problems:

### The four load-bearing problems

**P1 · Single-pillar financial dependency** (super-reviewer C1/C2, athena F7).
The v0.2 economic model is structurally dependent on AI-laboratory grounding revenue. The paper projects $695M–$2.78B/year Year-3 treasury inflow while `/ideas/05` scenario-projects $17M/year — a 40–400× gap. The paper's §10 treats "AI-laboratory non-integration" as one of ten equal risks rather than as the load-bearing revenue pillar. If AI-lab integration does not materialise at scale, the validator-compensation math collapses and the promise that "mutually-hostile frames produce investigations for each other" collapses with it.

**P2 · Permissionless-write vs operational-refusal contradiction** (athena F1, aegis wording drift, sentinel §6.1).
The paper simultaneously claims the write layer is permissionless and that a narrow operational-refusal list is enforced at the protocol layer. Both cannot be true in the strong sense. Items 3 (credible threats) and 4 (mass-casualty-weapon synthesis) of the initial refusal list are topical dressed as operational. The wording of the five refusals drifts across paper §7.1 / Appendix C / ideas/09 (including a silent broadening of item 4), evidence the team is not itself holding the operational-vs-topical line.

**P3 · Governance capture pushed one level down, not resolved** (super-reviewer C7, athena F3, sentinel §6.1 + §10).
Two independent reviewers enumerated the same **five editorial surfaces** the foundation de facto controls while the paper claims "no single authority":
1. Reference aggregator set (4–6, foundation-picked)
2. Starter CPML set (4–6, foundation-curated)
3. Hard-list of refusals (foundation-published, panel-revisable)
4. Chapter affiliation approval
5. Charter-registry maintenance

The design pushes capture down a level; it does not resolve it. The honest framing is "distributed authority with five disclosed editorial surfaces," not "no single authority."

**P4 · Adversarial-cost floors too low** (sentinel §§1.1, 2.1, 6.1, 7.1, 9.1; super-reviewer H10, H11).
Five CRITICAL attack classes at costs small relative to value-at-stake:

| # | Attack | Cost | Defeats |
|---|---|---:|---|
| 1 | State-actor legitimation (100 validators, 3 yr) | ~$2M/yr | reputation-math |
| 2 | Cascade-quorum whale (bribe 3-of-5 validators) | $500K–$5M | quorum |
| 3 | Trojan starter-CPML masquerading via SEO | low | default trust |
| 4 | Upstream-source retraction forgery | ~$100K | cascade authenticity |
| 5 | Foreign-agent validator criminalisation (RU/CN/TR) | legal, not $ | jurisdictional diversity |
| 6 | Write-layer DoS (1M spam attestations) | ~$2K | aggregator filtering |

All six share one root: **"permissionless-write with post-hoc reputation weighting is not defended against adversaries with patience and budget to earn reputation honestly before weaponising it"** (sentinel).

### Additional high-severity problems

- **P5 · Pluralism/relativism boundary not held** (athena F2). The Boghossian incoherence objection is not discharged. Under sophisticated reading the pluralism looks like relativism in disguise — alienates principled partners, attracts bad-faith ones.
- **P6 · CPML provenance-laundering** (athena F4). The JSON sketch (cardinal weights + conflict rules) is inconsistent with the cited VAF / Amgoud-Vesic formal framework (ordinal audience-orderings + Dung extensions). Using the framework's authority without implementing it.
- **P7 · Six-property composition tensions** (athena F5). Three pairwise and one three-way tension not derived in the paper: P2×P4, P2×P3, P3×P6, P2×P4×P6.
- **P8 · Falsification conditions all external** (athena F6). The four falsification conditions in `/ideas/11` all require external events; none names internal-failure conditions (editorial drift, CPML-default convergence, investigation-market capture in the wild).
- **P9 · Citation discipline + numerical drift** (super-reviewer C3–C6, H1–H3, aegis 7 inconsistencies). Seven uncited references; Phase II timeline drift 6–18 vs 24–36 months; Base tx cost $0.01 vs $0.0001–0.001; Phase I cost $300–500K vs $200–300K; validator-count drift 5–10 / 12 / 20; chapter-count drift 2 vs 4 day-one.
- **P10 · Cold-start period undefended** (super-reviewer H10). All defences are steady-state. During cold-start, default users have no CPMLs, reputation math has insufficient signal, sham domains can dominate before detection-math kicks in.

### What the v0.2 copyedit round already fixed (published in `24f4c37`)

- Stossel v. Meta citation corrected to N.D. Cal. 2021 (dismissal) + 9th Cir. 2022 (affirmance)
- W3C VC 2.0 date corrected to 15 May 2025
- Wojcik/PNAS replaced with verifiable Renault et al. PNAS 2024
- PICS factual clause rewritten ("labelled content, not users")
- v0.1 → v0.2 version labels on paper reader + brief page 4
- Agent codenames scrubbed from public surface
- Pluralism caveat added to §4 Principle 6
- Revenue-gap disclosed inline at §8.3 with pointer to the critical review
- Brief + ideas-reader nav bars both link to the critical review

**What the copyedit round did not address: the substance.** The 40–400× revenue gap is disclosed, not reconciled. The permissionless-write-vs-refusal contradiction is acknowledged, not resolved. The five editorial surfaces are still there. The pluralism-coherence section is still promised, not written. The VAF formal spec is still a sketch. The five CRITICAL attack classes are still not structurally defended. This is **v0.3 scope.**

---

## 2 · Classification of findings

For each finding, one of four classifications:

- **COPYEDIT** (minutes-hours): fix wording, correct date, restore marker, scrub identifier. Done in v0.2 copyedit round.
- **STRUCTURAL** (days-weeks): rewrite a section, add a section, reconcile numbers, revise architecture diagram, resolve a contradiction.
- **RESEARCH** (weeks-months): open research question requiring work outside scope (VAF formal spec; validator reputation mathematics; AI-lab pilot measurement).
- **DEFERRED** (v0.4+): work acknowledged as necessary but beyond v0.3's scope.

The full classification table follows the workstreams.

---

## 3 · v0.3 workstreams

Seven numbered workstreams (W1–W7). Each has named deliverables, estimated effort, dependencies, and the findings it closes.

### W1 · Economic model overhaul

**Closes:** P1, P9 (numerical drift related to economics).
**Status:** STRUCTURAL.
**Effort estimate:** 2–3 weeks.

**Deliverables:**

1. **`paper §8` complete rewrite.** Replace with:
   - AI-lab grounding revenue explicitly named as THE load-bearing pillar (not one of six equal streams).
   - Three-tier revenue projection: pessimistic ($5–15M Year-3), base ($50–150M), optimistic ($500M–$1B+). Each tier with a sourced/named methodology.
   - Fallback economic model: what happens if AI-lab integration does not materialise. (Grant-only operating floor; validator-institution in-kind-only contribution model; phased validator-count reduction.)
   - Single reconciled Phase I / Phase II / Phase III cost table — numbers identical across paper, brief, deck, `/ideas/05`.

2. **`/ideas/05-revenue-model.md` harmonisation.** Same numbers as paper §8 after rewrite; explicit note that the three-tier projection supersedes the prior single-scenario estimate.

3. **AI-lab pilot specification.** New `/ideas/12-ai-lab-pilot-spec.md`. What exactly does "AI-lab integration" mean? Named benchmark (e.g. HaluLens + TruthfulQA subset). Named target metric (e.g. "≥10% reduction in confidently-wrong responses on factual tasks vs ungrounded baseline"). Named pilot partner criteria. Measurable Phase II gate.

4. **Revenue-risk matrix.** Publish explicitly what happens at each revenue scenario: which Phase-II commitments become unreachable, which Phase-III milestones slip, at what point does the foundation recommend winding down.

5. **Treasury-model spreadsheet.** Open, auditable, published at `/economic-model.xlsx` or `/economic-model.csv`. Named assumptions and let anyone re-run the scenarios.

### W2 · Permissionless-write / refusal-list architectural resolution

**Closes:** P2 (fully).
**Status:** STRUCTURAL.
**Effort estimate:** 2 weeks + 1 week review.

**Deliverables:**

1. **New `paper §5.8` (Refusal enforcement model).** Specifies, for each refusal class, *where* enforcement happens:
   - **Protocol-level refusals** — refuse at write. Only for operationally-incoherent shapes (e.g. CSAM verification — no coherent validator can *possess* + *verify*). One item maximum on this list.
   - **Protocol-level flags** — write accepted; message is emitted that the attestation matches a flagged operation shape. Downstream consumers (aggregators, CPMLs, clients) can refuse to display.
   - **Aggregator-level filters** — each aggregator publishes its own filter policy. The foundation reference aggregators filter; third-party aggregators may or may not.
   - **CPML-level filters** — user's own filter preferences, fully under user control.

2. **Rewrite `/ideas/09-refusals-and-panel.md`.** Same categorisation. Honest framing of items 3 and 4 (threats, mass-casualty-weapon synthesis) as aggregator-level filters, not protocol-level. Item 1 (CSAM) is the only true protocol-level refusal.

3. **Refusal-list wording uniformity.** Paper §7.1, Appendix C, `/ideas/09`, brief — word-for-word identical. Fix silent broadening in Appendix C (aegis finding).

4. **Revise Principle 4** in §4. "Operation-based refusals only" is too strong; replace with "tiered-enforcement refusals: protocol-level refusals are narrow and operationally-incoherent; topically-sensitive content is handled at aggregator and CPML layers."

### W3 · Governance-surface reduction and honest framing

**Closes:** P3 (partially — full reduction is v0.4+), P9 (governance-related drift).
**Status:** STRUCTURAL + editorial.
**Effort estimate:** 1 week.

**Deliverables:**

1. **Rewrite `paper §7`.** Open with an explicit enumeration of the five editorial surfaces the foundation operates. Language: "The foundation controls five editorial surfaces. Each is disclosed; each has an appeals path; capture of any single one is visible in public audit logs. The protocol does not claim neutrality; it claims transparency."

2. **Surface-reduction roadmap.** Identify, per surface, whether v0.3 can reduce or whether reduction is v0.4+:
   - Reference aggregator set: can be reduced to *one* minimal-editorial-policy reference ("foundation-default"), with all curated aggregators operated by third parties.
   - Starter CPMLs: can be reduced to *one* minimal "new-user default" ("scientific-default + journalism-default with no composition preference"), with all other starter CPMLs community-published and the foundation maintaining only a discovery catalogue.
   - Hard-list of refusals: stays with foundation but panel composition is strengthened (see W5).
   - Chapter affiliation: stays; justification is legal accountability, not editorial authority.
   - Charter-registry maintenance: stays; justification is bibliographic consistency, not editorial authority.

3. **Capture-visibility mechanism.** Every editorial surface decision goes to a signed, timestamped, public-log entry with rationale. Third parties can audit foundation decisions without access to internal channels.

### W4 · Adversarial hardening (sentinel's 14 pre-launch gate items)

**Closes:** P4 (completely), P10 (cold-start).
**Status:** STRUCTURAL + RESEARCH (some items).
**Effort estimate:** 4–8 weeks (concurrent with W1–W3).

**Deliverables (each a named sub-item):**

1. **K-scaling cascade quorum.** Cascade confirmation requires K validators with K scaling in value-at-stake. Below $10K: K=3. $10K–$100K: K=5. $100K–$1M: K=9. ≥$1M: K=15 + 72-hour confirmation window.

2. **Commit-reveal cascade voting.** Validators commit hashed verdict first, reveal after commit window. Prevents validator-observer herd effects.

3. **Source-authenticated retraction.** Retraction events must carry a signed credential from the originating source (journal, court, publisher) OR a quorum K-of-M attestation that the source itself has retracted (public-archive-check by validators). Closes sentinel §7.1 ($100K attack).

4. **State-actor credential disclosure.** Validators with substantive state-affiliation must disclose the affiliation as part of credential metadata. Closes sentinel §1.1 by making state-actor-validator presence legible to CPML filters; does not block participation.

5. **Signed starter-CPML registry.** Foundation-published starter CPMLs are cryptographically signed by foundation key. Third-party CPMLs can be signed by their author. Clients refuse to auto-load unsigned CPMLs. Closes sentinel §6.1 (trojan-CPML via SEO).

6. **Short-lived validator signing keys.** Rotate every 90 days; revocation propagates via gossip + transparency log. Closes sentinel §10.1 (key-compromise long-tail).

7. **Write-layer minimum fee + rate limiting.** $0.05 per attestation + 100/day/validator-credential + exponential back-off per-source. Closes super-reviewer H11.

8. **Cold-start period governance.** For the first 12 months after launch, a named **cold-start validator pool** (≥ 20 credentialed institutions) anchors reputation math. After 12 months, cold-start rules sunset.

9. **Multi-person reference-implementation review.** All reference-implementation releases signed by ≥ 2 foundation reviewers. No single-maintainer push.

10. **Jurisdictional diversity requirement.** For a consensus domain to be recognised at the reference aggregators, ≥ 3 validators in ≥ 3 jurisdictions. Closes sentinel §9.1 (foreign-agent criminalisation on any single jurisdiction).

11. **Oracle-manipulation cap (Polymarket/UMA lesson).** Resolution stake cap at 10× expected attacker bribe; if value-at-stake exceeds the cap, protocol defers decision to extended-window human panel review.

12. **Signed CPML-registry discovery.** Users discover third-party CPMLs through a federation-signed registry, not via SEO-indexed URLs. Anti-phishing for the epistemic layer.

13. **Validator-cohort-withdrawal attack defence.** Coordinated withdrawal is detected (>20% of validators in a domain leaving within 7 days); cascade + attestation weights freeze pending panel review.

14. **Investigation-market muddy-pattern detection.** Commissioners flagged for patterns (single beneficial owner, repeated commissions on well-established claims, Sybil-wallet patterns). Public dashboard.

**New `paper §11 — Adversarial-cost floors`** section consolidates these.

### W5 · Pluralism-coherence section

**Closes:** P5 (the single most important finding per athena), P8 (adds internal falsification conditions).
**Status:** RESEARCH + STRUCTURAL.
**Effort estimate:** 3–4 weeks (requires philosophy consultation).

**Deliverables:**

1. **New `paper §4.A — Pluralism, not relativism`.** Size: ~1,500 words.
   - Explicit statement of the Lynchian position (frame-scoped evidence standards; world is not divided into incommensurable truths).
   - Explicit disavowal of Rortyan and Foucauldian framings.
   - Engagement with Boghossian's *Fear of Knowledge* incoherence objection; response derived from Lynch 2021 (*Logic of Relativism*) and Miller 2014 (*Incoherence in Epistemic Relativism*).
   - Named **architectural universals** — bare empirical claims, validator attribution, cascade events, refusal-list items — that compose identically across any CPML.
   - Named **frame-relative objects** — consensus domains, verdicts per-domain, CPML weights — that are honestly frame-scoped.
   - Test cases: Holocaust denial (not admissible as a frame; operation-shape refused); flat-earth (CPMLs may subscribe; mainstream consumers by default do not); religious-scientific conflict (different evidential standards; protocol admits both).

2. **Rewrite `/ideas/11-blockchain-debate.md` falsification conditions.** Add six *internal* falsification conditions to the existing four external ones (per athena F6):
   - Foundation editorial drift
   - CPML default-convergence above threshold X
   - Investigation-market capture observed in the wild
   - Boghossian pluralism coherence failing public test
   - Cascade-universality gap (bare empirical claims composing differently across CPMLs)
   - Legitimation-laundering in the wild (state narrative attestations appearing in reference aggregators)

3. **Public-communications companion.** A 500-word "what Veritas is and is not" page explaining plural verdicts to a policy audience without philosophical jargon. Reinforces the non-relativist position without using the word "relativism."

### W6 · CPML formal specification

**Closes:** P6 (provenance-laundering of VAF citation).
**Status:** RESEARCH + STRUCTURAL.
**Effort estimate:** 3–4 weeks.

**Deliverables:**

1. **Decision: VAF or alternative.** Choose:
   - **Path A** — implement CPML as a genuine Value-based Argumentation Framework audience (Bench-Capon 2003; Amgoud-Vesic 2011). JSON schema expresses an ordinal value-ordering over arguments; conflict resolution by Dung extension semantics. Requires substantial formal work.
   - **Path B** — different formal basis. Candidates: truth-maintenance systems (Doyle 1979; de Kleer 1986); trust-based recommendation (Guha 2003); probabilistic logic (Nilsson 1986).
   - **Path C** — repudiate formal claim. CPML is an engineering artefact, not a formal model; remove VAF citations; present as pragmatic composition rules.

2. **Deliver the chosen path.** For Path A: a worked example with test vectors and a reference resolver in TypeScript. For Path B: the same, under the new framework. For Path C: a rewrite of §5.4 and `/ideas/04-cpml.md` that makes no formal-framework claim.

3. **Subtype hierarchy + inheritance.** CPMLs can extend / fork / compose. Define the semantics. (Who owns a derived CPML? How do conflicts between extensions resolve?)

4. **Worst-case and degenerate cases.** What happens with an empty CPML? A CPML with contradictory weights? A CPML referencing a non-existent consensus domain? Specify behaviour for each.

### W7 · Editorial + consistency pass (residual aegis findings)

**Closes:** P7 (composition tensions — discussed openly), P9 (numerical drift), aegis residual findings.
**Status:** COPYEDIT.
**Effort estimate:** 1–2 weeks.

**Deliverables:**

1. **Single numeric source of truth.** One spreadsheet containing every numerical claim in v0.3 artefacts (Phase costs, validator counts, chapter counts, fee tiers, revenue tiers, grounding latency, cascade windows, cost per attestation). Every document draws from this.

2. **Terminology glossary.** Published at `/glossary/`. Includes:
   - consensus domain (= consensus frame = epistemic frame; pick ONE term, use consistently)
   - validator (= verification centre = signing centre; pick ONE)
   - CPML (always expanded on first use per document)
   - Chain / L2 / rollup (precise use)
   - Permissionless write (post-W2 rewrite; use in layered sense)
   - And ~20 other terms.

3. **Reference-audit pass.** Every reference in the paper's bibliography is either (a) cited in body with ≥1 marker, (b) removed. Every in-body citation marker resolves to a bibliography entry. Closes super-reviewer H6.

4. **[UNVERIFIED] marker restoration.** Five claims currently missing the hedge: "six orders of magnitude" generation-vs-review; "tens of billions" AI-hallucination market; $695M–$2.78B treasury inflow; sub-50ms edge-cache latency; sub-millisecond proof verification. Restore the hedge *or* replace with properly sourced numbers.

5. **Composition-tension discussion.** New `paper §4.B — Tensions in the composition`. Derive the three pairwise and one three-way tensions athena identified (P2×P4, P2×P3, P3×P6, P2×P4×P6). State honestly that the composition resolves each tension only at the cost of specific design trade-offs.

6. **Mailto destination.** Replace `mailto:nousaeternos@gmail.com` with a working-group-owned address once it exists. For v0.3: `mailto:veritas-wg@[domain]` where domain is chosen. Alternative: form-only contact with no email fallback.

---

## 4 · Sequencing and dependencies

```
┌──────────────────────────────────────────────────────────────────┐
│ Phase A  (weeks 1–2)        PARALLEL START                       │
│   W7 editorial pass         ← start: independent                 │
│   W5 pluralism research     ← start: independent (research)      │
│                                                                  │
├──────────────────────────────────────────────────────────────────┤
│ Phase B  (weeks 3–6)                                             │
│   W2 refusal-list resolution                                     │
│   W3 governance-surface rewrite                                  │
│   W6 CPML formal-spec path selection                             │
│                                                                  │
├──────────────────────────────────────────────────────────────────┤
│ Phase C  (weeks 5–10)       PARALLEL (long-running)              │
│   W1 economic model overhaul                                     │
│   W4 adversarial hardening                                       │
│   W6 CPML formal-spec implementation                             │
│   W5 pluralism section writing + response to Boghossian          │
│                                                                  │
├──────────────────────────────────────────────────────────────────┤
│ Phase D  (weeks 10–12)      INTEGRATION                          │
│   Merge W1–W6 into v0.3 paper                                    │
│   W7 final editorial pass on merged document                     │
│   Commission a follow-up critical review (super-reviewer)        │
│   Update brief + deck + ideas                                    │
│                                                                  │
├──────────────────────────────────────────────────────────────────┤
│ Phase E  (weeks 12–13)      PUBLISH v0.3                         │
│   Archive v0.2 at `/v0.2/`                                       │
│   Deploy v0.3 at `/` + `/paper/` + `/deck/` + `/ideas/`          │
│   Publish the follow-up critical review alongside                │
│   Updated roadmap + build-elements + call-for-partners           │
└──────────────────────────────────────────────────────────────────┘
```

Total calendar: ~12–13 weeks.
Can compress to ~8 weeks with two parallel editors and delegated research.

### Dependencies

- **W1** depends on **AI-lab pilot conversation** (even preliminary) — the pilot defines what "AI-lab revenue materialising" actually looks like. Cannot be fully written without.
- **W2** is prerequisite for **W4** (the K-scaling cascade and the source-authenticated retraction both assume a settled refusal model).
- **W4** #11 (oracle-manipulation cap) depends on the AI-lab pilot showing realistic value-at-stake.
- **W5** blocks public re-release. If the Boghossian engagement is not convincing, v0.3 loses principled partners.
- **W6** is independent but blocks any implementation work on the reference CPML resolver.
- **W7** is always-on; incremental commits.

---

## 5 · What v0.3 does *not* address

Explicitly deferred to v0.4 or beyond:

- **Full reduction of editorial surfaces.** Reducing from 5 to 3 (W3) is plausible in v0.3. Going to 1 (or 0) requires an operating protocol, not a paper.
- **Validator reputation mathematics.** W4 #1 (K-scaling) needs a validated reputation algorithm. v0.3 names the algorithm (EigenTrust variant); v0.4 ships a tested implementation.
- **Real investigation-market operation.** v0.3 specifies the mechanism; real operation (and the pay-to-muddy detection in W4 #14) requires live traffic.
- **Country chapters beyond the initial three.** EU, US, UK. Others deferred.
- **Real AI-lab pilot benchmark data.** Requires the pilot. Phase II deliverable.

v0.3 is the last paper-only release. v0.4 is a paper + implementation release.

---

## 6 · Resolution table — finding × workstream

| Finding | Severity | Workstream | Status in v0.2 | Status in v0.3 |
|---|---|---|---|---|
| P1 · Single-pillar revenue | CRIT | W1 | disclosed | resolved |
| P2 · Permissionless vs refusal | CRIT | W2 | acknowledged | resolved |
| P3 · Governance capture pushed down | CRIT | W3 | disclosed | partially (5→3); full to v0.4 |
| P4 · Adversarial-cost floors | CRIT | W4 | open | 14-item gate resolved |
| P5 · Pluralism coherence | CRIT | W5 | caveat added | resolved (Boghossian response) |
| P6 · VAF provenance-laundering | HIGH | W6 | caveat added | resolved (Path A/B/C chosen) |
| P7 · Composition tensions | HIGH | W7 §4.B | not addressed | derived openly |
| P8 · Falsification external-only | HIGH | W5 | not addressed | 6 internal conditions added |
| P9 · Numerical drift | HIGH | W7 | partially fixed | single source-of-truth table |
| P10 · Cold-start defences | HIGH | W4 #8 | not addressed | cold-start validator pool + 12-mo sunset |
| Super-review C3 (URDNA2015 missing) | CRIT | W7 | open | ref added |
| Super-review C4 (Stossel date) | CRIT | W7 | FIXED in v0.2 | — |
| Super-review C5 (PICS error) | CRIT | W7 | FIXED in v0.2 | — |
| Super-review C6 (Wojcik cite) | CRIT | W7 | FIXED in v0.2 | — |
| Super-review H1 ($40K vs $81K math) | HIGH | W1 | open | reconciled |
| Super-review H2 (Base tx cost 10×) | HIGH | W7 | open | reconciled |
| Super-review H3 (timeline 6-18 vs 24-36) | HIGH | W7 | open | reconciled |
| Super-review H6 (7 uncited refs) | HIGH | W7 | open | resolved |
| Super-review H7 (W3C VC 2.0 date) | HIGH | W7 | FIXED in v0.2 | — |
| Super-review H8 (six OOM claim rhetorical) | HIGH | W7 | open | hedged or replaced |
| Super-review H9 (sub-ms proof verification) | HIGH | W7 | open | hedged (1-5ms realistic) |
| Super-review H10 (cold-start undefended) | HIGH | W4 #8 | open | cold-start pool |
| Super-review H11 (write-layer DoS) | HIGH | W4 #7 | open | min fee + rate limit |
| Athena F1 (permissionless vs refusal) | CRIT | W2 | acknowledged | resolved |
| Athena F2 (pluralism coherence) | CRIT | W5 | caveat | resolved |
| Athena F3 (no single authority false) | CRIT | W3 | acknowledged | framing rewritten |
| Athena F4 (VAF laundering) | HIGH | W6 | caveat | path chosen |
| Athena F5 (composition tensions) | HIGH | W7 §4.B | not addressed | derived |
| Athena F6 (falsification external-only) | HIGH | W5 | not addressed | 6 internal added |
| Athena F7 (AI-grounding thesis premises) | HIGH | W1 | partial | 3-tier projection |
| Sentinel §1.1 (state-actor legitimation) | CRIT | W4 #4 | open | credential disclosure |
| Sentinel §2.1 (cascade-quorum whale) | CRIT | W4 #1, #11 | open | K-scaling + cap |
| Sentinel §6.1 (trojan starter-CPML) | CRIT | W4 #5, #12 | open | signed registry |
| Sentinel §7.1 (retraction forgery) | CRIT | W4 #3 | open | source-auth retraction |
| Sentinel §9.1 (foreign-agent criminal) | CRIT | W4 #10 | open | jurisdictional diversity |
| Aegis version-label drift | CRIT | W7 | FIXED in v0.2 | — |
| Aegis refusal-list wording drift | CRIT | W2 | FIXED in v0.2 (partial) | unified |
| Aegis internal identifiers leak | CRIT | W7 | FIXED in v0.2 (mostly) | mailto remains → v0.3 fix |
| Aegis numerical inconsistencies (×7) | CRIT | W7 | disclosed | reconciled (all 7) |
| Aegis unverified-marker strip (×5) | CRIT | W7 | disclosed | restored |

**Total resolved in v0.3:** 40 findings.
**Total deferred:** 5–6 (editorial-surface full reduction; real pilot data; real investigation-market data; additional country chapters; CPML formal spec implementation may spill into v0.4; pay-to-muddy detection cannot be validated without live data).

---

## 7 · Proposed immediate next actions

If the plan above is approved, the next working-group actions (this week or next):

1. **Split workstream ownership.** Suggest: W1 tokenomics/economics analyst; W2+W3 protocol-architecture lead; W4 security / red-team engineer; W5 philosophy consultant; W6 logic / formal-methods consultant; W7 editorial lead.
2. **Commission AI-lab pilot conversation.** Without a named lab contact, W1 stays aspirational. Start with Anthropic, Mistral, Hugging Face.
3. **Schedule the follow-up critical review for Phase D.** Same 4-reviewer protocol against v0.3-draft before publication. Publish the follow-up review alongside v0.3 as v0.2's review was published alongside v0.2.
4. **Begin W7 editorial pass immediately.** Editorial work is not bottlenecked on any other workstream; can start this week.
5. **Publish this plan.** Add a link from the brief nav and the critical-review reader. Makes the response legible to partners reading the critical review.

---

*Plan authored in response to the 2026-04-21 critical review.
Target publication of v0.3: approximately 12–13 weeks from start of Phase A.*