v3.2: the propagation pass

v3.2 is the canonical full edition (525 pages). It follows v3.1, the peer-review remediation pass of 11 September, and it is driven by a second, independent read of the v3.1 source, published alongside this post as Peer review of v3.1.

What the review found

The v3.1 body held. Every P0–P2 item from the v3.0 review was present and, where it was a mechanism, coherent: the componentwise MATMUL bound, the bounded-time no-loss exit, the disclosure-scope model, the single workload registry, the Tornado Cash sequence through Storm, the Kardashev Labs chapter, RL16–18 with their maturity gates and power calculation. The Part II sequence from fee incidence to denomination survived an adversarial read intact.

What v3.1 had not done was propagate. The remediation was applied at the site of each finding, and the document’s summaries and definitions were left as they were — so the body stood at v3.1 while the Thesis in Plain Language, the Chain at a Glance, the Key Definitions, the glossary, the instrument table, and the formal-definitions appendix stood at v2.x on exactly the points v3.0 and v3.1 had changed. The review counted three Critical and about twenty-five Major findings, every one of them mechanical, and two that rose above stale text: a passage in Part II denying that any red line covers the collateral–capacity spiral that Red Line 15 covers, and a governance-weight cap in Part V citing a Red Line 8 threshold that did not exist.

What changed in the record

• Summaries now match the body. The Plain Language chapter and the Chain at a Glance state the three link-9 mechanisms — pledgeability without due diligence, denomination, the state-contingent holder-side flow — with Red Lines 16–18 as their falsifiers. “Three bridges Medium” reads six, matching the chain-strength table. The chain walk’s link 9 names all three mechanisms rather than state-contingency alone.
• One definition each. Refund safety has one definition (sec:l5-safety) and every other site — Definition 5, the instrument table, the Reader’s Map, the glossary, the Part I stress test — states it in those terms; the two denominators (protocol-attributable at 99.9%, overall at 99.5%) are stated together once. VerifyReach has one definition of succ₂ and one threshold set. The fee partition is pinned once, in the Monetary Constitution (70% capacity providers and delegating stakers / 20% retired / 10% assurance), and quoted everywhere else; the Layer 0 Assurance Fund is the assurance budget’s largest line rather than a separate “2% of fees”.
• Glossary. Symmetric knowledge is stated as deterrence, replacing the “every participant has already produced it” wording that the v3.0 late refinement had retracted in the body and the glossary had kept. The take rate τ reads gross fees, as Red Line 14 defines it. Bronze/Silver/Gold is defined as the latency axis, with the assurance axis, the power-firmness classes, and Gold assurance distinguished; receipts carry an assurance tier beside the latency tier.
• Red Lines 8 and 12 gain numeric clauses — a single address family above 20% of time-locked governance weight for two quarters, emergency-path use above two invocations in four quarters; forced-disclosure incidence at 10% and a non-custodial share floor of 50% — so that the v3.1 claim that every red line carries a numeric trigger becomes true. Tightenings, not relaxations.
• The two promised tables. The collateral haircut table is now indexed on modality × grade (tab:haircut-modality-grade), with the modality increment exceeding the grade increment in every cell but one. The scope-drift gap named in the pledgeability section becomes a published series — off-ledger exposure disclosure — on the Value Capture Board, with no trigger of its own.
• The instrument table (Appendix B), which by its own rule treats any body metric absent from it as a defect, gains rows for the convenience-yield vector (RL16), cross-lender haircut dispersion (RL17), free-choice denomination share with the quote/settle matrix (RL18), the M1+M2 share, acceptance breadth, forced-disclosure incidence, governance-weight concentration, and the Native Monetary Buyer Map.
• Naming. The constitutional cost band on verifying receipts is the Real-Resource VerifyPrice SLO; “Physical VerifyPrice” now names only the FCR audit cost Red Line 10 reads. Ten sites that presented BTC↔ZEC as a deployed corridor read BTC↔XMR with shielded-ZEC gated.
• Accuracy minors. Gold’s lease rates are “ordinarily” negligible, with the early-2025 episode named; Fedimint’s federation posts no bond; the stablecoin-flight and lost-bitcoin claims carry sources; FCMP++ is stated to have no mainnet date set; the corridor time-to-finality target sits above the deployed corridor’s confirmation-bound floor rather than below it; twenty-two Sources entries that lacked a scope note carry one. The introduction’s vignette says what it shows — the stack closing — and what it does not: its batch is quoted in dollars and settled in the unit, the configuration Red Line 18 names as the null.

Unchanged, deliberately

Brief v3.1 and essay v1.3 ship unchanged; neither carried any of the defects above, and their content files stay frozen at their tags. The chain-strength table is unchanged: no rating rose because text was reconciled.

Reproducibility

The checksum file records the LaTeX source commit the PDF was built from (94d3eab, tag v3.2). Both reviews — of v3.0 and of v3.1 — are now published on this site verbatim rather than left in a private repository.

Boundaries held

No dated outcome predictions. No fund, token, or gate. No red line relaxed, no threshold moved, no mechanism added. A version whose entire content is making the document agree with itself is the least glamorous kind of release and, for a thesis that publishes its own kill conditions, the kind it owes.