v2.2 — The Instrument-Hardening Pass

v2.2 — The Instrument-Hardening Pass

Date: 2026-08-29

Version 2.2 makes the thesis’s two flagship measurement families agree with the standards the rest of the document already holds itself to. VerifyPrice and VerifyFlow were audited end-to-end: every equation re-derived symbolically and verified numerically. The core derivations hold. The aggregation, specification, and anti-gaming layers behind them did not. Ten defects, now fixed.

Aggregation

• The double-count in $Q^{\text{exp}}$. The asset-level flow vector defined total exposure demand as $Q^{RC} + Q^{RD}$, but $Q^{RD}$ consumed raw net creation — whose elasticity-driven component is already inside $\kappa$ in $Q^{RC}$. The same dollars were counted twice, and the overstatement propagated into MPR, the stress harness, and any red line reading on them. $Q^{RD}$ now consumes only the intercept-plus-residual component, per the decomposition rule Appendix H already stated — and the vector violated. Undecomposed creations are published flagged, not silently aggregated.
• WNG aggregation and deflation. The Wrapper–Native Growth Gap feeds the triple-divergence composite and Red Line 8 but had no aggregation specification, while its adjacent Red Line 14 quantity already carried one. WNG is now specified: trailing four-quarter window, components summed as deflation-adjusted turnover, both unit- and USD-denominated rates published. Without the adjustment, an unadjusted WNG rises on a valuation artifact alone — the composite can fire on a phantom.
• MPR gross and net. The Mechanical Pressure Ratio now publishes gross and net forms, mirroring the Wrapper Dominance Ratio’s stock/flow discipline: netting offsetting flows before the modulus erases the liquidity both sides consumed.

Domain

• WRR is undefined at $L = 1$ — the dominant wrapper class. The Wrapper Recycling Ratio divides by $L - 1$, so it does not exist for an unlevered wrapper. The spot ETF — plausibly this asset’s largest wrapper — is exactly that. The degeneracy is correct behavior, not a bug: an unlevered wrapper has no rebalancing obligation, so there is nothing to recycle. The $\kappa$–WRR identity is now stated, the $L{=}1$ publication rule fixed (publish $\varepsilon$; never a WRR column), and a wrapper-class applicability map added so classes without shares and NAV — perpetual futures above all — get named substitute constructions instead of fabricated readings.

Anti-gaming

• The $r(W)$ denominator had no contract. The verification numerator of the cost asymmetry ratio is produced under a multi-operator, signed-batch, adversarial-corpus regime. The production denominator was self-reported. A prover who inflates production cost drives $r(W)$ down and the network looks more verification-asymmetric than it is. The denominator now carries the mirror of the numerator’s contract: reference cost models pinned like the reference kernel, independently estimated, published derivation — and where independent estimation is impossible, $r(W)$ publishes as prover-favorable and unverified, unable to support a passing reading on chain link 3.

Estimation

• The OLS upper-bound rule was conditional, not absolute. Simultaneity bias inflates the fitted elasticity, so OLS was published as an upper bound on amplification. But for a stale-priced underlying, errors-in-variables attenuation biases the estimate downward — and an attenuated OLS value bounds nothing. A staleness diagnostic is now published alongside, and where staleness is material, the reading publishes as unidentified rather than as a bound.

Specification

• One canonical VerifyPrice tuple. Three tuple definitions lived across four sites. Appendix A now states the canonical five-field form and demotes the others to views of it, reconciling to the same receipts.
• VerifyPower took the reciprocal of a vector. It is now a scalar on a pinned reference workload, with the workload declared and changed only with notice — the signal cannot be dressed by re-picking.
• VerifyPrice_adj is a convenience summary. An expectation averages away the tail that resilience lives in. The per-state vector is load-bearing; no red line reads on the summary.
• AffordabilityRatio improves when fees rise. The ratio falls when the denominator grows, so a protocol can pass the SLO while becoming less affordable. The ratio now publishes with its components; a falling ratio driven by a rising denominator is a deterioration wearing an improvement’s clothes.

Artifacts

The full edition is 381 pages; the brief is 19 pages. Both are checksummed; the citation record identifies v2.2 as canonical. No red line was relaxed, no threshold was moved, and no falsification condition was withdrawn.