Claim, Exposure, Regime

The classification market has consolidated around GICS-style hierarchies. The index providers that actually back investable products (CoinDesk Indices, CF Benchmarks, FTSE Russell with Grayscale, MarketVector, Nasdaq) all use mutually exclusive sector assignment by primary use and quarterly review; the data vendors (Messari, CoinGecko, Token Terminal, Lukka) allow multi-tagging and go deeper; the regulators classify by legal form (MiCA's asset-referenced, e-money and other tokens; the SEC's March 2026 five categories; Basel's four prudential groups). Since the SEC's September 2025 generic listing standards, the binding classification for what a US ETF may hold is neither of these but an eligibility test: a regulated futures market for six months, or an ISG-member spot venue.

Three gaps matter for an investor building test criteria. First, no scheme distinguishes assets by how they co-move: a 2017 proof-of-work coin and a 2023 venture-backed L1 both sit in "Currency" or "Platform" while behaving as different factors. Second, the quantities that decided Zcash's decade — issuance, committed float, venue access, compliance optionality, security verifiability — appear, if at all, as pass/fail screens, so they cannot be compared across assets or tracked through time. Third, the regime is absent: every taxonomy is static, while the factor literature (Liu–Tsyvinski–Wu 2022; Borri, Liu, Tsyvinski and Wu 2026; the IPCA work of Bianchi and Babiak) finds that the priced dimensions — market beta, size, liquidity, downside risk, on-chain value — change sign and strength with the cycle.

The classes follow the value-accrual logic that GCCS's superclasses and the SEC's 2026 taxonomy both gesture at, rather than the sector logic of the index providers. An asset gets one class and any number of attribute flags; the flags are where privacy, meme, layer-2 status and similar cross-cutting features live, because the evidence says they interact with class rather than replace it.

Attribute flags, applied on top of the class: privacy (shielded or ring-signature transfers; drives venue and regulatory exposure), meme (attention-driven demand regardless of class), layer-2 (security inherited from another chain; token often optional), exchange-issued, proof-of-work (miner supply pressure, hash-rate crowding), vintage cohort (launch era, which the PCA shows is a factor in its own right), ETP-wrapped, and regulated-derivatives-listed. Zcash is Monetary with privacy, proof-of-work, 2016 cohort, ETP-wrapped and derivatives-listed flags.

Two groups. The priced factors are estimated from returns and are what the academic literature has found to carry premia or explain co-movement; they change monthly. The structural attributes are the ledger's slow variables: they rarely show up in a PCA because they are categorical or scheduled, but the event evidence says they dominate outcomes when they change.

The ledger's factors do not carry equal weight across classes. The proposed priors below come from the Zcash evidence and the literature; they are the thing the test programme in section 07 is designed to calibrate. Weights are relative (row sums to 10); the veto applies everywhere.

The logic above was written first. The test is whether return data agrees with any of it. Two panels: a deep daily panel of 53 assets with unbroken Coin Metrics history from January 2021 to May 2026, and a broad weekly panel of 62 assets from April 2023 to May 2026 that adds the newer L1s (SOL, AVAX, ATOM, NEAR, SUI, APT, ARB, FIL) from Kraken weekly closes plus ICP from Coin Metrics. Stablecoins and wrapped assets are excluded from both; daily moves are capped at ±50% so single prints do not dominate.

Concentration is cyclical and it is a regime variable. The first component carried 69% of variance in the 2018 and 2022 bears, 58% in the 2021 bull, 50% in 2023 and 54–59% since. Whenever the number is high the market trades as one asset and no classification pays; when it falls, dispersion appears. That is the same finding as the literature (James and Menzies 2021; Makarov and Schoar 2020 at the single-asset level) and it is the first Layer III variable.

Removing PC1 from each asset's daily returns leaves the part of co-movement that a class label could explain. The average residual correlation across all pairs is −0.019. Within-class averages tell you which labels carry information.

DeFi governance tokens (+0.055) and exchange tokens (+0.061) hold together after the market is removed; Monetary holds weakly (+0.032), driven by the legacy proof-of-work coins; Platform, Utility and Culture do not hold at all. The negative DeFi–Monetary residual (−0.067) is the second-order structure of this market: capital flows between "DeFi risk" and "legacy coins" on days when the market factor is flat. Ward clustering on the residual correlations at k=6 recovers this: a DeFi cluster (AAVE, UNI, SNX, CRV, COMP, SUSHI, 1INCH, plus FUN), a legacy cluster (BCH, LTC, BSV, BTG, ETC, EOS, NEO, XTZ, OMG), a 2018–20 large-cap alt cluster (XRP, XLM, ADA, DOT, ALGO, LINK, plus SNT), a 2017 ERC-20 utility cluster (BAT, ZRX, KNC, GNT, MANA, CVC, POWR), an ETH-adjacent trio (ETH, GNO, BAL), and a residual bucket holding BTC, XMR, ZEC, DASH, DOGE, BNB, TRX and the exchange tokens whose idiosyncratic risk swamps any shared component. Purity against the six Layer I classes is 0.49 and the adjusted Rand index 0.12: the labels are better than random and much worse than vintage.

The clearest demonstration that class plus attribute matters more than beta: the same period, sorted by 2025 calendar return. Every asset is coloured by Layer I class. Peak-to-May-2026 is measured from Bitcoin's 6 October 2025 top to the end of the Coin Metrics panel.

The Zcash study found that turnover, hash-rate crowding and drawdown depth predicted forward relative return, that shielded-supply growth did so in the change-based form, and that issuance set the decade. Six of those variables can be measured for every asset in the deep panel. Each January from 2021 to 2025 the 53 assets are ranked on the variable, and the forward twelve-month log return relative to the equal-weighted universe is recorded. Spearman rank correlations by year, and quartile means pooled across the five years, are below.

Four lessons for the ledger. The positioning veto generalises: turnover produced the widest quartile spread in the panel (−44 points against volatility's −37), though volatility has the marginally larger mean IC, exactly as it marked Zcash's 2018, 2021 and 2025 tops. A quality tilt has been rewarded in this cycle: large, low-volatility assets outperformed in four of five years, which matches Brigida's 2023–24 finding of a large-cap premium and inverts the 2014–18 small-cap effect in Liu–Tsyvinski–Wu; size is therefore a Layer III (regime) variable, not a constant. And dilution, on its own, does not rank assets: the lowest-issuance quartile is full of legacy chains whose supply is fixed because nobody is building on them. The ledger's Dilution factor should be read as a necessary condition paired with usage confirmation, not as a screen that can be applied alone. Address growth, the on-chain "value" variable of the academic models, carried nothing here, probably because the Coin Metrics active-address series is dominated by spam and exchange activity for several of these assets.

The Zcash ledger measured abnormal returns against Bitcoin alone. The framework says the right benchmark is the market factor, and possibly the cohort factor. Re-estimating every Zcash event since March 2021 under three market models gives a direct test of whether the ledger's conclusions depend on the benchmark.

Seven tests, ordered so that each one uses the output of the last. T1–T3 and T6 can run on data already assembled here; T2, T4, T5 and T7 need the additional sources named. Each has a pass condition written down in advance, so that the exercise cannot be rescued by reinterpretation after the fact.

Sequencing. T6 first, because it is already half done and it fixes the measurement instrument. T1 and T3 next on a small universe (four assets per class) to get the shape of the class-conditional weights. T4 before anything is used for capital, because the ledger was derived from a survivor. T2 in parallel as the external data is sourced. T5 starts the month the rules are frozen and is the only test that cannot be back-fitted.