Issue 3 — Bigger and less concentrated: $5.66B across 300 vaults, curator HHI below 0.25, and one Spark vault on Base absorbing +$102.4M in a month

Five bullets, the only thing some readers will read. This is the first monthly Sharpe Research Note: flow and yield deltas are measured over a 30-day window ending 2026-10-10, rather than the 7-day window the two weeklies used. It is also our first note since W22 — the weekly cadence lapsed after that issue, and rather than backfill weeks we didn't publish, we restart on a monthly rhythm and use the frozen W22 snapshot as the comparison point for the "since our last note" view (§03–§04). Numbers are sampled at the close of the window.

The Morpho Blue universe spans 300 active vaults holding $5.66B in TVL across 11 chains. Aggregate capacity utilization is 88.9% — $5.66B deployed against $6.37B of stated caps, against 77.7% at W22. Caps are fuller than at any point we've sampled; the data suggests stated headroom, rather than deposit demand, may increasingly be the binding constraint in the fuller books. Version split: v2 203 vaults / $4.25B, v1 97 vaults / $1.41B.

Chains. Ethereum holds 56.3% ($3189M, 165 vaults) and Base 38.6% ($2188M, 57 vaults) — together 95.0% of the universe. Below them the tail is thin: Monad 2.05% ($116.3M), Tempo 1.15% ($65.3M), HyperEVM 0.63% ($35.9M). The newcomer is Tempo, which entered the top five with $65.3M across 3 vaults — up from $2.4M at the W22 snapshot. Its 30-day flow reads exactly $0.0M, which we read as a history-coverage gap rather than a static month (see the caveat in §05).

Assets. USDC anchors at 56.4% ($3192M). PYUSD is #2 at 12.8% ($726M) and RLUSD #3 at 12.5% ($708M). The PYUSD + RLUSD stack is 25.3% of the universe — roughly 3.6× USDT's 7.1%. The regulated-stable lead we've tracked since issue 1 holds; within it, RLUSD gained share since W22 (9.5% → 12.5%) while PYUSD's slipped (14.7% → 12.8%) as USDC's base grew faster. Asset HHI is 0.357 (W22: 0.296) — asset concentration rose as curator concentration fell.

Curators. Curator HHI 0.215, below the 0.25 "highly concentrated" line for the first time in this series (W21 and W22 both sat above it). Steakhouse Financial (32.6%), Sentora (26.5%), Gauntlet (17.0%) hold 76% of universe TVL, down from 83% at W22. The easing came mostly from the leader: Steakhouse's share fell from 40.1% to 32.6% on roughly flat TVL, while SparkDAO grew from $9.2M to $489M and Sentora added $393M.

Our last issue (W22) was built on a snapshot sampled 2026-06-03; this one is sampled 2026-10-10 — a gap of 129 days in which we did not publish. We can't reconstruct week-by-week what happened in between, but the two frozen snapshots let us state the end-to-end change honestly. The table below is snapshot-to-snapshot; it is not a flow measurement and does not separate deposits from price effects on non-USD collateral.

The universe got bigger and more consolidated. TVL rose +$1.10B (+24.2%) from $4.56B to $5.66B, while the active-vault count fell by 29 to 300. Average vault size went from $13.9M to $18.9M. Set against a -0.12% 30-day flow, the data suggests most of that growth landed between June and early September rather than in this month's window.

Who grew. The largest curator-level gains since W22: SparkDAO +$479.4M, Sentora +$393.0M, Gauntlet +$119.3M, RockawayX +$92.4M. Who shrank: AlphaPing $-76.0M (absent from this snapshot), Felix $-64.1M, Galaxy Curation $-33.7M, Pangolins $-20.4M. SparkDAO is the standout — a $9.2M book at W22 is $489M today across 3 vaults.

Yield firmed. Universe-weighted APY moved from 4.00% to 4.63% (+0.63 pts), and the sustainable ratio from 98.5% to 99.97% — the rewards slice that was still visible at W22 is now effectively nil (weighted rewards APY 0.001%). By asset: USDC 4.39% → 4.66%, USDT 3.50% → 3.20%, WETH 2.45% → 1.85%. Stables up, ETH down.

W22 published five forward items. The format's rule is that every one gets an explicit answer — even after a gap, and even where the answer is that the data has moved on. These are snapshot-to-snapshot readbacks across the gap, not week-three readings.

(1) Gauntlet — repositioning complete, or structural decline? Gauntlet's TVL went from $844M to $963M since W22 — growth in dollars — but its share slipped from 18.5% to 17.0% as the universe grew faster. This month's 30-day flow was $-26.4M (-2.7%), with Vault Bridge WBTC the single largest outflow in the universe. Read: the W21-era wind-down did not become a collapse; Gauntlet is a slowly-diluting #3, not a shrinking one.

(2) AlphaPing's carry — hold or cliff? AlphaPing does not appear in this snapshot's curator list at all. At W22 it ran $76M across 14 vaults at a weighted 9.98%. The data does not tell us why — withdrawal, vault deprecation, or a change in how the indexer attributes the curator are all consistent with an absent row — so we record the fact and do not speculate on the cause.

(3) SingularV ETH liquidation watch. At W22 SingularV ETH was both the top yield anomaly (z=6.5) and the top risk-mover. Today the SingularV book is $0.25M across 3 vaults at a weighted 1.23% — down from $3.36M at 29.88% — and it no longer appears in the anomaly list. The carry is gone. Realized bad debt across the universe is $0.0M, so whatever unwound did so without a recorded loss to suppliers; the snapshot can't distinguish an orderly exit from a liquidation.

(4) HyperEVM fill-through. This one reversed. HyperEVM TVL is $35.9M, from $105.1M at W22, and lost $18.4M (-33.9%) in this window alone. Felix — whose cap-fill we called supply-constrained — is $31.0M, from $95.1M, and -36.1% over 30 days. The W22 read held for a week; it did not hold for a quarter. We said HyperEVM's lending universe was Felix's book — that concentration is what made the reversal chain-wide.

(5) The cross-chain drift — pause or inflection? It resumed. Base's share went from 32.5% to 38.6% and Ethereum's from 60.1% to 56.3%. This month Base gained +$76.5M and Ethereum lost $55.1M. W22's positive Ethereum print was a pause, not an inflection.

Net universe TVL was $-7.0M (-0.12%) over the 30-day window — flat. The ten largest inflows summed to +$114.9M and the ten largest outflows to $-82.9M; the headline is a near-exact offset of one very large inflow against a broad set of mid-sized outflows.

One vault carried the inflow side. Spark USDC Vault (SparkDAO, Base) took +$102.4M (+35.3%), from $290M to $392M. The next-largest inflow, Metronome msUSD Vault, was +$4.8M — the gap between #1 and #2 is more than 21×. Ex-Spark, the universe ran $-109.4M (-2.0%). SparkDAO as a curator netted +$92.9M (+23.5%) after Spark Blue Chip USDC Vault lost $9.5M (-67.6%).

The outflows were broad and Steakhouse-heavy. The top three: Vault Bridge WBTC (Gauntlet, Ethereum) $-12.4M (-44.6%); Steakhouse USDC (Steakhouse Financial, Base) $-11.5M (-8.7%); Felix HYPE (Felix, HyperEVM) $-10.9M (-59.5%). 5 of the ten largest outflows are Steakhouse vaults; at the curator level Steakhouse netted $-49.9M (-2.6%) and Gauntlet $-26.4M (-2.7%). None of these are large relative to the books they sit in — a ~2.6% monthly drawdown on the universe's largest curator is ordinary book management, not an event.

By chain: Base +$76.5M (+3.6%), Ethereum $-55.1M (-1.7%), HyperEVM $-18.4M (-33.9%), Katana $-8.1M (-30.9%). HyperEVM and Katana are small in dollars but lost roughly a third of their TVL each. By asset: USDC +$37.6M (+1.2%), USDT $-10.7M (-2.6%), WBTC $-12.4M (-39.8%) — the WBTC line is almost entirely Vault Bridge WBTC.

Coverage caveat — read before quoting the flat headline. When a vault's history doesn't reach back the full window, the aggregator falls back to a short-horizon reference TVL, which produces an exact-zero 30-day delta. 24 curators holding $2.05B (36.2% of universe TVL) report exactly $0.0M — including Sentora ($1499M). So do the PYUSD, RLUSD, USDS, pathUSD, OUSD, USDG asset cohorts and the Tempo and Stable chains. An exact zero across a multi-vault, nine-figure book is far more likely a measurement gap than a genuinely static month. The -0.12% is therefore best read as the net flow of the measured portion of the universe; it is not evidence that the unmeasured 36% was flat.

Universe weighted APY: 4.63%, sustainable 4.63%, sustainable ratio 99.97% — +0.03 pts over the 30-day window (4.60% a month ago). Effectively unchanged on the month, and the highest universe-weighted reading in this series (W22: 4.00%). One methodological note: the "then" sustainable figure is derived by applying today's sustainable/spot ratio to the historical APY, so the month-over-month sustainable ratio is flat by construction and carries no information — the informative comparison is the snapshot-to-snapshot one in §03.

Distribution. Vault-level APY percentiles: P10 0.07%, P25 2.05%, median 4.25%, P75 5.92%, P90 8.36%. Every percentile from P25 up sits above its W22 reading (median 3.63%, P75 4.77%).

By asset. USDC (56% of the universe) prints a weighted 4.66%, sustainable 4.66%, P25–P75 4.3%–6.3%. PYUSD is the standout large cohort at a weighted 6.28% across 5 vaults (P25–P75 5.9%–6.7%). USDT prints 3.20% weighted but a cohort median of only 2.49% — the weighted figure is pulled up by a few large or high-paying vaults (see Flowdesk in §07). WETH sits at 1.85% weighted (median 1.50%), down from 2.45% at W22.

Top APY movers over the window (TVL ≥ $1M): 5 up, 7 down among the 12 largest absolute moves.

Steakhouse High Yield USDC (Steakhouse Financial, Arbitrum, $2.4M): 4.50% → 22.41% (+17.92 pts). Outsized upside on a small vault — the §07 anomaly screen and the YELLOW flag in §08 both point at the same position; do not treat the new rate as a baseline.

Gauntlet DAI Core (Gauntlet, Ethereum, $2.1M): 4.64% → 7.25% (+2.61 pts). Material move worth tracking into next month's diff.

Yearn OG USDC (Yearn, Ethereum, $10.4M): 6.59% → 9.01% (+2.42 pts). Material move worth tracking into next month's diff.

MEV Capital HYPE (MEV Capital, HyperEVM, $1.0M): 4.02% → 2.28% (-1.74 pts). Material move worth tracking into next month's diff.

Gauntlet USDC (Gauntlet, HyperEVM, $1.5M): 5.35% → 3.74% (-1.61 pts). Material move worth tracking into next month's diff.

Felix USDC (Felix, HyperEVM, $4.0M): 5.39% → 3.78% (-1.61 pts). Material move worth tracking into next month's diff.

The downside movers cluster on HyperEVM — MEV Capital HYPE -1.74 pts, Gauntlet USDC -1.61 pts, Felix USDC -1.61 pts. Read alongside HyperEVM's -33.9% TVL move (§05), the data suggests falling borrow demand on the chain rather than a supply glut: rates and balances both moved down together.

Vault sustainable APY z-scored against its (loan asset) cohort flags positions paying materially above peers. A vault paying >2σ above cohort is being paid for risk we don't yet see — surface it, decide whether the risk is legible, size accordingly.

This month's standouts:

Flowdesk Confidential High Yield USDT (Flowdesk, USDT): sustainable 12.21% vs USDT cohort median 2.49% — z=4.4, anomaly score 80.

Steakhouse High Yield USDC (Steakhouse Financial, USDC): sustainable 22.41% vs USDC cohort median 4.62% — z=6.2, anomaly score 80.

KPK wARS Yield (KPK, wARS): sustainable 24.59% vs wARS cohort median 24.59% — z=0.0, anomaly score 75.

Bitget x Steakhouse USDC (Steakhouse Financial, USDC): sustainable 11.39% vs USDC cohort median 4.62% — z=2.2, anomaly score 47.

August USDC V2 (August Digital, USDC): sustainable 10.64% vs USDC cohort median 4.62% — z=1.9, anomaly score 40.

Steakhouse High Yield USDC is flagged on every screen we run. It is the top APY mover (4.50% → 22.41%), the highest cohort z-score (z=6.2 against a USDC median of 4.62%), carries the universe's only YELLOW warning (Unrecognized Collateral Asset), and is one of 2 risk-movers (composite 50, dominant factor liquidity=100). It is small — $2.4M — so this is a position-level, not a universe-level, concern. The combination of a new rate, an unrecognised collateral and a maxed liquidity factor suggests a vault lending against collateral the registries don't yet know, into a thin exit. That reading may be wrong, but every flag points the same way.

Flowdesk Confidential High Yield USDT pays sustainable 12.21% against a USDT cohort median of 2.49% (z=4.4). Flowdesk runs $27.4M across 2 vaults with 0 RED / 0 YELLOW warnings and weighted risk 38 (universe median 27). A clean surface with a ~5× cohort premium is exactly the case where the deterministic score can't see the mechanism — the "Confidential" label suggests the strategy detail isn't public, which is itself the underwriting question.

Data caveat: KPK wARS Yield (wARS, sustainable 24.59%, anomaly score 75) appears with z=0.0 because its cohort median equals its own APY — it is effectively a one-vault cohort, so the z-score is undefined rather than benign. The high anomaly score comes from the risk-factor model, not the cohort comparison. Treat it as unscreened, not as clean.

Beyond the top two, the list thins quickly: the remaining scored entries sit between z=1.3 and z=2.2 — mild premia rather than outliers. 3 vaults in the table clear z≥2 in total.

Credit: still clean. Bad debt realized $0.0M, unrealized $0.0M; 0 depeg elevations. Across $5.66B of supplied capital, and across a gap in which HyperEVM lost most of its TVL and SingularV's high-carry book unwound, the snapshot records no supplier loss. With PYUSD + RLUSD at 25.3% of the universe, a clean depeg surface on the regulated-stable stack remains the precondition for that share to hold.

Warnings: 6 RED, 1 YELLOW (W22: 0 RED, 2 YELLOW). 5 of the RED flags are short timelock — 5 on HyperEVM, 4 of them on idle vaults with no attributed curator (HyperEVM USDC Idle Vault; HyperEVM USDe Idle Vault; HyperEVM USDT0 Idle Vault; HyperEVM HYPE Idle Vault), plus Felix USDC V2 (Felix). The remaining 1 is SharpByte USDC High Yield (SharpByte, Monad, Timelock). These are governance-configuration flags, not credit events: a short timelock means depositors have less time to exit before a curator change takes effect. The YELLOW is Steakhouse High Yield USDC (Unrecognized Collateral Asset) — the same vault at the top of §07.

Pending governance changes: 20 listed — by curator: Felix 9, Gauntlet 8, Anthias Labs 1, Steakhouse Financial 1, Clearstar 1; kinds: SetCap, RemoveMarket (3 market removals). Two things stand out. First, every listed change has a validAt already in the past at sampling — the earliest 2024-09-12 (Moonwell Flagship EURC), the latest 2026-07-30 (Yield Clearstar USDC). Their timelocks have elapsed but they remain queued, i.e. executable-but-unexecuted; the data can't tell us whether curators intend to accept them or have abandoned them. Second, the list is capped at 20 in the snapshot, while curator-level counts sum to 23 — so a few changes aren't itemised below.

Risk distribution. Universe risk-score P25/P50/P75: 26/27/30 (W22: 26/28/34). The upper half compressed — P75 fell from 34 to 30 — consistent with the higher-complexity HyperEVM and long-tail books we flagged at W22 shrinking out of the universe. Only 2 vaults register as risk-movers this month: Felix USDC V2 (Felix, risk 50, maturity=53); Steakhouse High Yield USDC (Steakhouse Financial, risk 50, liquidity=100).

Sentora runs $1.50B across just 8 vaults — 26.5% of the universe — at a weighted APY of 5.31%, +0.68 pts above the universe-weighted 4.63%. It is the #2 curator by TVL and the highest-yielding of the top three: Steakhouse prints 3.81%, Gauntlet 4.22%. The average Sentora vault holds $187M, against $32M for Steakhouse and $16M for Gauntlet — a very different operating model: few, very large books.

Since W22 the book grew from $1.11B to $1.50B (+35.5%) on an unchanged vault count (8 → 8), and its weighted APY rose from 3.99% to 5.31%. Share went from 24.3% to 26.5%. In a period when the universe's top-3 share fell, Sentora was the one leader that gained.

Composition. The book is almost entirely USD-denominated: 7 vaults, $1.47B, weighted 5.34%; the other sleeve is 1 vault at $25.0M. Weighted risk 28.4 (universe median 27), 0 RED / 0 YELLOW warnings, 0 pending changes. A clean surface.

What we can't see. The snapshot doesn't carry curator-by-asset breakdowns, so we can't say which stablecoins Sentora's USD sleeve lends. We note only that the PYUSD + RLUSD cohorts ($1.43B) are of similar size to Sentora's USD book, and that both — like Sentora itself — show exactly zero 30-day flow, the coverage-gap signature from §05. That is a hypothesis for the next issue to test, not a finding. It also means we cannot report Sentora's flows for this month at all.

Spark USDC Vault on Base, now $392M TVL — up +$102.4M (+35.3%) in 30 days, the largest single inflow in the universe by a factor of more than 21. On its own it is larger than Base's net gain (+$76.5M) and larger in absolute terms than the universe's net change ($-7.0M).

The curator context. SparkDAO runs 3 vaults totalling $489M (8.6% of the universe), up from $9.2M at W22 — making it the fastest-growing large curator over the gap and the reason curator HHI fell below 0.25. Its weighted APY is 3.78%, below the universe-weighted 4.63% and below the USDC cohort's 4.66%; weighted risk 25.0, 0 RED / 0 YELLOW, 0 pending changes.

What the data suggests. Capital arriving at scale into a curator book whose weighted rate sits below the USDC cohort is not obviously yield-chasing (the snapshot carries curator-level, not vault-level, APY for Spark USDC Vault, but the vault is 80% of that book). The more consistent reading is allocation by mandate or relationship — capital choosing SparkDAO as a counterparty and accepting a below-median rate for it. We can't confirm the source of the deposits from the snapshot; we can say it isn't explained by rate.

The other side of the same book. Spark Blue Chip USDC Vault on Ethereum lost $9.5M (-67.6%) over the same window, to $4.5M. SparkDAO's book is consolidating into one very large Base USDC vault: Spark USDC Vault is 80% of the curator's TVL.

Five items the editorial team will track into the next monthly edition. Each will get an explicit answer there — the same discipline we applied to W22's list in §04.

(1) Spark USDC Vault — does the inflow continue at a below-cohort rate? +$102.4M in a month at 3.78% curator-weighted. A second month of nine-figure inflow would confirm mandate-driven allocation; a reversal would tell us the move was a single allocator. Either way, SparkDAO's share (8.6%) decides whether HHI stays below 0.25.

(2) Steakhouse High Yield USDC — does 22.4% hold, and does the YELLOW flag clear? Top mover, top z-score (6.2), unrecognised collateral, liquidity factor 100. We'll report whether the rate normalises, the collateral gets registered, or the position grows.

(3) HyperEVM — floor or further decline? $35.9M left, -33.9% this month, 5 short-timelock RED flags on the chain. Felix ($31.0M, -36.1%) still has 9 pending changes. If balances stabilise while the timelock flags persist, the chain's risk profile is getting worse per dollar even as it stops shrinking.

(4) The elapsed-timelock backlog. 20 listed changes past their validAt, the oldest from 2024-09-12; Gauntlet and Felix hold 17 of them. We'll track how many are executed, revoked, or still sitting there — and flag any that execute on vaults with material TVL.

(5) Closing the coverage gap. 36% of TVL — including all of Sentora and the PYUSD/RLUSD cohorts — showed exact-zero 30-day flow. By the next monthly we intend to measure these books from our own snapshot history rather than the indexer's per-vault history. If the "flat month" headline changes materially once they're measured, we'll say so.

Data sources. Per-vault state: Morpho Blue GraphQL indexer. Every number in this issue traces to the frozen snapshot for this slug, or — for "since our last note" comparisons — to the frozen W22 snapshot. This edition omits the cross-protocol and macro sections that ran in the weeklies: their external benchmarks (Aave/Compound rates, T-bill, stablecoin market caps, LST yields) are not part of the frozen snapshot, and we'd rather drop a section than print figures we can't trace.

Methodology. Risk score, complexity score, depeg composite, sustainable-APY computation, and yield-anomaly z-score — all documented at sharpe.finance/methodology. No black-box scoring. Re-running the methodology against the same sampled state produces the same composite — the payload hash below is the integrity check.

Sampling. Figures sampled at 2026-10-10T16:37:55.190Z UTC; flow and yield deltas computed against the matching time 30 days prior, using each vault's indexer history point closest to the lookback. Where that history doesn't reach back far enough, the aggregator falls back to a short-horizon reference, producing exact-zero deltas (§05 quantifies the affected share). The month-ago sustainable APY is derived from today's sustainable/spot ratio applied to historical APY. Vaults below $1M TVL are filtered out of cohort statistics (movers, anomalies); they remain in the universe-level footprint. Vaults with reported APY ≥50% are filtered out of weighted aggregates as pathological. Pending changes are listed oldest-first and capped at 20.

Cadence. This is the first monthly (SRN-YYYY-Mnn). After W22 the weekly cadence lapsed; monthly editions cover the 30 days to publication and compare against the most recent prior frozen snapshot where useful.

Glossary. Sustainable APY: spot APY net of the reward-emission slice — the rate that survives an immediate emission rolloff. Cohort z-score: vault sustainable APY normalised to its (loan asset) peer distribution; undefined for single-vault cohorts. Curator HHI: Herfindahl-Hirschman index of curator-level TVL shares (≥0.25 conventionally "highly concentrated"). Short timelock: a vault whose governance delay is below the threshold that gives depositors a practical exit window. Elapsed pending change: a queued change whose timelock has passed but which has not been executed or revoked.

Payload hash. 9b29bec3f70cc9cff92a2224… (sha-256 over the frozen data JSON).