Refresh, 20 October 2026 (block 26,233,101): economics, capacity, parameters, oracle, co-borrowers, wallet eligibility and the JAAA control plane were re-verified on-chain, superseding the 2 and 7 September figures and the same-day RLUSD borrow path, break-even yields and size limits were added (economics-latest.json from refresh-economics.py, plus snapshot-2026-10-20.json; the 7 September $1M table is kept as history).

Bottom line

The position is a self-managed leveraged JAAA carry, run from a NOCA wallet that is KYC’d and allowlisted by Anemoy. The wallet subscribes USDC into the JAAA fund, supplies the JAAA to Aave Horizon (the Aave Labs RWA instance) in eMode “JAAA GHO” (LTV 88% / liquidation threshold 90% / bonus 5%, unchanged), borrows GHO at its governance-set flat rate, 3.25% since 16 September, swaps it to USDC, subscribes again, and loops to a chosen leverage. A second path borrows RLUSD in Horizon’s base mode instead (LTV 87% / LT 89%), which swaps into USDC at almost no cost but pays a utilization-driven rate. Net return convention: leveraged spread between the JAAA NAV accrual and the borrow cost, minus the entry and exit swaps (GHO→USDC 22–24bp, RLUSD→USDC under 1.2bp, quoted 20 October) and zero venue fees. On 20 October, $1M at 4x models 3.92%, about $39,186 a year, on trailing 30-day JAAA accrual (3.46%), against 10.36% on 7 September; on 90-day accrual (4.42%) it models 7.78%. Unlevered JAAA earns 3.46% on the same 30-day basis, so leverage currently adds less than half a point while multiplying NAV exposure fourfold. Leverage beats holding JAAA only while JAAA yields more than the borrow cost: in year one, above about 3.53% on the GHO path (because of the entry swap) and 3.45–3.66% on the RLUSD path depending on size; after year one, above 3.30% on the GHO path. Trailing 30-day accrual (3.46%) sits at that line; 90-day accrual (4.42%) is well above it (see When it pays and how large it can be).

This is the same trade 3F sells (see ../3f-jaaa-leveraged-carry/memo.md), with the middleman removed: no performance fee, no bridge spread, no keeper who alone can execute our exit, no 2-of-3-behind-24h upgrade risk, a Chainlink OCR2 feed (16 transmitters) instead of Chronicle’s 4-of-8, and a 90% liquidation threshold instead of 98%. The borrow leg also differs: GHO is a flat 3.25%, while 3F’s Morpho market charges 4.35% and has inverted 3F’s 8x carry to −2.94% on the same inputs. What we take back in exchange: entry mechanics (about five sequential subscription rounds to loop to 4x, or a single round if the treasury fronts the full notional), self-managed liquidation risk, and Horizon’s own control plane (Aave Labs Safes that changed this reserve three times in five weeks with no timelock).

Immediately available exit: none at the position level. Debt can be repaid from treasury instantly (GHO is liquid), but the JAAA collateral always exits through the fund’s async redemption gate (Anemoy single-EOA approval, no forcing function; measured fulfillment since September: median 0.10 days, maximum 0.87 days, 11 of 11 requests). Dominant risks: a NAV markdown while levered, priced by a guarded feed that freezes at the last accepted value during anomalies, in a venue where two co-borrowers sit at health factor 1.02 and would hit the same redemption pipe; and, underneath every JAAA venue, a single Centrifuge BalanceSheet manager key that can move JAAA out of Horizon’s collateral reserve without Aave’s or the borrower’s signature (fork-verified; base memo R6).

The NOCA Safe is still not a JAAA member on-chain, which blocks the strategy at its first step. Capacity is no longer binding at $1M: the JAAA supply cap is 60M shares with 49.28M supplied, leaving 10.72M JAAA of headroom (about $11.3M of collateral, $2.8M of equity at 4x). That headroom, shared with every other supplier, is the strategy’s capacity at 4x.

Mechanism and true exposure

One unit of USDC travels: Anemoy subscription → JAAA → Horizon supply (eMode 10) → GHO borrow → swap → repeat.

Label-versus-exposure gap: “JAAA collateral on Aave” is AAA-CLO credit at chosen leverage where (i) collateral pricing is a guarded copy of a NAV one EOA produces, (ii) collateral mobility is zero (a non-transferable aToken over a KYC-gated token over an async fund), (iii) the collateral itself sits in a reserve that a single Centrifuge manager key can drain, and (iv) the venue’s other side is a handful of loopers, two at health factor 1.02, whose stress behavior converges on our exit pipe. Leverage layers: the CLO waterfall (inherent) plus one on-chain loop. No socialization layer applies to our debt, but a shortfall in the aJAAA reserve’s underlying would fall on its last withdrawers.

Net economics, benchmark and capacity

Return legs (block 26,233,101):

Measure2x4x (LTV 75%)5x (LTV 80%)8.3x (LTV 88%, eMode max)
Net carry, 30-day JAAA (3.46%)3.61%3.92%4.07%4.59%
Net carry, 90-day JAAA (4.42%)5.54%7.78%8.90%12.64%
NAV markdown to liquidation (LT 90%)44.4%16.7%11.1%2.2%
Equity loss per 1% markdown2%4%5%8.3%

Benchmarks: unlevered JAAA 3.46% (30-day) or 4.42% (90-day); 3F’s 8x product −2.94% (30-day) or +4.45% after fees (90-day) at a 5.7% liquidation distance. On today’s legs, the decision-relevant comparison is that 4x here out-earns 3F’s 8x on both bases while tolerating three times the markdown, because GHO costs 3.30% APY against 4.35% on 3F’s Morpho market.

Capacity and marginal capital: supply cap 60M JAAA (raised 40M → 50M on 16 September and 50M → 60M on 30 September), 49.28M supplied (82%): headroom 10.72M shares, about $11.3M of collateral and $2.8M of equity at 4x. A $1M 4x position needs 3.80M shares. GHO headroom of $25.7M covers the borrow leg. Marginal capital does not move the GHO rate (flat curve); it does move the RLUSD rate (below). The cap has been raised five times since April 2026, each time by a no-timelock operator Safe.

When it pays and how large it can be

Net equity yield at 4x = 4 × JAAA yield − 3 × borrow cost − the entry swap as a share of equity (year one only). Leverage therefore beats simply holding JAAA when JAAA yields more than the borrow cost plus one third of the entry swap. The position stays profitable in absolute terms down to a JAAA yield of about 2.6–2.7%.

Path, our equity at 4xBorrow cost (APY)Entry swapJAAA yield needed to beat holding JAAA (year one / after)Year-one net, 30-day JAAA (3.46%)Year-one net, 90-day JAAA (4.42%)
GHO, $1M3.30%$6,6523.53% / 3.30%3.25%7.12%
GHO, $2M3.30%$13,8953.53% / 3.30%3.22%7.09%
GHO, $2.82M (cap)3.30%$20,2863.54% / 3.30%3.20%7.06%
RLUSD, $1M3.45%about $03.45% / 3.45%3.48%7.35%
RLUSD, $2M3.56%$4263.56% / 3.56%3.14%7.00%
RLUSD, $2.82M (cap)3.64%$1,0143.66% / 3.64%2.86%6.73%
Holding JAAA, any size—none—3.46%4.42%

After year one the GHO path earns 3.92% (30-day) or 7.78% (90-day) at every size up to the cap, because its rate does not move and the swap is behind it. The same arithmetic at $1M across JAAA yields:

JAAA yieldHolding JAAAGHO path, year one (after)RLUSD path, year one
3.00%3.00%1.42% (2.09%)1.65%
3.50%3.50%3.42% (4.09%)3.65%
4.00%4.00%5.42% (6.09%)5.65%
4.50%4.50%7.42% (8.09%)7.65%
5.00%5.00%9.42% (10.09%)9.65%

Size. The supply-cap headroom is the hard limit: $2.82M of equity at 4x. Below it, the GHO path’s cost is flat in size apart from the swap (22.2bp rising to 24.0bp), while the RLUSD path’s rate rises with our own borrowing, so the GHO path earns more above about $1M. Lower leverage fits more equity under the same cap (up to about $5.6M at 2x), but earns less (GHO path at 2x: 3.61% on 30-day and 5.54% on 90-day inputs, before the swap). Other suppliers can take the headroom first, and cap raises are at Horizon’s discretion. Neither our size nor the path changes JAAA’s own yield, which is the variable that decides the result.

Inputs: Horizon reserve state and interest-rate parameters at blocks 26,233,742 and 26,233,759; single-shot CoW API quotes for the NOCA Safe, 2026-10-20 12:11–12:21 UTC (splitting the swap could cost less); figures in snapshot-2026-10-20.json under sizing.

Exit, cost and latency

Fast unwind (treasury self-bridge, the path 3F cannot offer us): repay the full GHO debt from treasury in one transaction (buy GHO at market; debt is par-priced), withdraw all JAAA, and requestRedeem the entire position. Latency is one to three days at the measured fulfillment cadence; the temporary treasury outlay is about three times equity at 4x. This also works as a fast entry: front the full notional, subscribe once, and borrow back 75%. That leverages the position in days with no bridge counterparty.

Looped unwind (no external capital): per round, redeem the withdrawable excess collateral, wait for settlement (typically hours, up to about three days), repay, and withdraw more. From 4x that is about five sequential rounds, roughly a week at the measured cadence, symmetric with looped entry. The position stays levered and exposed to markdown throughout.

What can block or degrade exit:

Concurrent exits: the four largest suppliers ($52.8M of combined collateral, two at health factor 1.02) redeem through the same operator-gated fund path. Under a markdown, liquidators seizing their JAAA also join that queue, ahead of or beside us. The fund’s measured record contains no episode of concurrent large redemptions under a markdown; the September down-prints (−8.3bp) were too small to test it.

Stress scenario (Mar-2020 shape, 3–5% markdown): at 4x we lose 12–20% of equity on paper, remain about 12 points above liquidation, and can choose to hold through recovery. The position survives without forced realization, which is the entire argument for 4x over 3F’s 8x and 10x. At 8.3x the same event is a near-certain liquidation at the marked-down NAV plus the 5% bonus. Between those bounds, leverage is a dial the committee controls at entry and we control thereafter (repay-down is always available at treasury speed).

Risks

Verified on-chain 2026-09-02 (blocks 25,890,991 and 25,891,011): reserve config and history, eMode categories, caps and headroom, IR data and GHO rate history, oracle chain (adapter → EACAggregatorProxy → OCR2, 16 transmitters), RwaAToken source and disabled paths, ACL role map, RwaATokenManager grants (none), whale account data, borrow/withdraw/supply simulations. Re-verified 2026-10-20 (block 26,233,101): caps, IR data, reserve configuration and its change transactions, the risk-admin Safe (3-of-4, isRiskAdmin = true), oracle versus vault, all aJAAA holders’ account data, NOCA Safe membership, fund settlement latency and NAV lag, and the BalanceSheet control path with fork tests. External: Horizon design doc (aave-v3-horizon repo), ARFC and governance forum, LlamaRisk weeklies, LlamaGuard release post, all read 2026-09-02. Base-memo JAAA risks import wholesale.

IDRiskMechanism and evidence
R1Leveraged liquidation into a thin, KYC-gated liquidator set, beside co-borrowers at the edgeeMode LT 90%, bonus 5%. The working liquidation path transfers underlying JAAA, so only Anemoy-allowlisted addresses can liquidate, and they then exit through the same operator-gated fund path. At the pin, eleven addresses hold aJAAA and the top four hold 85.6%: 0xcf25…68fb HF 1.071 ($24.6M collateral), 0xd5ef…4ae2b HF 1.022 ($9.8M), 0xb472…9112 HF 1.024 ($9.7M, new) and 0xcfed…b3fe HF 1.081 ($8.8M, new); 0xd838…823b has exited JAAA. A 2–3% markdown makes tens of millions liquidatable at once. The Horizon design doc itself flags that slow liquidations accrue bad debt. Our control: enter at 4–5x or below and hold repay capacity in treasury. Trigger: own HF < 1.5; any large holder HF < 1.01; aggregate market HF deterioration.
R2Guarded-NAV oracle freezes exactly when it matters; root is one EOALlamaGuard rejects out-of-bounds NAV updates and freezes at the last good value; bound widths and staleness for JAAA are unpublished; the bounds registry is owned by a 3-of-4 Safe. A freeze during a real markdown lets borrowers borrow against stale collateral value and delays liquidations market-wide, then gaps. The feed is Chainlink OCR2 (16 transmitters, stronger than Chronicle’s 4-of-8), but the value it publishes is the same Anemoy-EOA NAV (1.052831 on both venues and in the vault at the pin). The first NAV down-prints (late September) passed through without a freeze. Shared-root exposure across this position, the base JAAA position, 3F and every Horizon JAAA borrower (D6). Trigger: feed vs vault pricePerShare divergence >25bp; feed age >36h; params-registry change.
R3Operator control plane at Safe speed, no timelockAave Labs-side Safes (4-of-6 0x13b5…e1d3 and executors it owns) hold POOL_ADMIN / RISK_ADMIN / EMERGENCY_ADMIN: they can freeze or pause the reserve, cut LTV/LT and change caps with no timelock (BGD Labs’ unadopted objection in the ARFC). The 3-of-4 risk-admin Safe (0xe6ec…253c) used that power three times in five weeks: the GHO rate and a cap raise on 16 September, a further cap raise on 30 September, and base-mode JAAA LTV/LT 81/86 → 87/89 on 17 October. The 4-of-6 Safe also administers the ATOKEN_ADMIN forced-transfer power over aJAAA (HF-enforced; no grants made yet; recipients not validated against the allowlist). Aave DAO holds upgrade rights via Executor Lvl-1 (slow, standard). Open item: deployment contract 0x3eaf…d51e (AaveV3SetupBatch, owned by an EOA) still holds DEFAULT_ADMIN + POOL_ADMIN. Trigger: any ACL role change; any RwaATokenManager grant; reserve parameter change; the 0x3eaf roles being exercised.
R4GHO legs: policy rate and peg-at-exitThe borrow rate is set by governance: five changes since August 2025 within 2.75–4.75%, most recently 3.00% → 3.25% on 16 September (trigger fired). A 150bp hike at 4x costs 4.5 points of carry with no utilization signal preceding it, and at today’s 30-day JAAA accrual a further 15bp rise would erase the spread. Debt is oracle-priced at $1.00, but repayment buys GHO at market, so a GHO premium at unwind is an exit tax (on 2026-10-20 GHO traded slightly below par, which made entry cost 22–24bp and exit about 11bp cheaper). The RLUSD path replaces the policy rate with a utilization curve: its cost rises with our own size and with other borrowers’ demand, and reaches 5.13% APY at 90% utilization. Trigger: GHO rate-strategy update event; GHO market price >$1.005 sustained; on the RLUSD path, utilization above 80% or a borrow cost above trailing-30d JAAA accrual.
R5Self-managed operations; fund cadence is habit, not contractLooped entry to 4x is about five sequential subscription rounds, exposed to NAV and rate moves mid-build; fast entry or exit needs treasury fronting of about the full notional for days. The measured cadence (redemption median 0.10 days since September, 0.27 days over the prior six months, no stress observation) is operator behavior with no on-chain enforcement, so plan operations assuming T+3 or worse. No keeper exists to fail, and none exists to save us either: liquidation avoidance (repay-downs on markdown) is a NOCA operational duty, including through NAV-publication lags that reached 118 hours in October. Trigger: NAV staleness >72h while levered; markdown >1.5% in 24h; any redemption request pending >5d.
R6Underlying JAAA and venue concentration imports, including a single-key path to the reserve’s JAAASingle-EOA NAV, redemption and KYC gate, custody opacity, CLO mark-to-market and anchor concentration: base memo R1–R6. Base R6 applies directly: a Centrifuge BalanceSheet manager (the bare EOA 0x42b7…208f, two Safes, or any address the Anemoy NAV EOA appoints) can move JAAA out of the non-endorsed aJAAA reserve, which a fork at the pin confirmed. The ETH-leg JAAA supply fell 35% month on month to 278.5M, and Horizon now holds 17.7% of it. Venue side: the JAAA reserve is 82% of cap and dominated by four addresses (attribution UNVERIFIED); Horizon TVL drifted down from about $557M (Nov-2025 peak) to about $350M (Jul-2026). Trigger: supply-cap utilization >95%; any holder redeeming >25% of the reserve; any BalanceSheet TransferSharesFrom from the aJAAA reserve.

Trust and control

Evidence boundary

Pinned and reproducible: blocks 25,890,991 / 25,891,011 (2026-09-02) in snapshot.json, and block 26,233,101 (2026-10-20 10:08:59 UTC, hash 0xc116…1f8c) in economics-latest.json (refresh-economics.py 26233101) and snapshot-2026-10-20.json, all via cast and $ETH_RPC_URL. The BalanceSheet fork results and manager set are in ../janus-henderson-anemoy-jaaa/snapshot.json. Sources verified from Blockscout and GitHub: RwaAToken, RwaATokenManager, ScaledPriceAdapter, EACAggregatorProxy/OCR2, ACL role events, eMode configs, IR data and GHO rate-update history, reserve config history, whale account data, and the Centrifuge BalanceSheet, share token and hook.

External inputs (2026-09-02): Horizon design doc (liquidation and forced-transfer semantics); ARFC: Horizon’s RWA Instance (governance split, BGD and ACI dissent); LlamaRisk Horizon weeklies; LlamaGuard release; Chaos Labs on Horizon; Resolv $100M program; Aave launch and institutions blog posts. 2026-10-20: CoW Protocol quote API (api.cow.fi/mainnet/api/v1/quote) for GHO→USDC, USDC→GHO, RLUSD→USDC and USDC→RLUSD at $0.75M–$8.46M, signed as the NOCA Safe.

Unresolved (falsifiable, before any capital decision):

  1. LlamaGuard bound widths and staleness policy for the JAAA feed (feed contract unverified; params registry layout not yet reverse-engineered), and why the feed now prints exactly the vault price.
  2. Identity and intent of 0x3eaf…d51e (setup batch with live POOL_ADMIN + DEFAULT_ADMIN, owned by an EOA).
  3. Attribution of the large aJAAA holders and their deleveraging behavior or agreements.
  4. Stress-period swap depth and the worst observed GHO premium. Calm-market depth is measured: CoW quotes on 2026-10-20 cost 22–24bp for GHO→USDC up to $8.46M and under 1.2bp for RLUSD→USDC.
  5. Whether Horizon (or Aave Labs) has any arrangement with Centrifuge or Anemoy covering the BalanceSheet power over the reserve, for example endorsement of the aJAAA reserve.
  6. NOCA KYC and allowlisting with Anemoy (prerequisite, same as base memo; still absent on-chain at block 26,233,101) and jurisdictional eligibility for the fund.
  7. A fork test of the full loop (supply → borrow → swap → resubscribe → repay → withdraw → redeem request) from an allowlisted address, for both the GHO path in eMode and the RLUSD path in base mode; each GHO-path leg was simulated separately on 2026-09-02, but not the sequenced flow, and the RLUSD path not at all.

Supporting evidence

Reserve history (verified): listed Sep-2025 at LTV 71 / LT 78, cap 9.9M; raised to 81/86 by Dec-2025 and to 87/89 on 2026-10-17; cap 9.9M → 10M (Apr-2026) → 22M (May-2026) → 40M (Sep-2026) → 50M (2026-09-16) → 60M (2026-09-30); supplied about 1M → 2.5M (Apr) → 16.7M (May) → 32.7M (Sep) → 49.28M (20 Oct). GHO base rate: 4.75% (Aug-2025) → 4.25% → 3.25% → 2.75% → 3.00% (May-2026) → 3.25% (2026-09-16, tx 0x66ed…9dc8, risk-admin Safe).

aJAAA holders (block 26,233,101): 0xcf25…68fb 15.36M aJAAA, HF 1.071 (mixed collateral, base mode); 0xd5ef…4ae2b 9.29M, HF 1.022 (eMode 10); 0xb472…9112 9.21M, HF 1.024; 0xcfed…b3fe 8.33M, HF 1.081; 0x0f88…a077 4.05M, HF 1.138; six smaller holders. At block 25,890,991 the cohort was 0xd5ef HF 1.027, 0xcf25 HF 1.038 and 0xd838 HF 1.073, together about 97% of aJAAA.

Simulations (eth_call, 2026-09-02): borrow GHO from an existing supplier → clean; borrow RLUSD in eMode 10 → revert (eMode restriction); withdraw JAAA → clean; supply from a non-KYC address → reverts at the JAAA token. Fork (2026-10-20): a BalanceSheet manager moved JAAA out of the aJAAA reserve (transferSharesFrom, no Aave signature). The comparison with the 3F route is the table under Net economics; the 3F memo’s section “Running the same loop without 3F” covers the keeper-free Morpho alternative.

Fund settlement latency: measured 2026-09-03 over blocks 24,606,331→25,896,331, every RedeemRequest matched to the next RedeemClaimable per controller — 43/43 fulfilled, median 0.27d, max 3.22d (largest: Grove’s 318.5M JAAA on 2026-03-09, fulfilled in 1.92d); deposits n = 83, median 0.61d, max 4.18d. From block 25,896,331 to 26,233,101: redemptions 11/11, median 0.10d, max 0.87d (largest 3.27M on 10-14); deposits 33/33, median 0.20d, max 0.93d. The September outflow from the anchor holder bypassed this queue (base memo R6). This latency series is the basis for every entry and exit timing figure above; it is observed operator behavior with no on-chain enforcement and contains no stress episode.