The Structural Desk
Our own backtest proved plain crypto yield isn't an edge. The real $10M-scale edge is a structural role: the transparent measurement and underwriting desk for on-chain rates, collateral and risk — not picking APY.
How we got here
We ran three deep DeFi-edge researches and de-risked each on our own stack. All three converge on one truth: the money is not in what to hold — it is in occupying the seat that honestly measures and underwrites rates, collateral and risk.
Status
Paper research. Advisory — nothing trades live. Every verdict is deterministic (LLM forbidden in the risk gate). We publish what we built, what we gated, and what we killed.
Loading live desk summaries…
The three seats
Each thesis was de-risked on our stack and given an honest verdict. Two are built. One is killed. All three test the same hypothesis: the edge is a role, not a yield.
Live verdict inputs
loading the decision log…
loading the safety board…
Rates Desk
On-chain interest-rate & basis
The edge
Price & harvest the disagreement between Pendle implied yield, lending rates and perp funding — and REFUSE books where the high rate is just unpriced tail.
The de-risk finding
Refusal fired early on toxic LRTs (vetoed before the Aug-2024 / Apr-2026 depegs). In the backtest, the validated FixedCarry sleeve clears the 3.4% RWA floor in/out-of-sample and through every stress — ~6.09% net APY is a backtest/candidate figure, NOT realized carry (advisory, reporting-only). The live paper track is still short (~1 week, ~0% net so far) and does not yet beat the floor.
Open the Rates DeskRWA Repo Backstop
Liquidation-safe collateral
The edge
Lend against Liquidation NAV, not marketing NAV — underwrite the executable exit of tokenized-RWA collateral, not the assumed $1.00.
The de-risk finding
Built the live Collateral Safety Board: across the whole board, no tokenized-RWA asset is cash-like on an executable on-chain exit, and nearly all have ~$0 public on-chain exit (live counts on the board). We ship the measurement layer; the underwriting relationship/book is capital/legal-gated — off-code.
Open the Collateral Safety BoardCross-Domain Liquidator
Balance-sheet long-tail liquidator
The edge
Be the balance-sheet liquidator for long-tail collateral when MEV bots fail to clear it.
The de-risk finding
KILLED. The addressable long-tail liquidation opportunity is ~$2–4M/yr gross (top-20 ≈ $2.2M) — 5–10× below the $20M bar that would justify the custody + CEX build. A naive $18M estimate was an artifact of MetaMorpho curator vaults (not liquidatable).
Not built. Full reasoning & the numbers — below.
Rates Desk · sleeve promotion map
ADVISORY · reporting-only · not live
The four Rates-Desk sleeves are run through the SAME deterministic promotion rubric as the
Strategy Lab and given an honest stage. This is reporting:
the sleeves are IS_ADVISORY=True and are never routed to the live
tournament/allocator before go-live.
FixedCarry
PAPER_CANDIDATEbacktest net APY ~6.09% · advisory (not realized)
LeveredCarry
PAPER_CANDIDATEbacktest net APY ~4.96% · advisory (not realized)
RateMatrix
PAPER_CANDIDATEbacktest net APY ~6.09% · advisory (not realized)
BasisHedge
BLOCKED-NO-HEDGEno keyless forward-funding (Boros) · ~4.99% backtest-proxy only, live-blocked
Loading sleeve stages…
We publish what we kill
We tested the balance-sheet-liquidator thesis and KILLED it: ~$2–4M/yr addressable, 5–10× below our bar. We publish negative results because a track record of honest refusals is the moat.
$2–4M
addressable / yr (gross)
~$2.2M
top-20 pools
$20M
bar to justify custody + CEX
5–10×
below the bar
The naive $18M estimate that first looked attractive turned out to be an artifact: it counted MetaMorpho curator vaults as addressable, when those are not liquidatable. Strip that artifact out and the real addressable opportunity falls 5–10× below the bar that would justify the build. We killed it cheaply — at the estimation stage, before a single line of execution code.
Measured inputs behind the verdict
- addressable: ~$2–4M/yr gross (top-20 pools ≈ $2.2M)
- build bar: $20M/yr to justify custody + CEX execution
- naive $18M estimate rejected: MetaMorpho curator-vault artifact (not liquidatable)
- conclusion: 5–10× below bar → NO-GO, killed at estimation stage
reasoning: docs/LIQUIDATOR_DERISK.md · publishing the kill is the credibility signal
The shared engine
Three seats, one engine
All three theses rest on one core: a deterministic, risk-adjusted measurement engine — fair value + tail-risk haircuts + exit-capacity — that says what to harvest AND what to refuse. That engine is the real asset; the $10M is scale, trust and relationships — not strategy alpha. It is off-code: a multi-year game on top of the engine.
Fair value
Decompose every quoted rate into genuine carry vs. risk compensation.
Tail-risk haircuts
Penalize depeg, oracle, governance and credit tail — never ignore them.
Exit capacity
Measure what you actually realize on exit at size — not the marketing $1.00.
Go deeper