Fundability — the honest case, every claim checkable
The edge is not yield. The edge is honest measurement and refusal — being the party that correctly prices and declines the risk others don't see, and proves it. Below is the whole case, with the honest constraints stated up front, and a link to check every number yourself.
Rates Desk — refusal-first carry
Validated, live-paper. Carry leg real -> fundable. Capacity-bound.
RWA Backstop — book NO-GO
Measurement GO; underwriting needs custody + legal + capital (off-code).
Liquidator — published kill
Addressable market ~5-10x below the bar. We publish what we kill.
Honest constraints — up front, not buried
- The track is THIN. Only days with a real daily-cycle log count; the earlier backfill was reset out. The track is accruing to 30 — trustworthy risk-adjusted ratios land near day 30; until then they read THIN, never a fabricated number.
- Capacity is bounded. The standalone rates desk clears tens of thousands per year above the floor at today's market depth; combined across sleeves, after a correlation haircut, the honest figure is lower still. $10M = scale + decorrelation + AUM — NOT reachable today.
- The capital is paper. $0 real capital. This is a research / paper track on a virtual $100k book. Not investment advice, not a forecast.
Live proof
source: static / offlineChain decisions
—
hashed, verifiable
Refusals
—
published with reason
Evidenced days
— / 30
— ready · THIN · accruing
Chain verified
—
3-state · never green on absent
Head hash (fingerprint of the entire decision history): —
The validated edge
Same engine, same day: a toxic LRT book is refused on structure, a clean stable-carry book is approved. Refusal fires before economics — a great quote cannot buy its way past a toxic book. Both decisions are hashed and public. The fully worked example (real ezETH-refusal + sUSDe-entry hashes) lives in the DD pack.
Rates Desk
harvest carry · refuse tail-comp
RWA Backstop
liquidation NAV, not marketing NAV
Liquidator
measured -> published NO-GO
Between here and $10M — off-code
The code took each thesis to an honest verdict for free. But the boundary is the same across all three — the code can measure and refuse; the $10M is off-code. Named plainly, not hidden:
Custody / MPC
institutional key management for real capital
External audit
independent code + controls audit of the execution path
Legal
fund structure, collateral perfection, force-redemption rights
Real capital + relationships
whitelisting, redemption-queue access; the carry edge needs scale -> AUM
Check every number yourself
Don't trust us — check us. One zero-dependency verifier + SPA's public JSON, on a clean machine with none of our code: it re-derives every decision hash, every exit-NAV proof_hash, and the anchors — reporting the exact broken_at if a single byte of history was altered.
Command
python3 verify_spa.py data/ Verifier: scripts/verify_spa.py · recipe: docs/PROOF_CHAIN_SPEC.md · DD pack: docs/DD_PACK.md (the fully worked refused-vs-approved example with real proof_hashes; every number traced to its source).
Check each claim