# Release v0.5.65: Kinked utilization rates and directed rounding This release adds DeFiKinkedInterestRateUtilization as the 167th direct MasterSuite import. It proves the rational utilization admission criterion, continuity and monotonicity of a two-segment borrow curve, convexity under ordered slopes, supply-rate monotonicity and conditional spread, and fixed-point rounding bounds. For positive debt B and positive denominator D=Cash+B-Reserves, utilization U=B/D is at most one exactly when Reserves<=Cash. Zero debt has its own U=0 branch. Positive D alone is insufficient: Cash=10,B=100,Reserves=20 gives U=10/9. With nonnegative base and slopes and 0=0. On 0<=u<=1: rb-rs = rb*(1-u*(1-rho)); 0<=rs<=rb; rs0 and u*(1-rho)<1. At u=1,rho=0 the spread is zero. A decreasing-slope regression demonstrates that monotonicity and continuity do not imply convexity. ## Executable arithmetic and error bounds The natural implementation uses scale W>0, floors utilization, rounds the whole borrow numerator upward once, and floors supply once after multiplying all factors. The checked quote requires the exact Reserves<=Cash condition or zero debt, so a small excess U>1 hidden by utilization rounding still cannot pass admission. Successful quotes satisfy uInt<=W and supplyInt<=borrowInt. The rational reference uses the same encoded parameters divided by W. Utilization error is in [0,1/W); borrow rounding at encoded utilization is in [0,1/W). Net supply error at encoded utilization is two-sided, strictly between -1/W and 1/W, because borrow has already been rounded upward. With L=max(s1,s2), full borrow error against unquantized U lies in [-L/W,1/W). With S=(1-rho)*(rb(1)+L), full supply error lies strictly between -(S/W+1/W) and 1/W on admitted unit-interval states. These bounds include utilization quantization and cover intervals crossing the kink. [Detailed contract](../knowledge/01_defi/DeFiKinkedInterestRateUtilization.md) ## Scope and open obligations Rate order is an instantaneous algebraic property at fixed parameters. It does not prove overall pool solvency, default freedom, withdrawable liquidity or absence of liquidity crunches. Compound interest, debt indices, bad-debt writeoffs, reserve dynamics and governance changes require separate transition models. This is a reference rounding policy, not bit-exact refinement of deployed Compound or Aave contracts. Natural arithmetic is unbounded; machine overflow, gas, oracle truth, market execution and MEV are outside scope. Parameter quantization from an arbitrary real curve is not covered by the error bounds above. Tendermint Amnesia attribution still requires a complete observation or justification contract. Bulletproofs knowledge extraction, computational binding, HVZK, Fiat-Shamir/ROM and external Python, JSON/SHA-256 and SimLab correspondence remain open. ## Measured validation | Check | Measured public result | |---|---:| | Direct MasterSuite imports | 167 | | Lean files under Verification/ | 207 top-level + 2 auxiliary = 209 | | Root-inclusive sources identical to the reference snapshot | 210 | | Clean strict build | 3848 jobs; no warnings | | Complete verifier audit | 23749 declarations; no violations | | Pinned independent full audit | 23749 declarations; no violations | | Public live tests | 112; no failures or skips; 332.383s | | New kernel regression declarations | 46 | | New executable scenarios | 25 | Both full audits allow only propext, Classical.choice and Quot.sound. The registry audit and knowledge catalog passed separately. The independent auditor is pinned to source commit 46024e005996495c65ef609368e11ab39c4222e3 and executable SHA-256 8a045cbabfa078df6577541d9768d5971ce573915f40e60806f32a50c3ffc254. The project toolchain is v4.33.1; the auditor source toolchain file records v4.32.0-rc1. This is an independent axiom auditor, not a second proof kernel. The project build directory started empty; pinned dependency caches were copied into this isolated checkout. All 210 proof sources match reference snapshot 4b9e1579c5decc412d4d0ec48ca90e7726f85d08 byte-for-byte. Only English cards were exported; earlier public reports remain available. ```bash lake build --wfail python3 tools/verifier_skill.py verify-all LEAN_VERIFIER_LIVE_TESTS=1 python3 -m unittest discover -s tests -p "test_*.py" lake env lean -DwarningAsError=true Verification/AxiomAudit.lean lake env /tmp/simon-axiom-audit-source/.lake/build/bin/axiom-audit --root Verification --allow propext,Classical.choice,Quot.sound --json python3 scripts/check_knowledge.py ```