Article: RWA-PoB: A Credential-Based Proof-of-Backing Framework for Tokenized U.S. Treasury Products
Date: 2026-08-26
Publisher: arXiv / Rischan Mafrur, Gun Gun Febrianza, Sean Foley
Score: ₿₿
Read time: 20-25 min
Summary:
- Overview: The paper proposes RWA-PoB, a credential-based backing framework for tokenized U.S. Treasury products. Five institutional roles (custodian, administrator, valuation agent, trustee or legal role, and independent verifier) sign one canonical EIP-712 snapshot carrying reserve, liability, liquidity and policy data. A Solidity controller checks signer authorisation, epoch progression, freshness, liability continuity and replay protection, then applies two metrics: the Backing Coverage Ratio for solvency and Redemption Liquidity Coverage for short-horizon redemption capacity. Evaluation uses USDY-calibrated liabilities of approximately USD 2.162 billion, 100 synthetic positions per scenario generated with seed 42, and 31 deterministic tests. The authors state that the framework authenticates institutional claims rather than proving off-chain asset existence.
- Eligibility Gated Reserves: Each reserve position carries custody, legal title, encumbrance and settlement-finality flags, and a position failing any one is zeroed before the haircut is applied. In the encumbered-assets scenario, gross reserves of approximately USD 2.486 billion against USD 2.162 billion of token obligations still produce a BCR of 96.3408% against the 105% threshold, blocking issuance. The simplified aggregate baseline cannot represent encumbrance by construction, so the divergence is arithmetic rather than empirical.
- Solvency Liquidity Split: BCR and RLC fail differently. A BCR below threshold blocks minting outright; an RLC below threshold only changes the redemption disposition and never extinguishes the investor's claim. With liquid assets cut to 2% of token value and a proposed redemption of 5%, post-request RLC falls to 39.9999% while BCR holds at 114.1209%, producing a queue rather than a rejection. Solvent-but-illiquid is treated as a scheduling problem, not a solvency event.
- Burn To Pending: A redemption request burns tokens and moves the same USD value from token liabilities into pending-redemption liabilities, leaving total liabilities unchanged until a settlement-role account confirms payment. Atomic issuance consumed 132,573 gas against 153,006 for separate advisory authorisation and independent minting, 13.35% lower. Snapshot publication cost 362,241 gas versus 94,689 for the aggregate baseline, with subsequent updates 5.16 to 5.23 times more expensive.
- Supply Gate Tradeoffs: Supply becomes a function of post-mint BCR rather than issuer discretion, but the same gate governs liveness: one of five required signers withholding a signature halts issuance and queues every redemption. The 105% and 100% thresholds are experimental placeholders with no calibration against redemption flow or valuation volatility, the prototype fixes redemption price at 1 so accumulating variable-NAV designs are untested, and confirmed settlement does not propagate into cash, eligible or gross reserves between snapshots.
Article: ZAPs: A Reward Attribution Framework for DeFi Ecosystems with Adversarial-Robust Scoring via Parallel Anomaly Ensemble Detection
Date: 2026-07-30
Publisher: arXiv / Girish G N, Ashutosh Sahoo, Ajay Bhat, Akshay SP, Gurukiran S, Parag Paul, Dhanashekar Kandaswamy
Score: ₿₿
Read time: 15-20 min
Summary:
- Overview: The paper presents ZAPs, a reward-attribution framework built by Zeru Finance for points programs and retroactive airdrops, where volume, transaction-count and wallet-count heuristics let sybil operators farm distributions at scale. It combines a percentile-anchored composite activity score, a nested cross-domain weight equal to a protocol's global volume share, and a four-layer detection stack of transaction gates, an anomaly ensemble, post-distribution memory and funding-graph sybil clustering, feeding graduated rather than binary penalties. Evaluation covers 1,073 labeled malicious wallets across 124,638 transactions, eight constructed scenarios, and self-reported live campaigns. The authors state that the individual components are established and that the contribution is their composition into an attribution mechanism.
- Niche Protocol Bound: Attribution runs per wallet-protocol pair, weighted by protocol share within sector times sector share within ecosystem. The product telescopes to global volume share, bounding extractable reward at that share regardless of how cheaply local dominance is bought. A protocol holding 19.4% of the DEX sector, which carries 85.7% of volume, weights 0.166; one holding 71.3% of staking weights 0.0034. The protocol with over half of all wallets received 1.3% of rewards.
- Detector Polarity Inversion: The anomaly ensemble pairs a one-class reconstruction model trained on malicious wallets only with an isolation forest. Fit on the pooled population the forest scores 0.250 ROC-AUC, worse than chance, because sparse malicious wallets with a median 21 transactions define the dense region against benign wallets at a median 499. Fit on benign data it reaches 0.638, lifting the ensemble to 0.923 from 0.891 at a fusion weight of 0.90.
- Asymmetric Capture Reduction: Defended scoring cuts adversarial capture 90% for the spam bot, 89% for the diversity farmer, 43% for the flash-loan exploiter and 30% for the sybil operator, while legitimate controls move 1-8%. Live campaigns reported 56% less sybil allocation and 50% less post-distribution sell pressure. Every effectiveness figure rests on proxy or self-reported evidence, with exploit labels standing in for farmers, constructed rather than observed scenarios, and campaigns measured against undefined baselines, and Table 1's $300B aggregate volume does not reconcile with the $41.2B summed across Table 3's five sectors.
- Points Program Levers: Penalties activate only on multiple joint signals, leaving the flagged-but-unpenalized Suspicious tier, holding 84% of distributed rewards, untouched, because power users and market makers trip partial bot signals. Binary exclusion would confiscate that value. The mixing coefficients, fusion weight, percentile anchor and penalty breakpoints are withheld, so the scheme must be recalibrated rather than adopted.
Article: Methodology for Modelling Token Economies and Performing Event Impact Analysis with DeTEcT
Date: 2026-08-03
Publisher: arXiv / Rem Sadykhov, Geoffrey Goodell, Philip Treleaven
Score: ₿₿
Read time: 25-30 min
Summary:
- Overview: The paper formalises how to apply the DeTEcT agent-based framework to a real token economy and adds a procedure for measuring whether an event changed it. Agents are grouped into categories whose wealth flows are governed by an antisymmetric matrix of interaction rates, and backward propagation solves for those rates given a starting and an ending wealth distribution. Bitcoin is the case study, using CoinDesk daily wealth-distribution data from January 2009 to August 2025 split into 10 Gini-like buckets plus a Control Mechanism category holding unissued supply. Five economy-related BIPs, numbered 32, 42, 50, 141 and 341, are tested over six-month windows either side of their creation dates.
- Optimiser Cost Curve: Gradient descent with mean log error drove the final log-loss to 0.000002 and returned the same 55 interaction rates to five significant figures across 100, 500 and 1,000 iteration caps, at a cost of 200,417 objective function calls and 34,649 seconds. Nelder-Mead completed the same configuration in 4.5 minutes and stalled at a log-loss of 6.43, its random-jump variant reaching 5.38 once and worsening at 1,000 iterations. Gradient descent with mean absolute error crashed during numerical integration on every run.
- Non-Unique Rate Solutions: The recovered rates are one fit among many. Every optimisation is initialised at a matrix of zeros, so gradient descent returns whichever solution sits nearest the origin in a 55-dimensional parameter space, and the difference metrics built on those rates track movement within a chosen local solution rather than in the economy itself. Rotations are fixed at zero throughout because Bitcoin addresses are rarely reused, leaving all three rotation-difference metrics unexercised by the case study.
- One Standard Deviation Bounds: An event counts as significant when its rate difference falls outside the mean plus or minus one standard deviation of daily contextual simulations spanning 11-02-2012 to 19-01-2020, a band that under normality leaves roughly a third of ordinary days outside it; 5 of 55 event-by-category cells were flagged. Total difference for all five BIPs, ranging 0.041913 to 0.092249, sat inside the (0.026826, 0.132846) band. Both the six-month window and the benchmark results come from the authors' earlier Granger-causality paper on the same dataset, so the agreement is partly built in.
- Issuance Centralisation Finding: Interaction rates involving the Control Mechanism are the largest by magnitude and almost entirely outgoing, while the smallest rates sit between the four buckets below 1 BTC, placing issuance rather than peer trade at the centre of wealth circulation. BIP 32 and BIP 341, classed purely tokenomic, moved only the Control Mechanism; BIP 42 and BIP 141, classed monetary-like, moved the 0.1-1 and 1-10 BTC buckets. The deflation loop attached to that pattern is asserted, not tested.
Article: Redefining Stablecoins from Nominal to Real Value: A Maximum Likelihood Approach
Date: 2026-08-31
Publisher: arXiv / Tomonori Kanno, Kensuke Ito, Yushi Yoshimura, Kyohei Shibano
Score: ₿₿
Read time: 18-22 min
Summary:
- Overview: The paper argues that fiat-pegged stablecoins cannot be stable because the unit of account itself moves, and proposes pegging instead to the Maximum Likelihood Value (MLV): the most probable configuration of unobservable real-value log returns consistent with observed prices. With N assets only N-1 independent prices exist, leaving one degree of freedom, which an elliptical distribution assumption closes in analytical form. Two experiments follow, a synthetic 500-asset estimation study and an S&P 500 dividend-portfolio backtest running June 2019 to January 2026. MLV and its portfolio application are under patent application by VLUE Inc. in Japan and internationally.
- Identification By Assumption: Prices decompose as p_ij = v_j/v_i, an underdetermined system. Assumption 1 imposes an elliptical distribution on real log returns and Tyler's scatter matrix estimates the covariance, yielding a closed-form estimate that is a precision-weighted average of cross-sectional returns. The resulting unit is therefore fixed entirely by which assets are supplied to it. The locality and arbitrariness charged against CPI baskets returns as universe selection, unaddressed.
- Zero Real Return Theorem: Theorem 3.3 establishes the MLV as the numeraire under which the global minimum-variance portfolio earns exactly zero. This is an exact characterisation, not an empirical result: maximising the likelihood and defining away the minimum-variance portfolio's return are the same operation. Stability holds by construction relative to the chosen asset set, with no link to consumption, goods prices or purchasing power in the economic sense.
- Simulation Then One Backtest: Experiment 1 is synthetic throughout, with multivariate t returns drawn from a random correlation matrix, N set to 500 and ten runs, RMSE flattening past n≈200, and real-time feasibility inferred from O(n^3) scaling rather than timed on hardware. Experiment 2 spans one equity universe over six and a half years with no significance testing: USD-space return rises from 11.7% to 15.6%, Sharpe from 0.554 to 0.699, turnover falls from 0.529 to 0.133, and maximum drawdown deepens from -37.6% to -46.6%.
- Issuance Scheme Unbuilt: Issuance is a diagram. Users deposit assets for MLV-denominated coins, an operator computes and publishes the MLV and rebalances collateral every five weeks, and returns split among portfolio, users and operator. Redemption pays the then-current portfolio, so holders absorb composition risk and no redemption arbitrage anchors the peg. The split ratio, operator incentive, control of the published MLV and decentralisation are all deferred to future work.
Article: zLend: A Dual-Scope Cash-Flow Reconstruction Framework for On-Chain Credit Underwriting
Date: 2026-08-17
Publisher: arXiv / Girish G N, Ashutosh Sahoo, Akshay SP, Gurukiran S, Dhanashekar Kandaswamy
Score: ₿₿
Read time: 18-22 min
Summary:
- Overview: zLend reconstructs a wallet's daily balance from raw token transfers twice, once restricted to a fixed 15-symbol stablecoin basket and once over all fungible transfers, then derives repayment-capacity signals from each series: coverage against a $100 reference loan, inflow regularity, recurring-counterparty income cadence, maximum drawdown and recovery, and single-day outflow concentration. Divergence between the two views becomes a graded liquidity or flow mismatch flag. The stablecoin scope alone sets a four-tier primary signal; total wealth supplies a context signal that qualifies but never overrides it. Evidence is six synthetic fixture wallets, with no borrower outcomes or default data anywhere in the paper.
- Liquid Versus Aggregate: One fixture wallet climbs from $50,000 to $270,030 on the total-wealth series while its stablecoin balance never exceeds $40, a ratio near 6,668x with stablecoin inflow at 0% of trailing total, firing both mismatch flags at high severity. Neither series is an observed balance: cumulative net flow is shifted by the smallest offset keeping it non-negative, and the total-wealth view sums USD transfer values rather than marking holdings to market.
- Loan Size Governs: Coverage is defined against a reference loan L, fixed at $100 in deployment, and four of six wallets change tier across L in [$10, $25,000]. The thin-stablecoin wallet rates strong at $10 and weak at $50 because its balance sits near $30-40. Only the drawdown- and depletion-excluded wallets stay invariant, those criteria being scale-free. L functions as a product-scope declaration rather than a tuning constant.
- Verified Not Validated: The Python-to-TypeScript port is checked by golden master to 1e-9, replicating Kahan summation, population-denominator variance, round-half-to-even at ten decimals, and NaN propagation to the API instead of coercion to zero; an independent reimplementation agrees on 78 of 78 assertions. Nothing tests whether coverage, drawdown or income cadence predicts repayment. The balance-ratio threshold yields zero flag transitions from 1.5x to 10x because flagged wallets sit at 6,668x or the numerical floor.
- Wallet Quality Primitive: No token supply, emissions, governance or staking appears. The transferable structure is the conjunctive tier rule, where coverage and maximum drawdown bind on disjoint fixtures (one wallet at coverage 1.00 with drawdown 0.95, another at coverage 0.00 with drawdown 0.25), so neither statistic proxies the other, and any scoring system collapsing them into a single weighted index destroys that separation. The same split applies to airdrop weighting and reputation-gated allocation.
