Measure first. Say what was measured.

The questions

Runs

A run on a claim redeemable at face value is an equilibrium of expectations: holders who expect others to redeem redeem first (Douglas W. Diamond and Philip H. Dybvig (1983)), and the probability of a run can be tied to fundamentals rather than to a coin toss between equilibria (Itay Goldstein and Ady Pauzner (2005)). MONARK does not model the equilibrium. It measures the speed of the redemption flow on chain, once a day, for one population.

What the measurement found, as the served gate states it for the committed class stable-run-velocity-24h: a committed stable-run velocity calibration for the USDe synthetic-dollar-whitelisted-redeem population; over calm-window redemption flow; the calibration is measured non-stationary across half-years; no coverage is measured; for any other population, no stable-run velocity calibration is committed for this population; the gate abstains (under_calib). Because the calibration is not exchangeable across time, the served sentence rests on this result:

Until that departure is estimated, no coverage is measured, and the served sentence says so.

The tracker has stepped 6 times in the committed capture, and its printed bound is projected to reach its target at T = 1789. See the Narabi page.

Liquidations

Liquidations are where a price becomes a loss, and the scale of the problem is measured. The figures below are those of the one primary empirical study the site carries, read from the committed file of sourced figures, each with its qualifier and its page:

1 more figure of the same file is not shown here: the qualifier names a venue the site describes generically.

In a network of obligations, the payments that clear are a fixed point (Larry Eisenberg and Thomas H. Noe (2001)); with default costs the fixed point need not be unique (L. C. G. Rogers and L. A. M. Veraart (2013)), and conditions for a single equilibrium under liquidation costs are known (Hamed Amini, Damir Filipovic and Andreea Minca (2016)). Forced sales into an inelastic market move the price a book is marked at (Rodrigo Cifuentes, Gianluigi Ferrucci and Hyun Song Shin (2005)). And what a stress test flags is not what gets liquidated:

What MONARK measured, on one recorded episode, weth-2025-09-22: Population: 9034 mono-collateral WETH accounts (meets the floor of 100); class A calibration points: 205. Accounts liquidated by more than one call, against the pre-registered threshold of a share at most 5/100: liquidated class-A accounts 2 of 47 → yes; all liquidated accounts 4 of 72 → no. Liquidated amount in class A: 1847180.51809934; of it, after the first call: 48189.04377058; deficit reported apart: 0.00000000. The full report, stratum by stratum, is on the course page.

The oracle path

A liquidation is triggered by the price the protocol reads, not by the market’s price. Ukemi records the oracle path from chain events and checks, against a bound fixed before the run, that the values the protocol read at the liquidation blocks belong to that path. On the course episode:

Oracle values the protocol read at the liquidation call blocks and at the block before each: 58 values, 57 in the on-chain update series of 42 updates, 1 equal to the anchor, 0 outside. Measured maximum lag 2 updates (pre-registered bound 3 updates; 0 over it) → yes. At the call blocks themselves: 35 values, 35 in the series, maximum lag 1. Update blocks monotone: yes; phase change: no.

Anchor: block 23414669, price 4448.00000000, taken from the last oracle update before the reference block; lowest value read 4120.72000000.

Declared bias: these values are sampled at and just before liquidation blocks only.

Calibration

Every region the gate states is conformal. The method gives a finite-sample, marginal statement under exchangeability, with no assumption on the model:

Coverage conditional on one input cannot be promised in general (Vladimir Vovk (2012)), which is why the gate keeps one region per category fixed in advance:

A population with too few calibration points abstains; the floor follows the analysis of prediction sets for grouped data (Robin Dunn, Larry Wasserman and Aaditya Ramdas (2022)). Abstention itself is an old idea: the tradeoff between errors and rejections (C. K. Chow (1970)). And the served gate says plainly when a class is not a measurement: btc-dir-15m is served as declared synthetic — a plumbing fixture, not a measured predictor.

Off-hours gaps

Tokens that track U.S. equities trade while their market is closed, and empirical work measures how far and how often they stray from the last close, overnight and over the weekend (Cong, Landsman, Rabetti, Zhang and Zhao (2025)). The weekend effect in stock returns has an older literature, cited here by name only until the work is obtained (Kenneth R. French (1980)).

MONARK Bell publishes the gap session by session. In its latest record, a gap is computed for 9 of 9 session rows, and each row counts whether the gap exceeded each threshold named by the served rows (1 percent, 2 percent, 5 percent). A measurement over a full collection window is not published yet; until it is, no share per regime is claimed here. See the Bell page.

Open questions

• Does the tracker’s bound become informative at the projected T, and does the drift criterion fire before it?
• Is the next liquidation episode exchangeable with the one the course calibrated on? The distance to a new event is named, never estimated away.
• Why did the oracle serve its price with a delay on the design episode? On the course episode the lag is measured against a pre-registered bound, not explained.
• How often does the off-hours gap cross each threshold, per regime, over a full collection window?
• Does a third party call the gate without being pushed? Paid demand for a gate like this is not demonstrated, and we say so.

Bibliography

Every work the documentation cites, with the reading level the project reached. A work not yet obtained is cited by name only, and nothing is claimed from its content.

• Anastasios N. Angelopoulos and Stephen Bates (2023) Conformal Prediction: A Gentle Introduction. Foundations and Trends in Machine Learning, vol. 16, no. 4, pp. 494-591, arXiv:2107.07511. read in part Used for the split-conformal quantile and the meaning of marginal coverage.
• Vladimir Vovk (2012) Conditional validity of inductive conformal predictors. Proceedings of the Asian Conference on Machine Learning, PMLR vol. 25, pp. 475-490. read in part Used for why coverage conditional on one input cannot be promised in general, and why a region is kept per category.
• Vladimir Vovk, David Lindsay, Ilia Nouretdinov and Alex Gammerman (2003) title not reproduced: the title names the method with a word the site does not print; the paper is the working paper of the series below. On-line Compression Modelling Project (New Series), Working Paper 4. read in full Used for one region per category of a taxonomy fixed in advance, with validity per category and never per case.
• Anastasios N. Angelopoulos, Rina Foygel Barber and Stephen Bates (2024) Online conformal prediction with decaying step sizes. Proceedings of the International Conference on Machine Learning (ICML), PMLR vol. 235, pp. 1616-1630, arXiv:2402.01139. read in part Used for the quantile tracker Narabi steps each day and the long-run bound it prints.
• Rina Foygel Barber, Emmanuel J. Candes, Aaditya Ramdas and Ryan J. Tibshirani (2023) Conformal Prediction Beyond Exchangeability. The Annals of Statistics, vol. 51, no. 2, pp. 816-845, doi:10.1214/23-AOS2276. read in part Used for the sentence served with a class whose calibration is not exchangeable: no coverage is measured.
• Isaac Gibbs and Emmanuel J. Candes (2021) Adaptive Conformal Inference Under Distribution Shift. Advances in Neural Information Processing Systems (NeurIPS), arXiv:2106.00170. read in part Used for the online update the tracker descends from.
• Stephen Bates, Anastasios N. Angelopoulos, Lihua Lei, Jitendra Malik and Michael I. Jordan (2021) Distribution-Free, Risk-Controlling Prediction Sets. Journal of the ACM, vol. 68, no. 6, article 43, doi:10.1145/3478535. read in part Used for the origin of the budget vocabulary only; the served budget is not that mechanism.
• Robin Dunn, Larry Wasserman and Aaditya Ramdas (2022) Distribution-Free Prediction Sets for Two-Layer Hierarchical Models. Journal of the American Statistical Association, doi:10.1080/01621459.2022.2060112. read in part Used for the minimum number of calibration points below which a population abstains.
• C. K. Chow (1970) On Optimum Recognition Error and Reject Tradeoff. IEEE Transactions on Information Theory, vol. 16, no. 1, pp. 41-46, doi:10.1109/TIT.1970.1054406. read in part Used for abstention as a first-class answer: trading errors for rejections.
• Larry Eisenberg and Thomas H. Noe (2001) Systemic Risk in Financial Systems. Management Science, vol. 47, no. 2, pp. 236-249, doi:10.1287/mnsc.47.2.236.9835. read in part Used for the clearing fixed point of a network of obligations.
• L. C. G. Rogers and L. A. M. Veraart (2013) Failure and Rescue in an Interbank Network. Management Science, vol. 59, no. 4, pp. 882-898, doi:10.1287/mnsc.1120.1569. read in part Used for default costs, under which the fixed point need not be unique.
• Hamed Amini, Damir Filipovic and Andreea Minca (2016) Uniqueness of equilibrium in a payment system with liquidation costs. Operations Research Letters, vol. 44, no. 1, pp. 1-5, doi:10.1016/j.orl.2015.10.005. read in part Used for when liquidation costs still leave a single equilibrium.
• Kaihua Qin, Liyi Zhou, Pablo Gamito, Philipp Jovanovic and Arthur Gervais (2021) An Empirical Study of DeFi Liquidations: Incentives, Risks, and Instabilities. ACM Internet Measurement Conference (IMC), doi:10.1145/3487552.3487811. read in part Used for the scale of liquidations and why an attested price matters.
• Daniel Gatto (2026) title not reproduced: the title names the lending protocol studied, which the site describes generically. SSRN working paper 7157638. read in full Used for eligible debt is not executed volume, and executed volume is not protocol bad debt.
• Rodrigo Cifuentes, Gianluigi Ferrucci and Hyun Song Shin (2005) Liquidity risk and contagion. Bank of England Working Paper 264. read in full Used for how forced sales into an inelastic market move the price a book is marked at.
• Douglas W. Diamond and Philip H. Dybvig (1983) Bank Runs, Deposit Insurance, and Liquidity. Journal of Political Economy, vol. 91, no. 3, pp. 401-419, doi:10.1086/261155. read in part Used for the object Narabi senses: a run on a claim redeemable at face value.
• Itay Goldstein and Ady Pauzner (2005) Demand-Deposit Contracts and the Probability of Bank Runs. The Journal of Finance, vol. 60, no. 3, pp. 1293-1327, doi:10.1111/j.1540-6261.2005.00762.x. read in part Used for runs tied to fundamentals rather than to a coin toss between equilibria.
• Cong, Landsman, Rabetti, Zhang and Zhao (2025) Tokenized Stocks. SSRN working paper 5937314. read in full Used for overnight and weekend deviations of tokenized equities from the last close.
• Kenneth R. French (1980) Stock Returns and the Weekend Effect. Journal of Financial Economics, vol. 8, no. 1, pp. 55-69, doi:10.1016/0304-405X(80)90021-5. not yet obtained Used for the historical antecedent of the weekend gap, cited as a name only until obtained.
• National Institute of Standards and Technology (2015) Secure Hash Standard (SHS). Federal Information Processing Standards Publication 180-4. read in part Used for the hash every record, line and digest is recomputed with.
• S. Josefsson and I. Liusvaara (2017) Edwards-Curve Digital Signature Algorithm (EdDSA). IETF RFC 8032. read in part Used for the signature over each line of the Bell timeline.
• C. Bormann and P. Hoffman (2020) Concise Binary Object Representation (CBOR). IETF RFC 8949. read in part Used for deterministic canonical bytes.
• Steve Ellis, Ari Juels and Sergey Nazarov (2017) ChainLink: A Decentralized Oracle Network. Whitepaper v1.0, 4 September 2017, https://research.chain.link/whitepaper-v1.pdf. read in part Used for faults correlated across data sources, one source copying another (Sections 4.1 and 5.3), and the research it proposed on mapping the independence of data sources.
• B. Laurie, E. Messeri and R. Stradling (2021) Certificate Transparency Version 2.0. IETF RFC 9162. read in part Used for an analogy only: an append-only, hash-chained log makes a rewrite detectable.
• Jason Milionis, Ciamac C. Moallemi, Tim Roughgarden and Anthony Lee Zhang (2022) Automated Market Making and Loss-Versus-Rebalancing. working paper, arXiv:2208.06046. read in part Used for what a liquidity pool loses to better-informed arbitrage, with fees tracked apart.
• Lawrence R. Glosten and Paul R. Milgrom (1985) Bid, Ask and Transaction Prices in a Specialist Market with Heterogeneously Informed Traders. Journal of Financial Economics, vol. 14, no. 1, pp. 71-100. read in part Used for informed flow and the spread it forces.
• David Easley, Marcos M. Lopez de Prado and Maureen O'Hara (2012) Flow Toxicity and Liquidity in a High-Frequency World. Review of Financial Studies, vol. 25, no. 5, pp. 1457-1493. read in full Used for a volume-synchronized measure of order-flow toxicity.
• Robert Almgren and Neil Chriss (2000) Optimal Execution of Portfolio Transactions. Journal of Risk, vol. 3, no. 2. read in part Used for the trade-off between the expected cost and the variance of an execution.
• Andre F. Perold (1988) The Implementation Shortfall: Paper versus Reality. The Journal of Portfolio Management, vol. 14, no. 3, pp. 4-9, doi:10.3905/jpm.1988.409150. not yet obtained Used for the yardstick of an execution receipt, cited as a name only until obtained.
• Marco Avellaneda and Sasha Stoikov (2008) High-frequency trading in a limit order book. Quantitative Finance, vol. 8, no. 3, pp. 217-224. read in full Used for how inventory risk shifts the reservation price and widens the spread.
• Charles R. Nelson and Andrew F. Siegel (1987) Parsimonious Modeling of Yield Curves. The Journal of Business (first as NBER Working Paper 1594). read in part Used for a curve across maturities with few parameters.
• Jerome H. Saltzer and Michael D. Schroeder (1975) The Protection of Information in Computer Systems. Proceedings of the IEEE, vol. 63, no. 9, pp. 1278-1308. read in part Used for least privilege and separation of privilege.
• Mark S. Miller (2006) Robust Composition: Towards a Unified Approach to Access Control and Concurrency Control. PhD dissertation, Johns Hopkins University. read in part Used for object capabilities: authority passed as a reference, never assumed.