Navigation

Axis-2 · Cascade-Free Harvest — Iteration 5 15 NEW (+2 OOS)

Family swept: crypto-spot liquidity-quality & integrity + volume-clock bars · campaign regime-invariant-orthogonality · append-only

15new (spot-legal)
+2flagged out-of-scope
0dupes
101running total (spot)

The integrity & clock axis — crypto-native. Every prior measure assumes the tape is real. This iteration harvests the tools that test that assumption and the tools that resample the tape by activity. Integrity: the three Cong et al. wash-trading tests (Benford first/second-digit, trade-size roundness-clustering, power-law tail), the Deng–Zhou liquidity jump/diffusion decomposition, pump-and-dump detectors (Kamps–Kleinberg, Xu–Livshits, La Morgia), and CUSUM/Bai–Perron volume structural-breaks — anchored by the Nigrini forensic baseline. Clocks: the Volume Clock and the AFML information-driven bars (tick/volume/dollar + imbalance + run) — the liquidity-adaptive alternatives to this project's open-deviation (price-threshold) clock.

Provenance was load-bearing this iteration. The Aloosh–Li (2024) validation (H77) is the authority: ground-truthed against the Mt.Gox trader-ID leak, it finds Benford / size-clustering / lognormal / structural-break tests useful but the power-law-tail test (H76) misleading — so H76 is WATCH-flagged. On FOSS: mlfinlab is proprietary (Nov-2021 commercial click-through, not open source) → use mlfinpy; xujiahuayz/pumpdump carries no license → not FOSS-usable. Two on-chain/L2 detectors are out-of-scope per principle #8.

New candidates (iter 5)

#NameFamilyMeasures / samplesCascade / liquidity mappingSpot?ParamStable-IDFOSS
H73Benford first-digit test (Cong et al.)fake-volume testfirst-significant-digit distribution of trade sizes vs Benford (χ²/MAD)flags fabricated liquidity; authentic sizes obey Benford, wash trades deviate[T/O]PFCong, Li, Tang, Yang 2023, Mgmt Sci — DOI 10.1287/mnsc.2021.02709; arXiv 2108.10984VER benfordslaw (Py, MIT); benford_py (Py, BSD-3); benford.analysis (R, GPL-3)
H74Benford second-digit testfake-volume testsecond-digit distribution (tighter vs near-uniform fabrication)same integrity flag; more sensitive to crude size-generation[T]PFCong et al. 2023 · parent H73VER as H73
H75Trade-size roundness / clustering (Cong et al.)trade-size-distributionexcess mass at round multiples (×10, integer BTC) vs authentic baselineflags synthetic order generation; the roundness-ratio is Cong et al.'s main wash estimator[T]DIMCong et al. 2023 (as H73)VER benford_py (roundness); hand-rolled
H76Power-law tail of trade sizestrade-size-distributionPareto-tail exponent α of large trade sizes (KS-optimal x_min)integrity flag — WATCH Aloosh-Li find this gives the OPPOSITE verdict vs ground truth[T]PPfit: Clauset, Shalizi, Newman 2009, SIAM Rev — DOI 10.1137/070710111; application Cong et al.VER powerlaw-devs/powerlaw (Py, MIT)
H77Aloosh–Li wash-trade validationwash-trading-detection (validation authority)ground-truth wash via trader-ID (Mt.Gox leak), then benchmarks the indirect teststhe authoritative verdict: Benford / clustering / lognormal / structural-break useful; power-law-tail misleadingdirect OOS verdict spot-usableAloosh & Li 2024, Mgmt Sci — DOI 10.1287/mnsc.2021.01448; SSRN 3362153UNV
H78Liquidity Jump / Liquidity Diffusion (Deng–Zhou)liquidity-quality-decompositionJump = ratio(regular ret / liq-adj ret); Diffusion = ratio(regular vol / liq-adj vol)decomposes liquidity into level vs variance of price-pressure; high diffusion = fragile/manipulated → exit-difficulty[T]PPDeng & Zhou 2025, arXiv 2411.05803 (succ. 2404.07222)UNV
H79The Volume Clock (Easley–LdP–O'Hara)volume-clocksamples one observation per fixed volume bucket (time subordinate to activity)the canonical liquidity-adaptive alternative to the open-deviation (price-threshold) clock[T]PPEasley, López de Prado, O'Hara 2012, JPM — DOI 10.3905/jpm.2012.39.1.019; SSRN 2034858VER mlfinpy; BlackArbsCEO
H80Tick / Volume / Dollar bars (AFML)information-driven-baremit a bar every N ticks / N units volume / N dollars tradedactivity clocks → returns closer to IID-Gaussian; dollar bars robust to price/supply drift[T]PPLópez de Prado, AFML (Wiley 2018) ch.2 — DOI 10.1002/9781119482086VER baobach/mlfinpy (Py, MIT); BlackArbsCEO (Py, MIT)
H81Imbalance Bars (TIB/VIB/DIB, AFML)information-driven-barsample when cumulative signed order-flow imbalance exceeds EWMA-expected thresholdstrong cascade relevance: fires on imbalance build-up → anticipates trend break; is_buyer_maker gives direction directly[T]PPLópez de Prado, AFML §2.3.2 · parent H80VER mlfinpy (imbalance_bars); BlackArbsCEO
H82Run Bars (TRB/VRB/DRB, AFML)information-driven-barsample when a run of same-direction ticks/volume/dollars exceeds expectationdetects persistent one-sided pressure (metaorder / directional cascade)[T]PPLópez de Prado, AFML §2.3.2 · parent H80VER mlfinpy; BlackArbsCEO
H83Kamps–Kleinberg P&D detectorpump-and-dump-detectionmoving-window thresholds on % price rise + volume spike define a P&D pointflags acute manipulation / exit-difficulty windows[O]MNKamps & Kleinberg 2018, Crime Science 7:18 — DOI 10.1186/s40163-018-0093-5UNV
H84Xu–Livshits pump predictorpump-and-dump-detectionML on pre-pump features (volume, price, mktcap, age) → pump likelihoodpredicts manipulation onset; core features spot-legal, mktcap adds metadata[O] corePPXu & Livshits 2019, USENIX Security — arXiv 1811.10109UNV xujiahuayz/pumpdump — NO LICENSE (not FOSS-usable)
H85La Morgia et al. "Doge of Wall Street"pump-and-dump-detectionreal-time ML (chart/volume-window) detects P&D ~25s after start, F1≈94.5%real-time cascade / exit-difficulty flag; trade-only features[T/O]PPLa Morgia et al. 2023, ACM TOIT — DOI 10.1145/3561300; arXiv 2105.00733UNV SystemsLab-Sapienza/pump-and-dump-dataset; Bayi-Hu (SIGMOD'23)
H86CUSUM / Bai–Perron volume structural-breakfake-volume / structural-breakchange-point detection on volume/trade-count seriesregime shifts in liquidity supply (fabrication onset/offset); Aloosh-Li confirm useful for wash[O]PPPage 1954, Biometrika 41:100-15 — DOI 10.1093/biomet/41.1-2.100; Bai & Perron 2003 — DOI 10.1002/jae.659VER deepcharles/ruptures (Py, BSD-2); strucchange (R, GPL-2|3)
H87Benford forensic baseline (Nigrini)fake-volume test (foundational)general digit-law forensic framework (MAD conformity; 1st/2nd/first-two digits)foundational integrity test under H73/H74; volume/qty digit conformity[T/O]PFNigrini 2012, Wiley (ISBN 978-1-118-15285-0); Benford 1938, PAPS 78(4):551 (JSTOR 984802)VER benford.analysis (R, GPL-3); benford_py (Py, BSD-3); benfordslaw (Py, MIT)
H88Victor–Weintraud DEX wash tradingwash-trading-detectiongraph SCC over on-chain account trade-graphon-chain contagionOOS needs on-chain metadataVictor & Weintraud 2021, WWW'21 — DOI 10.1145/3442381.3449824; arXiv 2102.07001OUT-OF-SCOPE
H89Amiram–Lyandres–Rabetti "Cooking the Order Books"wash-trading-detectionlower-bound wash from order-book / cross-exchange competitionintegrityOOS needs-L2 / cross-exchangeAmiram, Lyandres, Rabetti (SSRN WP; weak provenance)OUT-OF-SCOPE

FOSS anchors verified this iteration

Scope & provenance notes
Next angle → order-book depth / imbalance / resiliency (spot L2) (Kyle depth / book-slope, order-book imbalance OBI/QI, micro-price [Stoikov], queue imbalance [Cartea–Jaimungal], XLM cost-to-trade, Obizhaeva–Wang resiliency, LOB-shape [Bouchaud]). · Data SSoT: axis-2-cascade-free/CANDIDATE-CATALOG.md · back to loop hub