Arati Uday Kamat

Independent Researcher

Algorithmic DEX trading · counterfactual analysis · post-rejection sampling methodology

Working papers

Hour-Aware Adaptive Risk Management for Autonomous Memecoin Trading: A Multi-Layer Intelligence Framework

Kamat, A. U. (2026) · arXiv:2606.08232 · SSRN abstract 6564803 · Manuscript concept DOI 10.5281/zenodo.19670718 · Dataset concept DOI 10.5281/zenodo.20043301

Introduces a multi-layer intelligence framework for autonomous memecoin trading that incorporates hour-aware adaptive risk thresholds. Analyzes historical trade outcomes against UTC hour-of-day to identify high-variance time windows and dynamically adjust position entry, stop-loss, and filter strictness. Backtests on Solana DEX data demonstrate reduced drawdown relative to static-threshold baselines.

arXiv ↗ SSRN ↗ Manuscript concept DOI ↗ Dataset concept DOI ↗

Post-Rejection Follow-up Sampling: A Methodology for Counterfactual Outcome Measurement in Algorithmic DEX Trading

Kamat, A. U. (2026) · arXiv:2606.08228 · SSRN abstract 6607301 · Manuscript concept DOI 10.5281/zenodo.19671656 · Dataset concept DOI 10.5281/zenodo.20043515

Introduces Post-Rejection Follow-up Sampling (PRFS), a methodology for measuring what would have happened had a rejected trade been executed. By tracking forward price trajectories of rejected tokens over a fixed window, PRFS enables quantitative filter-quality evaluation. Applied to live Solana DEX trading data, the method identifies filters that reject profitable opportunities at higher rates than unprofitable ones, surfacing pruning candidates.

arXiv ↗ SSRN ↗ Manuscript concept DOI ↗ Dataset concept DOI ↗ Code (GitHub) ↗

Outcome-Classified Precision Auditing of Filter Rules in Algorithmic DEX Trading: Evidence from 2,400 Rejection Events

Kamat, A. U. (2026) · SSRN abstract 6638259 · Manuscript concept DOI 10.5281/zenodo.19720040 · Dataset concept DOI 10.5281/zenodo.19987695

Reports an at-scale empirical precision audit of eight filter rules operating in a live multi-strategy DEX trading fleet. Classifies approximately 2,400 unique rejection events over a fourteen-day operational window using an outcome schema extended from prior work (PRFS). Introduces the early-death classification: a rejected token's disappearance from the price oracle within sixty minutes is treated as an implicit positive save signal, justified by a sharply bimodal age distribution of single-sample events. Under this refinement, every active filter rule shows a net-positive precision verdict with an aggregate save-to-miss ratio of approximately fifteen to one; the conservative alternative (excluding early-death) yields approximately four to one.

SSRN ↗ Manuscript concept DOI ↗ Dataset concept DOI ↗

RED-2400: A Public Benchmark of Algorithmically-Rejected Trading Events with Outcome Labels

Kamat, A. U. (2026) · arXiv 2605.12151 [q-fin.TR] · SSRN abstract 6702198 · Multi-platform deposit: Zenodo + Kaggle + IEEE DataPort

First public benchmark dataset of algorithmically-rejected DEX trading events with linked post-rejection outcome trajectories. 6,659 rejection events, 169,122 outcome observations, 1,836 graveyard snapshots over the window 2026-04-10 to 2026-05-02 UTC. CC-BY-4.0 licensed. Deposited across five platforms (arXiv, SSRN, Zenodo, Kaggle, IEEE DataPort) for maximum discoverability and replication. RED-2400 is the first window in a planned dataset series; subsequent windows extend the time horizon and enable regime-stratified analysis.

arXiv ↗ SSRN ↗ Zenodo dataset concept DOI ↗ IEEE DataPort DOI ↗ Kaggle ↗ GitHub reader ↗

Pump.fun Graduation Regime Windows: Survival Analysis of 832,941 Token Launches and the Social-Presence Effect

Kamat, A. U. (2026) · arXiv:2607.02823 (v3 full-paper Replace 2026-08-13) · SSRN abstract 6915560 (CC-BY-4.0) · Dataset concept DOI 10.5281/zenodo.20633486 (always resolves to the latest version, currently the v1.5 corrigendum published 2026-09-06 at 10.5281/zenodo.22286914)

Corrigendum v1.5 (2026-09-06): Supersedes the v1.4 corrigendum. Version 1.5 clarifies the coverage-window framing and updates the reference bibliography; the core empirical findings from v1.4 stand unchanged. Corrigendum v1.4 (2026-08-13): Supersedes the v1.3 corrigendum. The v1.3 ceiling-crossing sensitivity analysis is withdrawn following direct on-chain verification against pump.fun's live bonding-curve API: 0 of 100 sampled mints from the candidate late-graduation set showed evidence of graduation. The market-capitalisation field used to identify candidate mints reflects virtual bonding-curve reserves (which move with speculative price action), not real reserves that determine graduation. Withdrawn as of v1.4: the 0.333 percent lower-bound estimate, the 9.5x-10x Telegram-lift range, and the Cox concordance stress test. The v1.3 coverage-window relabeling itself is unaffected and stands: the headline 0.198 percent rate should be read as a lower-bound fast-regime rate. True 24-hour rate remains unknown from this dataset; v2 re-collection planned.

Kaplan-Meier and Cox proportional-hazards survival analysis of 832,941 Solana pump.fun token launches, observed continuously between 2026-05-08 and 2026-06-10. The pooled graduation rate is 0.198 percent (Wilson 95 percent CI [0.189 percent, 0.208 percent]; steady-state 0.207 percent), which should be read as a fast-regime lower bound on the true 24-hour rate per the coverage-window corrigendum above. Comparing this figure to Marino et al. (2026) 0.63 percent for September-October 2025 gives a ratio of 3.18x; because the reported figure is itself a lower bound, this ratio is an upper bound on the true decline, and the true decline is smaller by an amount this dataset cannot establish. Within the fast-regime window, launches advertising a Telegram channel graduate at 1.485 percent versus 0.166 percent without (8.94x lift; Cox HR 5.40, 95 percent CI [4.73, 6.17]), likely itself a lower bound on the true 24-hour effect. Top market-cap quartile graduates at 0.634 percent, close to though not conclusively matching the Marino et al. pooled rate. Cox concordance 0.858. A subsequent attempt to recover missed late graduations was tested against pump.fun's live API and did not hold up (0 of 100 sampled mints); the analysis is reported and withdrawn in full (see corrigendum v1.4 above). Companion dataset RED-PUMP-2026-v1 released on Zenodo under CC-BY-4.0.

arXiv ↗ SSRN ↗ Zenodo dataset concept DOI ↗

Coordinated Sniper Cohorts on Pump.fun: Detection of 1,012 Persistent Wallet Rings and a Contamination-Adjusted Estimate of Coordination-Specific First-Hour Buyer-Flow Lift

Kamat, A. U. (2026) · arXiv:2607.02795 (v3 revision 2026-08-03) · Dataset concept DOI 10.5281/zenodo.20978741 · Companion dataset RED-COHORT-2026 · USPTO Provisional 64/099,108 covers detection methodology

Public reproducibility-grade dataset of 1,012 persistent sniper-cohort detections on Solana pump.fun bonding-curve marketplace, derived from 1,578,333 buyer events across 166,098 token launches observed 2026-06-11 to 2026-06-25 (13.38 days). Two-stage detection pipeline: intra-launch first-buyer-window extraction, then cross-launch persistent-cohort surfacing via union-find on co-occurrence graphs. Under a contamination-adjusted estimator that excludes cohort wallets' own events from the outcome, 1:1 nearest-neighbour propensity-score matching (0.2-SD caliper on ten launch-quality covariates) yields a first-30-minute buyer-count lift of +16.1% (95% CI [+13.0%, +19.4%]) on 5,419 matched pairs, with a first-30-minute SOL-inflow lift of +6.3% (95% CI [−0.5%, +15.1%]) not distinguishable from zero at 95% confidence. The naive contaminated same-universe pooled contrast on this corpus is +130.9%; approximately half of that is arithmetic contamination from cohort wallets' own events being counted in the outcome, and most of the remainder is absorbed by propensity-score matching. A 100-seed activity-matched placebo estimator produces median +189.6% (p5-p95 [+166.7%, +217.9%]) and is retained as a bias diagnostic rather than a falsification test. Interpretation-honest release of the detection catalogue enables downstream causal-identification research.

arXiv ↗ Zenodo dataset concept DOI ↗

Counterfactual Rejection Filter Ablation: An Empirical Application of Post-Rejection Follow-up Sampling to Solana Memecoin Trading

Kamat, A. U. (2026) · Dataset concept DOI 10.5281/zenodo.21149175 · Extends PRFS methodology (Paper 2). SSRN submission (abstract 7045818) marked REMOVED by SSRN 2026-07-20; paper and companion dataset remain publicly available on Zenodo.

Applies the Post-Rejection Follow-up Sampling framework to eight active filter rules operating on a live Solana DEX trading fleet. Systematic ablation shows every filter rejects a median-loser distribution, validating the aggregate save-to-miss ratios reported in prior work. Documents 158 counterfactual >10x winners across the observation window, with dominant reject-reason failure modes identified for downtrend_24h, marketcap_out_of_range, and multi_tf_fade rules. Naive outcome measurement systematically inverts these findings, motivating counterfactual-first evaluation for algorithmic decision systems.

Zenodo dataset concept DOI ↗

A Multi-Venue Solana/DeFi Microstructure Data Corpus: The RED-2400 Family v2

Kamat, A. U. (2026) · SSRN abstract 7062579 · Corpus dataset paper covering five deposited datasets: RED-ORACLE, RED-ARB, RED-LIQ, RED-BASIS, RED-BRIDGE

Corpus-level dataset paper describing the RED-2400 family v2 of Solana / DeFi microstructure benchmarks. Five deposited datasets on Zenodo under CC-BY-4.0: RED-ORACLE-2026-v1 (Pyth oracle staleness for BTC/ETH/SOL), RED-ARB-2026-v1 (328,186 Solana CEX-vs-DEX spread observations across 32 DEX venues), RED-LIQ-2026-v1 (18,750 Ethereum Aave V3 liquidations + 4-chain lending-utilization panel), RED-BASIS-2026-v1 (BTC/ETH/SOL spot-perpetual basis on OKX), RED-BRIDGE-2026-v1 (360,714 Wormhole cross-chain messages). All five span a common 57.4-day 2026 window (2026-05-08 to 2026-07-05). Standardized schema conventions, cross-dataset lineage, and reproducibility manifests for external researchers extending the methodology.

SSRN ↗ RED-ORACLE Zenodo ↗ RED-ARB Zenodo ↗ RED-LIQ Zenodo ↗ RED-BASIS Zenodo ↗ RED-BRIDGE Zenodo ↗

Sniper Cohorts and Algorithmic Filter Rejections in Solana Memecoin Markets: Two-Window Replication of Lifecycle-Stage Population Separation

Kamat, A. U. (2026) · SSRN abstract 7128818 · Dataset concept DOI 10.5281/zenodo.21399918 · Companion dataset LIFECYCLE-SEPARATION-2026-v1

Two independent algorithmic detection streams applied to the same Solana memecoin market during the same observation windows produce nearly disjoint token-set outputs. In the June 2026 window, 2 of 20,162 sniper-cohort-detected mints (0.010 percent) appear in the algorithmic rejection stream. In the July 2026 window, 5 of 623 cohort-detected mints (0.803 percent) appear. Cross-window cohort overlap between v1 and v2 is zero mints. The separation is structural: cohort detection operates at the pump.fun bonding-curve stage, while rejection filtering operates on PumpSwap and other post-graduation venues. Four alternative explanations (time-window artifact, data-source mismatch, DEX venue asymmetry, pipeline-version artifact) are systematically ruled out.

SSRN ↗ Zenodo dataset concept DOI ↗

RED-REJECT-2026-v1: A 96-Day Public Corpus of Algorithmic Filter Rejections with Post-Rejection Follow-up Samples on the Solana Pump.fun Ecosystem

Kamat, A. U. (2026) · SSRN abstract 7129798 · Dataset concept DOI 10.5281/zenodo.21402476 · 716,762 sample records across 4,635 unique SPL token addresses

Public benchmark corpus of algorithmic filter rejections with post-rejection follow-up samples, collected from a single blockchain observer on the Solana pump.fun ecosystem between 2026-04-11 and 2026-07-16 (96 days continuous, 13.7 weeks). Corpus size: 716,762 post-rejection sample records across 4,635 unique SPL token addresses. Each row represents a scheduled follow-up observation of a token rejected by one of eight algorithmic trading filter families (downtrend_24h 56.11 percent, marketcap_out_of_range 21.37 percent, multi_tf_fade 11.98 percent, timeSlot_not_strong 5.17 percent, hour_blacklist 2.35 percent, ultra_fast_reject_exit 1.64 percent, utc_hour_blocked 0.91 percent, h6_sellers_dominating 0.41 percent). DEX venue coverage: pumpswap 91.79 percent, meteora 2.73 percent, raydium 2.46 percent, pumpfun bonding-curve 1.83 percent. Operationalises the Post-Rejection Follow-up Sampling (PRFS) methodology at production scale.

SSRN ↗ Zenodo dataset concept DOI ↗

Dev-Wallet Silence Is Not a Rug Predictor on Solana Pump.fun: Empirical Evidence from a Positive-Class and Control-Panel Study

Kamat, A. U. (2026) · Concept DOI 10.5281/zenodo.21434629 (always resolves to the latest version) · Corrected v1.3 (2026-08-26) · Positive class n=27 operationally classified rug events (26 unique dev wallets); eligibility-filtered control panel n=91 dev wallets

Null-result empirical study testing whether dev-wallet transaction silence (>=6 hours) in the 24-hour window before a rug distinguishes Solana Pump.fun rug tokens from an eligibility-filtered non-rug control panel. Under a corrected retrieval protocol that distinguishes empty-window retrievals from retrievals that terminated at the operational pagination cap of the collector, four of 118 observations (two positive, two control) were not completed within the operational pagination cap and are treated as missing. In the complete-case analysis (n = 114), the positive silence rate is 17/25 = 68.0 percent (Wilson 95 percent CI [48.4, 82.8]) and the control silence rate is 78/89 = 87.6 percent (Wilson 95 percent CI [79.2, 93.0]). Two-proportion z-test z = -2.33, p = .020 (two-sided); Fisher exact two-sided p = .031. Framing silence as a positive test for the rug class: balanced accuracy 0.40; Youden's J -0.20. The sign of the between-class difference (positive rate below control rate) is stable under all 16 possible assignments of the four incomplete observations and across silence thresholds from 1 to 24 hours in one-hour steps. Under this operational definition, silence is not a useful positive rug-prediction signal in this sample. Deposited as SILENT-KILLER-2026-v1.3 reproducibility bundle.

Zenodo concept DOI (latest) ↗

RED-2400 Replication Toolkit: A Python Package for Reproducible Filter-Precision Auditing

Kamat, A. U. (2026) · 18/18 byte-identical reproducibility tests passing

Python toolkit accompanying RED-2400. Packages the audit workflow as an installable library, includes a continuous-integration test suite that verifies the per-filter tables in the companion paper reproduce byte-identically against deposited reference outputs. License: MIT (toolkit) / CC-BY-4.0 (dataset).

GitHub ↗

Citing this work: each paper has a permanent DOI. GitHub repos have CITATION.cff files that generate BibTeX automatically via the "Cite this repository" button.

Code

red2400-replication-toolkit v1.0.0

MIT License · Python · 18/18 byte-identical CI tests

Full replication harness for the RED-2400 benchmark. pip install red2400-toolkit; verifies all per-filter precision tables byte-identically against deposited reference outputs.

GitHub ↗ Software Heritage ↗

SWHID: swh:1:rev:a85bf1ff1eb4af345752555a3a1844e5b172bf3e (archived 2026-06-08)

post-rejection-sampling v1.0.0

Reference implementation of PRFS · MIT License · Python 3.9+

Clean reference implementation with RejectionTracker, FollowupSampler, and FilterEvaluator classes. Runs on synthetic data; no external infrastructure needed.

GitHub ↗ Software Heritage ↗

SWHID: swh:1:rev:f4571edd99cb1ef6cc41d826cf1b28ffec542bb8 (archived 2026-06-08)

red-2400-reader MIT

Lightweight reader for the RED-2400 dataset · Python

Minimal-dependency loader for the three RED-2400 files (rejections, outcomes, graveyard).

GitHub ↗ Software Heritage ↗

SWHID: swh:1:rev:d1be7e8b98133e6a3daa346cf249e455c6df5d61 (archived 2026-06-08)

Profiles

ORCID 0009-0000-4781-312X arXiv author page kamat_a_1 SSRN 8 papers distributed; Author Rank 266,953 of 2,848,878 (top 9.4%) Zenodo 21 records (papers + datasets + code) GitHub @aartikamat Google Scholar Author profile Web of Science ResearcherID PYY-2705-2026 Academia.edu @AratiKamat2 IEEE DataPort RED-2400 dataset deposit Kaggle RED-2400 dataset deposit

Peer review service

Seven completed DOI-anchored peer reviews across three scholarly publishing portals:

Registered as a reviewer across the following programs and platforms:

Public writing

The Solana Memecoin Success Rate Fell 3x. The Story Isn't What You Think. (July 15, 2026)

Why the headline pump.fun graduation rate is misleading, and what a Cox model actually shows about 832,941 token launches. Cross-published on Medium and Substack.

Medium ↗ Substack ↗

Memberships, patents, and trademarks

Professional memberships:

USPTO patents and trademarks (Micro Entity status):

Contact

Academic correspondence: arati.kamat@ieee.org. Research collaboration inquiries welcome via my ORCID profile.