Delayed replication of selected Hyperliquid accounts

Price eligibility, weighting and capital constraints in an exploratory cohort study Objective: To examine price eligibility, aggregation and capital constraints when replicating selected Hyperliquid accounts. Methods: A preselected 22-account cohort was evaluated using delayed whole-episode copying and fill mirroring with a fixed 10-second delay, a 60-second maximum eligible-price gap and explicit follower fees. Saved downstream ledgers were reconciled; raw extraction and delayed-price search were not rerun. Separate hypothetical examples distinguish exposure, margin and external capital. Results: Of 707,802 complete episodes, 483,321 were mirror-price eligible and 540,452 were whole-episode eligible under different price requirements. Whole-episode equal-notional mean net return was -6.96 basis points; mirror P&L was USD 4.521 million at leader notional scale, net of modeled follower fees. Both estimators exclude funding and size impact, use different samples and weights, and are not capital returns. Mirror-excluded episodes contained USD 9.203 million of original net realized P&L versus USD 13.058 million in eligible episodes. Accounts 05 and 13 failed the archived individual mirror acceptance conditions but contributed more than the aggregate profit; the other 20 summed to approximately -USD 13,080. Account 17 supplied 90.06% of all complete episodes and strongly influenced the episode-weighted mean. Conclusions: Coverage, selection, weighting and access to additional collateral limit follower inference. A leverage setting is not an account exposure ratio, and deposits are not investment gains. The study does not validate a profitable fixed-capital strategy. Follower-to-leader inventory transfer is not identifiable with the available evidence. Independent working paper. Not peer reviewed. This deposit contains the public research manuscript only. Its historical experiment concerns 22 preselected accounts; the stated selection and acceptance thresholds define that experiment and are not production admission rules or recommendations. Capital illustrations are hypothetical accounting examples, not an additional observed-account study. Underlying study data and analysis code are retained privately and are not deposited here. They may be requested subject to source-use conditions and a separate disclosure review; access is not guaranteed. No application source code, operating specifications or business documentation is included. The manuscript discloses substantial assistance from Codex and Claude and the precise limits of computational reconciliation and AI-assisted textual critique.