Dated public notes and SHA-256 fingerprints to verify. A fingerprint identifies a file; it does not certify its interpretation or engine profitability.

Five years of 15-minute candles simulated (2021–2026). The audit found a selection bias at the close: the study is closed and kept exactly as published.
6,540 simulated positions · conclusion: bias identified
The current generation runs in simulation on Hyperliquid. All of its trades are published live, winners and losers alike. It is the execution showcase.
Loading the trade journal
The next engine reads Hyperliquid's order book. It is assembled from explicit mechanical parts, with no learned model. Every hypothesis is preregistered before any computation, then confirmed or refuted on the whole order book recorded since 6 September 2026. A research agent runs the lab; nothing changes without human approval.
Research checkpoint #1067 · October 29, 2026
Computed automatically from private commits (product and research) and from the lab register. Only numbers, dates and fingerprints are published: no research content.
Hover or tap a bar to see its detail.
Latest experiment recordedOctober 19, 2026
Last activityOctober 29, 2026
91 preregistration fingerprints anchored in Bitcoin with this state
Anchored in Bitcoin: published state of October 29, 2026, block #973,617.ots proofAnchored fileVerify on opentimestamps.org
To check it yourself: 1. download the .ots proof and the anchored file; 2. on opentimestamps.org, drop the .ots proof; 3. then drop the anchored file in the second area.
The file name carries the fingerprint of its content; OpenTimestamps shows the fingerprint of the whole file. They differ, and that is normal.
The site follows every new commit within 5 minutes. Public GitHub repositories receive at most one digest per day.
Documented release
F378
Paper observation
Recorded commits
2904
Since 2026-09-07
Latest activity
Engineering activity and technical checks do not establish profitability.
Rejected on examined development data
Rejected on examined development data
Rejected on examined development data
Rejected on examined development data
Rejected on examined development data
Rejected on examined development data
Evidence inconclusive
Rejected on examined development data
The previous engine runs in simulation on Hyperliquid: every trade, winning or losing, is published live. It is the execution showcase while the new engine is built in the lab.
Every simulated trade, live, with the simulation account balance.
x402scan counters unavailable.
On x402scan, select “All Time” to compare.
15m Historical Positions
6 540
Historical dataset 2021–2026 (Binance)
Simulated liquidations
0
Previous simulation only
Study conclusion
Bias identified
Selected at close, not predictive
Simulated liquidation rate
0.00%
Historical observation, no guarantee
Dated archives · previous models · 645 / n_fills=0 / 2026-09-03
These historical results belong to their original versions and datasets. They do not measure the deployed release and do not guarantee a future edge.