Ideas get tested here before they get trusted. Most of them fail, and we publish those too. You see the rigor, the evidence and the status of every program. The formulas stay private.
Rule-based strategies built and tested by the AURA lab on futures data, measured in R. These are hypothetical backtests and, from October 2026, forward shadow logs. They are not the Discord intelligence or Hologram Candles, and not any member's account.
The Results Vault shows what individual members and the founder chose to share, with permission. Real, and individual. Never mixed into research numbers.
An idea only reaches production after it survives all of them. Most stop at step two.
A specific, testable claim about how a market behaves.
Years of exchange futures data, every trade logged in R, fees and slippage included.
Neighboring settings, other markets, each year on its own.
Data the settings never saw. Many ideas die here.
Runs live without risk, logged before the outcome is known.
Only after the forward record agrees with the backtest.
October 2022 to October 2026 on Nasdaq, S&P 500 and Gold futures, ten automated programs in four families, micro contracts with fees and slippage, one open trade per market at a time, exactly how the live relay runs. Results are in R, one unit of fixed risk per trade, so they don't depend on account size.
Read these as in-sample research: the programs were built on this same history. In an October 2026 cross-market check, 6 of the 8 setups did not hold up outside Nasdaq, so the number below is a research benchmark, not a track record. The forward log is the real test.
Three New York session programs on Nasdaq and S&P 500 futures. Low win rate by design: many small losses, fewer large winners. Now skips quiet opens, which cut its drawdown by more than half.
Four selective Nasdaq and S&P 500 programs that only trade on qualifying days, about seven trades a month.
A structural reversal program on Nasdaq futures in the New York morning, one trade a day at most.
Two Gold programs built around liquidity in the New York morning. Small contributors, stronger since 2024.
How to read this: by-year figures are October to December 2022, 2023, 2024, 2025 and January to early October 2026. These are the conservative numbers: one open trade per market at a time (how the live relay runs), extra slippage on opening entries, half days left out. The tests run on one-minute exchange data for Nasdaq and Gold futures with micro contracts, commissions and slippage. Most programs were first built on 2022 to mid-2024 and then checked, unchanged, on mid-2024 to mid-2026; the three Nasdaq Flow programs were found in October 2026 and also checked on July to October 2026, which they had never seen, but those samples are small. A program is only added when it raises total R without making the worst drawdown worse relative to profit in both periods. Treat the figures as optimistic: forward shadow results will be published here separately, dated and unedited, and live fills can differ.
Published October 6, 2026 and kept here unchanged: AURA Core (Gold, Russell 2000, Dow) +119.6R, AURA Reversal (Nasdaq, Dow) +48.6R and AURA Edge (four selective S&P 500, Russell 2000 and Nasdaq programs) +168.3R, combined +336.5R with 33 of 36 positive months. These programs overlap only partly with the AURA Robot and stay in shadow. Their details are in the status board and the log below.
Four programs wired into the automation relay and running in dry run: live signals, no orders, no capital at risk.
ShadowThree programs and one more Session program passed the backtest and the portfolio test in October 2026. Next: dry run alongside the others.
ResearchThe full robot is too large for smaller evaluation accounts at minimum size, so a lower-drawdown subset was defined for them. Sizing is set by the worst drawdown, never by margin.
CheckBacktest passed. Forward shadow log starts October 2026: live signals, no capital at risk, logged before the outcome.
ShadowBacktest passed on Nasdaq and Dow. Running in shadow alongside Core.
ShadowA New York open program that only trades on selected days. +21.8R, positive every year. Forward shadow from October 2026.
ShadowA one-directional liquidity reversal, about two trades a month. +21.6R, positive every year. Forward shadow from October 2026.
ShadowAn opening-session reversal with resting limit entries. +45.8R, positive every year. Forward shadow from October 2026.
ShadowBuilt on AURA's proprietary level map, one-directional by design. +79.0R, worst drawdown −6.3R, positive every year. Forward shadow from October 2026.
ShadowSplit entries without raising maximum risk. Promoted into Core in October 2026: +119.6R versus +90.6R over the same three years, with a smaller worst drawdown. Forward shadow from October 2026.
ShadowThe programs above, running hands-free on our own accounts first. Opens to others only after a live track record. Join the waitlist.
In developmentFramework designed (below). Data collection has not started, so there are no numbers to show yet.
Study designedA lab that never rejects anything isn't testing anything. These ideas were built, tested and killed.
Sweep, break and retest variations on Gold and Nasdaq. All lost money over three years.
RejectedThe Core engine was tested on nine more futures markets. S&P 500, crude oil, euro and copper failed and were dropped.
RejectedReached an 85% win rate with very small targets, and cut total profit by about 70%. A win rate is not an edge.
RejectedCutting losers early and letting winners run longer. Every early cut also killed winners that dipped first. Net flat.
RejectedBox and compression breakouts looked good in-sample and failed out-of-sample.
RejectedFading and following the overnight gap on five markets. One great year, then losses. Not consistent enough to keep.
RejectedInitial-balance pullbacks and inside-day breakouts across five markets. Mostly losing over three years.
RejectedAdded on top of working programs, they removed more winners than losers.
RejectedAbout 50,000 combinations of price, volume and time conditions on Nasdaq and Gold. Real data passed no more often than shuffled data, so the survivors were luck.
RejectedModels trained to copy a human trader's entries and exits, including higher timeframes, other markets and cycles. Once measured without look-ahead, they did no better than random.
RejectedFourteen projection levels, alone and in clusters. Even picked with hindsight, they marked the morning turn no more often than levels taken from random days.
RejectedWaiting for price to prove the turn with a stop order at stretched levels. About 185,000 simulated orders: break-even on its own.
RejectedLooked strong on 2022 to 2025 data, then lost on four of five markets over the last ten months. Dropped before it went anywhere near a live account.
RejectedTrading the first close through a mapped level toward the next one. Lost on all five markets tested.
RejectedEach projection gets recorded at five points, so anyone can see whether price followed the mapped scenario, without exposing how the projection is built.
The naked chart and session context at the time.
The projected path, invalidation and objective, timestamped.
How price interacted with the path as it printed.
Where price actually resolved.
Counted outcomes with the sample size shown. No headline numbers before a dated sample exists.
We re-ran the robot's setups on other markets they were never built on. 6 of the 8 did not hold up, which means much of the +450R is fitted to Nasdaq history. The headline stays on this page as an in-sample research benchmark, clearly labeled, and nothing goes live with capital until the forward log proves it.
CheckEvery program re-run on the last four years only (October 2022 to October 2026), one open trade per market at a time exactly like the live relay, extra slippage on opening entries and half days removed. The headline moved from +704R to +450R. Every rolling 12-month period and every quarter is still positive. The bigger number counted overlapping trades the live system would never take.
CheckOne new rule for the Nasdaq Session family: skip quiet opens. Worst month-end drawdown went from −33.8R to −24.2R with about the same profit (+718R to +704R), better in both halves of the data and on July to October 2026, which it had never seen. Ten other drawdown ideas were tested and rejected.
Every program re-tested on one-minute exchange data from January 2022, micro contracts, fees and slippage: every calendar year positive, every rolling 12-month period positive.
Candidates were only kept if they raised total R without making drawdown worse relative to profit, in both the earlier and the later half of the data. Four passed; most candidates failed. The robot went from about 84R a year to about 160R a year.
Five research lines rejected, three of them after catching look-ahead mistakes in our own tests (see above).
Checked against the full portfolio first: about 11% more profit for the whole system, with risk growing less than profit. Core goes from +90.6R to +119.6R and stays positive in every year.
From roughly a hundred backtests across five markets, four selective programs held up in every year and across neighbouring settings. Added together with Core, they roughly doubled profit with no larger worst drawdown in testing.
Each Edge program was rebuilt as a live script and re-run over the same three years before going to shadow. Results reproduced the research to the trade.
Tested on five markets and rejected (see above).
Split entries with the same maximum risk. Improved every Core market and reduced Gold's drawdown. Moved to research status pending forward data.
Two programs with negatively correlated months: +139.3R combined versus +90.6R for Core alone, with a smaller worst drawdown.
A structural reversal entry with resting limit orders. Passed on Nasdaq and Dow; failed on Gold and Russell, so it doesn't trade them.
Gold's core logic held up on two more index markets. Four other markets were rejected.
High win-rate targets and early exits tested and rejected (see above).
The full multi-filter setup only broke even on earlier years, while the simpler Gold core held up. The simpler version became Core.
Specific formulas, filters, thresholds, timing logic and execution rules are intentionally withheld. They form part of AURΔBØT's proprietary research framework. Published here: test periods, sample results, drawdowns, monthly consistency, versions and status.
Every forward entry carries the instrument, the session, the strategy version and a timestamp. Win rates appear only as counted outcomes from a dated, versioned sample, with the sample size shown.
HYPOTHETICAL OR SIMULATED PERFORMANCE RESULTS HAVE CERTAIN LIMITATIONS. UNLIKE AN ACTUAL PERFORMANCE RECORD, SIMULATED RESULTS DO NOT REPRESENT ACTUAL TRADING. ALSO, SINCE THE TRADES HAVE NOT BEEN EXECUTED, THE RESULTS MAY HAVE UNDER- OR OVER-COMPENSATED FOR THE IMPACT, IF ANY, OF CERTAIN MARKET FACTORS, SUCH AS LACK OF LIQUIDITY. SIMULATED TRADING PROGRAMS IN GENERAL ARE ALSO SUBJECT TO THE FACT THAT THEY ARE DESIGNED WITH THE BENEFIT OF HINDSIGHT. NO REPRESENTATION IS BEING MADE THAT ANY ACCOUNT WILL OR IS LIKELY TO ACHIEVE PROFIT OR LOSSES SIMILAR TO THOSE SHOWN.
Trading futures involves substantial risk of loss and is not suitable for every investor. Nothing on this page is financial advice or a recommendation. Research programs described here are not offered as a managed or copy-trading service.