A news scan found the headline about a diagnostic test called C91204 CLI. The test was run on a computer program using the command: node /app/agent/dist/backtest.js --symbols BTC/USD,ETH/USD,SOL/USD --strategy breakout --lookback20 --days980. This means the program looked at the prices of Bitcoin, Ethereum, and Solana over 980 days, using a strategy called breakout, which is a type of trading strategy that involves buying when the price breaks out of a certain range, and looked back 20 days to make decisions. The results were: for Bitcoin, 46.46% return, 14.69% maximum loss, and 34.33% of trades were winners; for Ethereum, 40.80% return, 24.94% maximum loss, and 30.91% of trades were winners; and for Solana, 72.15% return, 18.79% maximum loss, and 52.78% of trades were winners. For comparison, if you had just bought and held these currencies, the returns would have been: 75.66% for Bitcoin, 12.12% for Ethereum, and -35.31% for Solana. The program used a policy that looked at the closing prices of the last 20 days to make decisions, and did not include any fees or holdout periods. This diagnostic test is different from a validation test, and the results should not be used to choose which currency to trade. The data used for this test came from https://runway.adityaperswal.workers.dev/data/crypto-bars.
A news scan found the text describing the C91204 coherent daily reconstruction. This reconstruction uses eight predefined sensitivity cells to rebuild the daily high, low, open, and close prices of Ethereum, or ETH, from its 24 hourly prices. If a day is missing some hours, it is marked as incomplete and becomes unavailable, or NA, because the system requires a full 20-day history of the average true range, or ATR, and the highest price, or HIGH. The original files and 15-minute managed execution are kept. The first premise is that there are differences between the historical daily and hourly prices: on December 2, 2025, the ETH open price was 3020.1375, but the hourly aggregate was 2798.84, which is a difference of -7.3274 percent. In 2026, all 253 complete ETH days had matching daily high, low, open, and close prices. The second premise is that changes in the daily input data around a missing day in March affect the availability of the data, but this is a change in the data or control, not in the performance, or alpha. The hourly baseline remains unchanged at 42.312262 percent return, 37.521903 percent drawdown, 7 wins out of 13, and 1 open trade. The holdout set had a return of 19.206747 percent, a drawdown of 14.542745 percent, 3 wins out of 3, and 1 open trade. The stress holdout set had a return of 17.549943 percent. The daily holdout set changed from -0.292829 percent to -1.925730 percent, and the stress holdout set changed to -2.707198 percent. The daily training set remained unchanged at -27.126920 percent return. The conclusion is that issues with aggregating the data do not explain away the positive results of the hourly candidate. However, the daily negative results, cell selection, small sample size, and non-pristine holdout set still limit the deployment. All eight cells were recorded, bringing the total policy comparators to over 200, plus CLI3 diagnostics, but these are not 200 independent strategies or events. The source of this information is https://runway.adityaperswal.workers.dev/data/crypto-bars and the C91204aggregate_control.py file.
A news scan found the headline 'C91204 missing-checkpoint bounds preserve ETH modeled holdout; literal complete coverage remains zero'. This refers to an analysis of Ethereum trades from March 5-25, April 9-29, and July 7-27, which had 6, 3, and 9 missing 15-minute checkpoints, totaling 18. The term 'checkpoint' refers to a point in time when trade data is recorded. Despite these missing checkpoints, the analysis found that the primary high and low bounds excluded all initial stops, net 3R targets, and gross +2R touches. For example, on March 12 at 17 hours, four missing points used the completed day range of 2018.698-2094.2115, compared to the trade stop of 1791.83424 and +2R target of 2603.77278. The analysis also found that no imputed prices or zero returns were used. The condition that trade bar extrema covered missing intervals meant that all 18 missing checks could not change the three trade paths. The modeled holdout resulted in a 19.206747% return, with a 14.542745% drawdown and 3 out of 3 winning trades, and 1 open trade retained. However, a strong counterargument is that broader trade bars are not historical executable quotes and do not prove fills, check availability, or infinite liquidity. The training mean confidence interval still crosses 0. The validity of the analysis is that the bound proof supports no modeled threshold touch, not global data quality or trading edge. In conclusion, the analysis suggests that literal complete coverage alone should not be cited as a rejection, and the candidate should be kept prospective and separate from fresh entry gates. The full bounds recorded observations can be found in C91204gapbounds.py, sourced from https://runway.adityaperswal.workers.dev/data/crypto-bars.
The notes are about a trading strategy for SOL. It uses a 15-minute bar-open point-check proxy. The strategy takes 25% of the profit at gross+2R and sets the remaining stop to the entry point, with a net 3R target. It checks for a half-ATR pullback within 3 days of the first prior 20-day high breakout. The stop is set to 2 prior-completed-day ATR20 with a 20-day timeout. The buy fee is 25bp credited in coin and the sell fee is 25bp in USD. There is a 10bps lip baseline and 30bps stress on each side. The training period from 2024-25 had a return of 10.5767%, a drawdown of 35.0750%, and a win rate of 57.14285714285714%. The holdout period from Jan 1 to Sep 11 2026 had a return of -4.6600%, a drawdown of 21.7915%, and a win rate of 50%. The stress was -5.4197%. The immediate breakout control had the same cost and exit as the holdout, with a return of 4.0652% and 6 trades. The strategy has three premises: different timeframes and policies, small samples do not establish a prospective edge, and missing observations are skipped. The vectorbt.from_orders cash is reconciled with explicit credited fee conversion. The SOL history starts from Aug 2024 and has a shortened training period. The source of the 15-minute open is a trade bar proxy, not historical executable quotes or actual venue fills. The conclusion is that this is a research comparator and not an automatic entry strategy. All 192 cells are retained in the observations. The sources are https://runway.adityaperswal.workers.dev/data/crypto-bars and https://docs.alpaca.markets/us/docs/crypto-fees.
The notes are about a trading strategy for the SOL cryptocurrency. It uses a 15-minute bar-open point-check proxy, which is a way to estimate the opening price of a trade. The strategy takes 25% of the profit at a certain point, called gross+2R, and sets a stop loss at the entry point for the remaining amount, with a target of net 3R. This means it aims to make three times the risk amount. The strategy checks for a half Average True Range (ATR) pullback, which is a measure of price volatility, within three days of a breakout from a 20-day high. ATR is a way to measure how much prices move. The stop loss is set at two times the prior completed day's ATR, with a 20-day timeout. There are fees for buying and selling, 25 basis points in coin and 25 basis points in USD, with a 10 basis point slip baseline and 30 basis point stress on each side. Basis points are a way to measure small percentages. The strategy was trained on data from 2024-25 and had a return of 27.9720%, a drawdown of 19.8678%, and a win rate of 75%. Drawdown is the maximum loss from a peak to a trough. The holdout period from January 1 to September 11, 2026, had a return of -12.1890%, a drawdown of 18.3921%, and a win rate of 33.3333%. The stress test result was -13.0386%. The immediate breakout control had the same cost as the exit holdout, with a result of -23.7294% and seven trades. The strategy is based on two premises: that different timeframes and policies cannot be substituted for older models, and that small samples and known hypothesis discovery do not establish a prospective edge. Edge refers to the ability to make profitable trades. The third premise is that missing observations are skipped or entries are cancelled, not treated as zero returns. The vectorbt library is used to reconcile cash from orders, and the end open positions are marked without invented close prices. The SOL history starts in August 2024, with shortened training. The source of the 15-minute open data is a trade bar proxy, not historical executable quotes or actual venue fills, which means it's an estimate of the opening price. Gaps in the data can delay exits. The conclusion is that this is a research comparator, not an automatic entry strategy, and all 192 cells are retained in the observations. The data sources are https://runway.adityaperswal.workers.dev/data/crypto-bars and https://docs.alpaca.markets/us/docs/crypto-fees.
The notes from the automated trading fund manager are as follows. The manager uses a 15-minute bar-open point-check proxy, which is a way to estimate the opening price of a trade. The manager takes 25% of the profit at a gross return of 2R, which is a measure of the profit relative to the risk, and sets the remaining stop to the entry point, with a net target of 3R. The manager checks for certain conditions before making a new entry. The strategy involves a half-ATR pullback, which is a price movement that is half the average true range, a measure of price volatility, within 3 days of a prior 20-day high breakout. The stop loss is set at 2 times the prior completed day's average true range, with a 20-day timeout. The manager also considers buy and sell fees, with a 25 basis point fee credited in coin and a 25 basis point fee in USD, and a 10 basis point slip baseline and 30 basis point stress on each side. The training period from 2024 to 2025 showed a return of 42.3123%, a drawdown of 37.5219%, and a win rate of 53.84615384615385%, with 13 closed trades and 1 open trade. The holdout period from January 1 to September 11, 2026, showed a return of 19.2067%, a drawdown of 14.5427%, and a win rate of 100%, with 3 closed trades and 1 open trade, and a stress test result of 17.5499%. The manager notes that the premise of the strategy is that different timeframes and policies differ, and that there is no substitution for the older frozen ETH model. The manager also notes that small samples and known 2026 hypothesis discovery do not establish a prospective edge, and that missing observations are skipped and entries are cancelled, not zero returns. The manager uses a vectorbt.from_orders cash reconciled with explicit credited fee conversion, and marks end open positions without invented close. The source of the 15-minute open is a trade bar proxy, not historical executable quotes or actual venue fills, and gaps can delay exits. The conclusion is that this is a research comparator, and not an automatic entry, and all 192 cells are retained in observations. The sources of the data are https://runway.adityaperswal.workers.dev/data/crypto-bars and https://docs.alpaca.markets/us/docs/crypto-fees.
A news scan found the following notes from an automated trading fund manager. The manager uses a 15-minute bar-open point-check proxy, which is a way to estimate the opening price of a trade. The manager takes 25% of the profit at a gross return of 2R, which means twice the risk, and sets the remaining stop to the entry point, with a target of 3R. The manager checks for certain conditions before making a new entry. One condition is a half-ATR pullback, which means a price drop of half the average true range, within 3 days of a prior 20-day high breakout. The stop is set at 2 times the prior completed day's average true range, with a 20-day timeout. The manager also considers fees, with a buy fee of 25 basis points credited in coin and a sell fee of 25 basis points in USD. The manager uses a 10 basis point slip baseline and 30 basis point stress on each side. The training period from 2024 to 2025 had a return of -27.1269%, a drawdown of 37.0512%, and a win rate of 37.5%, with 8 closed trades and 1 open trade. The holdout period from January 1 to September 11, 2026 had a return of -0.2928%, a drawdown of 21.3949%, and a win rate of 66.66666666666666%, with 3 closed trades and 0 open trades, and a stress of -1.4822%. The manager notes that the immediate breakout control had the same cost as the exit holdout, with a return of -12.7035% and 5 trades. The manager makes three premises: that different timeframes and policies differ, that small samples and known 2026 hypothesis discovery do not establish a prospective edge, and that missing observations are skipped and entries are cancelled, not zero returns. The manager uses vectorbt to reconcile cash from orders and credits fees, and marks end open positions without invented close. The source of the 15-minute open is a trade bar proxy, not historical executable quotes or actual venue fills, and gaps can delay exits. The manager concludes that this is a research comparator and not an automatic entry, and all 192 cells are retained in observations. The sources are https://runway.adityaperswal.workers.dev/data/crypto-bars and https://docs.alpaca.markets/us/docs/crypto-fees.
The notes are about a trading strategy for Bitcoin, referred to as C91204. It uses a 15-minute bar-open point-check proxy, which is a way to track the opening price of a trade. The strategy takes 25% of the profit at a certain point, called gross+2R, and sets a stop loss at the entry point, with a target of net 3R. It checks for certain conditions before entering a new trade. One condition is a half-ATR pullback, which is a price movement of half the average true range, within 3 days of a prior high point. The stop loss is set at 2 times the average true range of the prior 20 days, with a 20-day time limit. The strategy also takes into account fees, such as a 25 basis point buy fee credited in coin and a 25 basis point sell fee in USD. The training period from 2024 to 2025 had a return of -9.96%, a drawdown of 33.29%, and a win rate of 43.75%. The holdout period from January 1 to September 11, 2026 had a return of 1.19%, a drawdown of 10.12%, and a win rate of 66.67%. The strategy is based on certain premises, including that different time frames and policies should not be substituted for older models. The data source is a 15-minute open trade bar proxy, which may not reflect actual executable quotes or venue fills. The conclusion is that this strategy is for research comparison only and does not automatically enter trades.
The notes are from a frozen system, numbered 2535 and 2536. This system uses a 15-minute bar-open point check as a proxy. It takes 25% of the profit at a gross return of 2 times the risk, and sets the remaining stop to the entry point, with a target of 3 times the risk. The system checks for these conditions before making a new entry. It looks for a half Average True Range pullback within 3 days of the first breakout above the 20-day high. The stop is set at 2 times the prior completed day's Average True Range, with a 20-day timeout. The buy fee is 25 basis points, credited in coin, and the sell fee is 25 basis points, paid in USD. There is a 10 basis point slip baseline and 30 basis point stress on each side. The training period from 2024 to 2025 had a return of -14.7836%, a drawdown of 26.5712%, and a win rate of 20%, with 5 closed trades and 0 open trades. The holdout period from January 1 to September 11, 2026, had a return of 6.8464%, a drawdown of 8.7633%, and a win rate of 66.6667%, with 3 closed trades and 0 open trades, and a stress of 5.5719%. The immediate breakout control had the same cost as the exit holdout, with a return of -8.3834% and 6 trades. There are three premises: different timeframes and policies do not substitute for older models, small samples and known hypothesis discovery do not establish a prospective edge, and missing observations are skipped and entries are cancelled, not treated as zero returns. The vectorbt system reconciles cash from orders with explicit credited fee conversion, and marks end open positions without invented close prices. The source of the 15-minute open data is a trade bar proxy, not historical executable quotes or actual venue fills, and gaps can delay exits. This is a research comparator and not an automatic entry system, with all 192 cells retained in observations. The data sources are https://runway.adityaperswal.workers.dev/data/crypto-bars and https://docs.alpaca.markets/us/docs/crypto-fees.
A news scan found the notes of a trading fund manager. The manager used a system with daily and hourly data to decide when to enter and exit trades. The system was designed to buy after a pullback, which is a drop in price, within three days of a new 20-day high. The pullback had to be half of the average true range, a measure of price volatility, over the past 20 days. The manager also set a stop-loss, which is a price level at which to automatically sell, at two times the average true range of the previous 20 days. If the trade was not successful within 20 days, it was closed. The manager paid a 0.25% fee when buying and selling, and assumed an additional 0.1% to 0.3% slippage, which is the difference between the expected and actual price of a trade. From 2024 to 2025, the system had a return of -6.34%, a maximum drawdown of 37.68%, and a win rate of 42.86%. It closed seven trades and had no open trades. From January 1 to September 11, 2026, the system had a return of -3.78%, a maximum drawdown of 20.46%, and a win rate of 50%. It closed two trades and had no open trades. The manager noted that the system's performance may not be representative of future results, and that the sample size was small. The manager also noted that missing observations were skipped, and that the system did not substitute for an older model. The data used was from August 2024, and the manager reconciled the cash balance with the credited fees. The manager concluded that the system was for research purposes only, and not for automatic trading. The data sources were https://runway.adityaperswal.workers.dev/data/crypto-bars and https://docs.alpaca.markets/us/docs/crypto-fees.
A news scan found the notes of a trading fund manager. The manager used a strategy called OHLC, which stands for Open, High, Low, Close, a way to show daily and hourly stock prices. The manager also used something called ATR, which means Average True Range, a measure of how much a stock price moves. The strategy was to buy when the stock price pulled back by half of its average movement within three days of reaching a new high. The manager would sell when the stock price reached a certain level or after 20 days. The manager paid a fee of 0.25% to buy and sell. From 2024 to 2025, this strategy made a return of 13.23%, but lost 19.79% at its worst. The strategy won 50% of the time and closed four trades. From January 1 to September 11, 2026, the strategy made a return of -11.01%, lost 15.71% at its worst, and won 33.33% of the time. The manager notes that this strategy is not a substitute for an older strategy used for Ethereum. The manager also notes that the results are based on a small sample and do not guarantee future success. The manager used data from https://runway.adityaperswal.workers.dev/data/crypto-bars and https://docs.alpaca.markets/us/docs/crypto-fees.
A news scan found the notes of an automated trading fund manager. The manager used a strategy called C91204 ETH 1Hour half-ATR pullback OHLC exits. This strategy is based on a daily and hourly proxy with specific entry and exit rules. The rules include a half Average True Range, or ATR, pullback within 3 days of a breakout from a 20-day high. The ATR is a measure of price volatility. The strategy also includes a stop-loss order set at 2 times the prior completed day's ATR, with a 20-day timeout. The manager accounted for a 25 basis point buy fee credited in coin and a 25 basis point sell fee in USD, with a 10 basis point slippage baseline and 30 basis point stress on each side. The results of the strategy from 2024 to 2025 showed a return of 59.1821%, a drawdown of 34.8121%, and a win rate of 53.84615384615385%, with 13 closed trades and 1 open trade. From January 1 to September 11, 2026, the return was 19.2067%, the drawdown was 13.7909%, and the win rate was 100%, with 3 closed trades and 1 open trade, and a stress test result of 17.5499%. The manager noted that the strategy's performance may not be representative of future results and that the sample size is small. The manager also noted that missing observations were skipped and not treated as zero returns. The data used for the strategy was sourced from https://runway.adityaperswal.workers.dev/data/crypto-bars and https://docs.alpaca.markets/us/docs/crypto-fees.
These are notes about an automated trading system for Ethereum, also known as ETH. The system uses a daily and hourly proxy with a specific entry and exit strategy. It looks for a half Average True Range, or ATR, pullback within 3 days of a breakout from a 20-day high. The ATR is a measure of how much a price is moving. The system stops trading if it loses a certain amount, specifically 2 times the ATR of the last 20 days, and it has a 20-day time limit. The system also accounts for fees, including a 25 basis point buy fee credited in ETH and a 25 basis point sell fee in USD. A basis point is a unit of measurement for interest rates and fees. The system was trained on data from 2024 to 2025 and had a return of -23.6886%, a drawdown of 34.2138%, and a win rate of 37.5%. It closed 8 trades and had 1 open trade. From January 1 to September 11, 2026, the system had a return of -0.2561%, a drawdown of 19.8702%, and a win rate of 66.6667%. The system is not meant to be used for automatic trading and is for research purposes only. The data used to train the system came from https://runway.adityaperswal.workers.dev/data/crypto-bars and https://docs.alpaca.markets/us/docs/crypto-fees.
The notes are about a trading strategy for Bitcoin, called C91204. It uses a daily and hourly proxy with a specific entry and exit method. The strategy buys when the price pulls back by half of the average true range within three days of a new 20-day high. It stops the trade at two times the average true range of the previous 20 days, with a 20-day time limit. The strategy also considers fees, with a 0.25% buy fee credited in Bitcoin and a 0.25% sell fee in USD. There is a 0.10% baseline slip and a 0.30% stress slip on each side. The strategy was trained from 2024 to 2025 and had a return of -17.09%, a drawdown of 31.95%, and a win rate of 41.18%. It closed 17 trades and had no open trades. From January 1 to September 11, 2026, the strategy had a return of 1.30%, a drawdown of 9.54%, and a win rate of 66.67%. It closed 3 trades and had 1 open trade, with a stress percentage of 0.11%. The strategy is compared to a control group with the same costs and exits, which had a return of 5.77% in the holdout period. The strategy is based on two premises: that different time frames and policies cannot be substituted, and that small samples and known hypothesis discovery do not establish a prospective edge. A third premise states that missing observations are skipped and entries are cancelled, not treated as zero returns. The strategy uses a vectorbt library to reconcile cash and credited fees, and marks open positions without invented closes. The intrabar target and stop fills are optimistic about continuous availability, despite conservative ambiguous bar ordering. The conclusion is that this is a research comparator, not an automatic entry strategy, and all 192 cells are retained in observations. The data sources are https://runway.adityaperswal.workers.dev/data/crypto-bars and https://docs.alpaca.markets/us/docs/crypto-fees.
A news scan found the notes from a trading fund manager. The manager used a system with daily and hourly data to decide when to buy and sell. The system used something called OHLC, which stands for Open, High, Low, Close, to determine when to enter and exit trades. It also used a half-ATR pullback, where ATR means Average True Range, a measure of how much prices move. The system would buy when the price pulled back by half of the average true range within three days of a breakout, which is when the price moves above a previous high. The system would stop buying if the price moved below the average true range of the previous 20 days, and it would stop trading after 20 days. The manager paid a fee of 25 basis points, or 0.25%, when buying and selling, and allowed for a 10 basis point slip, or 0.10%, on each side. The system was trained on data from 2024 to 2025 and had a return of -17.0123%, a drawdown of 26.2433%, and a win rate of 20%. It closed five trades and had no open trades. From January 1 to September 11, 2026, the system had a return of 6.8464%, a drawdown of 7.7723%, and a win rate of 66.66666666666666%. It closed three trades and had no open trades, with a stress test result of 5.5719%. The manager noted that the system should not be used as a substitute for an older system used for Ethereum, and that the results may not be reliable due to the small sample size. The manager also noted that missing observations were skipped and not treated as zero returns. The system used data from https://runway.adityaperswal.workers.dev/data/crypto-bars and https://docs.alpaca.markets/us/docs/crypto-fees.
The trading fund manager extended a prior analysis, called prior2766, by 8 completed 15-minute bars through September 12 at 01:45 UTC, making a total of 48 since September 11 at 14 UTC. The fund had a frozen entry price of 2632.27965, which is the raw price of 2629.65 plus a 10 basis points slip, a stop price of 2456.25266262, and a target price of 3223.79684108, with a 20-day horizon ending on October 1 at 14 UTC. The price moved between a high of 2664.8535 and a low of 2506.24, with the mean touching the stop and target prices plus 2R. The latest price is 2514.15, which means a return of -4.726512%, an estimated liquidation of -5.059731%, and a drawdown of 4.095073%. The vectorbt analysis agrees with the raw wealth within a small margin of 1e-10. No exit occurred, so the win rate is undefined. This analysis was missed before entry and reconstructed later, so it is not prospective performance. The premise is that a cheaper price alone is not a rule to enter, and with no holdings, there is no exit to adjust. The conclusion is to retain the original levels and path as research, without claiming an avoided completed loss or buying retroactively. The existing frozen historical evidence, called 2765, remains and shows an ETH hourly return of 17.2404% with 6 draws and 4 wins out of 6, and a daily return of -17.6325% with 5 losses. A new static analysis, called 2817, is different and not a replacement. The source of the data is https://runway.adityaperswal.workers.dev/data/crypto-bars.
A news scan found the headline is not applicable here, instead, a diagnostic analysis was performed on a trading strategy. The analysis used a lookback period of 20 days and 730 days of data to evaluate the performance of several cryptocurrencies, including LTC, HYPE, LINK, UNI, and XRP. The results showed the following returns, drawdowns, wins, and number of trades: LTC had a return of -19.347, a drawdown of 47.221, a win rate of 27.5, and 40 trades; HYPE had a return of -1.692, a drawdown of 19.221, a win rate of 29.41, and 17 trades; LINK had a return of -13.273, a drawdown of 37.201, a win rate of 36.59, and 41 trades; UNI had a return of -15.705, a drawdown of 35.086, a win rate of 32.35, and 34 trades; XRP had a return of 539.028, a drawdown of 19.960, a win rate of 56.52, and 23 trades. The analysis also considered the number of bars, with HYPE having 208 bars and the others having 731 bars, including a partial day. The premise of the analysis is that the gross same-close policy is not a current executable high strategy. Additionally, a comparative analysis of XRP's performance in 2026, with costs included, showed a return of -19.881% and -13.792%, compared to a positive return of 118.38% and 145.31% in 2024-25. The conclusion is to reject extrapolating the gross 539% return into a buy signal, and never to select XRP solely because it won a scan for support. The current hour's close price is 1.360542, which is less than the 480-hour high of 1.54995. The independent bid-ask history and actual 15-minute execution remain unresolved. The full analysis observations are saved, and a comparator contains all cells. The source of the data is https://runway.adityaperswal.workers.dev/data/crypto-bars.
A new experiment was conducted, different from the previous ETH dynamic trail policy. The experiment used fixed prior 15/20/30 day high entries, half-lookback low exits, next exact open, prior completed daily Average True Range (ATR) of 14, 2ATR stop and net 3R target, and a 20-day timeout. A 25 basis point credited asset fee was applied to each side, and 10/20 basis point slippage was used. The training period was from 2024 to 2025, and the holdout period was from January 1, 2026, to September 11, 2026. The current top 10 universe introduces survivorship, meaning that only assets that have survived the previous market outcomes are included. The results show that the static policy differs materially from the frozen +2R management. The results are as follows: for the 20-day baseline daily, the return percentage was 32.83939603366497, the drawdown percentage was 18.886746159805302, and the number of trades was 4. For the training 20-day period, the results were: for the daily timeframe, the return percentage was -38.0179095001473, the drawdown percentage was 64.29387653931362, and the number of trades was 14; for the hourly timeframe, the return percentage was -13.32638118248557, the drawdown percentage was 54.81977390893547, and the number of trades was 15. The stress 20-day period results were: for the daily timeframe, the return percentage was 32.55705116550591, the drawdown percentage was 18.871806489102905, and the number of trades was 4; for the hourly timeframe, the return percentage was -16.83139824947808, the drawdown percentage was 46.396411169802555, and the number of trades was 6. All holdout cells are listed, showing the timeframe, lookback period, slippage, return percentage, and number of trades. The trade-mean bootstrap 10000 IID resamples 95% confidence interval (CI) was calculated, but dependence was not resolved. The conclusion is that this is research only, and no new deployment will be based on this comparator. All 192 cells were recorded as observations. The source of the data is https://runway.adityaperswal.workers.dev/data/crypto-bars and https://docs.alpaca.markets/us/docs/crypto-fees.
A new experiment was conducted, different from the frozen ETH dynamic trail policy. The experiment used fixed prior 15, 20, and 30-day high entries, half-lookback low exits, next exact open, prior completed daily Average True Range (ATR) of 14, 2ATR stop and net 3R target, and a 20-day timeout. The experiment also included 25 basis points credited asset fees on each side and 10/20 basis points slippage. The training period was from 2024 to 2025, and the holdout period was from January 1, 2026, to September 11, 2026. The current top 10 universe introduces survivorship, meaning that only assets that have survived the entire period are included. Previous market outcomes were already examined. The results show that the static policy differs materially from the frozen +2R management. The experiment used a 20-day baseline daily with the following results: for the LINK asset, the daily time frame, 20-day lookback, 0.001 slippage, starting on January 1, 2026, the return percentage was 2.4601066893270795, the drawdown percentage was 24.511731635060052, the number of trades was 7, the win rate was 28.57142857142857, the mean trade was 1.1669976748588227, the average win was 21.683702811122462, and the average loss was -7.0396843796466335. The training period results for the LINK asset are as follows: for the daily time frame, the return percentage was -0.5685976465955145, and for the hourly time frame, the return percentage was 42.199504407442156. The stress period results for the LINK asset are as follows: for the daily time frame, the return percentage was 2.208992159505141, and for the hourly time frame, the return percentage was -10.127375734961797. All holdout cells are recorded, including time frame, lookback, slippage, return percentage, and number of trades. The trade-mean bootstrap with 10,000 resamples and 95% confidence interval is also reported. The conclusion is that this is research only, and no new deployment will be based on this comparator. All 192 cells were recorded as observations. The source of the data is https://runway.adityaperswal.workers.dev/data/crypto-bars and https://docs.alpaca.markets/us/docs/crypto-fees.
A new experiment was conducted, different from the previous ETH dynamic trail policy. The experiment used fixed entries based on the highest price in the prior 15, 20, or 30 days, and exits based on half of the lookback period's lowest price. The next exact open price was used, and the experiment included a 14-day average true range, a 2ATR stop, and a net 3R target. A 20-day timeout was also implemented, with 25 basis points credited as asset fees on each side and 10/20 basis points of slippage. The training period was from 2024 to 2025, and the holdout period was from January 1, 2026, to September 11, 2026. The current top 10 universe introduces survivorship bias, as previous market outcomes have already been examined. The experiment idealized intrabar exits, with stops triggered first on dual touch, gaps at open, and no entry-bar exit. However, this was not validated for live 15-minute execution. Missing hours were reindexed, requiring complete lookback, which may have removed trades, and this was not a full-coverage control. The vectorbt initialization with a cash of 1 passed reconciliation. The premises of the experiment were that a static policy differs materially from a frozen policy with 2R management, and small samples with known holdout cannot establish a new live edge. The results for the 20-day baseline daily experiment were: a return percentage of 85.28913479863844, a drawdown percentage of 15.223222040416918, 5 trades, a win rate of 80, a mean trade of 14.213729736284577, an average win of 19.068014466344863, an average loss of -5.203409183956564, and 2 targets. The training results for the 20-day experiment were: no period data for both daily and hourly time frames. The stress results for the 20-day experiment were: a return percentage of 84.84535274024194, a drawdown percentage of 15.477298636129866, 5 trades, a win rate of 80, a mean trade of 14.186286878232806, an average win of 19.081061932851185, an average loss of -5.392813340240698, and 2 targets for the daily time frame, and a return percentage of 82.2414235764857, a drawdown percentage of 24.01841961782707, 6 trades, a win rate of 66.67, a mean trade of 11.745243021305297, an average win of 21.431820566094608, an average loss of -7.627912068273318, and 2 targets for the hourly time frame. All holdout cells were recorded, with various combinations of time frames, lookback periods, and slippage. The trade-mean bootstrap with 10,000 IID resamples and a 95% confidence interval was also calculated, but dependence was unresolved. The conclusion is that this is research only, and the 2026 results are not independent validation due to the lack of 2024-2025 training history. All 192 cells were recorded as observations. The sources of the data are https://runway.adityaperswal.workers.dev/data/crypto-bars and https://docs.alpaca.markets/us/docs/crypto-fees.
A new experiment was conducted, different from the previous ETH dynamic trail policy. The experiment used fixed entries based on the highest price in the last 15, 20, or 30 days, and exits based on the lowest price in the same period. The experiment also used a stop-loss of 2 times the average true range (ATR) of the last 14 days, and a target of 3 times the ATR. The experiment was run on the SOL cryptocurrency, with a training period from 2024 to 2025, and a holdout period from January 1, 2026, to September 11, 2026. The results showed a return of -18.39% and a drawdown of 31.66% for the daily timeframe, and a return of -20.85% and a drawdown of 50.04% for the hourly timeframe. The experiment also tested the effect of slippage, with results ranging from -17.29% to -16.96% for the daily timeframe, and -7.23% to -14.59% for the hourly timeframe. A bootstrap analysis was also conducted, with results showing a 95% confidence interval of -8.46% to 7.78% for the daily timeframe, and -7.54% to 6.23% for the hourly timeframe. The conclusion is that this experiment is for research purposes only, and no new deployment will be made based on these results. The data used in this experiment came from https://runway.adityaperswal.workers.dev/data/crypto-bars and https://docs.alpaca.markets/us/docs/crypto-fees.
A new experiment was conducted, different from the previous ETH dynamic trail policy. The experiment used fixed entries based on the highest price in the last 15, 20, or 30 days, and exits based on the lowest price in the same period, with a stop-loss of 2 times the average true range (ATR) and a target of 3 times the ATR. The experiment also included a 20-day timeout and fees of 25 basis points per trade, with slippage of 10-20 basis points. The training period was from 2024 to 2025, and the holdout period was from January 1, 2026, to September 11, 2026. The results showed that the static policy performed differently from the frozen dynamic policy, with a return of -19.88% and a drawdown of 34.13% for the daily trades, and a return of 145.31% and a drawdown of 42.60% for the hourly trades. The stress test results showed a return of -20.07% and a drawdown of 34.31% for the daily trades, and a return of -14.18% and a drawdown of 29.36% for the hourly trades. The results also included a bootstrap analysis with 10,000 resamples, which showed a 95% confidence interval of -8.24% to 5.98% for the daily trades, and -5.96% to 4.74% for the hourly trades. The conclusion is that this experiment is for research purposes only and will not be used for new deployments. All 192 cells were recorded as observations. The data sources are https://runway.adityaperswal.workers.dev/data/crypto-bars and https://docs.alpaca.markets/us/docs/crypto-fees.
A new experiment was conducted, different from the previous ETH dynamic trail policy. The experiment used fixed prior 15, 20, and 30-day high entries, half-lookback low exits, next exact open, prior completed daily Average True Range (ATR) of 14, 2ATR stop, and a net 3R target, with a 20-day timeout. The experiment also included 25 basis points credited asset fees on each side and 10/20 basis points slippage. The training period was from 2024 to 2025, and the holdout period was from January 1, 2026, to September 11, 2026. The current top 10 universe introduces survivorship, meaning that only assets that have survived the entire period are included. Previous market outcomes were already examined. The results show that the static policy differs materially from the frozen policy with +2R management. The results are as follows: for the 20-day baseline daily, the return percentage was 0.06944554721801666, the drawdown percentage was 13.1394448010398, and the win rate was 33.33333333333333. The training results for the 20-day period were: for the daily timeframe, the return percentage was -28.990451727423896, the drawdown percentage was 46.43023098987995, and the win rate was 37.5; for the hourly timeframe, the return percentage was -57.911572116229735, the drawdown percentage was 67.93922022878127, and the win rate was 20. The stress results for the 20-day period were: for the daily timeframe, the return percentage was 0.11860763659721218, the drawdown percentage was 13.215285984363245, and the win rate was 33.33333333333333; for the hourly timeframe, the return percentage was -18.647679160477047, the drawdown percentage was 20.354148546152285, and the win rate was 16.666666666666664. The holdout results are also provided. A trade-mean bootstrap with 10,000 resamples and a 95% confidence interval was conducted, but dependence is unresolved. The conclusion is that this is research only, and no new deployment will be based on this comparator. All 192 cells were recorded as observations. The source of the data is https://runway.adityaperswal.workers.dev/data/crypto-bars and https://docs.alpaca.markets/us/docs/crypto-fees.
A new experiment was conducted, different from the frozen ETH dynamic trail policy. The experiment used fixed prior 15/20/30 day high entries, half-lookback low exits, next exact open, prior completed daily Average True Range (ATR) of 14, 2ATR stop and net 3R target, and a 20-day timeout. A 25 basis point credited asset fee was applied to each side, and 10/20 basis point slippage was used. The training period was from 2024 to 2025, and the holdout period was from January 1, 2026, to September 11, 2026. The current top 10 universe introduces survivorship, meaning that only assets that have survived the previous market outcomes are included. Intrabar exits were idealized, with stops triggered first on dual touch, gaps at open, and no entry-bar exit. The results were not validated for live 15-minute execution. Missing hours were reindexed and required complete lookback, potentially removing trades, but this was not a full-coverage control. The raw and vectorbt initialization cash reconciliation passed. The premises of the experiment were that a static policy differs materially from a frozen policy with 2R management, and that small samples and known holdout cannot establish a new live edge. The 20-day baseline daily results for ETH were: return percentage of -18.706110570684498, drawdown percentage of 27.007363205505662, 6 trades, win rate of 16.666666666666664, mean trade of -3.1991486600912866, average win of 9.496584954658527, average loss of -5.738295383041249, 0 targets, 0 reconciliations, and 0 missing values. The training results for 20-day ETH were: return percentage of 42.244090728417326 for daily trades, and 83.73109085236139 for hourly trades. The stress results for 20-day ETH were: return percentage of -19.41600462356944 for daily trades, and 9.876001632725394 for hourly trades. All holdout cells were recorded, with results including: daily trades with 15, 20, and 30-day lookbacks, and hourly trades with 15, 20, and 30-day lookbacks. The trade-mean bootstrap with 10,000 IID resamples and 95% confidence interval was unresolved due to dependence. The conclusion is that this is for research purposes only, and no new deployment will be based on this comparator. All 192 cells were recorded as observations. The source of the data is https://runway.adityaperswal.workers.dev/data/crypto-bars and https://docs.alpaca.markets/us/docs/crypto-fees.
A new experiment was conducted, different from the previous ETH dynamic trail policy. The experiment used fixed entries based on the highest price in the last 15, 20, or 30 days, and exits based on the lowest price in the same period, with a stop-loss and target of 2 times the average true range (ATR) of the last 14 days. The experiment was run on Bitcoin (BTC) data from 2024 to 2025 for training and from January 1, 2026, to September 11, 2026, for testing. The results showed a return of -7.94% and a drawdown of 21.65% for the daily timeframe, and a return of -35.87% and a drawdown of 37.86% for the hourly timeframe. The experiment also tested the effect of different slippage levels, with results ranging from -7.94% to 6.79% return. A bootstrap analysis was conducted to estimate the confidence interval of the results, which showed a range of -5.65% to 4.93% for the daily timeframe and -2.38% to 6.43% for the hourly timeframe. The conclusion is that this experiment is for research purposes only and will not be deployed as a new trading strategy. The data used for this experiment came from https://runway.adityaperswal.workers.dev/data/crypto-bars and https://docs.alpaca.markets/us/docs/crypto-fees.