A news scan found the headline about a retrospective review of the C91200 trading fund's Friday ETH entry replay. This review is a limited path replay, reconstructed after observations, and not a prospective out-of-sample or full-policy validation. The signal was frozen on September 11 at 13 UTC with a close price of 2628.4075, which was above the known 14 UTC price of 2565.9. The next open price was 2629.65, and the entry price was 2632.27965 after a 10 basis point slip. The average true range over 14 days from September 10 was 88.0134936902. The initial stop was set at 2456.25266262, and the target was 3223.79684108, which would result in a net 3R with 25 basis points credited fees on each side and a 10 basis point exit slip. The replay was completed over 40 15-minute bars to 23:45. The peak price was 2664.8535, and the low was 2506.24, resulting in a 2R gain, and the initial stop and target were not touched. No dynamic transition was needed in this interval. The vectorbt portfolio was initialized with a cash value of 1, an effective entry price of fill divided by 0.9975, a quantity of 1 divided by the effective price, and a fee of 0 because it was encoded in the price. The raw wealth equality had an error of 0. The mark was -4.653905788%, the hypothetical liquidation was -4.987378753%, and the close drawdown was -4.079390856%. There was 1 open trade and 0 closed trades, resulting in an undefined win rate. At 23:45, the return was -0.6602036R, and the 20-day horizon on October 1 at 14 UTC had not elapsed. There was no observed stop break, and there was no reason to tighten a hypothetical red trade. A prior illustration with a price of 29.4825 resulted in a mark of -1.372088 and a net stop risk of 2.136323 USD, but the actual result was 0. The strongest counterargument is that a cheaper late entry may work, but it has untested conditioning. It is recommended to avoid recasting the missed signal as an avoided completed loss. The source of the data is https://runway.adityaperswal.workers.dev/data/crypto-bars, and the observation number is 25230.
A news scan found the text of an automated trading fund manager's notes. The manager did not do a new historical train and holdout sweep. They read the original data from 2605, 2606, and 2469 with a query for research on a crypto fund. The manager used frozen prior 480 hour high and 240 hour low, 2 ATR 14, and net 3R, with a 20 day timeout and a 2R cost BE, and 2 ATR trail. The next exact open had 25 basis points credited fees and 10 basis points slip, and 20 basis points stress. ATR stands for Average True Range, a measure of price volatility. The manager trained the system from 2024 to 2025 and used the period from January 1, 2026, to September 11, 2026, as a holdout. The hourly ETH train had a 33.250114 percent N 19, and the test had a 17.240439 percent DD 19.830197 percent win 4/6, with a stress of 15.841958 percent. The daily 20 high and 10 low train had a 2.029153 percent N 15, and the test had a -17.632466 percent DD 27.353641 percent win 1/5. The prior 15 minute test had a 22.171852 percent N 6, with a stress of 20.714547 percent. The policies for daily and hourly trading differ, and a daily failure does not automatically veto an hourly watch. The manager reported a prior raw and vectorbt init cash 1 reconciliation of 2.45e-15. The premises are that six reused trades leave expectancy uncertain, a later Friday signal was observed but its next open passed, live plus 2R activation and delayed stop fills remain different from the simulated policy. The conclusion is that there will be no retroactive entry or retuning. The source of the data is https://runway.adityaperswal.workers.dev/data/crypto-bars and https://docs.alpaca.markets/us/docs/crypto-fees. The existing sample is not counted as fresh validation.
A news scan found the headline 'C91200 CLI close-breakout diagnostic remains unsuitable for orders'. A computer program called CLI, short for Command Line Interface, was run on September 12 at 00:02 UTC. The program tested a trading strategy called breakout, which means buying when prices break out to new highs, on three cryptocurrency pairs: BTC/USD, ETH/USD, and SOL/USD. The test used 730 days of data and 20 days of lookback, meaning it looked at the last 20 days to make decisions. The results were: BTC had a return of 25.666493%, a drawdown of 13.536724%, and a win rate of 32.653061%; ETH had a return of 28.850813%, a drawdown of 24.943256%, and a win rate of 21.052632%; and SOL had a return of 80.474020%, a drawdown of 18.795394%, and a win rate of 53.125%. The test assumed that trades were filled at the same closing price, and that the breakout and signal exit were based on the prior closing price. It also assumed zero explicit costs and allowed partial fills on the current day. However, these assumptions differ from the actual execution, which uses the highest and lowest prices, known as high and low, and has a net return of three times the risk, referred to as 3R. The conclusion is that these results are only for diagnostic purposes and should not be used for actual trading or to calculate the optimal bet size, known as the Kelly criterion, based on the gross returns. The data came from https://runway.adityaperswal.workers.dev/data/crypto-bars and the code is located at /app/agent/dist/backtest/{engine,bars,cli}.js. The full diagnostic observation is numbered 25231-33, and a separate audit of the costed experiment was retained.
A review of existing experiments 2605 and 2606, and CLI 2545, was conducted via a query of the research fund for crypto. No new backtest was run, in order to preserve the 2024-25 training and the reused 2026 January 1 to September 11 00 UTC test. The experiment used a hourly frozen 480 hour high and 240 hour low, with a 2 ATR 14 risk, a net 3R target, a 20 day timeout, and a +2R cost BE and peak -2 ATR trail. The next exact open execution was used, with a 25 basis point credited asset fee and a 10 basis point slip, and a 20 basis point stress fixed target. The Vectorbt initialization cash was 1 and the raw reconciliation was less than or equal to 2.45e-15, as reported in the original 48 cells. The ETH hourly training resulted in a 33.250114% return, with 19 trades, and the test resulted in a 17.240439% return, with a 19.830197% drawdown, a 66.6667% win rate, and 6 trades. The stress test resulted in a 15.841958% return. The daily 20 high and 10 low comparator training resulted in a 2.029153% return, with 15 trades, and the test resulted in a -17.632466% return, with a 27.353641% drawdown, a 20% win rate, and 5 trades. Different policies were used, with no automatic daily veto or agreement. The CLI gross return was 28.905427%, with a 24.943256% drawdown, a 21.6216% win rate, and 37 trades, but this did not include fees or holdout, and signal close fills were used. The current completed entry was false, and no new rule was traded. There were unknown historical bid ask prices, and exact venue 15 minute trail activation and 2025 anomalies remained. This record is an evidence audit, and not an additional independent sample, and no reoptimization was done. The source of the data is https://runway.adityaperswal.workers.dev/data/crypto-bars, and the original records are 2545, 2605, and 2606.
A news scan found the text of a report on an automated trading fund manager's notes for C91106 SOL. The report discusses the fund's hourly frozen net replication and cost stress. The fund uses a premise with a frozen rolling 480 hour high and 240 hour low, and an ATR14 reconstructed from full 24 hour UTC days. The fund has a 2 ATR risk, cost-aware 3R target, 20 day timeout, and a peak 2 ATR trailing target. The report states that signals occur after a completed bar, at the next exact open, and exits are sampled at hourly opens. The fund credits a 25 basis point fee per side and 10 basis point lippage baseline, with a 20 basis point stress that freezes the baseline target. The report also discusses the fund's training and holdout returns, with a training return of -0.4796588637% and a holdout return of 1.1724554945%. The report concludes with a research rejection due to negative hourly training, stress, and prior 15 minute execution test results, resulting in no order. The report cites sources https://runway.adityaperswal.workers.dev/data/crypto-bars and https://docs.alpaca.markets/us/docs/crypto-fees, and notes observations 24627-24674 and specification 2231.
A news scan found the text of a daily report from an automated trading fund manager for SOL. The report has three premises. Premise 1 describes the trading strategy, which uses a 20-day high and low, a 14-day average true range, or ATR, which is a measure of price volatility, and a 3R target, where R is a risk unit. The strategy also includes a 20-day timeout and a 2R cost, with a peak 2ATR trailing stop. The target is retained until the next exact open. Exits are sampled at daily opens, and missing opens are skipped. The report notes that stop fills are not guaranteed. The strategy credits a 25 basis point fee per side, which is a 0.25% fee for buying or selling, and a 10 basis point lippage baseline, which is a 0.10% expected price difference. A 20 basis point stress freezes the baseline target. The report uses a cash-only vectorbt, which is a backtesting library, with an initial cash value of 1.0. Premise 2 compares the training period from 2024 to 2025 with a holdout period from 2026 January 1 to September 11. The training return was 29.23%, the maximum drawdown, or DD, was 27.02%, and there were 9 trades. The holdout return was -8.22%, the DD was 31.79%, and there were 6 trades, with a win rate of 16.67% and a mean return of -0.64%. The stress return was -9.31%, the static return was -8.22%, and the buy and hold, or BH, return was -18.41%. Premise 3 notes that all current frozen entries are false, and daily and hourly policies are distinct, with no equal return requirement. The report concludes that the research is rejected due to negative hourly training, negative stress, and negative prior 15-minute execution test results. The accounting is valid within assumptions, but there are limitations due to unknown historical bid-ask spreads, 2025 bar anomalies, small sample size, and off-run exact trail implementation. The sources of the data are https://runway.adityaperswal.workers.dev/data/crypto-bars and https://docs.alpaca.markets/us/docs/crypto-fees. The observations are from 24627 to 24674, and the specification is 2231.
A news scan found the report of an automated trading fund manager. The report discusses the performance of an Ethereum trading strategy. The strategy uses a 480-hour high and 240-hour low, with a 14-day average true range, or ATR, which is a measure of price volatility. The strategy aims to make a 3R return, which means making three times the risk amount, with a 20-day timeout and a 2R cost basis, or breakeven, and a peak 2ATR trailing stop. The strategy enters trades at the next hourly open after a signal, and exits are sampled at hourly opens, with missing opens skipped and no guaranteed stop fills. The strategy credits a 25 basis point fee per side, which is a 0.25% fee, and 10 basis points of lippage, which is a measure of the difference between the expected and actual price of a trade, as a baseline, with a 20 basis point stress test that freezes the baseline target. The strategy uses a cash-only vector and initializes with a cash amount of 1.0, with raw wealth and closed trade fee denominators reconciled to less than 1e-10 across 48 cells, with a maximum of 2.45e-15. The report compares the performance of the strategy in two time periods: from 2024 to 2025, and from January 1, 2026, to September 11, 2026. The training return was 33.25%, with a maximum drawdown of 39.72%, and 19 trades. The holdout return was 17.24%, with a maximum drawdown of 19.83%, and 6 trades, with a win rate of 66.67% and a mean return of 2.87%. The stress test return was 15.84%, with a static return of 17.24%, and a buy-and-hold return of -17.60%. The report notes that all current frozen entries are false, and that daily and hourly policies are distinct, with no equal return requirement. The report reproduces a prior return of 17.24% with six trades exactly, but notes that repeated calculation is not new independent proof. The report concludes that it is for research watch only, and no order should be placed. The report notes that the accounting is valid within assumptions, but that unknown historical bid-ask spreads, 2025 bar anomalies, small sample size, and off-run exact trail implementation limit the soundness of the results. The report cites two sources: https://runway.adityaperswal.workers.dev/data/crypto-bars and https://docs.alpaca.markets/us/docs/crypto-fees. The report is based on observations from 24627 to 24674, with a specification of 2231.
A news scan found the report of an automated trading fund manager for C91106 ETH. The report discusses the daily frozen net 3R replication and cost stress. The premise uses a native daily high and low with a 14-day average true range, or ATR, which is a measure of volatility. The cost-aware target has a 20-day timeout and a 2R cost breakeven, or BE, with a peak 2ATR trailing stop. The target is retained and the signal is given after the bar is completed, with the next exact open. Exits are sampled at daily opens, and missing opens are skipped, with no guaranteed stop fills. The credited asset has a 25 basis point fee per side, which is a measure of the cost, and a 10 basis point lippage baseline, which is a measure of the difference between the expected and actual price. A 20 basis point stress freezes the baseline target. The cash-only vector backtest, or bt, has an initial cash of 1.0, and the raw wealth and closed trade fee denominators are reconciled to be less than or equal to 1e-10 across 48 cells, with a maximum of 2.45e-15. The training period is from 2024 to 2025, with a holdout period from 2026 January 1 to September 11 00 UTC. The training return is 2.0291530569%, with a drawdown, or DD, of 34.9426662999%, and 15 trades. The holdout return is -17.6324659487%, with a DD of 27.3536412981%, a win rate of 20%, and 5 trades, with a mean return of -3.5218771014%. The stress return is -18.452038509%, the static return is -17.6324659487%, and the buy and hold, or BH, return is -16.7705527594%. All current frozen entries are false, and daily and hourly policies are distinct, with no equal return requirement. This report is for research watch only, and no order should be placed. The accounting is valid within assumptions, but there are limitations due to unknown historical bid ask, 2025 bar anomalies, small sample size, and off-run exact trail implementation. The source of the data is https://runway.adityaperswal.workers.dev/data/crypto-bars and https://docs.alpaca.markets/us/docs/crypto-fees. The observations are from 24627 to 24674, and the specification is 2231.
A news scan found the report: Premise 1 uses a frozen rolling 480 hour high and 240 hour low, with a 14-day Average True Range (ATR), which measures price volatility, reconstructed from full 24 hour UTC days. It has a 2 ATR risk, cost-aware 3R target, 20 day timeout, +2R cost breakeven (BE) and peak-2 ATR trailing, target retained. Signals occur after a completed bar, at the next exact open, and exits are sampled at hourly opens, with missing opens skipped and no guaranteed stop fills. The report credits a 25 basis point fee per side and 10 basis point lippage baseline, with a 20 basis point stress that freezes the baseline target. The vectorbt init cash is 1.0, with raw wealth and closed trade fee denominators reconciled to less than or equal to 1e-10 across 48 cells, with a maximum of 2.45e-15. Premise 2 compares the chronological 2024-25 train, with SOL available from August 26, 2024, to the reused 2026 January 1 to September 11 00 UTC holdout, with a flat reset. The training return is 20.5581742344%, with a drawdown (DD) of 28.4815768457% and 19 trades. The holdout return is 7.6357530493%, with a DD of 12.8707598205%, a win rate of 50%, 6 trades, and a mean of 1.4019896435%. The stress is 6.3518399027%, the static is 7.6357530493%, and the buy and hold (BH) is -12.2163852845%. Premise 3 states that all current frozen entries are false, with daily and hourly distinct policies and no equal return requirement. The conclusion is for research watch only, with no order. The report notes valid accounting within assumptions, but unknown historical bid ask, 2025 bar anomalies, small N, and off-run exact trail implementation limit soundness. The sources are https://runway.adityaperswal.workers.dev/data/crypto-bars and https://docs.alpaca.markets/us/docs/crypto-fees, with observations from 24627 to 24674 and specification 2231.
A news scan found the headline 'C91106 BTC daily frozen net3R replication and cost stress'. The report discusses a trading strategy for Bitcoin (BTC) with specific rules: it freezes the native daily high and low, uses a 14-day average true range (a measure of price volatility) to set a 2 times risk target, and has a 20-day timeout. The strategy also includes a cost-aware target return and a peak 2 times average true range trailing stop. The report notes that signals are generated after a completed bar, and exits are sampled at daily opens, with missing opens skipped and no guaranteed stop fills. It also mentions that a 25 basis point fee per side and 10 basis point lippage (the difference between the expected price and the actual price) are credited to the asset, with a 20 basis point stress freeze on the baseline target. The report references two premises: the first discusses training data from 2024 to 2025, with a return of 1.4238283647% and a drawdown (the maximum loss from a peak to a trough) of 32.3728386495%, and the second discusses holdout data from 2026, with a return of -3.355266979% and a drawdown of 21.0292788335%. The report concludes that all current frozen entries are false, and that daily and hourly policies are distinct, with no equal return requirement. It also notes that this research is for watching only, and no orders should be placed based on it. The report cites two sources: https://runway.adityaperswal.workers.dev/data/crypto-bars and https://docs.alpaca.markets/us/docs/crypto-fees, and references observations 24627-24674 and specification 2231.
A test was run on September 11 at 06:02 UTC using a computer program called node and a file named backtest.js. The test looked at the prices of three cryptocurrency pairs, BTC/USD, ETH/USD, and SOL/USD, over 730 days using a strategy called breakout. The results were: BTC had a return of 25.66649%, a drawdown of 13.53672%, a win rate of 32.6531%, 49 trades, and a buy and hold return of 34.63505%; ETH had a return of 28.90543%, a drawdown of 24.94326%, a win rate of 21.62162%, 37 trades, and a buy and hold return of 5.45225%; SOL had a return of 80.47402%, a drawdown of 18.79539%, a win rate of 53.125%, 32 trades, and a buy and hold return of -24.66375%. This test, called Premise1, assumed perfect trading with no fees or slippage, and ideal stop levels. Another test, Premise2, used a different method with hourly high and low prices and accounted for costs. The conclusion is that these results are not suitable for making actual trades, but they do satisfy a required diagnostic test. A separate study using a tool called vectorbt is also being run. The data used for this test can be found at https://runway.adityaperswal.workers.dev/data/crypto-bars, and information on fees can be found at https://docs.alpaca.markets/us/docs/crypto-fees.
A news scan found the text of a report from an automated trading fund manager. The report discusses three premises. Premise 1 uses a 20-day high and 10-day low, with a 14-day average true range, or ATR, which is a measure of volatility, for risk management. It also has a 3R target, which refers to a three times risk reward ratio, and credits a 25 basis point, or 0.25%, asset fee, with a 10 basis point, or 0.10%, slip, and a 20-day timeout. The signal is known after the bar is completed, and the next open is exact. There is a 2R cost basis, or break-even point, and a peak minus 2 entry ATR trail, with a retained target. A 20 basis point slip stress leaves the target fixed. The report uses a native calendar-hourly timeframe, which is based on the actual calendar hours, and a rolling-hourly timeframe, which looks back 480 hours. Premise 2 uses a chronological train from 2024 to 2025, and a reused test from January 1, 2026, to September 11, 2026, with a reset to flat, or zero, at the start of the test. The report also discusses hourly and daily open sampled exits, with no guarantee of stop price, and raw wealth reconciliation using vectorbt, a backtesting library, with a precision of 1e-10. The full native calendar-hourly train has a return of 21.988671%, with 12 trades, and the holdout, or test, period has a return of -10.135134%, with a maximum drawdown of 35.033237%, and a win rate of 28.571429%, with 7 trades. The same-target stress test has a return of -11.384479%. Premise 3 discusses attribution, which refers to the process of assigning returns to specific factors, with a ladder of full native daily and daily, daily and hourly, and calendar and hourly timeframes. The report also discusses matched-full 24-hour daily and daily, and rolling 480-hourly timeframes. The matching process drops the incomplete hour day on March 12, and suppresses later entries, which is a selection effect, not a price alpha, or return, effect. The native versus aggregate matched 2026 returns coincide, but the source construction is not the explanation for the ladder. The report concludes that positive matched variants cannot replace previously frozen rejected policies, and there is no switching based on this reused sample. The current 20-day entry is absent. The validity of the report is based on controlled changes that attribute differences within implementation, and the soundness is limited by tiny reused samples, historical bar anomalies, and sampled quotes. The report uses 80 cells per asset, with 168 matched and 72 full native, for a total of 240, but these are not 80 independent trials. The source of the data is https://runway.adityaperswal.workers.dev/data/crypto-bars, and the method notes are 2175/2180.
A news scan found the text of a report from an automated trading fund manager. The report discusses the performance of a trading strategy, called Premise1, which uses a 20-day high and 10-day low, with a 14-day average true range, or ATR, for risk management, and a net target of 3R, with a 25 basis point credited asset fee and a 10 basis point slip. The strategy has a 20-day timeout and a signal is known after a completed bar, with the next exact open. The report also mentions a cost basis, or BE, of +2R and a peak entry ATR trail, with a retained target. A 20 basis point slip stress leaves the target fixed. The report uses a native calendar-hourly timeframe, with 20 completed UTC days, and a rolling-hourly prior 480 hours. The report is based on two premises, or assumptions, with the first being a chronological train from 2024 to 2025, and a reused holdout from 2026 January 1 to September 11, at 00 UTC, with a reset to flat. The report also discusses the performance of the strategy, with a 41.228668% return on the full native calendar-hourly train, and a 7.324178% return on the holdout, with a 42.857143% win rate and a 19.830197% drawdown. The report also mentions a same-target stress test, with a 5.832105% return. The report breaks down the attribution of the strategy, with a full native calendar-hourly train, and a daily and hourly open sampled exit, with no guarantee of stop price. The report also mentions a raw wealth reconciliation, using a vectorbt 1e-10. The report concludes that the validity of the results is due to controlled changes, and that the soundness of the results is limited by tiny reused samples, historical bar anomalies, and sampled quotes. The report is based on 80 cells per asset, with 168 matched and 72 full native, for a total of 240, but notes that these are not 80 independent trials. The source of the data is https://runway.adityaperswal.workers.dev/data/crypto-bars, and the method notes are 2175/2180.
A news scan found the text describing an automated trading fund manager's notes. The notes discuss a trading strategy for Bitcoin (BTC) with specific rules. The strategy has a 20-day high and 10-day low premise, with a 14-day average true range (ATR) risk, which is a measure of price volatility. The target return is 3 times the risk, with a 25 basis point (0.25%) asset fee and a 10 basis point (0.10%) slip, which is the difference between the expected and actual price. The strategy also has a 20-day timeout. The signal to buy or sell is known after the bar is completed, and the next exact open price is used. The strategy also includes a 2 times risk cost and a peak minus 2 times ATR trail, which is a way to adjust the stop-loss price. The notes also discuss the training and testing of the strategy using historical data from 2024-25 and 2026. The results show a return of 30.44% for the training period and 2.08% for the testing period, with a win rate of 42.86%. The notes also discuss the attribution of the returns to different factors, including the native calendar-hourly and daily returns. The conclusion is to watch the strategy, but not to switch to it based on the results. The validity of the results is limited by the small sample size and historical bar anomalies. The source of the data is https://runway.adityaperswal.workers.dev/data/crypto-bars, and the method notes are numbered 2175/2180.
A news scan found the headline about a built-in CLI breakout diagnostic that is not executable replication. The diagnostic ran a test using the command node /app/agent/dist/backtest.js with the following parameters: symbols BTC/USD, ETH/USD, and SOL/USD, strategy breakout, lookback 20, days 984, stop percentage 8, and target percentage 24. The results were: for BTC, 985 bars, return 46.46075, drawdown 14.69243, win 34.32836, and number of trades 67; for ETH, 985 bars, return 24.55541, drawdown 24.94326, win 29.09091, and number of trades 55; for SOL, 747 bars, return 48.32999, drawdown 18.61323, win 52.63158, and number of trades 38. The source code audit showed that the engine.js file enters a trade at the signal bar close and exits when the closes cease exceeding 20 prior closes, using the bar low and high as exact stop and target, with no explicit fee or slippage, and including the current partial day. The bars.js file is single page only. This means that the gross full sample CLI output cannot validate the 20 day high and 10 day low next open net 3R rule and is not out of sample. The positive headline returns were rejected as deployment evidence. A vectorbt study was recorded separately, with chronological costed daily and hourly matched input. The source of the data is https://runway.adityaperswal.workers.dev/data/crypto-bars and the files /app/agent/dist/backtest/{engine,strategies,bars}.js. The validity of the measures implemented in the CLI was confirmed, but the soundness for cash deployment failed due to execution, cost, and window premises.
A trading strategy called C91102 SOL has been tested. This strategy uses a 20-day high breakout and a 10-day low exit, with a 2ATR14 risk and a 20-day timeout. The target is to make a net return of 3 times the risk, considering 25 basis points of credited-asset fees and 10 basis points of slip. The strategy also includes a 2ATR absolute trailing stop. The premise of the strategy was trained on data from 2024 to 2025 and tested on data from January 1, 2026, to September 11, 2026. The results showed a return of -9.58405% and a maximum drawdown of 33.33683%. Another test with a 20 basis points execution stress showed a return of -10.84106%. The strategy did not have any current daily or hourly trigger. The conclusion is to reject the deployment of the strategy due to negative principal holdouts. The validity of the results depends on the arithmetic of the rules and costs. The soundness of the results is limited by the small sample size and missing or zero-volume bars. The source of the data is https://runway.adityaperswal.workers.dev/data/crypto-bars.
The automated trading fund manager has issued a report on the C91102 ETH strategy. This strategy involves a 20-day high breakout and a 10-day low exit, with a 2ATR14 risk and a 20-day timeout. The target is set to solve net3R at 25bp credited-asset fees and 10bp slip, with a +2R full-position breakeven and 2ATR absolute trailing. The report is based on two premises: Premise1, which used 2024-25 training data with 15min checks and resulted in a 76.33020% success rate, and Premise2, which used the same target with a 20bp execution stress and resulted in a 20.71455% success rate. Premise3 used holdout data and resulted in a mean of 3.56329% and an average win/loss of 6.88094%/3.07202%. The report also includes controls, such as a 15min no-trail return of 18.74207% and 10/30day neighbors with returns of 14.16928% and -20.01429%. The conclusion is to retain the research watch, despite daily failure and neighboring 30day failure, due to the absence of an entry bar deployment. The report notes that the validity is conditional on rule/cost arithmetic and that the soundness is limited by reused small samples, missing/zero-volume bars, and unverified live quotes/off-run dynamic trail. The report is based on data from https://runway.adityaperswal.workers.dev/data/crypto-bars.
A trading strategy called C91102 for Bitcoin, which uses a 20-day high breakout and a 10-day low exit, with a 2ATR14 risk management system and a 20-day timeout, has been evaluated. The strategy aims to achieve a net return of 3R, with a target that includes 25 basis points for credited-asset fees and 10 basis points for slippage. The strategy also includes a 2R full-position breakeven and a 2ATR absolute trailing stop, with a fixed target retained. The evaluation was based on three premises: Premise 1, which used 15-minute checks from 2024 to 2025 and showed a return of 9.17702%; Premise 2, which used a 20 basis point execution stress test and showed a return of -3.82831%; and Premise 3, which used a holdout period and showed a mean return of -0.12225%. The results also showed that the strategy did not have any current daily or hourly triggers. The controls used for comparison showed a return of -2.81737% for a 15-minute no-trail strategy, and 4.81000% and -15.25060% for 10-day and 30-day neighbor strategies, respectively. Based on these results, the conclusion is to reject the current deployment of the strategy due to negative principal holdouts. The validity of the results is conditional on the rule and cost arithmetic used, and the soundness of the results is limited by the small sample size and missing or zero-volume bars. The results are based on data from https://runway.adityaperswal.workers.dev/data/crypto-bars.
A test called C91102 CLI breakout smoke test was run. It used a computer program to simulate trading with the symbols BTC/USD, ETH/USD, and SOL/USD, using a strategy called breakout. The test looked back 20 days and ran for 730 days, with a stop loss of 6% and a target profit of 18%. The data for the test came from https://runway.adityaperswal.workers.dev/data/crypto-bars. The results were: for BTC, a return of 24.5117%, a drawdown of 15.4848%, a win rate of 32.6531%, and 49 trades; for ETH, a return of 38.7083%, a drawdown of 24.9433%, a win rate of 23.6842%, and 38 trades; and for SOL, a return of 25.1113%, a drawdown of 28.5687%, a win rate of 48.5714%, and 35 trades. Each symbol had 731 bars of data. A code audit found that the test used the same close for entry, the prior close high rather than the intraday high, and exits on the next bar unless there is another breakout. It also used ideal in-bar stop and target, assumed zero fees and slippage, did not have a separate holdout, and did not paginate when fetching data. This means the test used a different rule than the frozen 20/10 day channel and cannot validate current 15-minute execution. The test assumes that attractive gross returns are plausible diagnostic output, but notes that missing costs, timing, and holdout can reverse the edge. The test results are not suitable for use in placing orders. A separate study using vectorbt long only credit-fee train/holdout is underway. No gross CLI return was used for Kelly.
A news scan found the headline 'C91100 SOL/USD fixed 3R breakout: not robust after costs'. This refers to a trading strategy for the SOL/USD currency pair, where 3R breakout means a price movement that exceeds three times the average true range, a measure of price volatility. The strategy was tested on historical data from 2024 to 2025 and then applied to a holdout period from January 1, 2026, to September 10, 2026. The results showed that the strategy was not profitable after accounting for costs, such as fees and slippage, which is the difference between the expected and actual price of a trade. The strategy had a return of 7.389004% during the training period, but a loss of 2.192042% during the holdout period. When stress-tested, the strategy had a loss of 3.358726%. The daily holdout period showed a loss of 20.758684%, and when the strategy was reconstructed, the loss was 7.586258%. The results suggest that the strategy is not currently effective, and the allocation to this strategy should be rejected. The validity of the results is conditional on the accuracy of the sampled quotes and the cost model used. The soundness of the results is limited by the reuse of the holdout period, missing bars, correlated trades, and different venue exit cadence. The strongest counterexample to the results is that a new regime or execution can change the results, but this is not evidence of a current entry. The mathematical half-Kelly criterion, which is a measure of the optimal fraction of wealth to allocate to a trade, is 0.0019008411572001005, but this is not an authorized allocation. The portfolio totals remain unchanged after an equity-denominator correction, and all 52 saved holdout trades reconcile to 2.84e-14pp. The sensitivity of the results to changes in the neighbors, which are the surrounding data points, is only considered for 10 and 30-day periods. The source of the data is https://runway.adityaperswal.workers.dev/data/crypto-bars.
A news scan found the headline about an automated trading fund manager's notes. The manager is watching ETH/USD with a fixed 3R breakout strategy, which is currently positive on an hourly basis, but there is no entry signal. The strategy has rules such as closing a position if the price is higher than the prior 20-day high, and exiting if the price is lower than the prior 10-day low. It also uses a 2 ATR 14 stop and a 6 ATR target, with a 20-day timeout. The manager trained the model from 2024 to 2025 and tested it from January 1, 2026, to September 10, 2026. The results show a 61.457476% return on the training data, 23.420985% on the holdout data, and 21.948781% under stress. However, there is currently no daily or hourly signal to enter a trade. The manager will retain this strategy as a candidate for further validation, but will not place an order. The validity of the results depends on the accuracy of the sampled quotes and the cost model used. The strongest counterexample that could change the results is a new regime or execution, but this does not provide evidence for a current entry. The mathematical half Kelly value of 0.26153098515283013 is an approximation of the binary-average payoff, but it is not an authorized allocation. The portfolio totals remain unchanged after an equity-denominator correction, and all 52 saved holdout trades reconcile to 2.84e-14pp. The manager also checked the sensitivity of the results over 10 and 30 days. The data source is https://runway.adityaperswal.workers.dev/data/crypto-bars.
A news scan found the headline about a BTC/USD fixed 3R breakout with a positive hourly watch, but no current entry. Before results, the close price was higher than the prior 20-day high, and the exit close price was lower than the prior 10-day low. The strategy used a 2 ATR 14 stop and a 6 ATR target, with a 20-day timeout, and the next actual open price. This was based on hourly rolling 480 and 240 hours, with completed daily ATR known before use. There were no partials or trailing stops, so this is not the current live exit policy. The analysis used Vectorbt long only, with fees of 25 basis points on the received asset and a 10 basis point slip on the base 20 stress. The data was trained from 2024 to 2025 and held out from January 1, 2026, to September 10, 2026, which has already been reused by broader research, with no parameter selection. The premises had a 1 hourly train of 18.046070%, an N19 holdout of 5.762346%, and a stress of 4.291986%. The supplied daily holdout was -5.685453%, and the matched and reconstructed holdout was 4.831708%. The mask control isolates the coverage effect. There is currently no daily or hourly signal, and the N7 holdout mean confidence interval is [-3.476200, 5.778872]%, which crosses 0. The conclusion is to retain this as a candidate for independent forward and 15-minute trailing/net 3R validation, but with no order. The validity is conditional on sampled quotes and the cost model, and the soundness is limited by the reused holdout, missing bars, correlated trades, and different venue exit cadence. The strongest counterexample is a new regime or execution that can change the results, but this is not evidence of a current entry. The mathematical half Kelly is 0.06392224669948926, which is a binary-average payoff approximation, but not an authorized allocation. The portfolio totals are unchanged by the equity-denominator correction, and all 52 saved holdout trades reconcile to 2.84e-14pp. The neighbors have a 10/30-day sensitivity only. The source of this information is https://runway.adityaperswal.workers.dev/data/crypto-bars.
A computer program executed a test of a trading strategy called breakout on three cryptocurrency pairs: BTC/USD, ETH/USD, and SOL/USD. The test used 731 days of data and a 20-day lookback period. The results showed gross returns of 25.66649% for BTC, 26.22267% for ETH, and 55.50067% for SOL. However, the test did not account for fees or slippage, and the entry and exit rules were based on the same closing price. Because of this, the positive results do not necessarily mean the strategy is good. The test was run to check the computer program's performance, and the results are limited by the way the data was put together. A different version of the test that includes costs and uses the next day's opening price could potentially lose money, even if the gross return is positive. The full results are available in a file called cli.json. The data used for the test came from https://runway.adityaperswal.workers.dev/data/crypto-bars.
A trading fund manager ran a test of a breakout strategy on a computer program called CLI. The test used data from the last 730 days and looked back 7 days to make decisions. The results showed that the strategy would have made 30.1279% profit with Bitcoin, 33.8668% with Ethereum, and 23.4302% with Solana, with a total of 214 trades. The manager also looked at the engine that made the trades and found that it used a daily same-close entry, did not account for fees or slippage, and did not hold out any data for testing. The manager concluded that these results cannot be used to confirm the success of the strategy or an order. The manager also noted that the results are not adjusted for costs and are only for diagnostic purposes. A separate test of a buy-and-hold strategy showed profits of 35.4970% with Bitcoin, 3.4305% with Ethereum, and a loss of 25.4054% with Solana. The data used for the test can be found at https://runway.adityaperswal.workers.dev/data/crypto-bars and the code for the engine is at /app/agent/dist/backtest/engine.js.
A news scan found the note: C91008 SOL compression breakout fails daily/hourly validation. The strategy was to enter a trade when the price closed above the prior 7-day high, but only if the prior 7-day fractional range was less than the trailing 60-day median. The plan was to enter the trade at the next open, with a stop loss at 1 times the average daily range of the last 20 days and a target at 1.5 times the average daily range of the last 20 days, with a 7-day timeout. The daily and hourly close threshold observations were followed by next-open fills, with one lot and no filled-forward execution. The fees were 25 basis points per side on the received asset, with a slippage of 10 basis points as the baseline and 20 basis points as the stress test. The training data was from 2025, the holdout data was from January 1, 2026, to September 9, 2026, and the warmup data was from October to December 2024. The premises were: 1. The supplied daily and hourly holdout returns were -3.5641% and -6.6888%, and the hourly training return was -10.9322%. 2. The reconstructed daily and hourly holdout returns were 2.8607% and -1.9359%, and changing the completeness of the data changed the sample, not the pure price attribution. 3. All 36 hourly holdout cells across 3 majors, 40, 60, and 80 median windows, two histories, and two costs lost. The conclusion was to reject the strategy because there was no positive independent executable edge. The baseline supplied hourly data showed 8 trades, with a win rate of 50.0000%, an average win of 3.8428%, an average loss of 5.3172%, and a mean return of -0.7372%, with a 95% confidence interval of -4.2210% to 2.6310% using 10,000 independent and identically distributed bootstrap samples. The windows and costs were diagnostics, not selected winners, and history faults, reused holdout, and 15-minute plus 1R exit gaps remained. All 144 cells were stored in 24 groups from 21505 to 21550. There is no current signal. The source of the data is https://runway.adityaperswal.workers.dev/data/crypto-bars, and the test and verification scripts are test.py and verify.py.