A news scan found the report: Premises: three $29.785 USD hypothetical quarter-fund probes total $89.355; synchronized observations preserve cross-coin dependence; only 36 complete known 2026 Sunday 08–24 UTC windows exist. Seed 91308 resamples 36 matched return vectors with replacement 10,000 times, computing each sample mean endpoint profit and loss, or PnL, which is a measure of the difference between the total value of assets and the total value of liabilities. The gross mean is −$0.18809 USD; the 95% bootstrap confidence interval, or CI, of the mean is [−$0.78683, +$0.40165] USD. The worst actual sample endpoint is −$4.18314 USD, which is not the worst possible path. This is a confidence interval for the mean, not a Monte Carlo terminal-loss interval or stop-sampled 20-day forecast. Endpoint 25 basis points, or bp, which is a unit of measurement, credited asset fees per side reduce coin means to −0.62053%/−0.70287%/−0.80307%; net wins are 30.56/33.33/30.56%, average wins are 1.1061/1.8399/2.1299%, and average losses are 1.3802/1.9742/2.0936%. The approximate binary mean half Kelly, a formula used to determine the optimal bet size, max(0, 0.5 * (p/l - (1-p)/w)) is 0 for each. No extra slippage, which is the difference between the expected price and the actual price, buy price uncertainty, serial week dependence, novel jumps, or intra-hour stop fills are modeled; this is descriptive cost sensitivity, not a new trading backtest. Conclusion: no sizing from unconditional window means; no conviction upgrade from small modeled losses. Actual size is 0 because 3208 signals are absent and frozen 3190–95 actual policy rejected. Source https://runway.adityaperswal.workers.dev/data/crypto-bars; fee assumptions https://docs.alpaca.markets/us/docs/crypto-fees.
The automated trading fund manager made some premises. These are that quiet weekend prices do not affect 20-day risk, and cross-coin paths must consider common shocks. The manager also decided that unconditional resampling does not validate an entry. The manager used 10,000 synchronized paths, which is called a seed, and 480 hours of data. The data was sampled in 24-hour blocks, starting on Sunday, across Bitcoin, Ethereum, and Solana. The manager trained the data from 2024-2025 and compared it to known data from 2026. The manager also held out 254 complete joint days of data. The data included a 10 basis point buy slip and a 1.001 initial spot, as well as 25 basis point credited asset fees per side and a 10 basis point sell slip. The current ATR20, which is a measure of volatility, was frozen at the September 12 value. The manager checked for exits every hour and every two hours, using a proxy, not the actual manager engine. The full net 3R targets required rises of 20.5241%, 24.8197%, and 34.5551%. The holdout full target probabilities were 5.73%, 9.77%, and 3.07%, with stops at 54.72%, 57.10%, and 45.70%. The means were -0.9304%, -0.6575%, and -1.2271%. The quarter at +2 gross R, stop entry, residual target solved for combined net 3R required gains of 23.3401%, 28.1130%, and 38.9300%. The final target probability was 3.56%, 6.92%, and 1.76%, with means of -0.9019%, -0.6034%, and -1.2217%. The approximate binary mean half Kelly was 0 for all three holdouts. The training positive Bitcoin or Solana plug-ins were not adopted. Three probes of $29.785 were made, totaling $89.355. The managed proxy mean PnL was -$0.8123, P1 was -$9.2638, and the worst sample was -$10.8530, which is a 1.275% account loss. It was noted that no loss greater than 3% in 10,000 paths is not a hard bound, and empirical blocks omit new jumps, structural changes, liquidity failure, and intrahour excursions. The individual managed P99 loss was $2.6158, $3.2233, and $4.2212. A counterargument was made that trend-conditioned fresh entries can outperform unconditional blocks, but these odds neither prove nor veto a valid conditional edge. The actual size was 0 because fresh entry was absent and the same policy manager holdout was weak. All 16 scenario rows were recorded. The data was sourced from https://runway.adityaperswal.workers.dev/data/crypto-bars and https://docs.alpaca.markets/us/docs/crypto-fees.
A computer program called NumPy with the seed 91304 created 10,000 examples of 20-day price paths for Bitcoin, Ethereum, and Solana from 235 overlapping time periods in 2026. It assumed equal initial investments and no changes to the investments. The results showed that the worst 1% of examples had a loss of 40.2133%, which would be a loss of $47.9102 for a full investment of $119.14 or $11.9775 for a quarter investment of $29.785. The daily price changes of these three cryptocurrencies were highly correlated, with correlation coefficients of 0.911160 for Bitcoin and Ethereum, 0.878579 for Bitcoin and Solana, and 0.879436 for Ethereum and Solana. The program assumed that risk increases over time and that these three cryptocurrencies are highly correlated. However, it noted that unexpected events like exchange failures could be worse. These scenarios are based on historical data and do not predict future outcomes. They also do not take into account factors like fees, stops, or rebalancing. A 25% joint crash would cost $29.785 for a full investment, which is 3.5% of the account, or $7.44625 for a quarter investment, which is 0.875% of the account. The program requires current executable entry and policy-specific stops before using size and should not be used to predict future quiet periods. The data source is https://runway.adityaperswal.workers.dev/data/crypto-bars.
A news scan found the text describing the C91302 paired-window weekend basket downside and crash boundary. The text mentions a NumPy seed, which is a way to get the same random numbers, and 10000 resampled contiguous 2-hour joint weekend return blocks with hypothetical 119.14 equal weights. The majors, which are the main items, had 1754 blocks with a loss of 2.332858% or -2.779367 USD for the first two hours. Seven had 1418 blocks with a loss of 1.856590% or -2.211941 USD. Comparing different windows would falsely suggest improvement. The same 1418 block window for majors had a loss of 1.636146% or -1.949304 USD, and for seven, the loss was -0.849714% and the gain was +0.992534%, compared to majors which had a loss of -0.791907% and a gain of +0.765652%. The text also mentions drawing samples using an independent random number generator, not paired Monte Carlo draw indices, with the date window matched. Seven is not automatically safer despite having a lower correlation with HYPE and BTC. There were no sample losses beyond 25.5297 USD, which is 3% of the account, and this is not a bound. The explicit 25% joint crash costs 29.785 USD, which is 3.50% of the account at full fund, and the quarter fund loses 7.44625 USD, which is 0.875% of the account. The text assumes no entry conditioning, fees, stops, slippage, or longer crisis, and the first hour can start on Friday at 23 UTC. The premise is that correlated 24/7 risk needs a joint budget, and a counterexample of an unseen exchange gap dominates quiet weekend history. There is no order, and the actual exposure is 0. The source is https://runway.adityaperswal.workers.dev/data/crypto-bars.
A news scan found the text describing a simulation called C91300, which is a two-hour joint weekend bootstrap and unmodeled crash boundary. This simulation used a NumPy seed of 91300, 10000 paths, and 1752 contiguous two-hour joint weekend blocks from 2026 hourly log returns through September 12. It considered a hypothetical basket of 119.14 units of Bitcoin, Ethereum, and Solana, with a mean return of -0.014703 percent. The simulation found that the first percentile was -2.332657 percent, which is equivalent to -2.779128 USD, the fifth percentile was -0.919514 percent, and the 95th percentile was 0.776395 percent. The simulation did not find any losses beyond 25.5297 USD, which is 3 percent of the account, but this is not a guaranteed bound because the historical sample is limited and may not include new regimes. A 25 percent joint crash would cost 29.785 USD, or 3.50004 percent of the account, and the quarter fund would lose 7.44625 USD on the same shock. The simulation preserved the two-hour serial and cross-asset dependence, but the first hour may start on Friday at 23 UTC for a Saturday 00 UTC ending block. The simulation did not include fees, stops, targets, entry conditioning, or path-longer regimes, and it is not a backtest or Kelly edge. The premises of the simulation are that a correlated book needs a joint budget, and observed quiet weekends do not promise future quiet weekends. The strongest counterexample is exchange or security gaps outside historical blocks. The conclusion is that there is no order based on the simulation, and the actual size is 0, with entry tests separate. The source of the data is https://runway.adityaperswal.workers.dev/data/crypto-bars, with an observation number of 26595.
A simulation was run with a starting crypto allocation of 119.14 USD in a hypothetical basket of equal parts Bitcoin, Ethereum, and Solana, with no actual position. The simulation used 10,000 paths over 16 hours, with 6-hour blocks sampled from 6,091 complete blocks from January 1 to September 12. This method, called a joint bootstrap, preserves the dependence between assets within each block, but ignores longer-term trends. The average return of the basket was -0.037462%. The worst intermediate simulated loss was -11.815274%. The simulation found that the worst possible loss at the end of the period was 25.5297 USD, which is 3% of the account equity. If the entire fund were invested, a 25% crash would cost 29.785 USD, or 3.50004% of the account. If only a quarter of the fund were invested, a 25% crash would cost 7.44625 USD. The simulation did not include fees, stop-loss orders, or profit targets. It also did not consider the conditions under which an investment would be made. The conclusion is to use a joint exposure budget and set stop-loss orders when a qualified investment opportunity arises. The actual investment size is currently zero because no such opportunity has occurred, not because the simulation predicts a loss. The data for this simulation came from https://runway.adityaperswal.workers.dev/data/crypto-bars.
A fund manager started with $850.99 in an account and $153.18 in cryptocash, with a position venue cap of $212.75 and a fund cap of $153.18. The manager considered a hypothetical portfolio with $51.06 in major USD currencies and $38.29 in Ethereum, but did not place any orders. The manager used two premises to study risk: the first premise looked at a 48-hour book mean of -1.011511, a worst-case scenario of -15.532567 USD, and a P1 value of -13.055955. The second premise examined 20-day periods with 15-minute intervals, with a P1 value of -14.906163 and a worst-case scenario of -16.577630 USD, and 60-minute intervals, with a P1 value of -15.606363 and a worst-case scenario of -17.917748 USD. The manager found that a 3% equity breach occurred in 25.5297 out of 0 sample paths, but noted that zero sample breaches do not guarantee safety. The manager also probed a 20-day Ethereum scenario with a P1 value of -3.799714 and a worst-case scenario of -4.333579 USD, and found that a 30% gap in Bitcoin, Ethereum, and Solana prices could result in a loss of $45.95 USD, or 5.4% of the account on a full fund, and $11.49 USD, or 1.35% of the account on a quarter fund. This scenario exceeded historical bootstrap support. The manager concluded that correlated exposure should be capped and that half of the Kelly criterion should be derived from independently observed strategy returns. The manager also noted that an unconditional negative bootstrap mean alone supports a zero Kelly criterion, but cannot prove a conditional rule of zero. The current entry was absent, which set the actual size to 0, and a shortfall of $349.01 did not enter the formula. The data source was https://runway.adityaperswal.workers.dev/data/crypto-bars.
A simulation called C91206 uses a set of historical data, specifically 15-minute log returns from March 1 to September 12, to test trading strategies for Bitcoin, Ethereum, and Solana. This simulation uses 18,744 calendar rows of data, but is missing some data for Bitcoin, Ethereum, and Solana. The simulation creates 10,000 two-day paths and 2,000 twenty-day paths to test how these cryptocurrencies perform together, taking into account how their prices are related within and between days. The simulation also tests a stop-loss strategy, where a trade is stopped if a certain loss is reached, and a target price is set to meet a net return of 3R. The target price is calculated using a formula that takes into account the volatility of the asset and the desired return. The simulation uses point prices at 15-minute, 30-minute, and 60-minute intervals, rather than the highest or lowest price within a given time period. The results of the simulation show that for Ethereum, the mean return over a two-day period is -0.67%, and the target price is 0%. Over a twenty-day period, the mean return is -0.39%, and the target price is 3.35%. The simulation also found that the return for Ethereum over a twenty-day period is -9.92% when using a 1-minute interval, compared to -10.47% when using a 60-minute interval. The simulation is useful for testing the size of potential losses, but it does not provide a reason to override the timing of trades. The data used for the simulation is available at https://runway.adityaperswal.workers.dev/data/crypto-bars.
A 48-hour Ethereum block simulation was run with 20,000 paths, each with two resampled 24-hour blocks of 15-minute log returns, using 16,000 overlapping complete historical blocks from March 1 to September 12, 2004, UTC, resulting in 14,658 unique samples. This method preserves within-day dependence, breaks between days, omits missing blocks, and cannot invent unseen crashes. It is not conditioned on a pullback. The simulation started with a hypothetical completed spot price of 2513.175 at 03 UTC, with an entry price of 2515.688175 and a 10 basis point slip. The prior 20-day Average True Range (ATR), a measure of volatility, was 98.98817058. The stop loss was set at 2317.71183383 and the target at 3170.24364866, with a net gain of 3 times the risk after fees and slippage. The simulation checked every 15 minutes, with a quarter of the gross gain plus 2 times the risk if the stop loss was hit, and a full stop if the target was hit, with a 48-hour marked liquidation. The stop loss was hit first 3.9% of the time, the target was hit first 0.01% of the time, and a partial gain of 0.345% was achieved. The net 48-hour mean gain was -0.580697%, with a worst sample loss of -11.323239%. These 48-hour frequency figures cannot be used to price a 20-day thesis. However, using 965 overlapping daily windows from 2024 onward, the 20-day maximum high gain at the 90th percentile was 33.743953%, and the target gain of 26.0199% is within the measured range, although not guaranteed. The training data showed 7 wins and 6 losses out of 13 trades, with a win probability of 0.16101921 and a loss probability of 0.09851387, resulting in a half-Kelly bet of 1.299744. The capped risk notional was 153.18, with a quarter ceiling of 38.295 risks and a simulated worst-case loss of 3.224981 nominal USD, although the actual loss was 0 because the entry was fresh and the late policy was untested. The simulation used no leverage and no need-based sizing. The data sources were https://runway.adityaperswal.workers.dev/data/crypto-bars and https://docs.alpaca.markets/us/docs/crypto-fees.
A research simulation was conducted with 20,000 paths per asset over 48 hours, sampling two 24-hour blocks from 4,634 overlapping complete blocks from March 1 to September 12 at 01 UTC. This historical resampling preserves within-day clustering, but destroys longer dependence and cannot invent unseen crashes. It is not based on 20,000 independent historical trades. The simulation used close-only hourly exits, which approximate a 1-hour delay, not actual 15-minute fills. For ETH spot, the price was 2,514.15, with a prior-day Average True Range (ATR) of 95.3601, resulting in a stop at 2,323.4298 and a net 3R target at 3,157.0172, with 25 basis points credited for fees and 10 basis points for slippage on each side. The net stop loss was 8.2311%, with a stop at 3.09% and a target at 0.04%. The 48-hour net 1st percentile was -9.0551%, with the worst sample at -12.3155%. For BTC, the stop was at 3.365% and the target at 0.01%, while for SOL, it was 1.10% and 0%, respectively. The sizing uses old frozen ETH trade odds, with an average win of 0.05848427 and an average loss of 0.03072019. The formula f = (p*w - q*l) / (w*l) = 16.001716 was used, with half at 8.000858. Prohibited leverage resulted in a clip at a minimum of 153.18 fund, 212.7475 venue, and 25.5297 normal risk budget, divided by 0.0823108663 to 153.18. A quarter-ceiling of 38.295 would risk 3.1521 nominal USD or 4.7161 at the simulated worst, about 0.55% of the account. This is a model ceiling, not a current recommendation, as the actual value is 0 since there is no fresh entry and policy uncertainty. The full fund nominal risk is 12.6084 USD, but unseen market gaps may exceed the model's predictions. The premises of the simulation are that a small reused sample does not justify a raw Kelly, and limiting dollars is preferred over assuming precision. The sources used were https://runway.adityaperswal.workers.dev/data/crypto-bars, frozen research 2765, and https://docs.alpaca.markets/us/docs/crypto-fees.
A news scan found the text of an experiment with the title C91200. The experiment is based on three premises. Premise 1 uses 90 days of hourly log returns, with a seed of 91200 and 20,000 paths per asset. Each path concatenates two sampled 24-hour blocks, resulting in 2,136 overlapping historical blocks per asset, rather than 60,000 independent market observations. Premise 2 sets illustrative stops at 5.7564%, 7.5801%, and 9.7048% below the prices of BTC, ETH, and SOL, and targets at 17.2692%, 22.7403%, and 29.1145% above. Premise 3 analyzes the frequencies of first-stop and first-target events, which are 1.125%, 0.990%, and 0.320% for first-stop, and 0.030%, 0.225%, and 0% for first-target. The conclusion is that the measured ranges cannot justify expecting a net 3R payoff in two days, so a multiweek thesis horizon should be retained if entry later validates. The source of the data is https://runway.adityaperswal.workers.dev/data/crypto-bars, with all measured path figures in observations 25,189, 25,191, and 25,193.
A news scan found the following text: Premise 1 is based on 552 hourly bars per asset from August 19, 08 UTC, to September 11, 07 UTC, with 547 overlapping 5-hour return blocks. The seed 91108 draws 20,000 blocks per asset, totaling 60,000 draws, but only 547 historical blocks each, preserving within-block dependence and not introducing new independent evidence. Premise 2 freezes the current 480-hour thresholds. The required gains for BTC, ETH, and SOL are 6.5140%, 4.03979%, and 10.82051%, respectively, with hourly-close fixed-threshold touches of 0%, 3.375%, and 0% in the bootstrap. The ETH terminal 5-hour mean is 0.263294%, with the 1st and 99th percentiles being -3.221144% and +8.031913%, and the worst observed 5-hour return being -3.733345%. Premise 3 samples include one recent market regime, are not conditioned on CPI events or breakout conditions, and exclude intrahour excursions and new unseen shocks. The conclusion is to not anticipate entry, as the rolling production threshold can change while the scenario keeps it fixed. The descriptive bootstrap supports monitoring relevance only, not order size, CPI probability, or costed policy return. A prior 90k 20-day study, numbered 2670, remains the risk reference and was not rerun. The source is https://runway.adityaperswal.workers.dev/data/crypto-bars, with local study.py and study.json, and observations from 25017 onward.
A news scan found the headline is not available, but a data line was found. The data line describes a trading strategy called C91106. The strategy was tested on the ETH asset with a unique 20-day hourly path. The test ran from September 11, 2025, to September 11, 2026. The strategy used a static and trailing share path with 180,000 possible outcomes. The strategy aimed for a 3R target return, which means three times the risk, with a 25 basis point fee and a 10 basis point lip, or slippage, which is the difference between expected and actual trade prices. The strategy also used a 2R cost-benefit ratio and a peak-2 ATR trail, which is a type of stop-loss order that uses the average true range, or ATR, to set the stop price. The first sampled close exit was used, with a 20-day timeout. The results showed a mean net return of -2.00844%, -2.08325%, and -2.38655% for the static 1, 24, and 72-hour targets. The trailing mean returns were -1.87692%, -1.97279%, and -2.25903%. A separate test was run on 20,000 independent and identically distributed, or IID, bootstrap samples of ETH trades. The results showed a mean return of 2.874945% with a 95% confidence interval of -1.587870 to 8.390857. The win rate was 66.667% with an average win of 5.848427 and an average loss of 3.072019. The best compound return was 1.735528%. A cash-only log growth grid maximizer was used to determine the optimal allocation. The results showed that a 50% allocation would result in a terminal loss of 13.14% with a 5th percentile of -3.12814 and a 95th percentile of 23.84138. The same test was run on the BTC asset with similar results, but the SOL asset showed negative results and was not deployed. The conclusion is to retain cash now and retain a conditional watch. The results do not disprove the conditional breakout edge, and the empirical bootstrap does not give the probability of a true edge or unseen crisis coverage. The source of the data is https://runway.adityaperswal.workers.dev/data/crypto-bars, with observations from 24796 to 24819.
A news scan found the following text: 1. The mean payoff of ETH rolling 480 hour holdout 6 trades is 2.874945%, with a 95% confidence interval of -1.587870% to 8.390857%. The best trade had a payoff of 15.2404%. If the best trade is excluded, the mean payoff is 1.735528% compounded. The mean payoff of calendar hour 7 trades is 1.206384%, with a 95% confidence interval of -2.735977% to 6.438506%. If the best trade is excluded, the mean payoff is -6.869318%. The mean payoff of daily 5 trades is -3.521877%, with a 95% confidence interval of -9.559165% to 3.335655%. These are reused trades, not new independent proof. 2. A cash-constrained log growth grid with a step of 0.001 gives daily 0, calendar hour 1, and rolling hour 1, with half wealth fractions of 0, 0.5, and 0.5. The fund started with 235.86 and ended with 117.93, with a full ETH allocation of 29.4825 and a quarter probe of before eligibility. The fixed 50% notional bootstrap terminal loss is 13.39% rolling and 34.195% calendar, with rolling 5th and 95th terminal returns of -3.128142% and 24.133260%. The model uncertainty is large and not forecast. 3. Separate unconditional path stress draws of 10,000 paths per asset, with 20 complete 24 hour blocks from the last 365 days, were analyzed. The number of blocks for BTC, ETH, and SOL are 363, 362, and 361. The current 2 ATR stop is 5.9418%, 7.1973%, and 10.0209%, with a net 3R price up of 20.6377%, 24.4043%, and 32.8751%. The target first is 4.25%, 8.12%, and 3.94%, with a stop first of 59.81%, 64.69%, and 55.19%. The mean static net is -1.8309%, -2.0018%, and -2.4120%. The ETH current net stop is 7.845261%, which implies a half notional risk of 9.251916 USD per quarter, or 2.312979 USD, which is 0.271799% of the account. However, the actual size is 0 due to the absence of a signal. The validity of the results is based on seeded 91104 empirical bootstrap and conditional payoff arithmetic. The soundness of the results is limited by small N, IID trades, block stationarity, unseen jumps, and close-mark execution limit forecast, and cannot use unconditional negative paths to disprove conditional edge. The conclusion is to keep hypothetical size illustration distinct from order and act only on actual validated signal. The source of the data is https://runway.adityaperswal.workers.dev/data/crypto-bars.
A news scan found the headline 'C91102 60000 sampled paths: fees enlarge 3R targets and gaps exceed nominal stops'. This refers to a report on an automated trading fund. The report is based on 10,000 paths per asset per regime, with a 20-day horizon and contiguous 24-hour blocks, comparing 2024-25 to 2026 data. Historical empirical tails were retained, but cross-block dependence was broken. The current mid-starts are 76,905.29, 24,52.03, and 15,99.3295, with 2 ATR risk of 5.9387, 7.1788, and 10.0173 percent. Fees of 25 basis points credit asset and 10 basis points lip side imply a net stop loss of 6.5954, 7.8269, and 10.6455 percent. The net 3R target is 92,769.56, 30,49.07, and 131.9734, with a gross of 20.628, 24.349, and 32.864 percent. The 2026 regime means are -1.150, -1.276, and -1.301 percent, with target frequencies of 4.14, 6.46, and 3.55 percent. The training means are +0.731, +0.059, and +0.071 percent. The conclusion is that wider range gains are possible, but unconditioned entry has no evidenced positive current expectancy. An hourly check fills proxy with +2R breakeven and 2 ATR trail, with no entry signal conditioning, so probabilities are not signal forecasts. The worst sample quantiles are not bounds. At 212.7475 notional, the training 5th percentile loss is 16.779, 21.543, and 26.739 USD, with SOL exceeding 25.5297 normal 3 percent account budget. The strongest counterexample breakout conditioning could improve odds, but absent trigger cannot borrow it. There is no MC-derived Kelly order. The source of this data is https://runway.adityaperswal.workers.dev/data/crypto-bars and https://runway.adityaperswal.workers.dev/data/crypto-snapshots.
A simulation was run with certain premises. The first premise was to create paths of 20 days per asset from 253 supplied daily returns, using observations from days 23079 to 23081, with a circular 3-day block, and using numpy with 10000 seeds. The second premise was to use the latest 2 ATR stop and 6 ATR target, with daily close being the first stop and first target, resulting in odds of 47.50% and 7.27% for BTC, 51.91% and 10.29% for ETH, and 42.68% and 4.11% for SOL, with the remainder not meeting these conditions. The third premise was that these scenarios only consider closing prices and do not account for intraday thresholds, crisis gaps, costs, and breakout conditioning. The conclusion is that there is no inferred live win probability or Kelly from this simulation. The 5th percentile account loss at 25% allocation was -4.29%, -5.73%, and -6.02%, indicating that venue capacity alone cannot substitute for risk control. The validity of the simulation is conditional on resampling arithmetic, but its soundness fails as a calibrated trading forecast if regime changes or aggregation errors matter. A counterexample shows that trend-conditioned returns can be much better, tested separately. The stop distances were 5.967%, 7.224%, and 10.091%, and the nominal targets were 17.902%, 21.671%, and 30.273%, but no actual stop or target was set because there was no holding or entry. Fees also reduce the nominal 3R to less than 3R net, and a live policy requires new validation. The source of the data is https://runway.adityaperswal.workers.dev/data/crypto-bars.
A four-day risk simulation for the C91008 fund has changed its outlook when the recent drift is removed. The simulation is based on two premises. First, it generates 10,000 joint 96-hour paths in 24-hour blocks from the last 60 days of log returns, with a seed of 9100804, and preserves cross-asset dependence. It stops when the target is 1.5 times the average daily range, observes and fills the hourly close, and times out after 96 hours, with a 25 basis point fee per side and a 10 basis point slippage. Second, the observed drift means that the net returns for Bitcoin, Ethereum, and Solana are +0.3551%, +0.8279%, and +1.2283% respectively. However, when the drift is removed, these numbers change to -0.6744%, -0.9049%, and -0.5181% respectively. The probabilities of reaching the target first are 19.58%, 16.22%, and 11.75% for the demeaned targets, and the timeouts are 49.21%, 51.48%, and 71.91% respectively. For the full fund with an equal sleeve, the demeaned mean is -$1.649, the 5th percentile is -$13.197, and the loss probability is 62.42%. For the 25% and 50% funds, the 5th percentiles are -$3.299 and -$6.598 respectively. The conclusion is that the drift assumption drives optimism, and these are risk scenarios, not backtest win probabilities or edge. The fact ends at observed returns, and stationarity, block choice, and the next four days are thought-experiment assumptions. The model misses intra-hour tails, next open fill, +1R partial, and breakeven, and venue 15-minute checks, and nominal stops are not loss bounds. The result was approved with no timeframe or target widening for shortfall. The source of the data is https://runway.adityaperswal.workers.dev/data/crypto-bars, and the statistics were generated using stats.py, with observations from 21562 to 21573.
A joint monthly sizing bootstrap analysis for C91006 showed losses at every nonzero illustrative allocation. The analysis was based on three premises: first, there were no closed 2026 holdout trades, so an out-of-sample trade-payoff Monte Carlo analysis was not possible, which is a statistical method to estimate potential outcomes. Second, a separate 2025 training exercise used 17 completed trades, divided into three equal cash sleeves, with 12 monthly observations and 10,000 joint circular three-month resamples, using a seed value of 9100602 for reproducibility. Third, the actual training portfolio return was -4.8278%, with a monthly mean confidence interval of -1.0938% to +0.2693%, which is a range of expected average returns. The simulated median annual return was -1.0327% at 25% of fund capital, -2.1304% at 50%, and -4.5639% at 100%, with a loss probability of 85.21%, 86.55%, and 88.38%, respectively, and a 95th percentile monthly-observed drawdown of 3.9160%, 7.7296%, and 15.1195%. The conclusion is that no nonzero illustrative allocation is supported, and these results are training diagnostics, not predictions of future probabilities. The full-fund median dollar loss is $10.76 on $235.86, which is not an estimate of this month's result. The analysis has limitations, including repeated overlapping months, few regimes, supplied-bar faults, fixed sleeve rebalance, and no intramonth drawdown or venue exit model. The source of the data is https://runway.adityaperswal.workers.dev/data/crypto-bars.
A news scan found the text: Premises: Each asset contributes 2160 recent hourly returns over 2154 overlapping 6-hour blocks, with 10,000 paths per model, over 28 days, and a seed of 9100601. The stop is set at 1 ADR 20, with a target of 1.5 ADR 20, using the gross reference of the September 9 close, and fees of 25 basis points and a slip of 10 basis points per side. Empirical drift versus demeaned target hit BTC at 54.51 and 40.09 percent, ETH at 58.63 and 39.87 percent, and SOL at 63.99 and 40.96 percent, with net means of plus 0.6116 and minus 0.6963 percent, plus 1.4099 and minus 0.6895 percent, and plus 3.3789 and minus 0.4621 percent. The conclusion is that a positive path forecast depends heavily on assuming recent drift persists, and it cannot be used for Kelly probability. The validity is based on common random draws and an isolated drift assumption. The soundness is based on unconditional recent returns, overlapping blocks, and a close fill on the first hourly threshold cross, not the next open, with no 15-minute venue or plus 1 R partial, and forecasts are conditional scenarios. A no-gap nominal 1 percent equity risk notional of 211.70, 176.73, and 122.79 is not a loss guarantee, and 5th percentile losses of 5.08, 6.32, and 7.98 percent exceed ADR plus cost. The sources are https://runway.adityaperswal.workers.dev/data/crypto-bars and https://docs.alpaca.markets/us/docs/crypto-fees.
A fund manager analyzed trading outcomes based on three premises. First, daily returns from 14 trades of Bitcoin, 14 trades of Ethereum, and 8 trades of Solana remained negative. Second, a simulation created 10,000 possible trade paths for each asset, starting at $235.86 with $50 positions, to isolate the effects of price construction. Third, the results showed loss frequencies of 84.51% for Bitcoin, 52.15% for Ethereum, and 91.81% for Solana, with median ending capital of $232.71, $235.46, and $223.35, respectively. The 95th-percentile drawdowns were $10.25, $12.45, and $28.53. The results were compared to actual daily losses, which showed a similar loss frequency of 87.79% for Bitcoin. The conclusion was that no approved allocation was made due to limitations such as small samples and repeated use of holdout data. The analysis was based on data from https://runway.adityaperswal.workers.dev/data/crypto-bars and backtests from 1991-1993.
A fund manager analyzed data and found that all hourly holdout means were negative and the quarter-Kelly allocation was 0. Using a numpy seed of 91003, the manager drew 3 trade circular blocks from each asset's actual holdout returns, creating 10,000 paths with 10 trades and a fixed amount of 50 USD, starting at 235.86. The whole capital returns were adjusted for a vbt entry fee denominator of 1.0025. The probabilities of a loss below the starting point were: BTC 99.95%, ETH 98.63%, SOL 99.86%, XRP 99.05%, LINK 98.12%, HYPE 83.70%, LTC 99.99%, and UNI 88.53%. The median end points were between 231.47 and 233.86. The loss odds were based on the empirical sample, not future calendar risk or probability bound. The 95th maximum drawdown was: 5.93 USD for BTC, 7.94 for ETH, 12.20 for SOL, 8.98 for XRP, 7 for LINK, 8.21 for HYPE, 8.16 for LTC, and 11.85 for UNI. No joint portfolio simulation was done and unseen regimes or 15-minute overshoots were not represented. The daily quarter-Kelly point estimate for HYPE was 6.66%, which would not qualify without training, cross cadence, and a positive confidence interval. The daily quarter-Kelly point estimate for ETH was 0.50%, which is approximately 1.19 USD below the 10 minimum. The approved allocation is 0 and cannot be increased to meet the minimum or 349.01 shortfall. The validity bootstrap reflects negative means, but the soundness is limited by selection, serial structure, and block length. The data sources are https://runway.adityaperswal.workers.dev/data/crypto-bars and backtests from 1949 to 1956.
A news scan found the headline about a block-bootstrap cash challenge that loses in 82–88 percent of paths. The test used a fixed rule to make trades every hour, and it did not work well. It used real trade returns from 2013, 2016, and 2017, and it used a special method called numpy with a seed of 910 to make the test results more realistic. The test was run 10,000 times, with 10 trades, and each trade was worth $50. The test started with a price of $235.86. The results showed that the median end price was $231.83 for Bitcoin, $231.08 for Ethereum, and $230.81 for Solana. There was a high probability, 88.16% for Bitcoin, 87% for Ethereum, and 81.95% for Solana, that the price would be below the starting price. The test also showed that the price could range from $229.10 to $242.53 for Bitcoin, $227.96 to $246.14 for Ethereum, and $227.35 to $251.19 for Solana. The biggest loss was $6.81 for Bitcoin, $7.99 for Ethereum, and $8.58 for Solana. The test did not try to predict when to buy or sell based on the time of year, and it did not work well when the price changed suddenly. A method called Quarter Kelly estimated that the test would not work well, and another estimate showed a small positive return of 1.2678% for Solana, but this was not approved because it did not meet certain conditions. The test was not scaled up to make it more affordable. The results showed that the test was not valid because it did not have an edge, and it was not sound because it relied on past history and the choice of blocks. The source of the data is https://runway.adityaperswal.workers.dev/data/crypto-bars.
A simulation of a trading book over ten cycles kept its initial investment size at zero. The simulation was based on certain premises, including that the actual payoffs from exiting trades after 15 minutes were all negative, with a quarter of the Kelly criterion being 0.2. The simulation only looked at how changes in certain variables affected the outcome, using an initial book value of 235.86 and an initial investment of 50, which was 21.199% of the total, rebalanced over ten cycles, with 10,000 random paths generated for the simulation, using a seed of 90309. The median, or middle, value of the terminal prices was 232.708 USD for Bitcoin, 232.959 USD for Ethereum, and 234.934 USD for Solana, with the fifth percentile being 224.267, 220.746, and 221.932 USD respectively, and the 95th percentile being 241.655, 246.676, and 249.107 USD respectively, and the frequency of losses being 72.8, 63.86, and 53.83%. The simulation did not try to predict what would happen over the next 22 days or what the best time to enter the market would be, and it did not take into account things like correlated market regimes, unobserved liquidity, gaps, and cost uncertainty. The simulation used a stop-loss order, which is a type of order that sells a security when it falls to a certain price, with a maximum daily overshoot of 2.2121, 1.8739, and 1.4293 times the average true range, and a maximum hourly overshoot of 0.6485, 0.4115, and 0.3794 times the average true range, and a maximum 15-minute overshoot of 0.4898, 0.2225, and 0.2099 times the average true range, which means that even with a 15-minute stop-loss order, the simulation could still lose 1.49 times the average true range before fees and slippage. The conclusion of the simulation was that the planned loss limit was 8.5099 USD, which is not a guarantee, and the selected allocation was zero. One argument against this conclusion is that the 95th percentile of the simulation was positive, which shows that there is a possible upside, but it does not mean that the expected payoff is positive. The data for the simulation came from https://runway.adityaperswal.workers.dev/data/crypto-bars and https://docs.alpaca.markets/us/docs/crypto-fees.
A trading fund manager made some simulations to test how well a trading strategy works. The manager used real data from March 1 for Bitcoin, Ethereum, and Solana. The strategy was to buy and sell at certain times based on the prices of these cryptocurrencies. The manager found that the strategy did not work well, with losses of 42.24% for Bitcoin, 11.31% for Ethereum, and 21.93% for Solana. The manager also found that the strategy was sensitive to changes in the market and that more data is needed to make good decisions. The manager used a technique called resampling to test the strategy, which involves repeating the simulation many times to see how well it works. The results were not good, with high losses and low gains. The manager concluded that more data is needed to make good decisions and that the strategy should not be used without more testing. The manager used data from two sources: https://runway.adityaperswal.workers.dev/data/crypto-bars and https://docs.alpaca.markets/us/docs/crypto-fees.
A test was done to see if changing the rules for buying and selling would improve results. This test, which is called a counterfactual and not a proposal for changing the size of trades, looked at what would happen if the rules for trading with USDC were applied to trades made in 2026. The test used 10,000 different paths, each with 10 cycles, and started with $235.86, with $78.62 in each major trade. The test also took into account fees and the difference between the price at which a trade was made and the price at which it was executed, which is called slippage. The average result of these trades was a loss of 3.21270%. The test also looked at how the prices of different cryptocurrencies, including BTC, ETH, and SOL, were related to each other. The results showed that the prices of these cryptocurrencies were highly correlated, with correlation coefficients of 0.91259 for BTC and ETH, 0.85103 for BTC and SOL, and 0.88099 for ETH and SOL. The test also found that the ending values of the trades were $125.56 at the 5th percentile, $168.44 at the 50th percentile, and $222.01 at the 95th percentile, and that the frequency of losses was 98.28%. The median maximum drawdown was 30.43% and the 95th percentile maximum drawdown was 47.46%. The test used only 8 empirical vectors and did not take into account the fact that the results may be different over time, which is called serial dependence, or the fact that the market may change, which is called regime change, or the risk of extreme events, which is called tail risk. The test also did not use stops or targets, and trades were executed every 15 minutes. The idea that trading in three different cryptocurrencies would reduce risk is not strong, and the results of the test show that the outcomes of the trades are still highly dependent on each other. The Quarter Kelly criterion, which is a measure of the optimal fraction of wealth to invest, was zero because the individual means were negative, so no dollars were approved for investment. The simulation shows what would happen if the rules for overriding the gate, which is a mechanism for controlling risk, were not followed, and it does not take into account the probability of future losses. The sources used for this test include https://stablecoins.llama.fi/stablecoin/2 and https://runway.adityaperswal.workers.dev/data/crypto-bars, and the files research/audit_sim.py and json.