Stats

A news scan found a report about Darden event bootstrap. The report used data from DRI SEC releases and a website to look at cash returns from 16251 to 1656. It used a computer program called NumPy with a seed of 20260909 to create 50000 different paths of ten events, starting at 49.14 dollars and reinvesting at each event. The results showed that the median, or middle, outcome was 53.8006 dollars, with 11.128 percent of the paths resulting in a loss. The worst loss in the top 5 percent of paths was 7.4522 percent. Another calculation showed that the average outcome could be between -0.8285 percent and 2.9660 percent, which includes zero. This is based on the idea that the six observed outcomes are representative and independent. However, this simulation does not take into account things like large gaps in data, changes in the market, taxes, or actual execution failures. No real money was invested. One example that goes against this simulation is an unobserved event where there was a realized loss of 45.86 percent. The simulation should not be used to predict actual losses or probabilities. It is meant to inform research, not authorize investment. A quarter Kelly estimate of 0.073922 was overridden by other evidence to 0.

A news scan found the text of a hypothetical deployment analysis. The analysis started with an account balance of $850.99 and $245.69 in cash, with no positions held. The goal was to achieve repeatable positive profit, with a binding constraint of negative validated payoff. The analysis used a frozen trading history to sample joint portfolio returns, rather than independent stock returns. The first simulation, seed 9090201, used 20,000 circular 3-session block paths to approximate the remaining September 9-30 trading window, with the entire $245.69 quant book supposedly deployed into 8 equal initial fractional sleeves. The base scenario showed a profit range of -$1.87 to $1.12, with a loss frequency of 49.8%, and a stress scenario showed a loss range of -$2.30 to $0.82, with a loss frequency of 58.45%. No sample achieved a profit of $349.01, but this does not prove that the true likelihood is zero. The analysis also found that flat days in the historical strategy explained the median return of 0, with no missing returns converted to zero. A longer diagnostic simulation, seed 9090200, used 20,000 joint 3-month block paths over 60 months, with a constant weight between the quant book and account. The base scenario showed a return range of -3.15% to -0.64%, with a loss frequency of 99.325%, and the stress scenario showed a return range of -5.27% to -2.68%, with a loss frequency of 100% in an infinite Monte Carlo sample. The analysis corrected the actual horizon to 60 months, rather than the originally contemplated 100 months, and excluded fixed operating costs, new regimes, whole lots, 15-minute fill gaps, and future cash rates. The resampling assumed stationarity, but short blocks did not preserve the monthly entry calendar, and the hypothetical fresh schedule was not an approved late entry. The Quarter Kelly binary-average-loss approximation floored at 0 in both the base and stress scenarios, with a notional value of 0. The caps were set at $212.75 per position and $8.51 nominal risk limit loss plans, but these were not guaranteed gap loss plans. Neither the need for $349.01 nor the parent positive returns supplied an approved edge. The sources of the analysis were archived in horizon_sim.py/json and diagnostics, and can be found at https://runway.adityaperswal.workers.dev/data/stock-bars.

A three-day risk bootstrap analysis was performed. The process involved generating 10,000 joint overlapping three-day log return blocks from the last 60 completed daily bars through September 8, using a NumPy seed of 2026090902. This resulted in only 58 distinct blocks, which preserved cross-major dependence within each block. A fee haircut of 0.9975 squared was applied, with no slippage, stops, targets, or 15-minute execution. The analysis provided unconditional risk estimates, not entry-conditioned forecasts. The simulated mean returns were 0.72379% for BTC, 1.36121% for ETH, and 1.28556% for SOL. The loss probabilities were 49.84% for BTC, 46.94% for ETH, and 47.79% for SOL. The simulated fifth percentile losses were -3.36973% for BTC, -3.90019% for ETH, and -5.85392% for SOL, with dollar losses at 212.7475, capped at 7.16901 for BTC, 8.29756 for ETH, and 12.45408 for SOL. SOL exceeded the equity risk budget of 8.5099. The exact linear-interpolated fifth percentiles across the 58 distinct blocks were -3.21421% for BTC, -3.66867% for ETH, and -5.10359% for SOL. The worst three-day losses were -4.13597% for BTC, -4.26211% for ETH, and -7.24518% for SOL. The quantile difference from the previous seed was due to the tail having few discrete blocks, not new information, and there was no confidence from 10000 repeats. Since there were no new daily bars since the prior run, dependence and volatility remained unchanged. Positive unconditional drift and lower BTC historical tail did not calibrate entry win rate. Kelly gate already binds at 0, and stop slippage could exceed the risk budget. The source of the data is https://runway.adityaperswal.workers.dev/data/crypto-bars, with test.py and simulation.json providing complete observation. The conclusion is size 0, with no invented exit odds.

A recorded trade ledger, called 1302 costed 24 trade, was used instead of a bullish stock path. This analysis used a random number generator, called Numpy RNG, with a seed of 2026090900, to make 20,000 mean bootstrap draws. The analysis was done in blocks of 3 consecutive trades, with 7 block starts per path, and each path was trimmed to 20 trades, making a total of 20,000 paths. The initial cash was included in the peak. When the equity exposure was 10% per path, the probability of loss was 88.865%, the median loss was -4.502515%, and the 95th percentile drawdown was 10.590189%. When the equity exposure was 25%, the probability of loss was 89.61%, the median loss was -11.144542%, and the 95th percentile drawdown was 24.447057%. These results are based on repeated independent block starts and an unchanged payoff distribution, and do not forecast the probability of future stock losses. The average win was 16.456597% and the average loss was 7.115503%, which means the strategy is not profitable. The quarter-Kelly support is zero, which means the strategy does not have a good expected return. The data came from https://runway.adityaperswal.workers.dev/data/stock-bars. The analysis notes that small history, unseen fat tails, and price gaps can make outcomes worse. Since the base strategy did not work, no trade stop or target was invented, and no expensive 15-minute optimization was done to try to rescue it. A different, independently tested rule may work, but it would need to be preregistered and tested on new data.

A news scan found the text of a block of notes from a fund manager. The notes start with the current equity at 850.99 and quant cash at 245.69, with the goal of achieving repeatable net profit while considering edge uncertainty. The manager makes two premises: first, using 219 completed fixed rule returns from 2021 to 2025 with a base of 6 basis points per side and stress of 16 basis points per side, and second, using a NumPy seed of 9090000, 20,000 paths, 100 sequential trades, and circular 5 trade blocks, with each trade hypothetically investing 25% of the evolving account equity. The base terminal percentiles are -5.54344%, +1.53083%, and +9.92274%, with a loss probability of 36.68%, a 10% drawdown probability of 0.85%, and a 95th percentile drawdown of 7.78848%. The stress percentiles are -10.13895%, -3.41932%, and +4.55102%, with a loss probability of 77.075%, a 10% drawdown probability of 11.14%, and a 95th percentile drawdown of 11.33923%. The conclusion is that the simulated upside is not robust to an extra 10 basis points of slippage on each side. The claim of 100 trades in 23 days is not valid, as the runway window differs from the 5-year horizon of approximately 219 trades. The sampling assumes that the past block distribution will repeat, but it does not model new crashes, 15-minute exit overshoot, whole lot rounding, fixed monthly costs, or censored missing fills. The scenario of 25% notional is not approved risk and does not guarantee a 1% loss limit. The binary payoff uses a normalized loss quarter Kelly formula, which is 0.01145882 for the base and floored at 0 for stress. This quantity is a fraction at risk in average-loss units, not automatically position notional. The full variable payoff Kelly with unbounded tail risk would need a separate model. The adopted size is 0 because the confidence interval includes 0 and the stress is negative. The hard position cap is 212.7475 and the nominal risk is 8.5099, which are caps, not sizing recommendations. A counterexample could be that the cost could actually remain at 6 basis points, but unknown execution cannot be assumed from daily liquidity, and the operator needs 349.01 changes with no probability. The source of the observation is 16076, with backtests QTR9090000, and the data can be found at https://runway.adityaperswal.workers.dev/data/stock-bars.

A new test using a random number generator, called a NumPy seed, created 10,000 groups of three-day blocks of daily returns from the last 60 days up to September 8. The test kept the relationships between different assets and applied a 25 basis point fee to each part of the trade, without considering slippage or stops, and assuming execution within 15 minutes. This simpler model does not account for intraday stops. The average three-day returns were 0.73676% for Bitcoin, 1.39588% for Ethereum, and 1.25803% for Solana, with a 50.29%, 47.23%, and 48.09% chance of loss, respectively. The worst 5% of returns were -2.78008% for Bitcoin, -3.62782% for Ethereum, and -5.01531% for Solana, with potential losses of $5.91455, $7.71809, and $10.66996 at a $212.7475 cap. Solana's risk exceeds its budget by 8.5099 even without considering slippage. A test using cash as a counterexample did not find a positive historical drift, which does not establish a condition for entering a trade. The 58 overlapping blocks used in the test are not independent observations, and extreme events can worsen the results. A risk-only cap is not approved due to the binding of a negative selected strategy Kelly. The data sources are https://runway.adityaperswal.workers.dev/data/crypto-bars and https://docs.alpaca.markets/us/docs/crypto-fees, with this being observation 16043 from simulation.json. The test is limited by historical daily revisions and missing extreme values.

A news scan found the report of a trading fund manager. The manager decided to keep a frozen 15-minute risk simulation and not create new paths. The simulation was based on prior data from June to August, with 10000 joint 72-hour paths and 8109 complete 4-hour blocks. Each path had 18 blocks and used a seed from September 8, 2026. The simulation did not use interpolation. It stopped when it reached 1 ADR, with a target of 1.5 ADR, and had a 15-minute overshoot. There was a 25 basis point fee and a 5 basis point slip for each leg. On the same day, the manager kept the ADR for 14 BTC at 3.013842%, ETH at 3.751064%, and SOL at 5.786846%. The 95th percentile loss was 212.7475, with a notional value of 8.63794, 10.77460, and 14.60955 USD, which exceeded the planned risk of 8.5099. The mean returns were -.566133%, -.355353%, and -.190563%, which are outputs of the risk model, not the condition for entering a trade. The manager did not run new simulations. The report cited limitations due to future volatility, missing bars, and unseen crashes, which affect the accuracy of tail inference. A 15-minute stop cannot guarantee a 1% realized max loss. The Kelly criterion is already binding, so no risk-only cap is approved. The source of the data is https://runway.adityaperswal.workers.dev/data/crypto-bars and https://docs.alpaca.markets/us/docs/crypto-fees.

A review of a 15-minute risk simulation was done at 19:40. The simulation used 10000 joint 72-hour paths, with 8109 complete 4-hour blocks from June to August, and 18 blocks per path. The seed used was 2026090806, with no interpolation. The simulation had a stop at 1 ADR, a target of 1.5 ADR, and a 1 R half or break-even point, with a 15-minute overshoot and a fee of 25 basis points plus a 5 basis point slip for each leg. The simulation was run on the same UTC day, so the ADR for BTC, ETH, and SOL was preserved at 3.013842%, 3.751064%, and 5.786846% respectively. The 95th percentile loss was calculated at 212.7475, with a cap of 8.63794, 10.77460, and 14.60955 USD, which exceeds the risk allowance of 8.5099. The unconditional means were -.566133%, -.355353%, and -.190563%, which does not provide a conditional entry edge. No rerun was claimed, and the latest negative Kelly controls size 0 regardless of the risk-only cap. The simulation notes that counterexample future volatility may differ, and incomplete bars and unseen crashes may limit the tails. Stops monitored every 15 minutes do not guarantee a 1% maximum realized loss. The data used can be found at https://runway.adityaperswal.workers.dev/data/crypto-bars, and the crypto fees are explained at https://docs.alpaca.markets/us/docs/crypto-fees.

The 15-minute risk model used to determine the size of trades is still in effect as of 16:30. This model uses a type of simulation called Monte Carlo, which is a method to estimate the outcome of different scenarios. The original simulation had 10,000 joint 72-hour paths, with 8,109 complete 4-hour blocks from June to August, and 15-minute intervals, with 18 blocks per path, and a seed of 2026090806, meaning a specific starting point for the simulation, and no interpolation, which is a way to fill in missing data. The model has a stop at 1 ADR, which stands for Average Daily Range, a measure of how much a price moves in a day, and a target of 1.5 ADR, with a 1 R, or risk unit, half of the Breakeven point, and 15-minute checks with overshoot, which means checking every 15 minutes if the price has moved too far, and a fee of 25 basis points, or 0.25%, plus a 5 basis points, or 0.05%, lip per leg, which is a small additional fee. The model also looks at the prices of BTC, ETH, and SOL, which are types of cryptocurrencies, and their prices changed by 3.013842%, 3.751064%, and 5.786846% respectively. The original 95% loss at 212.7475, with a cap of 8.63794, 10.77460, and 14.60955 USD, exceeds the 8.5099 risk budget, which is the maximum amount of risk allowed. The unconditional means, which are the average returns without any conditions, are -0.566133%, -0.355353%, and -0.190563%, but these are not used to determine if it's a good time to enter a trade, because they are not conditioned on the current market conditions, and the risk-only caps, which are limits on the amount of risk taken, do not approve the sizes of the trades. The model also notes that future volatility, which is a measure of how much prices move, may differ from the past, and that missing data and unseen crashes can limit the accuracy of the model. The Quarter Kelly, a method to determine the optimal size of a trade, is 0, because the validation controls failed. More information can be found at https://runway.adityaperswal.workers.dev/data/crypto-bars and https://docs.alpaca.markets/us/docs/crypto-fees, which are websites with data and documentation on cryptocurrency trading.

A risk simulation was applied at 13:50 with no changes. The simulation used original data from 1103 with 10000 joint 72-hour paths, 8109 complete 4-hour blocks from June to August, 15-minute intervals, 18 blocks per path, and a seed of 2026090806, with no interpolation. The simulation included a stop loss of 1 average daily range (ADR) and a target of 1.5 ADR, with a half risk and break-even point, and 15-minute checks for overshoot, with a 25 basis point fee and 5 basis point slip per leg. On the current UTC day, no new paths were completed, with ADR values of 3.013842% for BTC, 3.751064% for ETH, and 5.786846% for SOL. The original 95% loss was at 212.7475, with a cap of 8.63794, 10.77460, and 14.60955 USD, exceeding the risk budget of 8.5099. The unconditional mean of -.566133%, -.355353%, and -.190563% is not an entry-conditioned edge. An extra hour did not justify a rerun. Risk-only caps were not approved, with quarter Kelly 0 controls in place. Note that future volatility may change and unknown crashes can exceed historical extremes. The sources of the data are https://runway.adityaperswal.workers.dev/data/crypto-bars, https://runway.adityaperswal.workers.dev/data/crypto-snapshots, and https://docs.alpaca.markets/us/docs/crypto-fees.

A 12:30 application of a frozen 15-minute risk simulation was performed, with no new Monte Carlo simulation. The simulation, which is a method to predict possible outcomes, used 10,000 joint 72-hour paths, with 8,109 complete 4-hour blocks from June to August, and 15-minute intervals, with 18 blocks per path, and a seed of 2026090806. The simulation did not use interpolation, and had a stop loss of 1 ADR and a target of 1.5 ADR, with a half and breakeven point of 1 R, and 15-minute checks to fill overshoot, with a 25 basis point credited asset fee and a 5 basis point slip per leg. There were no new completed UTC days, and the average daily returns for BTC, ETH, and SOL were 3.013842%, 3.751064%, and 5.786846% respectively, which were unchanged. The original unconditional means were -0.566133%, -0.355353%, and -0.190563%, with a 95% loss at 21.27475 position cap, and risk budget of 8.63794, 10.77460, and 14.60955 USD above 8.5099. The unconditional path statistics are not conditional trade edge, and the risk-only notionals are not allocations. No justified simulation rerun was performed after four hours. The Quarter Kelly from failed selected holdout remains 0. A counterexample of a favorable future regime could alter the distribution, and requires independent positive entry validation before sizing and current 15-minute execution test. 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 headline about AVO empirical path sizing. The automated trading fund manager made notes about a sample of 16 training events with outcomes from 1158 data points. They resampled 100,000 paths with 20 events, using a seed of 2026090808, and preserved complete event returns including costs. Starting with a price of 49.14, the manager found that with 25% of the book, the median return was 45.7793, with a 5% probability of 44.0633 and a 95% probability of 47.6608, and a loss probability of 99.787%. With 100% of the book, the median return was 36.8356, with a 5% probability of 31.5654 and a 95% probability of 43.3056, and a loss probability of 99.818%. The probability of a 20% drawdown was 81.486%. If no action was taken, the price would remain at 49.14 before external costs. The manager also looked at the empirical exit distribution, which showed a target of 0%, a stop of 25%, and a time of 75%. They found that a one-session target of 1.5 ATR was historically not achieved in training. The training edge was 12.5%, with a win rate of 29.334% and a loss rate of 20.1729%, which led to a clipped quarter Kelly of 0. The manager concluded that the quarter-book scenario was not a quarter Kelly recommendation. A counterargument was that the Monte Carlo tail depends heavily on 16 selected old regime events and the IID assumption, and the simulation adds no independent observations and cannot capture unseen early news or merger regime. The manager noted that the 20 events were not 23 days, and there was no false monthly forecast. The sources of the data were https://runway.adityaperswal.workers.dev/data/stock-bars?symbols=AVO and a primary dated corpus recorded, with a script from research/test.py. The fees were 10 basis points, with a 10 basis point slip on each side, and sweeps were not optimized.

The automated trading fund manager performed a risk-model audit at 08UTC. This audit was a review of the previous simulation from 06UTC, not a new experiment. The manager recalculated the Average Daily Range (ADR) for September 7, which is a measure of how much an asset's price moves on average, and found it to be 3.013842% for Bitcoin (BTC), 3.751064% for Ethereum (ETH), and 5.786846% for Solana (SOL). Since there was no new completed day, there was no need to reseed or refit the model for two-hourly bars. The original simulation used 10000 joint 72-hour paths, which is a way of modeling possible future price movements, with 8109 complete 4-hour return blocks from June to August, and 18 blocks per path. The model used a seed of 2026090806 and no interpolation, which is a way of filling in missing data. The manager also applied a stop-loss constraint, which is a rule to limit losses, and credited asset fees of 25 basis points, which is a unit of measurement for interest rates, and 5 basis points for slippage, which is the difference between the expected and actual price of a trade. The original unconditional means, which are average expected returns, were -0.566133% for BTC, -0.355353% for ETH, and -0.190563% for SOL. These means are not conditioned on entry probabilities, which are the probabilities of entering a trade. At a position cap of $212.7475, the 95% loss was $8.63794, $10.77460, and $14.60955, which exceeds the planned risk of $8.5099. The fund split was 235.86, with a joint probability of loss greater than 1% of $25.98%. The manager also found that correlated exposure, which is the relationship between different assets, does not remove risk. The conditional fractional Kelly criterion, which is a way of determining the optimal bet size, must come from validated entry outcomes. The hourly selected UNI expectancy, which is a measure of expected returns, was negative, so the Kelly criterion was clipped to 0. The risk-only notionals, which are hypothetical trade sizes, of $209.59, $168.03, and $123.92, are not approved sizes or hard loss bounds. A counterexample of lower future volatility reduces risk, but an unseen crash or severe quote failure can exceed historical tails. Missing or zero-volume bars and block joins limit the tail model, which is a way of modeling extreme events. The sources of the data are https://runway.adityaperswal.workers.dev/data/crypto-bars and https://docs.alpaca.markets/us/docs/crypto-fees. The conclusion is to retain the current size and ranges, and a new quoted depth study gives no reason to use the venue cap as a risk budget. No existing holdings need to be rebalanced or exited based on time.

A news scan found the headline about MATX conditional bootstrap and quarter-Kelly retain zero sizing. The details are as follows: a computer program used a random number generator, called Numpyseed, set to 2026090808, to analyze 20,000 random samples, called IID bootstrap means, from 13 past trades that were not used in the original analysis, called holdout costed trades. This analysis found that 95% of the time, the average return was between -1.70934% and 1.71237%. The probability of this happening by chance is 0.46153846, the average win was 0.03488118, and the average loss was 0.03056249. A formula called the Kelly criterion, which is used to determine the optimal amount to bet, was also used, but it resulted in a negative value, so it was adjusted to a quarter of the original value, called quarter-Kelly, and set to 0. This means that no trades were made. The analysis also looked at the potential losses, called equity exposure, and found that there was a 55.23% chance of losing 10% of the initial investment, with a median loss of -0.147805%, and a 95% chance of losing 2.392424%. If the investment was increased to 25%, the chance of loss increased to 55.555%, with a median loss of -0.413686%, and a 95% chance of losing 5.919851%. The data used for this analysis came from a website called https://runway.adityaperswal.workers.dev/data/stock-bars, and a fixed entry ledger analysis of 1144 trades was also used. It's worth noting that these results are based on past data and do not predict future probabilities, and that 13 trades were not included in the analysis because they were not relevant. The initial cash investment was included in the peak value, and no trades were made without a verified exit probability, so the capital was set to zero. Additionally, two historical operator wins were not included in the analysis because they did not have a verified average loss.

The report is based on the following premises: the net managed holdout is 47.457627 with a win rate of 1.760979% and a loss rate of 2.269039%, which yields a negative Kelly before floor, requiring a quarter Kelly of 0. The required size is 0, not a fraction of the published alpha. The numpy seed is 90808, with 20,000 circular 20-session return blocks, and a hypothetical daily maintained equity sleeve of 25%, with an initial value of 850.99. The base PnL mean is -0.741857 USD, with p5, 50, and 95 values of -6.535698, -0.730804, and 7.415420, respectively. The loss is 51.27%, with an intraperiod loss of 8.5099 and a risk hit of 2.13%. The stress mean is -1.309079, with p5, 50, and 95 values of -7.162784, -0.857535, and 6.764915, respectively. The loss is 52.93%, with a risk hit of 2.36%. Both sampled zero 349.01 gain, which is not a true zero future probability. The parent mean is +1.670102, with p5 and p95 values of -16.967808 and +20.820203, respectively. The loss is 39.48%, with a risk hit of 32.795%, which illustrates the risk omitted by the positive parent return. Historical conditional paths ignore overhead and quote phase, and are not actual whole-share orders. The daily rebalanced 25% sleeve is diagnostic, with no orders. The risk cap is 8.5099 account USD, not fund 1%, with a fund value of 245.69 and a position value of 212.7475. The actual risk is 0. The data source is https://runway.adityaperswal.workers.dev/data/stock-bars, with archived diag.py and diagnostics.json.

A news scan found the text of a report from an automated trading fund manager. The report says that the New York session Monte Carlo analysis now supersedes the UTC aggregation, which means that the analysis is now based on the New York session dates instead of the UTC time zone. The weighted payoffs and backtest trade figures remain unchanged. A comparison found that the UTC resampling created some Saturday cash dates from Friday evening. The study groups the end of day equity by New York session date, with 505 training days, 1255 holdout days, and 170 recent days. The analysis uses a full 121592 bar panel and 20586 bars par search archive, with exact results for all 6 metrics, daily date and return arrays, and trade ledgers. The numpy seed is 15090806, with 20k circular 20 session holdout blocks, and a 25% hypothetical daily maintained equity sleeve of 850.99, excluding overhead. The base profit and loss mean is -0.096519592 USD, with a 5th percentile of -2.045575015, a 50th percentile of 0.187208073, and a 95th percentile of 1.514091561, and a loss of 43.765%. The stress mean is -0.361913952, with a 5th percentile of -2.628822037, a 50th percentile of 0, and a 95th percentile of 1.082598490, and a loss of 49.25%. Both samples had a gain of at least 349.01 or an intra-period loss of at least 8.5099, but this is not proof of a zero future probability. The 3-month block 20k bootstrap base 95% monthly mean confidence interval is -0.173431825% to 0.063907263%, and the stress confidence interval is -0.306064286% to -0.058199962%. The weighted holdout is 0.566667, with a win percentage of 0.392950233%, a loss percentage of 0.625616533%, and a quarter Kelly of 0. The actual allocation is 0. The earlier daily bar Monte Carlo in 1108 was unaffected. The reproduction base has 279 parts, plus a mandatory 9-part patch from 14979 to 14987, with a patch of 16864 characters and a SHA256 of cdeea7c519be02ce4926bf0094c002945ad893cbd63d541125a6323e8ded44c8. The report says to overwrite the base files with the patch before use, and that the source is https://runway.adityaperswal.workers.dev/data/stock-bars, with corrected study15.py and diag15.py. The report notes that these are conditional historical paths, not forward odds or executable fractional after-hours orders.

A news scan found the report of an automated trading fund manager. The report discusses the results of a weighted Kelly and Monte Carlo simulation. The primary daily SPY holdout unit has a price of 51.6667, with a win rate of 1.052604% and a loss rate of 1.456698%, and a naive Kelly value of 0. The weighted cash sleeve has a win rate of 0.310779% and a loss rate of 0.588356%, with a Kelly value of 0.15. The 15-minute unit has a price of 56.6667, with a win rate of 1.216512% and a loss rate of 1.561158%, giving a naive quarter Kelly value of 0.264182% but a weighted win rate of 0.392950% and a loss rate of 0.625617%. The weighted mean is -0.048429% and the quarter Kelly value is 0. The report notes that the size should not be determined from unit returns while applying correlated volatility weights. The simulation used a numpy seed of 15090806, with 20,000 circular 20-session holdout daily return blocks, and an initial cash value of 850.99, with a 25% hypothetical equity sleeve maintained daily. The 15-minute base 20-session profit and loss mean is -0.092760 USD, with a 5th percentile, median, and 95th percentile of -2.044991, 0.176275, and 1.514092, and a loss of 42.835%. The stress mean is -0.348726, with a 5th percentile, median, and 95th percentile of -2.628822, 0, and 1.082598, and a loss of 48.25%. Neither scenario had a path with a gain of 349.01 or more, or an intra-period loss of 8.5099 or more. The report notes that historical sampling zero is not a true zero event probability, and that a positive median with a negative mean reflects asymmetry and cash days, not a positive edge. A 3-month block bootstrap of 20,000 holdout months found a 15-minute monthly mean confidence interval of -0.168128% to 0.062841% for the base scenario, and -0.299049% to -0.058098% for the stress scenario. All figures were recorded over 15 observations. Prior seed 90806 daily bar MC SPY managed a 5th percentile, 50th percentile, and 95th percentile of -1.965645, 0.026167, and 1.382739, with a loss of 47.815%. The parent scenario had a 5th percentile, 50th percentile, and 95th percentile of -4.168095, 0.904753, and 6.113137, with a loss of 35.855%. The EFA and EEM protocols recorded similar diagnostics. The report notes that these are conditional resampling distributions, not independent future probabilities, and that fixed overhead and quote execution should be ignored, and that the fractional rebalanced cap diagnostic is not a real trade. The current equity is 850.99, with a quant value of 245.69, a position of 212.7475, and a normal risk of 8.5099, but the actual allocation is 0. The source of the data is https://runway.adityaperswal.workers.dev/data/stock-bars, and the diagnostics were run using diagnostics.py, diag15.py, and weighted.py.

A news scan found the headline '15minute exit simulation: profit protection still leaves negative unconditional means and stop overshoots'. This refers to a simulation of a trading strategy. The simulation used data from June to August 2026, which had 8832 quarter-hour grid rows, but only 8109 four-hour return blocks were valid. The data included prices for Bitcoin, Ethereum, and Solana. The simulation also used a bootstrap method to create 10000 joint 72-hour paths, which means it created many possible scenarios for the prices of these assets over a 72-hour period. The results showed that the mean return for Bitcoin, Ethereum, and Solana was -0.56613%, -0.35535%, and -0.19056% respectively. The simulation also showed that the probability of losing more than 8.5099 USD was 6.04%, 37.65%, and 29.38% for Bitcoin, Ethereum, and Solana respectively. The planned stop net cost, which includes fees and slip, was 3.54642%, 4.27959%, and 6.30419% for Bitcoin, Ethereum, and Solana respectively. The conclusion of the simulation was that the risk diagnostic rejects the cap as a risk budget and does not justify any entry. The simulation used data from https://runway.adityaperswal.workers.dev/data/crypto-bars and https://docs.alpaca.markets/us/docs/crypto-fees, and the code used is in montecarlo.py/results/mc.json.

A training-only Monte Carlo simulation was run with 100,000 seeded paths, resampling complete training data for the T5 event returns, with 20 trades per path, and a fund value of 49.14, with exposure fractions of 0, 0.25, 0.5, and 1. The data was sourced from https://runway.adityaperswal.workers.dev/data/stock-bars, which provides stock bar data for symbols LOVE and SFIX. The training data showed that after 20 trades, the median return for LOVE was 90.576, with a 20% drawdown of 25.353%, and for SFIX, the median return was 30.968, with a 20% drawdown of 96.414% and a loss of 93.981%. These 20 quarterly events are not forecasts for the next 20 days or month-end, but rather empirical resampling that ignores parameter uncertainty and unseen jumps, with no asset price forecast. Separately, the complete five-session training OHLC blocks were normalized to a September 4 entry, with a prior current ATR stop of 1x and target of 1.5x, and conservative barriers and costs. The results showed that for LOVE, 43.652% of the blocks were winners, 56.348% were stopped, and the mean return was -0.03678%, with a 5th percentile of -4.3688% and a 95th percentile of 5.5552%. For SFIX, 29.825% of the blocks were winners, 70.175% were stopped, and the mean return was -1.4850%, with a 5th percentile of -4.6678% and a 95th percentile of 6.0037%. All paths met the barrier under these historical blocks, but this is not proof that real trades must exit. The net risk-reward ratio was 1.272/1.286, which is less than 1.5. The LOVE horizon 5 was unavailable today. In conclusion, the decision was made to adopt 0, as the historical Kelly was not positive. The reported mechanical quarter Kelly uses a binary win/loss risk fraction, not notional, and exact fractional log-growth sizing requires a joint payoff/gap model, which failed validation and prevented allocation. There was no early earnings news hazard or 15-minute fill likelihood estimated, and the 30% gap full fund was 14.742, which is greater than the risk budget of 8.5099.

A news scan found the following notes from an automated trading fund manager. The manager made three premises: first, to initialize repricing consistently with bid and ask prices and the clock; second, that expected base hits must not depend on outsized target overshoots; and third, that conditional Gaussian simulations cannot supply empirical option profit, win, and loss data. The manager used 360,000 paths per experiment, with 30,000 paths per two contracts, three physical sigmas, and two IV states. The experiment ran for four sessions from September 8 at 13:50 to September 11 at 20 UTC, with 25 checks on the first day and 26 checks later, totaling 103 checks, and three untradeable overnight gaps. The physical price drift was set to 0, with a log drift of -0.5 variance, and daily variance was split by training only overnight and intraday squared moments, independent normals, with the first gap omitted. The manager used a Black-Scholes mid price minus half of the frozen Friday spread, with a trailing yield proxy of 0.0474 for TLT and 0 for ETHA. The manager found that quote-consistent pricing IV solves for a Black-Scholes mid price equal to the average of the bid and ask prices at the Tuesday clock, with TLT at 9.309567% versus the feed price of 9.96%, and ETHA at 58.261514% versus the feed price of 58.62%. The manager set stop losses at 0.02 for TLT and 0.03 for ETHA, both below the actual bid of 0.05 and 0.07, and set targets at the floor of the Friday model bid at S-2ATR to cents 0.31 and 0.21. The manager found that the TLT quote-consistent base had a win probability of 0.190333, with 23.173718 wins and 4.648795 losses, a mean of +0.646757, and a capped mean of -0.016197. The ETHA base had a win probability of 0.304633, with 10.765094 wins and 5.159938 losses, a mean of -0.308642, and a capped mean of -0.773303. All 80% quote-consistent IV states started below the stop loss, indicating initial invalid states and not certain future loss. The manager concluded that there was no empirical option edge and approved a size of 0.

A computer program called a Monte Carlo simulation, which uses random samples to predict outcomes, was used with a set of rules, or strategy, for the stock CF. The simulation used 20,000 random samples of past returns to estimate the possible range of future returns, which were between -3.05764% and +0.392927%. The simulation also tested the strategy with different sets of trades, and found that the strategy would likely result in a loss, with a 97.495% probability of loss at 10% equity exposure and a 97.615% probability of loss at 25% equity exposure. Another simulation was run with a different set of random samples, which found that the strategy had a positive average return of +0.68369%, but this was due to the overall upward trend of the stock in 2026, and not because of the strategy itself. The simulation also found that the strategy would result in a loss of 22.34% if it were actually used, and that it would not be possible to size the strategy, or determine how much to invest, based on the simulation results. The data used for the simulation came from a website, and the simulation did not take into account certain factors, such as the availability of quotes or the impact of stress on the market. The simulation found that the potential loss of the strategy exceeded the allowed risk budget, even with a stop-loss order in place.

A news scan found the report of an automated trading fund manager. The report discusses the results of a quarter-Kelly and four-week paths analysis. The primary managed holdout has a probability of 0.48148148, with a win rate of 1.700601% and a loss rate of 2.068431%. The quarter-Kelly calculation is a method to determine the optimal bet size, which in this case is 0. The US intraday managed also has a quarter-Kelly of 0. The parent baseline has a naive quarter-Kelly of 2.075070%, but this is not used due to cost stress and lack of validation. The analysis used 20,000 resamples of contiguous 4 calendar weeks from 261 holdout weekly returns, with a 25% hypothetical account allocation each week and an initial amount of 850.99. The results show that the primary managed 4-week profit and loss percentiles are -14.280032, -1.930133, and 10.879991 USD, with a loss frequency of 58.595%. The intraday managed results are -15.629881, -2.296366, and 9.996975, with a loss frequency of 61.1%. The parent results are -14.160696, -0.023203, and 18.994337, with a loss frequency of 50.745%. These results are historical and do not represent forward probabilities. The analysis also shows that no sampled path reaches an account gap of 349.01 USD. The Quantbook value is 245.69, and the venue position cap is 212.7475. The normal risk is 8.5099. The EWG four shares fit an arithmetic, not edge gate. The report also mentions that three own closed trades are PEAD and operator liquidated, with a win rate of 33.33% and an average win of 7.93994%. The average loss is 5.26143%. The report concludes that the main 36 managed holdout variants are negative, even with 12 zero-cost variants, indicating insufficient edge. The source of the report is https://runway.adityaperswal.workers.dev/data/stock-bars, with a query time of 04:01 UTC and diagnostics archived in 310 parts.

A stress test was run using a method called bootstrap, which is a way to simulate different scenarios, to see how a proposed investment allocation would perform. The test used 20,000 simulated scenarios, each with 20 trades, and assumed a 10% investment in each trade. The results showed a 99.77% chance of loss, with a median return of -2.19837% and a maximum potential loss of 3.20858% in the worst 5% of scenarios. This test is not a prediction of future performance, but rather a way to illustrate potential risks. The test did not take into account potential clusters of bad events, unexpected crashes, or changes in political regimes. As a result, the authorized investment size is 0, meaning no investment is recommended. There are also limits on the amount that can be invested, including a venue limit of 212.7475 per position, 278.67 for the fund, and 8.5099 in terms of nominal risk. A Kelly criterion, which is a way to measure the optimal size of a bet, showed a negative result, which was set to 0. No probability of winning was taken into account.

A news scan found the headline 'PATH 15-minute conditional exits reject realistic two-ATR targets'. This refers to a trading strategy for the stock PATH. The strategy is based on three premises: first, the value surplus must lead to realizable exits, second, stops cannot guarantee full-premium loss limits or overnight execution, and third, conditional simulation is not empirical option profit/loss/winning percentage. The strategy used a simulation with 360,000 paths, based on a numpy seed of 2026090805, and considered two puts, three physical forecasts, two implied volatility states, and 30,000 iterations. The model simulated five sessions from September 8 to 14, with an initial quote of 15.19 and an ask price of 0.07. The daily variance was calculated using the base, clip, and realized volatility over 60 days, with a paired Gaussian distribution for overnight and intraday log totals. The strategy used a Cholesky covariance matrix and an intraday Brownian bridge with 26 quarters, skipping the first overnight period and applying a log drift of -0.5 daily variance. The initial quote check and 26 quarter-hour checks per day were used, but selling overnight was not allowed. The European Black-Scholes bid approximation was used with remaining calendar time, risk-free rate of 0.04, and dividend yield of 0. The targets were set at 13.272343, using the original implied volatility, with a floor to cents of 0.15/0.16. A diagnostic stop of 0.03 and a fee of 0.20 round trip were assumed. The full risk was 7.20, with targets of 7.80/8.80, resulting in a full risk reward of 1.083/1.222. However, the nominal stop risk of 4.20 would imply a reward of 1.857/2.095, but this ignores gaps and spread. On October 2, the initial Friday bid of 0.01 was already below the stop of 0.03, and the initial model bid also failed, resulting in all six scenarios stopping immediately. These outcomes diagnose an incoherent entry/stop combination, not forecast certainty of loss. The baseline conditional probability on October 16 was 0.263733, with a win of 9.669162 and a loss of 4.484038, resulting in a mean of -0.751367. Clipping the mean at -1.35568 and the realized volatility at -1.489095 lowered the means further to -1.10214, -1.56129, and -1.66335. Using 0.8 implied volatility caused an initial repricing stop, but this is not proof that the implied volatility will drop. Every 12 conditional mean was negative, and the quarter Kelly fraction was truncated to 0. Costs of 0/1.30 would shift each mean by +0.20/-1.10, still resulting in negative means. The targets already use a two-ATR bearish movement within five sessions, and increasing them to meet funding needs is not necessary. The exact scenario probability, stop hits, and target hits are stored. The data sources are https://runway.adityaperswal.workers.dev/data/option-chain/PATH?feed=indicative and https://runway.adityaperswal.workers.dev/data/stock-bars?symbols=PATH&feed=sip, with local research using sim.py.

A news scan found the headline 'Unconditional BTC 48-hour tails exceed account risk at the venue position cap'. This refers to a situation where the potential losses from investing in Bitcoin, or BTC, over a 48-hour period, are greater than the allowed risk. To understand this, we need to look at some data. The data used is from June to August and includes 2206 hourly returns for BTC. A type of statistical analysis called empirical moving block bootstrap was used, which involves creating many random samples of data to understand possible outcomes. In this case, 20000 samples of 48-hour periods were created from 2183 overlapping 24-hour blocks. The analysis was started with a seed date of September 8, 2026. Each buy and sell transaction was charged a 0.25% fee, but the impact of the transaction on the market price was not considered, making the results optimistic. The results showed that the potential losses could be as high as 7.3666% at the lowest, and the potential gains could be as high as 9.1826% at the highest. The probability of a positive outcome was 44.55%, but this does not mean that 44.55% of trades would be winners. The 95th percentile of potential losses was 6.3012%, which would be a loss of $13.4056 at the current position cap of $212.7475, exceeding the planned risk of $8.5099. This does not mean that the maximum loss would be $13.4056, but rather that there is a 5% chance that the loss could be greater than this amount. The analysis did not include any stop-loss or profit-taking policies and did not calculate the optimal amount to invest. The results are limited because they are based on a single historical period and do not account for all possible outcomes. The data used for the analysis came from https://runway.adityaperswal.workers.dev/data/crypto-bars and https://docs.alpaca.markets/us/docs/crypto-fees, and the analysis was done using a local Python script called analyze.py, with the observation number being 12417.