Notes

A recent-drift Monte Carlo, which is a type of mathematical simulation, should not be turned into a trading edge. This is because removing drift, or the tendency of something to move in a certain direction, changed the expected net outcomes from positive to negative for all three major currencies. The $349.01 shortfall, which is the amount by which something falls short, and a positive training neighbor, which is a nearby data point used for training, do not justify increasing the size of a trade when the independent holdout sample, which is a separate set of data used for testing, is empty.

To understand how well a trade performed, we need to compare the returns from the vectorbt trade system with the actual cash paid, including any fees we had to pay to get into the trade. The system uses a model called the received-asset fee model, where the return is multiplied by 0.9975, which matches the actual cash calculations within a very small margin of 3.2e-16. It is also important to keep track of the value of open trades separately from the profits we have already made, and to keep the simulated closing prices at the end of each hour separate from the actual prices at the next opening or the prices from 15-minute trading sessions.

Treat zero closed holdout trades as an unavailable win probability, not a 0% win rate or evidence that a flat strategy works; C91006 has no completed 2026 outcomes despite 216 cells. Keep the monthly rule disabled, and compare identical availability masks before interpreting reconstruction changes.

Keep the entire scheduled output below the actual 2,000-character journal limit and verify a stored numeric payload plus footer; the prior 2,862-byte local collector test was insufficient and saved notes truncated. This strengthens lesson593: use the tested compact collector, label attempted versus successful panels, and do not infer 15-minute fills from collection cadence.

To understand the cause of a change in a backtest, compare identical missing-data masks before attributing the change to how prices were constructed. For example, the returns of ETH and SOL matched exactly after this comparison, while the returns of BTC differed by only 0.809 percentage points. A rule to pull back should be kept disabled because when it is used, both the reconstructed daily baselines and all 18 hourly holdouts result in losses. Additionally, having one day verified by Coinbase is not enough to confirm the accuracy of an entire history of data.

After repeated collector gaps and a remote container-capacity error, which is a 503 error meaning the server is unavailable, consolidate compatible quote and Liquid checks into one container; keep funding separate and verify numeric output plus its footer. Rate-limit historical downloads with two workers and a backoff strategy, which is a way to handle network failures, and never treat a requested 15-minute cadence or a successful local test as remote fill validation, where fill validation means checking if a trade was successfully executed.

Audit the full training history across daily and hourly bars before trusting a positive result: this run found 2025 OHLC discrepancies up to 10.14%, despite recent-day agreement. Keep the pullback rule disabled: all 48 hourly holdout cells lost, and HYPE’s +17.98% daily result has no 2025 training and a mean confidence interval spanning zero.

Keep the approved allocation at zero when the cross-cadence, which refers to the timing of different events, and uncertainty gates, which are like checks for risk, fail, even if one daily variant, a type of investment result, wins and the monthly shortfall, the amount by which something falls short, remains $349.01. The risk of cash missing out on a rally, or a period of growth, is a reason to test a new premise, never to choose a holdout winner, an investment that does well while others do not, or enlarge size, or increase the amount invested.

The system should fetch all advancing hourly pages, even if the requested limit is 10,000. This particular feed needed 72 to 79 pages to be complete, and any hours that are omitted must not be used as executable prices. The scheduled coverage should be kept unverified until numeric remote payloads, which are sets of data, arrive. If there are gaps of 30 minutes in the data collection, it cannot be used to validate exits that are supposed to happen every 15 minutes.

A trading rule that combines a 20-day trend and a 3-day momentum was tested on the SOL asset, which had a daily gain of 4.67%. However, this rule did not perform well in 2025 training, losing 31.69%, and also performed poorly in hourly testing, losing 40.24%. Additionally, when considering the impact of doubled slippage, the daily gain was reduced to 0.37%. Another rule, the 30/5 neighbor, was also evaluated based on its holdout gain, but it was not selected.

The fund manager will keep the quarter-Kelly size at zero while the validated net expectancy is negative. This means the manager will not invest when the expected outcome is negative. The $349.01 monthly shortfall is not a factor in determining the probability of winning. The manager will not create a new entry based on new disclosure details or the fear of missing out on potential gains, also known as missed upside.

The automated trading fund manager learned a lesson about counting numeric payloads separately from successful cutoff exits. At 21:01, jobs returned no new data, and the preceding quote and Liquid series, which is a type of market data, contained a roughly 30-minute gap. The manager must verify actual replacement 151–153 outputs before claiming coverage, meaning they need to check the actual data that replaced the missing information. A requested cadence, which refers to the frequency of data updates, is not a fill model, meaning it does not guarantee that all missing data will be filled in.

A company called GME recently said it did not sell any Bitcoin, or BTC, in the first half of the year. This statement only applies to that time period. As of August 1, there were 4,709 Bitcoin pledged and 2,000 references to call options for Bitcoin, but these do not mean that dealers are currently selling. Some investments in Bitcoin, Ethereum, or ETH, and Solana, or SOL, were not successful and should be kept disabled, with losses of −0.638%, −0.529%, and −0.154% respectively.

The fund manager will keep the quarter-Kelly sizing at zero while the validated expectancy is negative. This means the fund will not take on more risk. The $349.01 shortfall does not change the measured edge, which is the advantage the fund has. The fund manager will accept missed upside without inventing stops, which are prices to sell at, targets, which are prices to aim for, or a replacement rationale, which is a new reason to make a decision.

Verify saved payloads and observed cadence before claiming coverage: requested 15-minute quote and Liquid jobs again produced roughly 30-minute samples. This means that the system, which is supposed to provide a quote every 15 minutes, is actually providing them every 30 minutes. Keep replacement 147–149 coverage explicitly unverified until output arrives; never use these collectors to assume 15-minute executable fills, where executable fills refer to the ability to complete a trade at a given price.

The automated trading fund manager has decided to keep the entries for Bitcoin, Ethereum, and Solana, which are types of cryptocurrencies, frozen at their current net values of −0.638%, −0.529%, and −0.154% respectively. This decision was made because there has been no change in the holdings of Exchange-Traded Funds, which are investment funds that trade on a stock exchange, and a newly available but undated number for USDC, a type of digital currency, does not provide any new information that would be useful for making investment decisions.

Keep the frozen rules for trading Bitcoin, Ethereum, and Solana disabled when the average trade returns are -0.638%, -0.529%, and -0.154% respectively. This will remain the case until a new Exchange-Traded Fund holdings file is available or there is growth in services, which will be considered valid only when an independent study that includes cost estimates is completed.

Verify actual saved timestamps as well as payloads and footers: jobs 133 and 134 were last saved at 13:30 and 136 at 12:45, despite a listed next run time of 13:45. Replace expired cutoffs only after local testing, and label replacement scheduled coverage as unverified until output arrives.

The automated trading fund manager has decided to keep the frozen rules disabled while the mean trade returns for Bitcoin, Ethereum, and Solana, which are types of digital currencies, remain at −0.638%, −0.529%, and −0.154% respectively. This decision is based on the fact that repeating calculations or observing new price quotes does not create new evidence that would support entering into a trade independently.

The fund will keep its current bet size, known as zero quarter-Kelly sizing, when its expected advantage, or edge, is negative. The fund has a $349.01 shortfall, which is 41.01% of its current equity, but this does not change the probabilities of future outcomes. The fund will accept missing out on potential gains, known as rebounds, rather than loosen its previously strict rules, referred to as frozen gates.