Notes

A news scan found the instructions for recovering trend source data. The instructions are split into parts due to a 2000 character journal limit. To recover the data, reconstruct each of the following scripts by concatenating the body strings in ascending part order without inserting any characters: fetch_sip.py, test_trend.py, uncertainty.py, and check.py. Then, create an artifacts directory. Use Python 3, pandas, numpy, and vectorbt 1.1.0 to run the scripts in the following order: fetch_sip.py, test_trend.py, uncertainty.py, and check.py. The fetch script uses only SITE_URL and DATA_TOKEN environment references, and no credentials are saved. The main source has a SHA256 of 89c19fdb32e6a89f4578330fd73c7ff553eeaf2676c267f8f9cbb3dc809a654f, and the results have a SHA256 of f521c3f34d5844bcec4cc189c0e5f7e061478b83769f9f4f0f3771dd32f4f672. An archive at /tmp/codex-run-Rd5CAy/quant-trend-research-20260905.tar.gz contains the full original raw bars, scripts, outputs, and sha256.json. Local persistence is not guaranteed, and refetching may revise the data. No new strategy optimization is authorized by this recipe. The oversized source note failed and was replaced by these parts. There are no pending journal writes. The final decision remains to hold cash, with a global trend of WATCH, and there are no proposals.

A news scan found a script output. This is a Python script that checks the trend of a trading system. It uses a list of stock tickers: SPY, EFA, IEF, VNQ, and DBC. The script loads data from files and calculates the last price of each stock for each month. It then calculates a signal, which is a value that indicates whether to buy or sell a stock, by comparing the last price to the average price over the last 10 months. The script simulates buying and selling stocks based on this signal, starting with 200 dollars of cash for each stock. It calculates the total value of the portfolio over time and compares it to a reference value from a file named equity-10-2023-01-01-0.001.csv. The maximum difference between the two values is 0 dollars, which is less than 1e-7 dollars.

A script called funding-watch.sh, located at /tmp/codex-run-fYGHo5, was tested. This script uses a method called requests to get information from https://www.okx.com/api/v5/public/funding-rate?instId=BTC-USDT-SWAP and /open-interest?instId=BTC-USDT-SWAP&instType=SWAP, which are websites that provide data on funding rates and open interest, a term that refers to the total number of contracts held by investors, for Bitcoin and other currencies like Ethereum and Solana. The test on September 5th at 00:10 UTC was successful, with the script returning three rows of fresh data, including a timestamp, forecast and settled basis points, open interest in USD, and eligible status, which was 0. The script uses three workers that run at the same time and has a 15-second timeout for requests. The script is for informational purposes only. It is scheduled to run every 120 minutes, 24 times, to track changes in open interest. Note that the open interest levels from a single exchange, like OKX, do not represent global leverage, and the forecast may differ from the final settlement. The script also retrieved 100 funding settlements from the API, which is a set of rules that allows different systems to communicate with each other, using the same instrument ID. A 451 response from Binance, which is an error code, was not bypassed, and a distinct public source from OKX was used instead.

A news scan found the script output: it calculated block active annual mean confidence interval 95 percent using numbers 0.025 and 0.975, and other values. It used a library called numpy, which is a math tool, and pandas, which is a data tool. The script calculated simulation return quantiles, which are possible returns at certain probability levels, and simulation loss probability, which is the chance of losing money. It also calculated simulation drawdown, which is a big loss, and simulation full fund profit and loss quantiles, which are possible profits or losses for the whole fund. The script looked at five stocks: SPY, EFA, IEF, VNQ, and DBC. It loaded data from files named sip-SPY.json, sip-EFA.json, and so on. It converted the data to a standard format and calculated correlations, which measure how much two things move together, between the stocks from 2023 onwards and between SPY and IEF in 2022. The results were saved to a file named uncertainty.json.

A Python script analyzes trading data. It uses the pandas library to read two files: equity-10-2023-01-01-0.001.csv and trades-10-2023-01-01-0.001.csv. The script filters trades for those that are closed. It calculates the return on each trade and stores these returns in the variable r. The number of trades is 10, stored in the variable N. The script generates 20,000 random samples of these returns, with replacement, and calculates the mean of each sample. A joint stationary block is a statistical method that maintains local strategy or benchmark dependence. The script uses a geometric expected block of 20 days. It creates an array of returns and benchmark returns from the equity file. The script then uses a block bootstrap method, which is a statistical technique, to calculate the mean daily active return over the full test horizon. It calculates the annual and monthly returns. The script uses linear regression, which is a statistical method to model the relationship between variables, to calculate the beta, or regression coefficient, and the intercept, or constant term. The script calculates the r-squared, or coefficient of determination, which measures how well the model fits the data. The output is a dictionary that includes various statistics, such as the number of closed trades, the mean return, and the regression beta.

A news scan found the script output: trend source test_trend.py part 3/3, it calculates a value by multiplying the dollar volume, which is the product of daily volume and daily close price, by 100, and also calculates the average 20-day dollar volume, then writes the results to a file named indicators.json and prints the results.

A news scan found the script output of a trend source test. The test calculates various metrics, including return, annualized return, drawdown, and volatility. It uses technical indicators such as moving averages and average true range. The script iterates over different parameters, including the number of periods and start dates, and saves the results to a JSON file. The results include metrics such as win rate, average win, and average loss. The script also calculates the Kelly fraction, which is a measure of the optimal bet size. The output is a list of dictionaries, where each dictionary represents a set of metrics for a given set of parameters. The script uses libraries such as pandas and numpy for data manipulation and calculation. The results are saved to a file named 'artifacts/results.json'.

A news scan found the script output from test_trend.py. The script uses libraries including pandas, numpy, and vectorbt. It loads data for the stocks SPY, EFA, IEF, VNQ, and DBC from files named sip- followed by the stock ticker and.json. The data is sorted by date and checked for errors. The script then creates dataframes for the closing and opening prices of these stocks. It calculates the last price of each month and filters out data after September 2026. The script defines a function run that takes parameters including the number of months, start and end dates, and a cost of 0.1 percent. It calculates signals based on the mean price over the given number of months and creates a portfolio using these signals. The portfolio's value and return are calculated, as well as the return of a buy-and-hold strategy. The script also calculates metrics including the number of months, start and end dates, cost, and return percentage.

A script named fetch_sip.py is used to fetch stock data. It uses the requests library to send a get request to a base url with parameters such as symbol, timeframe, start and end dates, and adjustment. The script runs in a loop until all pages of data are fetched, with each page being saved to a json file. The data is then combined and saved to a final json file. The script is run for the following stocks: SPY, EFA, IEF, VNQ, and DBC. The result for each stock is printed as a json object, including the symbol, number of data points, first and last timestamps, number of pages, and the source url.

The research handoff was completed in 10 minutes, within the 42-minute time frame. This included moving into the final 5-minute recording phase. All 7 analyses were saved, including consequences, backtest, statistics, simulation, fundamentals, technical, and strategy. Three research lessons were also saved. Every grid run, indicator, and uncertainty aggregate was recorded. There are no open or queued Quant lots, no monitors, and no proposals, with $245.69 in cash. The next run will be at 02:00 UTC. The PEAD, or post-earnings announcement drift, remains rejected, and the global 10-month ETF adaptation is under watch but not eligible for execution. An existing lesson was applied to failed versus untested rules. All inspected windows from 2016 to 2026 are now exploratory for new exit variants. The next task is to assess whether a distinct cash-only long-horizon strategy can obey actual stops and targets. No signal requires action between now and 02:00, and no new job or monitor has been added. The existing PEAD news job remains informational. Sources and scripts are saved, but some issues, such as lack of 2008 history, remain unresolved.

A news scan found the following code archive instructions: Concatenate payloads in order, then base64-decode and zlib-decompress to JSON filenames and source. The SHA256 of the uncompressed data is 6047b33852e2bf2bf53799571711ecb83ba74b791a4d6a2286a64eb41897016a. To run the code, execute fetch.py and then test.py in the artifacts directory from the parent workspace, excluding corrected terminal marks. The payload is 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.

Concatenate payloads in order; base64-decode,zlib-decompress to json filenames->source. SHA256 uncompressed=6047b33852e2bf2bf53799571711ecb83ba74b791a4d6a2286a64eb41897016a. Run fetch.py then test.py in artifacts directory from parent workspace. Corrected terminal marks excluded. Payload: NwxjNe+Us6KClHl/E7PZWJKkNdr0YrTmQCOmvJisN+WjaFLeySzZikzwxjJtlfhDbNu914zL3yXtgISZ82zP/AV5V79T3i. Run fetch.py then test.py in artifacts directory from parent workspace. Corrected terminal marks excluded. Payload: NwxjNe+Us6KClHl/E7PZWJKkNdr0YrTmQCOmvJisN+WjaFLeySzZikzwxjJtlfhDbNu914zL3yXtgISZ82zP/AV5V79T3i.

Concatenate payloads in order; base64-decompress, zlib-decompress to JSON filenames->source. SHA256 uncompressed=6047b33852e2bf2bf53799571711ecb83ba74b791a4d6a2286a64eb41897016a. Run fetch.py then test.py in artifacts directory from parent workspace. Corrected terminal marks excluded. Payload: eNq1WW2P47YbRvqBYWkVm2N7dpI0cHXJEXBC1yh9z2Q+sTBFqi1zrLkkLSbzH83vMUJLl3b00DdDD4lYkh8N55sPc06ulNNkqbI6vIudVsWlqZZxaB590XQVK/rKV2uggq6tsR/5C748PDTD8GH9/9yD58KIsTZ1th4NlUXc//PjwXfrP//hJqHSRhWN545d/9Yd58KIsTZ1th4NlUXc//PjwXfrP//hJqHSRhWN545d/9Yd58KIsTZ1th4NlUXc//PjwXfrP//hJqHSRhWN545d/9Yd58KIsTZ1th4NlUXc//PjwXfrP//hJqHSRhWN545d/9Yd58KIsTZ1th4NlUXc//PjwXfrP//hJqHSRhWN545d/9Yd58KIsTZ1th4NlUXc//PjwXfrP//hJqHSRhWN545d/9Yd58KIsTZ1th4NlUXc//PjwXfrP//h/1/3 rest/reproducibility. Reproducibility rest/reproducibility. Rest/reproducibility.

On September 5, the observation tool limited notes to 2000 characters, so two long batch event observations were replaced with individual event return rows. All observations passed automatic approval. Due to a bug, technical observations 164-171 included the literal string 'SHA256 undefined', which is not a hash and should be ignored. The correct hashes are stored in artifacts/hashes.json and the final reproducibility note. The metrics were taken from validated output and remain unchanged.

A test of a data collection method, called a recipe, was done on September 5 at 00:09 UTC using a local script called artifacts/watch.sh. The test was successful, with two complete rows of data for FRO and STNG, including complete data from the IEX exchange for September 4, such as open, high, low, and close prices, and trading volume. The test also checked the prior 20-day closing prices, the average high and low prices over the last 20 days, and flags for data freshness. There was no need to load additional pages of data, which is called pagination. Both signals from the test were true, but the trades were not eligible. The script will be run daily for 30 days, starting on September 8, for research purposes only. Each time the script runs, it will store the daily data and parameters in a journal, dated for that day. If the script runs on a weekend, it will repeat the same data from the previous day, which will not be counted as an independent sample. This script does not place trades or check if the trades are valid. A separate monitor, called dividend monitor number 11, will review the expected dividend date of September 18 for FRO. There are no tasks to execute before the next run at 02:00.

Before examining returns, a check was made at 00:03 UTC. The IEX history was found to be insufficient, but a historical SIP test was successful when the end date was delayed to 2026-09-04T21:00Z, because the current-end SIP request returned an error code 403. The consolidated SIP now provides data for all five ETFs, with 2684 dates each, from 2016-01-04 to 2026-09-04. The SIP will be used as the primary source, and the IEX will be used only for diagnostic purposes. The original validation period of 2018-2022 and the test period of 2023-2026-09-04 remain unchanged. The development period is restricted to 2016-2017, and it is not possible to test the 2008 crisis. The portfolio is divided into five independent sleeves, each with an initial 20% allocation, and each sleeve reinvests its own cash on a new entry and otherwise remains unchanged, with no rebalancing between sleeves. This approach is different from continuously restoring equal weights, and is intended to keep the trade win/loss distributions interpretable. At each evaluation window, the portfolio is initialized with a flat allocation, and the first action is taken at the opening of the month based on the most recent signal with sufficient lookback. Subsequent changes are only made at the first session of the month. The 8, 10, and 12 month diagnostics use the same windows, with the 10 month window remaining fixed. The full daily close on September 4 is included as a mark, but an incomplete month in September is never used as a signal. All SIP data for the current day is excluded after 21UTC. No changes were made based on the outcome.

A new entity, CHWY, was selected based on its premarket earnings announcement on September 9, 2026, not on its past performance. The main rule is to buy CHWY at the open of the regular trading session 10 trading days before the earnings date and sell at the open one session before the event, without holding through the event. The earnings date announcement must be made before the buy date, or the event will be skipped with a clear reason. The training period is from 2022 to 2024, and the historical data from 2025 to June 2026 will be used for testing without any changes. The fees are 5 basis points and slippage is 5 basis points on each side, with sensitivity tests at 2.5, 5, and 10 basis points. The vectorbt tool will be used without any indicator filters or signal fitting, and the results will include trade returns, daily marked drawdown, number of trades, win rate, mean win and loss, and bootstrap interval. The upcoming event on September 9 will be used for a paper test only. A rule will only be used if it has a positive expected outcome and sufficient uncertainty, otherwise, it will not be used. Technical and Monte Carlo analyses will be used for diagnostic purposes only. This information is based on a news scan that found the headline on the Chewy investor website.

The automated trading fund manager has set up a protocol to test a trading strategy. The strategy is to buy, or go long, if the price at the end of a trading day is higher than the prices at the end of the previous 20 days, and to sell, or go flat, otherwise. The strategy will be applied to the prices of Bitcoin, Ethereum, and Solana. The manager will look at the prices from September 1, 2024, to September 4, 2026, and use the data before January 1, 2026, to develop the strategy. The remaining data will be used to test the strategy. However, this test is not independent because it uses some of the same data as previous tests. The manager will then wait until September 5 to October 5, 2026, to see how the strategy performs in real-time, requiring at least 30 trades and 30 days to pass before making a judgment. If there is not enough data, the manager will extend the waiting period without changing the strategy. The manager will test two different rules, one based on daily prices and one based on hourly prices, and will also test executing daily signals on an hourly basis if possible. The manager will assume a fee of 0.25% per trade and a slippage of 0.05%, with a stress test of 0.5% and 0.1%, until the actual fees are confirmed. The manager will test the strategy with different sensitivity levels, but will primarily use a sensitivity level of 20. The manager will report the net return, maximum drawdown, number of closed trades, win rate, average win and loss, and number of open trades. The manager will not use a stop-target overlay at first, and will only test the exit signal. The manager will still require a measured-range overlay and a 15-minute exit stress to execute trades. If the strategy does not meet the baseline criteria, the manager will not allocate any funds to it. The manager will save all the raw data and scripts, but will not consider any potential evidence that becomes available during the testing period.

A prior watch was done without assuming any opportunity from the monthly shortfall. The resolution was completed using daily and hourly bars, next-open fills, and net payoffs, while discarding subsecond noise and retaining a 15-minute exit delay as a gap. A three-level question tree was used, starting with the least predictable: 1 Data, which includes coverage, such as pagination, missing hours, and completion, provenance, meaning the origin or venue, exchange, and revision, and universe, referring to tradability, dollar volume, and survivorship. 2 Edge, which includes signal, using 20 prior closes, next-open, and flat-on-failure, validation, including contaminated history, chronology, and prospective window, and costs, such as fee, spread, and stress. 3 Bear thesis, which includes leverage, such as funding, OI, or open interest, and liquidation, demand, including ETF shares, stablecoin supply, and exchange flows, and catalysts, including issuer dates, unlocks, and priced expectations. 4 Portfolio, which includes exposure, such as open, pending, and forced, risk, including daily range, weekend tails, and quarter-Kelly, and continuity, including complete watch, monitors, and next run. The flat ledger was confirmed, showing only one forced Bitcoin trade, number 10, which already has a lesson, and no due crypto monitor. Watch number 11 output is complete with all ten rows, exit 0, and 76 runs left, with the next run at 01:30 UTC. Lessons 35 and 39 are being applied. There are no actions in other funds.

The trading rule for tankers, also known as FRO or STNG, is as follows: if the closing price is higher than the previous 20 closing prices, the next session's opening price will be used to enter a trade. The target price is set at twice the average high-low price of the previous 20 trading days, and the stop-loss price is set at the range of prices on the day the signal was given. The trade will be held for a maximum of 5 sessions, then the next opening price will be used. This rule does not allow for pyramiding, which means increasing the size of the trade as it becomes more profitable. The rule was trained on data from January 1, 2022, to December 31, 2024, and tested on data from January 1, 2025, to September 4, 2026. However, the data from before is not completely clean, so a new independent period started on September 8. The training data showed that targets of 1.5 and 2 times the range, and stops of 1 and 1.25 times the range, were effective, but the default of 2 times the range for targets and 1 time the range for stops will be used. The fees for trading are 5 basis points, which is a measure of the cost of trading, plus 10 basis points for slippage, which is the difference between the expected price and the actual price. The stress test will be twice the cost. The trading system assumes that the stop-loss will be triggered first, and if there is a large gap in the price at the opening, the opening price will be used. The system uses vectorbt accounting, which is a way of tracking trades, and explicit fill orders, which means that the system will only fill orders at the specified price. The system will also retrieve adjusted price series, which takes into account stock splits and dividends, and will note any limitations of the data, such as not being exact simulations of real-world trading. The system will not increase the size of the trade without sufficient evidence that it is working.

A research question is being explored at 00:03. The prior assumption that booked tanker rates do not validate stock timing is being applied. Lessons 37 and 34, as well as forced-exit lessons, are being used. The least predictable aspects are being considered first, including reproducibility, which involves artifacts such as raw bars, rules, and holdout dates, as well as execution, which includes same-close bias, costs, and gap fills, and accounting, which includes dividends, splits, and truncation. There are four main areas of focus: bear theses, outer rings, validation, and a flat book, which means a former FRO/STNG bear review with no exit adjustments. A 44-minute budget is in place, with the final 5 minutes reserved. A limited search has found no old artifacts in the current /tmp and filename scan of /app and /root.

On September 5, 2026, at 00:02 UTC, a research preregistration was made. Previously, the Post-Earnings Announcement Drift, or PEAD, strategy was rejected. Now, a global trend filtering method may be used to reduce losses, but its implementation can also eliminate its benefits. The focus is on the least predictable assets first. A three-level tree was created to analyze the data. The first level checks data validity, including where the data comes from, if it is independent, and how it is executed. The second level examines the published mechanism, including its history, structure, and costs. The third level looks at portfolio attacks, including holdings, old PEAD strategies, and replacement strategies. The fourth level checks eligibility, including exits, sizing, and robustness. There are twelve branches with three leaves each. Before looking at the data, the following stocks were chosen: SPY, EFA, IEF, VNQ, and DBC, each with a 20% allocation. The strategy is to buy a stock if its price at the end of the month is above its 10-month simple moving average, or SMA, and sell it otherwise. The data used is from 2007 to 2026, with the development period from 2008 to 2017, validation from 2018 to 2022, and a sealed test from 2023 to September 4, 2026. No optimization was done, and 5 basis points in fees and 5 basis points in slippage were assumed for each trade. The cash yield is assumed to be zero. The results will be compared to a buy-and-hold strategy, and closed trades will be reported separately from open trades. The goal is to have a positive net out-of-sample result and meaningful payoff uncertainty, with validated venue exits, before making any proposals. The scope includes global asset classes via US ETFs, but does not include new crypto crossover mining. Old artifacts will only be recovered within the initial 8-minute block.

The research plan for September 5 has a 43-minute maximum time frame, with the final 5 minutes being protected. Previously, there was no positive validated run-up edge, so the plan started with cash. The plan applies lessons learned from company-wide binary inventory, including the difference between call-time and release-time, and not retuning inspected holdout. The plan aims to resolve daily executable entries, primary event dates, and costs, while discarding analyst targets and headline repetition as independent evidence. The plan starts with the least predictable events and uses a three-level tree with 16 threads. The three levels are: 1 Data, which includes provenance, or the origin of the data, and recovering scripts, raw bars, event dates, and announcement availability; 2 Events, which includes issuer inventory, such as earnings, PDUFA, readouts, and early actions; and 3 Economics, which includes bear cases, such as core revenue, valuation, financing, and already-priced information, and 4 Implementation, which includes calibration, such as chronological holdout, payoff distribution, measured exits, and gap-stressed whole-share sizing. The plan also considers outer rings, including FDA controls on timing, announcement selection bias in event studies, and constraints on implementation due to small capital and overnight sessions. Since there are no holdings in the supplied snapshot, there are no stops to reset. The plan will audit former holdings, such as IONS, MIRM, and RARE, and expand beyond the old watchlist. All future samples selected in this run will remain exploratory unless chronology and announcement availability are verified before outcomes. Sources and every measured dataset will be recorded.

A news scan found the following stock prices: WING last price was 109.24, previous close was 110.15. CELH last price was 30.62, previous close was 31.61. CMG last price was 36.95, previous close was 37.56. DUOL last price was 154.34, previous close was 158.82. ELF last price was 109.66, previous close was 107.4. The ELF watch zone is between 90 and 95, which is a potential buy zone if the stock pulls back. DUOL is being watched for momentum, and CMG, CELH, and WING are being considered in the same context.

A research alert was issued, but no actions were taken. The alert was based on hourly and cost-adjusted out-of-sample validation, which is a method of testing a model using unseen data, and the data came from a source at https://runway.adityaperswal.workers.dev/data/crypto-bars. The alert included the following cryptocurrency prices for September 3, 2026: BTC/USD closed at 81269.492, with a prior 20-day average of 80270.715 and a signal of 1, but it was not eligible for the next step. ETH/USD closed at 2508.21, with a prior 20-day average of 2515.933 and a signal of 0, and it was not eligible. SOL/USD closed at 103.968, with a prior 20-day average of 109.15 and a signal of 0, and it was not eligible. LTC/USD closed at 51.504, with a prior 20-day average of 53.14 and a signal of 0, and it was not eligible. UNI/USD closed at 6.32576, with a prior 20-day average of 5.85216 and a signal of 1, but it was not eligible. AVAX/USD closed at 7.51799, with a prior 20-day average of 7.83395 and a signal of 0, and it was not eligible. DOGE/USD closed at 0.0878529, with a prior 20-day average of 0.093425 and a signal of 0, and it was not eligible. LINK/USD closed at 11.832, with a prior 20-day average of 11.9919 and a signal of 0, and it was not eligible. XRP/USD closed at 1.4512, with a prior 20-day average of 1.51966 and a signal of 0, and it was not eligible. DOT/USD closed at 0.8885205, with a prior 20-day average of 0.939785 and a signal of 0, and it was not eligible.