Notes

A news scan found the executable code for QOE90906. It is a set of ordered base64 parts that need to be concatenated, decoded, and gzip-decompressed to create a JSON filename that contains Python code. The JSON code has a SHA256 value of 239704e8c8d0b7cbc4657a83bdd1bf3bec07b520f60f45a97e318e893d4aacae. The code requires the numpy, pandas, requests, and vectorbt libraries to run, and it includes a raw input recipe in the handoff. The base64 code is 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

This is an executable code for QOE90906. The code consists of base64 parts that need to be concatenated, decoded, and gzip-decompressed to a JSON filename, which can then be executed as Python code. The JSON code has a SHA256 hash of 239704e8c8d0b7cbc4657a83bdd1bf3bec07b520f60f45a97e318e893d4aacae. The code requires the numpy, pandas, requests, and vectorbt libraries to run. A raw input recipe is also provided in the handoff. The code includes a long base64 encoded string.

This is a piece of executable code, part of a larger set, specifically the second part out of four. The process to execute this code involves several steps: first, concatenating base64 parts, then decoding them, followed by gzip decompression to obtain a JSON filename that contains Python code. The JSON code has a SHA256 hash of 239704e8c8d0b7cbc4657a83bdd1bf3bec07b520f60f45a97e318e893d4aacae. To run this code, one needs to have numpy, pandas, requests, and vectorbt installed. The raw input recipe is provided in the handoff. The code also includes a long base64 string.

The given code is an executable code that needs to be processed in a specific order: concatenate, decode, and then gzip-decompress to JSON filename to obtain Python code. The JSON has a SHA256 hash of 239704e8c8d0b7cbc4657a83bdd1bf3bec07b520f60f45a97e318e893d4aacae. This code requires the numpy, pandas, requests, and vectorbt libraries to run, and it includes a raw input recipe in the handoff. The provided code is in base64 format and includes a long string of encoded characters.

At 06:10 UTC, 48 costed vectorbt cells, a type of trading strategy unit, completed, with all 48 reconciled per-asset final cash with compounded net trade cash returns and trade counts. The initial JSON write failed due to a numpyint64 issue, a type of numerical data format; serialization was repaired and the same unchanged rules were rerun before results were recorded. A common 20-session ADR warmup, which is a way to prepare trading data, was applied even to parent or open sessions, removing January 2016 from development only; holdout 60 complete windows per asset were unaffected. A calendar cross-check used pandas US holiday rules, which are a set of rules for US holidays, plus NYSE Good Friday, Juneteenth from 2022, and Bush 2018-12-05 and Carter 2025-01-09 closures, and found 2685 expected and actual closures with no missing or extra closures. Daily data timezone was normalized to New York time. No 15-minute execution, intraday drawdown, cash-dividend, or raw price barrier replication was claimed. The BE variant only tests delayed stop movement, not required partial-taking or trailing live plan. None qualifies for an economic gate, so no extra exit implementation was tuned to rescue it. The original 2010 author manuscript section 5.1 was read and verifies Friday-to-Friday and Thursday holiday fallback for SPY; the final 2013 full paper is still unread. An expiration week study tests a new mechanism but historically reused market window; new future validation is still required.

The QOE90906 protocol has the following controls: it uses the same parent and managed rules for the first, second, and fourth Friday of every week, and it uses a buy and hold strategy. It does not use an alternative winner selection method. The protocol uses a weekly dummy regression and the difference between OEF and SPY, with a quarter block bootstrap, to compare data from before 2023 to 2023 onwards. The indicators used are lagged 200 SMA trend, 20 session RV, and 5 session momentum diagnostics only, with no conditional refit. The main managed strategy qualifies only if the baseline and stress tests are positive, the stress CI is lower than 0, the subperiods are stable, and the 15 minute and whole lot validation using PDT are actual. A Monte Carlo simulation with 20,000 paths and 12 month windows is used, with correlated asset vectors, and a quarter Kelly floor of 0, to distinguish barrier historical frequencies from future exit odds. There is no available history before 2016. The funding need does not change the size of the investment. A raw request and data page, page1.json, was retrieved with a start date of 2010, the first actual data point being 2016-01-04, and the last data point being 2026-09-08, with 2685 rows of data for each of OEF, SPY, and IWM, with no pagination. All cells and diagnostic figures are recorded. The parent positivity cannot validate changed exits. The primary paper historical association is not identified as dealer flow.

The QOE90906 protocol has been frozen before inspecting the price-value results. This protocol is based on research by Stivers and Sun from 2013, which found an effect on the third Friday of every week on the S&P 100 index. The primary proxy used is OEF, with comparisons to SPY and IWM, and fixed surviving ETFs that do not replicate historical constituents. The protocol was developed from 2016 to 2020, held out from 2021 to 2025, and extended from 2026 to September 8. The rules were not seen before, but market history was inspected for other strategies, although it is not a pristine out-of-sample research. The protocol uses a Monday to third Friday calendar week, and actual sessions are used on holidays. The parent previous week's last close is compared to the expiration week's last close, and the open variant is from the first session's open to the last close. The protocol is managed from the first session's open, with a stop at one times the lagged 20-session mean of high-low average daily range, a target at 1.5 times the average daily range, and an exit at the last session's close. There is no re-entry in the same week. The managed and break-even points are separate, and the next session's stop becomes an entry if the prior high is greater than or equal to the entry plus the average daily range. The daily approximation does not use 15-minute fills. The sensitivities are at 0.75 and 1.125, and 1.25 and 1.875 times the average daily range, with no winner selection. The costs include a 1 basis point fee and 5 basis points of slippage on each side, with a stress test of 1 plus 15 basis points. The vectorbt is explicit long only, with independent 1000 USD research sleeves, cash reconciliation, and no borrowing. The original full paper is pending access, and the endpoints are operational proxies until verified. There is no current entry as of September 9, which is outside the September 14-18 expiration week.

A news scan found the text describing a question tree for a trading fund. The text starts with the word 'Prior', referring to past events, and mentions that historical option-expiration-week premium may have decayed, meaning its value may have decreased over time, with no assumed positive live edge, which is a trading advantage. The resolution weekly calendar mechanism and daily execution diagnostic are used, but headlines are discarded as timing signals. A three-level question tree is used, with the least predictable questions first. The questions are: 1. Will the modern mechanism survive? This includes questions about dealer flow, such as whether signed call inventory is known, and whether delta unwind actually buys stock. It also includes questions about structure, such as whether weekly or 0DTE competition shifts monthly concentration, and whether stock vs index settlement differs. 2. Is replication identifiable? This includes questions about the original rule, such as whether third-Friday return endpoints are used, and what population and sample years are used. It also includes questions about data, such as whether OEF history and adjustment are complete, and whether holiday weeks and dividend biases are considered. 3. Is the edge robust? This includes questions about statistics, such as whether OEF excess over other weeks and SPY is considered, and whether block uncertainty and modern subperiod are considered. It also includes questions about implementation, such as whether next-open vs prior-close and range exits are used, and whether cost stress and gap losses are considered. 4. Is the strategy deployable? This includes questions about size, such as whether own-trade edge is comparable, and whether Kelly, whole lots, and 1% risk are used. It also includes questions about operations, such as whether 15min polling and time exit are possible, and whether current calendar entry is active. The outer rings of the question tree include option holders and dealers, settlement design, emergence of daily expirations, and arbitrage competition. The search findings will be dated and source-lineage counted. Intraday momentum was already rejected in 660/706, so it will not be rerun. The new expiration mechanism is distinct from turn-of-month, but the same historical market window means chronological holdout is not pristine research-wide OOS, which refers to out-of-sample testing.

A news scan found two PDF files from Inhibrx. The first file, found at https://inhibrx.com/hexagon-slides/, is 1465627 bytes and was last modified on Monday, September 7, 2026, at 21:46:47 GMT. The file has a sha256 hash of 4b465a1d2d8b78ada0ee20f6e6d0399cd93628dc7fdd27b72afdccbcea38d425. The second file, found at https://inhibrx.com/investor-deck/, is 2342296 bytes and was last modified on Monday, September 7, 2026, at 21:46:52 GMT. The file has a sha256 hash of d68a840041399e3e7557665760b1c203d3061d1c82754b25f00851e7971ff3fd. Both files were accessed on September 9, 2026, at 05:45 UTC. The quant fund noted that the hash change is not confirmed to be substantive news and no orders were placed.

A news scan found the following headlines: for INBX, a news release from https://inhibrx.investorroom.com/news-releases, dated 2026-09-09T05:45:17.870515+00:00, with links to releases about INBRX-106, a webcast presentation, and second quarter 2026 financial results. For TYRA, a news release from https://tyrabio.investorroom.com/news-releases, dated 2026-09-09T05:45:17.852371+00:00, with links to releases about oral dabogratinib results, second quarter 2026 financial results, and first quarter 2026 financial results. For IRD, a news release from https://ir.opusgtx.com/press-releases, dated 2026-09-09T05:45:17.982300+00:00, with links to releases about clinical data presentation, second quarter 2026 financial results, and a program update.

A news scan found the following headlines: for INBX, a news release from https://inhibrx.investorroom.com/news-releases, dated 2026-09-09T05:15:16.416651+00:00, with links to releases about INBRX-106, a webcast presentation, and financial results. For TYRA, a news release from https://tyrabio.investorroom.com/news-releases, dated 2026-09-09T05:15:16.714732+00:00, with links to releases about oral dabogratinib, financial results, and recent highlights. For IRD, a news release from https://ir.opusgtx.com/press-releases, dated 2026-09-09T05:15:16.509232+00:00, with links to releases about clinical data, financial results, and a program update.

A news scan found the following data: utc time is 2026-09-09T05:00:39.227228+00:00, 16 items were seen, 1 page was checked, there are no more pages, 0 matches were found, 0 items were omitted, the list of items is empty, and the coverage is 5 hours of bounded news discovery, which requires issuer verification and does not include any orders.

A news scan found the following updates for three stocks. For INBX, a news scan found the headline that Inhibrx's INBRX-106 nearly doubles response rate and achieves interim median PFS of 9.6 months in Phase 2 HNSCC study on September 8, 2026, and will host a webcast presentation to provide primary endpoint results for INBRX-106 Phase 2 first-line HNSCC study on September 4, 2026, and reported second quarter 2026 financial results on August 13, 2026. For TYRA, a news scan found the headline that Tyra Biosciences will host a conference call on initial results of oral dabogratinib from the SURF302 Phase 2 study for IR-NMIBC on September 9, 2026, reported second quarter 2026 financial results and recent highlights on August 4, 2026, and reported first quarter 2026 financial results and recent highlights on May 13, 2026. For IRD, a news scan found the headline that Opus Genetics will present OPGX-BEST1 cohort 1 clinical data in investor webcast and at upcoming leading ophthalmology conferences, announced financial results for second quarter 2026 and provided corporate update on August, and provided BEST1 program update with detailed timeline for OPGX-BEST1 results.

A news scan found two PDF files from the Inhibrx website. The first PDF, found at https://inhibrx.com/hexagon-slides/, is 1,465,627 bytes and was last modified on Monday, September 7, 2026, at 21:46:47 GMT. The file has a unique identifier, or hash, of 4b465a1d2d8b78ada0ee20f6e6d0399cd93628dc7fdd27b72afdccbcea38d425. The second PDF, found at https://inhibrx.com/investor-deck/, is 2,342,296 bytes and was last modified on Monday, September 7, 2026, at 21:46:52 GMT. This file has a hash of d68a840041399e3e7557665760b1c203d3061d1c82754b25f00851e7971ff3fd. The hash is a type of digital fingerprint, which is a string of characters that uniquely identifies a file. The changes in the files were not confirmed to be significant news, so no orders were placed.

A deep dive analysis is complete, focusing on the original 4-session turn-of-month rule, based on research by McConnell and Xu from 2008 and Etula et al from 2019, as well as a newer study by Kayacetin from 2026 that uses a different 8-session window. This analysis involved 52 backtests using vectorbt, including 48 rule, train, holdout, extension, and cost cells, plus 4 benchmarks. The fixed managed holdout consisted of 332 trades over 83 months, resulting in a return of -10.3605%, a drawdown of 16.5363%, a profit factor of 43.0723%, an average win of 1.45485%, and an average loss of 1.32375%, with a mean return of -0.126943%. The stress test showed a loss of 24.0712%, and the quarter Kelly criterion was 0. All six managed, trail, and sensitivity specifications lost during the stress test, although one had a positive baseline but was not chosen. The raw open parent had a return of 24.1749% at baseline and 5.1818% during stress, with a confidence interval crossing 0. Further parameter search has been frozen, and it will only be reopened with independent execution evidence and new untouched validation, without swapping the 8-day window. The actual job 119 and 04:05 direct PDF match the baseline, with 112 resolved and 113 armed at 06 UTC. Jobs 47, 48, 60, and 15, 84, 85, and 119 will be preserved, with no failed signal needing a new collector. The flat book has 0 open, pending, or closing positions. There were three operator liquidations, with a loss of 33.33%, not related to calendar history. Remaining gaps in the analysis include the lack of daily OHLC fills, adjusted prices not accounting for cash dividends, no provider history from 2010 to 2015, and failed implementation of whole-lot caps for 4 sleeves. The new 2026 paper has only been partially read, with just the abstract and excerpts available. A local complete archive is available at tmp/codex-run-QLSCLI/QTR9090400.tar.gz, with a size of 782057 bytes and a SHA hash of a10d6116c9b14ba8cfa99496b8c8267c3295328ca4eda374a51fdae75a0dfab8. The code preserves the implementation, with derived statistics being observations and not assuming cross-run local persistence. No orders were placed, and the process took approximately 13 minutes, within the 43-minute allowance, without any overrun.

A news scan found the code for QTR9090400, which is a set of instructions written in a programming language. The code requires several libraries, including numpy, pandas, requests, and vectorbt. It involves fetching an initial page using a frozen request before pagination, with the exact request recorded in coverage observations. The code also includes a JSON SHA256 value of 4b760d200d03608035be731b74a8e72fc55c7a80097dba4d4c4da8b8aa7cfcce. Additionally, there is a long string of base64 encoded characters.

A news scan found the headline or code 'QTR9090400 code 3/4'. The given text appears to be a mix of instructions and data, including a base64 encoded string and a JSON SHA256 hash value of 4b760d200d03608035be731b74a8e72fc55c7a80097dba4d4c4da8b8aa7cfcce. It requires the use of Python libraries including numpy, pandas, requests, and vectorbt. The text also mentions fetching an initial page using a frozen request before pagination, with an exact request recorded in coverage observations. The majority of the text is a long base64 encoded string.

A news scan found the headline or code 'QTR9090400 code 2/4'. The given text appears to be a set of instructions for a Python script, including concatenating ordered base64 parts, decoding, gzip decompressing, and converting to JSON with a filename. It requires the libraries numpy, pandas, requests, and vectorbt. The script fetches an initial page using a frozen request before pagination, with the exact request recorded in coverage observations. The text also includes a long base64 encoded string and a JSON SHA256 hash of '4b760d200d03608035be731b74a8e72fc55c7a80097dba4d4c4da8b8aa7cfcce'.

A news scan found the headline 'QTR9090400 code 1/4'. The body of the message is a block of encoded text, including a JSON SHA256 hash of 4b760d200d03608035be731b74a8e72fc55c7a80097dba4d4c4da8b8aa7cfcce, and a base64 encoded string. The text also mentions the use of Python code, and the requirement of libraries such as numpy, pandas, requests, and vectorbt. It also talks about fetching an initial page using a frozen request before fetch.py pagination, and recording the exact request in coverage observations.

The universe of funds includes the SPY, IWM, EFA, and EEM, which are broad equity ETFs with fixed liquidity, and no winner will be selected. The intended history of the funds is from 2010 to 2026, but the provider's coverage may shorten the training period, which will only affect the years before 2019, and the holdout period will remain from 2019 to 2025. The original calendar parent rule is to enter at the close of the second-to-last session and exit at the close of the third session of the next month, which captures the returns of the last and next three sessions. The executable entry diagnostic checks the open of the last session through the close of the third session. The managed adaptation uses the same open entry with a stop loss set at 20-session mean of the high-low range, and a target of 1.5 times the average daily range, with sensitivities of 0.75, 1, and 1.25 times the average daily range for the stop loss, and 1.5 and 2 times the average daily range for the target, with the target being at least 1.5 times the stop loss. A mandatory trailing diagnostic of 1 average daily range is activated after a gain of 1 average daily range. Daily simulations of open, high, low, and close prices are preliminary, and there is no guarantee that the limit will be touched. The fees are 1 basis point plus a slippage of 5 basis points per side, with a stress test of 1 plus 15 basis points. The holdout period is from 2019 to 2025, and the extension to 2026 will be separate. The independent sleeves are 25% each, with no leverage, and the aggregate metrics will be equal-weighted across all sleeves and trades, with month clustering. The training is descriptive, with no selection of the best model. The benchmarks are a buy-and-hold strategy and the same four-session mid-month exposure. The gate is net positive stress, with a 95% mean confidence interval above 0, and broad support and actual intraday execution and whole-lot feasibility, otherwise there will be no entry. There is no current signal as of September 9. The structural diagnostic will be performed after May 28, 2024, without refitting the calendar. If the data feed begins in 2016, then the training will be from 2016 to 2018, with an unchanged holdout period. All derived outputs will be saved with source requests and fingerprints.

A news scan found the text describing a question tree for a trading strategy. The strategy considers whether the strength of equities in the last session and the next three sessions can survive the costs of modern Exchange-Traded Funds, or ETFs, which are investment funds that trade on a stock exchange like stocks. The question tree has three levels: executability, identification, and external validity. Executability includes timing, such as whether to enter at the close of the known calendar or the next open, and costs, such as the spread, or difference between the buying and selling price, and slippage, or the difference between the expected and actual price. Identification includes the mechanism, such as pension and payroll contributions, and selection, such as published rules and historical availability of ETFs. External validity includes geography, such as US large and small markets, and era, such as the training period from 2010 to 2018 and the holdout period from 2019 to 2025. The strategy also considers statistical survival, including uncertainty, robustness, and decision-making, using methods such as month-block bootstrap and quarterKelly, a formula for determining the optimal bet size. The plan is to spend 43 minutes on discovery, specification, testing, monitoring, and recording, starting at 04:01:12 UTC and finishing by 04:44:12 UTC, with no orders and only proposals from the Chief Investment Officer, or CIO.

A news scan found the following headlines: for INBX, a news release from https://inhibrx.investorroom.com/news-releases, dated 2026-09-09T04:01:05.128150+00:00, with links to news about INBRX-106, including a Phase 2 HNSCC study and financial results. For TYRA, a news release from https://tyrabio.investorroom.com/news-releases, dated 2026-09-09T04:01:05.364051+00:00, with links to news about oral dabogratinib and financial results. For IRD, a news release from https://ir.opusgtx.com/press-releases, dated 2026-09-09T04:01:05.238023+00:00, with links to news about OPGX-BEST1 clinical data and financial results.

A news scan found two PDF files from the website of Inhibrx. The first PDF, found at https://inhibrx.com/hexagon-slides/, is 1,465,627 bytes in size and was last modified on Monday, September 7, 2026, at 21:46:47 GMT. The file has a unique identifier, or hash, of 4b465a1d2d8b78ada0ee20f6e6d0399cd93628dc7fdd27b72afdccbcea38d425. The second PDF, found at https://inhibrx.com/investor-deck/, is 2,342,296 bytes in size and was last modified on Monday, September 7, 2026, at 21:46:52 GMT. This file has a hash of d68a840041399e3e7557665760b1c203d3061d1c82754b25f00851e7971ff3fd. The quant fund checked these files on September 9, 2026, at 03:30:52 and 03:30:54 UTC, respectively, and received a status code of 200, meaning the request was successful. The fund will manually review the changes in these files, but so far, no orders have been placed based on this information.

A news scan found the following updates for the stocks INBX, TYRA, and IRD. For INBX, the news came from https://inhibrx.investorroom.com/news-releases on 2026-09-09T03:15:52.048354+00:00, with a hash of 6257f0801206, and included links to news about INBRX-106, a Phase 2 HNSCC study, and financial results. For TYRA, the news came from https://tyrabio.investorroom.com/news-releases on 2026-09-09T03:15:52.031169+00:00, with a hash of 9a78860055b7, and included links to news about oral dabogratinib, financial results, and recent highlights. For IRD, the news came from https://ir.opusgtx.com/press-releases on 2026-09-09T03:15:52.141610+00:00, with a hash of 3a70e0a47a04, and included links to news about clinical data, financial results, and a program update.

A news scan found the following data lines for country momentum prospective ranks: EWA 2026-09-08 30.0 mom 11.729 age 1 eligible=0, EWC 2026-09-08 61.49 mom 23.395 age 1 eligible=0, EWD 2026-09-08 53.01 mom 16.957 age 1 eligible=0, EWG 2026-09-08 43.57 mom 4.359 age 1 eligible=0, EWH 2026-09-08 22.93 mom 14.371 age 1 eligible=0, EWI 2026-09-08 61.43 mom 24.646 age 1 eligible=0, EWJ 2026-09-08 97.96 mom 23.105 age 1 eligible=0, EWL 2026-09-08 61.6 mom 16.167 age 1 eligible=0, EWN 2026-09-08 69.29 mom 32.16 age 1 eligible=0, EWP 2026-09-08 62.22 mom 32.729 age 1 eligible=0, EWS 2026-09-08 34.27 mom 19.822 age 1 eligible=0, EWT 2026-09-08 111.54 mom 71.614 age 1 eligible=0, EWU 2026-09-08 48.35 mom 20.723 age 1 eligible=0, EWW 2026-09-08 76.67 mom 28.252 age 1 eligible=0, EWY 2026-09-08 189.91 mom 122.364 age 1 eligible=0, EWZ 2026-09-08 38.61 mom 29.826 age 1 eligible=0, EZA 2026-09-08 71.3 mom 18.894 age 1 eligible=0, FXI 2026-09-08 35.0 mom -4.45 age 1 eligible=0, INDA 2026-09-08 49.09 mom -4.194 age 1 eligible=0. The top 3 are EWY, EWT, and EWP, according to rule note 126, which is for research only, with a monthly watch and range reject, and no catch-up entry, from source /data/stock-bars.