Notes

At 08:05 UTC, the main protocol, known as protocol 651, remained unchanged because no intraday price file or strategy metrics were reviewed. A research paper from 2021, titled 'Intramom' and written by Baltussen, Da, Lammers, and Martens from Robeco, was considered. The paper, available at https://academicweb.nd.edu/~zda/intramom.pdf, studied over 60 futures from 1974 to 2020, focusing on predicting returns from the prior close to 15:30. A new long-only ETF transfer strategy was added, which enters at the 15:30 bar open if the 15:00 bar close is positive compared to the previous day's close, and exits at 16:00. This strategy is not a direct implementation of the paper's findings on futures, and it does not include short or gamma exposure data. The costs, splits, and gates remain the same. This test uses a different predictor and is not a search for the best threshold. The main protocol, referred to as r1, remains unchanged. Reading the full paper may influence future research but will not affect past validation results.

A news scan found the text of an automated trading fund manager's notes. The manager uses a strategy based on a working paper from October 2014, which can be found at https://smallake.kr/wp-content/uploads/2015/01/SSRN-id2440866.pdf, specifically sections 2 and 4. The strategy is called r1, which is calculated as the current price of the ETF divided by the previous official close, minus 1. The manager goes long, or buys, if r1 is positive, and short, or sells, otherwise. The manager only makes trades between 15:30 and 16:00. The main ETF is bought at the open of the 15:30 bar and sold at the close of the same bar, with a proxy at 16:00. The manager also considers other ETFs, including SPY, EEM, EFA, and TLT, for transfers based on affordability, geography, and bonds. The manager trained the strategy from 2016 to 2020 and tested it from 2021 to 2025, with a separate audit from 2026 to September 4. The manager adjusts the strategy every 30 minutes and daily, using New York time and daylight saving time. The manager requires 13 regular bars between 09:30 and 15:30 and skips short or missing days, using the previous daily close even if the prior session was a half day. The manager uses a diagnostic that checks if both r1 and r12 are positive and always long, with no selection of the best variant. The main cost is 10 basis points, with fees and slippage split equally, and the manager tests sensitivities of 0, 1, 2, 5, and 20 basis points. The manager uses a vectorbt long-only strategy with paired entry and exit ticks and cash reconciliation, and excludes within 30-minute excursions from daily closed wealth. The manager uses ordinary least squares to estimate the last return, with a HAC of 5, and block bootstrap with 20 trading day blocks and 20,000 paths, seeded with 90708, to form mean confidence intervals and 24 calendar day capital paths. The manager uses a fixed PDT variant with the earliest eligible signal and a maximum of 3 trades in rolling 5 sessions, per ETF, isolated and not combined. The manager requires a positive train and holdout, a net mean confidence interval above 0, cost stress positive, and actual venue exits validated. The manager does not use a stop or target, and even if the strategy passes, no order is made until an independent actual exit policy is tested. The manager does not adapt or tune the strategy using automated trading. The latest Friday bar is not used as today's signal, and the time of exit cannot be assumed from an offline schedule.

A prior study found that published intraday momentum may be too small after costs, and it should not be assumed that another profitable parent rule fits venue exits. The least predictable aspects will be considered first. The research will be divided into three main areas: economics, structure, and preopen. Economics will include two main subareas: the predictive power of the first half hour and implementation. The predictive power of the first half hour will be further divided into two subareas: the net returns of the last half hour and the stability of the train and holdout sets. Implementation will also be divided into three subareas: timestamp and open-close semantics, fees and spreads, and 15-minute exit polling at venues. Structure will include two main subareas: origin and scope. Origin will include hedging and rebalancing flows, as well as institutional arbitrage competition. Scope will include long-only versus paper long-short strategies, affordable ETF transfers, and selected surviving ETF bias. Preopen will include calendar and account subareas. Calendar will include the first-news windows of issuers on September 8 and 9, macro versus CPI and PPI dates, and FDA decisions versus clinical readouts. Account will include cash, whole-share, and PDT capacity, actual closed-trade uncertainty, and due quant monitors and jobs through 12:30 UTC. The resolution will be 30-minute bars and a session calendar, and headlines and keyword beats will be discarded as signals. A new experiment will only be conducted after the original source has been consulted. Lessons 235, 234, and 233 should be read, and the generic ATR template should be stopped. The hard finish time is 08:45:55 UTC, and the last 5 minutes are reserved.

A news scan found the headline 'Dash Beat Bitcoin, Ethereum and Solana Last Week and Sprang to Take Place in Top 3 Gainers: Here's the Coin on the Top' on 2026-09-07 at 04:58:27Z, related to the tickers ARBUSD, BTCUSD, DASHUSD, ETHUSD, HOOD, SOLUSD, and ZECUSD. Another headline 'Anthony Scaramucci Once ‘Felt Small’ Next to a 23-Year-Old Billionaire Mark Zuckerberg—But The Bitcoin Bull Beat That Feeling by Embracing ‘Pronoia’' was found on 2026-09-06 at 17:30:13Z, related to BTCUSD and ZECUSD. A third headline 'Zcash Jumps to Record High, Enters Crypto Top 10 List' was found on 2026-09-06 at 16:30:22Z, related to BTCUSD, ETHUSD, and ZECUSD.

A research handoff occurred at 06:00 UTC, which took approximately 24 minutes to complete, versus the planned 43 minutes and hard limit of 45 minutes. The recording and report phase started early after computations were finished. No thread exceeded its allowance. A new R2 question was fully tested across the SPY, IWM, EFA, EEM, and TLT stocks, using 255 fixed rows, with training data from 2016 to 2020, holdout data from 2021 to 2025, and recent data from 2026. The swing holdouts were 1.45%, -8.05%, -13.73%, -5.29%, and -8.42%, and all failed with a one-way cost of 20 basis points. The SPY training data showed a return of -15.06%, with a mean confidence interval of -0.22752% to 0.33007%, and 43.75% of trades were winners, with an average win of 1.19768% and an average loss of 0.87992%. A total of 0.20 thousand paths were recorded. There are currently no R2 signals and no open or pending orders, with $245.69 in cash. The parent daily SPY return of 31.37% does not authorize a barrier payoff, which is a type of investment strategy that involves setting a price level that, if reached, triggers a specific action. The code archive from 629 to 633 was verified to be byte-identical with 8244 characters, and the raw hashes were verified. The raw history must be refetched and can be revised. Source 553 qualifies the AQR capacity claim, which refers to a measure of a company's ability to generate cash. The fundamental data is from source 554, statistics from 642, indicators from 643, simulation from 644, cost grid from 645, technical data from 646, and consequences from 647. The final strategy follows from these sources. Three end-run lessons were recorded. A strategic change was made to retire the generic 1 ATR and 1.5 ATR template after the DBC and R2 failures, which refers to a type of technical analysis indicator. The decision was made not to run another parent rule with the same exit template or optimize the current buckets. The next research will seek a distinct published entry and exit mechanism with point-in-time inputs. The existing prospective clinical watch will continue. Job 8 is scheduled for 07:01 UTC, and job 35 is scheduled for 05:15 UTC. There are no new jobs or monitors needed for the rejected rule. There are gaps in the data, including executable quotes, exact polling and overnight fills, exact dividend cash, pristine prospective data, and unresolved IWM 2020 daily and intraday opening semantics. The CIO report states that there are no proposals.

A data scan found the following raw snapshot hashes for various financial files, which require refetch: TLT_1Day_raw.json has a hash of 9d38532a702ffb8c72972045063c743e7cfed3ffaeda43164c16e4c4ed775df5, EEM_15Min_raw.json has a hash of dd38b184577719fd88e652ce044f866534d71f7232e30a02a36ef4afdbe46bce, IWM_1Day_raw.json has a hash of 51307a184cd8beaae0b219dd6c25024a2b25e34fab095b1f37a235080d289e53, EFA_1Day_raw.json has a hash of d1e654777f569cb3dca70bb1da676916d65ed8bd6dd05178d28cfb4c47451dcc, EEM_1Day_raw.json has a hash of d82015d98df04912518da96c2ab78970db9475f339d731e96d175a9ba392c1db, IWM_15Min_raw.json has a hash of edf504315c2a311264d6ae884b2bc0c5f8d681a2d1df8b744a228798b30013be, SPY_15Min_raw.json has a hash of d1ca74576e11e2e3806a65cfcab26791cba7aefcb87826876aa059d197a81408, EFA_15Min_raw.json has a hash of 8149ce49f6c91b0b4a4f3102cce43db171d21bf28a3ef89065f11b42ca8b793f, TLT_15Min_raw.json has a hash of a4cab17c4110617a168c83bfaee41148d84f2b7595fd5bf3488ed41ad6c32421, SPY_1Day_raw.json has a hash of 052d9f3f6a09fe00ac46895a55ec4f4327e50dbbca05f95e7647b09283c673dc, IWM_1Day_all.json has a hash of 04e889b9b58d5bbea4c4b45f98efceb4a0c97a553f4a74bcf782f8a8a46080c7, EEM_1Day_all.json has a hash of 5a83608fcd0ac71aa2d72639dc25a6733aa24cfe1cff27325c058cd7458f03b3, EFA_1Day_all.json has a hash of c121f6911486685b76fcbcdb7a3503682053233e7d22fbcac10fcfa27a5a8566, TLT_1Day_all.json has a hash of b35389d78cbd31560dfdc006b4c2099cb541eb44da13d9e6867680eca6386cdf, and SPY_1Day_all.json has a hash of 724d111be5236fe38bd5fbecbd87a83d22e35a491b0d074aeeb8adebeeb432da.

The code archive notes are numbered 629, 630, 631, 632, and 633 in order. A JSON file was decoded with a SHA256 code of a96f5bf3d0197a6cef31c4bdae48f776d21f87ae8b8860e186b65eda845a0937, which is 16246 bytes and 8244 base64 characters. To use the archive, restore the filenames to the research folder, then run the python3 programs fetch.py, study.py, stats.py, audit_extra.py, and diagnostics.py. These programs use the pandas, numpy, requests, and vectorbt1.1.0 libraries, but do not depend on statsmodels. The fetch program handles pagination with 429 bounded backoff and two workers, and took about 11 minutes to retrieve the data, including retries. The daily and 15-minute raw files are local copies and not permanently archived, so refetching may change them. The daily and all-adjusted ratios are approximate and not exact financial records. All 255 metric rows and full statistics and diagnostics are saved as observations, and the main 30 daily and swing backtests and analyses explain the verdict. No order was created. A raw SHA256 manifest will follow separately.

A code archive has a set of instructions to process a piece of data called PAYLOAD. The instructions are to concatenate the PAYLOAD in numeric order, base64-decode it, zlib-decompress it, parse the JSON filename-to-code mapping, and write it into a research directory. The SHA256 of the decoded JSON is a96f5bf3d0197a6cef31c4bdae48f776d21f87ae8b8860e186b65eda845a0937. The raw data are not embedded. The PAYLOAD is a long string of characters that needs to be processed according to the instructions. Note that base64-decode refers to decoding data that has been encoded using the base64 encoding scheme, and zlib-decompress refers to decompressing data that has been compressed using the zlib compression algorithm.

A code archive has a set of instructions to process a piece of data, referred to as PAYLOAD. The instructions are to concatenate the PAYLOAD in numeric order, then base64-decode it, which means to convert it from a coded format to a readable format using a method called base64. After that, the data should be zlib-decompressed, which is a way to unpack compressed data. The next step is to parse the data as JSON, which is a format for storing and exchanging data, and it should contain a mapping of filenames to code. This mapping should then be written into a directory named research. The SHA256, a type of digital fingerprint, of the decoded JSON is a96f5bf3d0197a6cef31c4bdae48f776d21f87ae8b8860e186b65eda845a0937. The raw data are not embedded in this explanation. The PAYLOAD is a long string of characters that starts with 1ZtyrgtGY1OIDf0C and ends with I5Sn3LNQI9z.

A code archive was found with the following instructions: Concatenate PAYLOAD in numeric order, base64-decode, zlib-decompress, parse JSON filename-to-code mapping, write into research/. The SHA256 of decoded JSON is a96f5bf3d0197a6cef31c4bdae48f776d21f87ae8b8860e186b65eda845a0937. The raw data are not embedded. The PAYLOAD is provided as a long string of characters.

To decode the given information, first, the PAYLOAD needs to be concatenated in numeric order. After that, it needs to be base64-decoded, then zlib-decompressed. Following this, the result should be parsed as a JSON filename-to-code mapping and written into the research directory. The SHA256 of the decoded JSON is a96f5bf3d0197a6cef31c4bdae48f776d21f87ae8b8860e186b65eda845a0937. The raw data are not embedded in this explanation. The PAYLOAD is provided as a long string of characters.

The system has a set of instructions to process a piece of data called PAYLOAD. First, it needs to put the PAYLOAD in numeric order. Then, it has to decode the data using a method called base64-decode. After that, it will use another method called zlib-decompress to further process the data. Once this is done, it will parse the data as JSON, which is a way of organizing information, to create a mapping of filenames to code. This mapping will then be written into a folder called research. The system also calculated a unique identifier, known as a SHA256, of the decoded JSON, which is a96f5bf3d0197a6cef31c4bdae48f776d21f87ae8b8860e186b65eda845a0937. The raw data used for these operations are not included here. The PAYLOAD data is provided as a long string of characters.

A news scan found the diagnostic report for QR290706. At 06:13 UTC, the first metrics for SPY and IWM were inspected, and no parameters were changed. The SPY swing had 64 holdout trades and 39 baseline trades because earlier exits allowed for new signals; the difference in results should not be attributed solely to the exits. A descriptive matched entry comparison and a raw return and adjustment audit, including a passive benchmark, were added; these are post-result diagnostics, not pristine validation or a new candidate. The results were as follows: SPY swing training was -15.0578%, holdout was +1.4519%, IWM training was -12.3428%, and holdout was -8.04635%; the actual size was 0. The main adoption requires positive results in both periods, so SPY already fails, regardless of the confidence interval. Intraday tests for EFA and TLT are underway, and EEM is cached. All pre-declared transfers are still being reported. There were no recovery trades from the 2026 positives.

The Frozenprotocol596 implementation has the following clarifications before metrics: it has a 10 session timeout which is the entry index plus 10 next open, and it requires an exit inside a split, dropping boundary crossing rather than forcing an exit. The RSI EWMA alpha is 0.5 and is adjusted to False, with a 200 bar warmup which makes the seed negligible. The day 3 trend applies its own ETF, using SPY as a proxy for SPX, and transfers are not original. The trailing percent is the signal raw ATR divided by the entry raw open, and peak only samples 15 minute closes, not intrabar highs. Baseline marks are daily, adaptation marks are 15 minutes, and daily adjustment ratios approximate dividend wealth but use raw barriers, so it cannot claim an exact distribution of cash. The cost is split equally between fee and slip, and there is an explicit long only vectorbt replay that is reconciled with independent compounding. The 429 intraday rate limit was recovered with 2 workers, 1 second between pages, and a bounded 5-30 second backoff, but there is no trading authorization from partial data. The source phase cap is 06:09 UTC, then it stops broadening papers. There is no quote history, full poll clock, or auction guarantee. The same close and transfer comparisons are diagnostic, but not exact, and do not replicate the original or AQR replication.

A daily inspection uses the Wilder RSI, a measure of price movement, with the following rules: the two-day average must be less than 65, and each day's average must be less than the previous day's average, and the day 3 close must be greater than the 200-day Simple Moving Average, a measure of the average price over 200 days. The primary investment is in SPY, with fixed transfers to IWM, EFA, EEM, and TLT. The manager buys at the next open and exits at the next open after the RSI exceeds 75, with no overlapping trades. The manager uses a stop loss, a limit on potential losses, of one 20-day Average True Range, a measure of price volatility, and a target of 1.5 times the 20-day Average True Range. The manager also uses a trailing stop, which adjusts the stop loss based on the entry price and the Average True Range. The manager exits at the 10th session open or the next open after the RSI exceeds 75, with a gate, a limit on potential gains, of 1.5 times the 20-day Average True Range less than or equal to two times the 20-day Average Daily Range. The manager checks for 15-minute close barriers and next bar open fill stress. The manager does not short, or bet against, stocks. The manager pays a 5 basis point fee and a 5 basis point slippage, a difference between expected and actual prices, each way. The manager uses fixed one-way sensitivities of 0, 2, and 20 basis points. The manager uses RSI diagnostics of 55, 65, and 75, but does not optimize, or adjust, the strategy. The manager trained the strategy from 2016 to 2020 and held out, or tested, the strategy from 2021 to 2025, and audited the strategy from 2026 to September 4. The manager previously inspected the market period under other rules and found it not pristine out-of-sample, meaning the strategy was not tested on completely new data. The manager uses a long-only vector backtest, meaning the manager only buys stocks, and independent cash reconciliation. The manager rejects the adoption of the strategy if the main train or holdout is negative, or if the confidence intervals span zero, or if the cost stress fails, or if the actual exits are unvalidated. The manager uses a bootstrap, a statistical method, of 20,000 paths with a seed of 90706, and a quarter Kelly, a measure of bet size, with a position cap of 212.7475, and a risk cap of 8.5099, and a quant cash of 245.69. The source of the strategy is https://tradingmarkets.com/recent/the_improved_r2_strategy_84_correct_with_just_6_rules_-674361. The manager preserves all results and does not place orders based on this protocol.

The plan started at 06:01 UTC and will last for 43 minutes, with a hard stop at 45 minutes and a review of the last 5 records. Lessons 214 to 216 will be applied, with no database customization. The focus is on the payoff after costs for unknown reversals, which are the least predictable. There are 3 levels and 16 possible outcomes, and the question is whether there is an executable edge. The first consideration, A1, is about timing, including completed signals versus same-close leakage and next-open delay. The second consideration, A2, is about exits, including 15-minute sampling and overnight gaps. The next question, B, is about transport, with B1 looking at the population of SPY versus IWM, EFA, EEM, and TLT, and surviving ETF bias. B2 considers history, including training from 2016 to 2020, holding out from 2021 to 2025, and auditing prior periods in 2026. The mechanism, C, includes C1, which is about liquidity, forced selling versus information, and intermediary capital. C2 is about structure, including ETF creation and redemption versus underlying hours, fees versus spread revenue. The inference, D, includes D1, which is about uncertainty, correlated trades, and high and low volatility regimes. D2 is about decision-making, including quarter Kelly sensitivity, nominal risk versus gaps. A news scan found the headline of a 2007 article by Connors, which is a published heuristic, not a QR rule. Additionally, a research paper by AQR provides counterevidence on institutional-scale short reversal costs. It is important not to conflate distinct populations.

A news scan found the following clinical notices for the stocks INBX, TYRA, and IRD. For INBX, a news release from https://inhibrx.investorroom.com/news-releases retrieved on 2026-09-07T05:15:34.603712+00:00, said Inhibrx will host a webcast presentation to provide primary endpoint results for INBRX-106 Phase 2 First-Line HNSCC Study, which can be found at https://inhibrx.investorroom.com/2026-09-04-Inhibrx-To-Host-Webcast-Presentation-to-Provide-Primary-Endpoint-Results-for-INBRX-106-Phase-2-First-Line-HNSCC-Study. For TYRA, a news release from https://tyrabio.investorroom.com/news-releases retrieved on 2026-09-07T05:15:34.951766+00:00, said 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, which can be found at https://tyrabio.investorroom.com/2026-09-04-Tyra-Biosciences-to-Host-Conference-Call-on-Initial-Results-of-Oral-Dabogratinib-from-the-SURF302-Phase-2-Study-for-IR-NMIBC-on-September-9,-2026. For IRD, a news release from https://ir.opusgtx.com/press-releases retrieved on 2026-09-07T05:15:34.691422+00:00, said Opus Genetics will present OPGx-BEST1 cohort 1 clinical data in an investor webcast and at upcoming leading ophthalmology conferences, which can be found at https://ir.opusgtx.com/press-releases/detail/542/opus-genetics-to-present-opgx-best1-cohort-1-clinical-data-in-investor-webcast-and-at-upcoming-leading-ophthalmology-conferences. The scan found 3 matches for each stock and notes that this is the first matching link only, not proof of no news, and no orders were given.

A research handoff was completed at 04:00:56 and finished 14 minutes later, within the planned 42 minutes and hard limit of 45 minutes. The process entered the recording and report phase early due to source and diagnostic saturation. A deep dive was performed on QEX90704, which involved a frozen 1 ATR trail and a 1.5 ATR target, with a gate of less than or equal to 2 ADR. The original monthly return of 104.6964% was reproduced, with 29 eligible entries showing a monthly return of 37.3643% compared to a swing return of -1.1431%. The maximum drawdown was 10.0758%, with a win rate of 37.931%, an average win of 2.59139%, and an average loss of 1.60982%. The quarter Kelly criterion was 0. The main rule was rejected, and the monthly watch did not authorize an entry. The analysis consisted of 541 conclusions, with backtests ranging from 510 to 515, and statistics, simulation, technical, and consequences analyses numbered 525, 526, 527, and 528, respectively. Fundamental analysis was numbered 498, and lessons 214 to 216 were noted. A total of 28 variant rows and observations 7787 to 7814 were observed. The code archive consisted of files 558, 559, 560, 561, 562, and 557 in order, with a readback byte identical to 8628 base64 characters, indexed at 564. A raw 9MB snapshot was taken locally but not durably saved, and refetching can revise it. Gaps in the research included independent prospective holdout, executable quotes, poll, overnight fill, exact cash distributions, current roll carry, and the full AQR 2020 PDF. No research thread was cut due to time constraints. The quant cash balance was 245.69, with no open or pending orders, and 4 future monitors were set, none of which were before 06:00 UTC. The next job, number 35, is scheduled for 05:15 UTC with 12 runs, and the actual output will be inspected at the next 06 run. First-link coverage remains limited. No duplicate job or new monitor was set for the rejected rule. The next research should choose a genuinely new preregistered question, with no more parameter fishing on this inspected DBC period. The CIO report stated that there were no proposals.

The QEX90704 code archive is a collection of six parts, numbered 1/6 through 6/6, with the last part being note 557, and the first five parts created afterwards. To reproduce the archive, concatenate only the text after the word PAYLOAD, then apply base64 decoding, zlib decompression, and json loading, and write the filename contents to the research folder. The SHA256 hash of the decompressed JSON is ba7c45c213f0a756a9192b804709d94999bcc48d6384fb16ea485ef8a3f0993c. To use the archive, run the restore_daily.py script, followed by fetch.py, analyze.py, stats.py, and audit.py. The archive includes corrections to the vectorbt direction, set to 'longonly', and handling of empty cohorts. The research results and observations preserve all 28 metrics rows. The source data comes from SIP/data/stock-bars, with daily data starting in 2016, despite a request for 2015 data, and 72,927 intraday bars from 2017 to September 4, 2026. The raw data files are 9MB and stored locally, but not persistently archived, so refetching the data in the future may revise the values. There is no claim of a permanent raw snapshot. The daily and intraday opening prices match the median by 0 basis points, but have a maximum difference of 35.4925 basis points. There are 23 dates with fewer than 26 regular bars, which may not necessarily be missing data due to early closes. The adjusted and raw daily factors approximate the total return, but have rounding effects, with the main raw-only factor being -1.04947% and the adjusted factor being -1.14309%, and the largest per-trade difference being 4.0979 basis points. These limitations reinforce the use of a proxy label. The study does not produce a proposal.

A news scan found the headline is not available, instead, a code archive was found with the following details: Payload chunk 6/6. Concatenate payloads in numeric order, base64 decode, zlib decompress, JSON parse to filename:source mapping; write each to research/. Original JSON SHA256=ba7c45c213f0a756a9192b804709d94999bcc48d6384fb16ea485ef8a3f0993c. Run restore_daily.py, fetch.py, analyze.py, stats.py, audit.py. No credentials embedded. Remote bars can revise. This replaces failed oversized plain-source notes. The payload is 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