[Finance AI / AI Agent Research] If stop-loss width is 3,500 JPY, do losses decrease? Six conditions to carry over to next week

Next week, would widening the stop-loss to 3,500 JPY or entrusting the AI with full discretion improve the losses this week? The conclusion is that we cannot confirm improvement, and we do not treat this as a live change. The final editing materials at 18:30 JST on October 25 leave the stop-loss read-only comparison, investigation of H1 and H4, and cross-reference with natural crossings and records as separate stages. As a weekly retrospective for Sunday, I will整理 the six confirmation conditions to carry over to next week.

October 25 evening update:Reflected the DAILY_BRIEF.md with a 18:30 deadline on the same public URL in the morning. The target range is from 00:00 October 19 to 18:30 JST on the 25th. Since no new transaction evidence for the same day has been found in the materials, the weekly figures inherit the confirmation values from 10/24. This is not new implementation or approval of live parameters, and no changes to order placement code, EA, AutoTrading, OBS, or delivery status were made through article edits. It is also not a report that recalculated each verification’s base records independently this time.

Conclusion: Do not mix framework, comparisons, implementation, and live permissions

This article is a record of financial AI, AI agents, self-made indicators, and EA AI research and development. The sources are the final DAILY_BRIEF.md at 18:30 on the day and LOCAL_CODEX_HANDOFF_2026-10-25_0610_JST. The latter covers 06:10 JST on the 24th to 06:10 JST on the 25th, and the idea for AI to make overall judgments is distinguished as a morning concept draft. The weekly comparison of final materials and the morning concept are not treated as being at the same progress stage. This article’s editor did not try real orders in production.

Turning research ideas into prose, comparing with past prices, building programs, deploying to a real account, and verifying operation there are separate steps. If you move to the next step with conditions not yet in place and say “considered” became “deployed,” readers may misconstrue the current risks and decision-making authority. This is a week’s result with a proposed improvement, so I will first indicate how far we have progressed at each step.

Basis: The scope given to AI is different from the current authorized role

In this morning’s handover, a plan where AI takes the lead in monitoring, ordering, holding, and settlement has been organized as a user’s concept. The current description—“EA proposes candidates, AI approves, EA places orders”—does not mean that a full-discretion system has been completed as is. If the decision-maker changes, freshness of decision materials, authority, stop on anomalies, avoidance of duplication, and responsibility for records must be reconsidered.

This article does not add new trading permissions, Gemini settings changes, AutoTrading switches, or re-deploy EA/indicators. In this morning’s local handover, there is no record approving new implementations from these tasks. Presenting concepts is part of research sharing, but to prevent readers from interpreting prose as live-change instructions, I clearly state the unresolved boundaries.

The six-month weekly materials’ 18 settlements, net after costs of -14,747 JPY, are treated as results that warrant improvement consideration. However, this loss is not evidence that it occurred solely due to insufficient AI authority. Because entry, held decision, exit, execution slippage, costs, and cross-checked supporting media are not all in place, we should not jump to the conclusion that a new method would have improved things. The point is to consider improvement options for this week’s results, not claim an outright fix.

3,500 JPY comparison is not a live stop-loss width change

In yesterday’s materials, for the initial nine trades, a read-only comparison using 3,500 JPY was recorded. In the comparison that maintains BE 1,250 JPY, two recoveries, six reach 3,500 JPY, one unsettled at the end. In the comparison that makes BE 1,750 JPY (0.5R), one avoidance candidate, seven reach 3,500 JPY, one unsettled at the end. The cases here are categories of comparison, not trades executed in a live account.

If the threshold in Japanese yen is widened, in some price paths the initial stop may not be touched and can revert. But on other paths, losses may increase, and holding longer could affect the next order. The materials did not replicate all strategies, check the interference with the next order, trailing/partial close, or slippage. Therefore, do not claim that the number of recovery candidates alone will reduce losses.

The unsettled-at-end is not a conclusion that can be determined within that comparison range. Classifying as win or ignoring it by dropping from the denominator changes the impression of the results. Because the BE threshold differs between the two tables, you cannot simply stitch good parts of both into one method. Evaluations including costs and concurrent holdings are incomplete and should not be treated as a completed optimization result.

Yesterday’s production settings in the documents do not include changes to EA v1.26, 0.02 lot, risk 2,500 JPY, BE 1,250 JPY, etc. These are configuration values, and there is no guarantee that actual losses will stay within these values. Also, this morning’s handover does not confirm the current VPS effective settings. Distinguish between “no changes in yesterday’s materials” and “re-read the same settings independently this morning.”

H1 and H4 plans separate judgment examples from operational results

H1 and H4 are not yet created or in operation in the weekly materials. Proposed ideas include a condition where the direction of a confirmed H4 20EMA and 30SMA agree, and a condition to follow only when H1 is profitable using 20EMA. EMA and SMA use different averaging methods, and the period settings included in the plan are recorded as-is. They are not recommended as currently valid trading conditions.

An example where the BUY on October 21 was rejected due to opposite direction on H4, and the SELL on the 22nd was aligned in the same direction, is a judgment example for direction conditions. Just having these two examples does not mean that all rejected trades’ losses could have been prevented or that the adopted trades were profitable. Additional validation is required, including opportunity losses after rejection, multiple market states, holding periods and costs, and overlaps with other entry points.

The initial stop on H1, the margin above EMA, price anchoring, the usual Scopebox period and thresholds remain undecided. We do not fill in undecided values for convenience of article creation. We will not revive the M5 range gate just because similar wording exists. Studies on different timeframes should be kept separate with distinct identifiers and conditions, and must not be secretly mixed into the current entry as a prerequisite for comparison.

Limitation: Do not oversimplify causes while audits or recordings are missing

The weekly materials still show a lack of score freshness, audit ledger gaps and locks, and unresolved closed_24h_valid=false. Even with new concepts, evaluations are difficult unless you can trace which decision led to which order and ensure fresh information and valid status continue. This article does not confirm those recoveries.

The original media for 18 settlements is also missing in the materials. Since there is no verified trade recording to match, this article will not include other trades’ videos or apparently fictitious charts. The diagrams here explain the next week’s six conditions and the verification stage. It’s an important limitation of sharing research not to present an attractive feature diagram as resolved when the cause of losses is unknown.

Next checks: records that can be evaluated before improvement ideas

The confirmation order carried over to next week fixes the scope and freshness of materials, linking candidates, approval responses, trade numbers, actual entries/holdings/exits/costs, and original media. It then clearly states comparison conditions, unsettled handling, and interference with other orders. AI full discretion, H1/H4, and the 3,500 JPY comparison remain incomplete, and a draft proposal is not to be treated as implemented or approved for live production.

The first condition of the final materials is to trace from the entrance to storing the records the first natural definite cross in v1.42. Next, the 3,500 JPY plan will re-examine the order of reach without changing the live, using tick precision or adequate natural samples. When a single bar might have both favorable and unfavorable levels, do not place the more favorable order first. Separate evaluations by extracting only the recovery candidates and the overall strategy’s P/L including interference with subsequent orders.

The H1 plan uses no future bars, and the next conditions are to shadow-evaluate by fixing in advance the confirmed bar conditions, H4 direction, initial SL margin, and entry price start conditions. H1’s buy/sell program is not created or in operation, and there are no H1-derived changes to the current M5 v1.42 / EA v1.26. Until the initial SL, EMA margin, price anchoring, normal scope integrations, and box period/threshold are determined, do not fill values for convenience and do not revive the old M5 range gate that was previously rejected.

Audits should confirm the freshness, gaps, and locks of the account score and runtime audit ledger, and ensure that the structure shadow’s closed_24h_valid=true continues. Permanent causes or recovery from a current false state are not confirmed, and temporary displays are not considered recovery. Media should not be considered complete merely by existing locally; videos corresponding to the same ticket and signal_id, entry/exit images, and trade_meta on Drive original must be present.

The sixth item is a separate matter from research itself: the quality of public operation. We will verify whether the final materials were saved to the article, whether the owner and target URL match, and whether the actual page and timeline were reflected. FAILED in past audit materials must be rechecked for each day and closed; this will not be declared fully corrected here. Corrections to saved reflections and unachieved publish times from past are treated separately and will not be retroactively changed to meet a specified time.

In research activities, we handle a 24-hour weekday live trading drive, but this does not mean we independently verified the current operation or original trade videos this time. Specifications and availability areGogoJungle sales pagePlease refer to this. The most recent inspection records show a stopped-sale indicator. This article shares research verification of self-made indicators and EA and does not guarantee purchase availability, trading recommendations, or profit/loss avoidance.

Final source: DAILY_BRIEF.md at 18:30 JST on October 25, 2026. Nine items, 2/6/1 items, 1/7/1 items are categories for comparisons using prior prices, not counts of new live account trades. 2,500, 3,500, 1,250, 1,750 JPY are the Japanese yen thresholds in the materials, and 0.5R is half of the risk 3,500 JPY for comparison. The final profit/loss difference for the entire strategy remains undecided and will be carried over to next week’s verification conditions.