Financial Instruments Business Operator Kanto Finance Bureau Director (Kinsho) No. 1960 / Membership Association Investment Trusts Association, Japan Membership No. 012-02323 Go to GogoJungle Home
icon List
柴亮太
09/21/2026 19:04
Release： 09/21/2026 19:04
Update： 09/21/2026 19:04
柴亮太
Investment style : Trading System
Investment career : More than 10 years
Investment amount : 10 million - 50 million JPY
Representative of RO LLC. My main line of work is AI agent research and development. I have about 25 years of investing experience, but it’s not a brag-worthy track record. I am not the type to sit still in front of a chart, and I would often cut losses when I looked away during critical moments. Through that repeated pattern, I have melted tens of millions of yen. What remains is knowledge and a record of my failures. “If I can’t watch it myself, I’ll build something that can watch for me.” With that in mind, I am also working on developing AI agent EAs. Not to predict the market, but to tell me when not to enter a position. To avoid silently expanding losses when prices move against me. Listing on GoGoJungle is a testing ground for that research and development. The tools I publish here are, as of now, completed products, but not the final finished version. They are things that run on actual accounts, record losses, and are cultivated while being improved each time. What didn’t go well is also shown as is. Therefore, for those who purchase, it isn’t a “buy and end” relationship, but a partnership to accompany the research. It isn’t that you are not winning because of your efforts. I have faced the same points and repeated the same losses many times. Those who stumbled in the same place, this time I hope we can turn to the winning side together and continue with verification. If you have questions or requests, please don’t hesitate to share. The feedback we receive will be reflected in the next version.
Article cover image

Financial AI and AI Agent R&D: Deploying in production the "up to 4" rule to stop slow re-entries

FinTech AI / AI Agent Research and Development

Financial AI / AI Agent Research and Development: Deploying the “4-entries max” rule to stop slow re-entries

To curb the repeated re-entry in the same direction toward the end of a trend, we added an upper limit to the new-entry evaluation for GOLD13: “same-direction warn0 allowed for 1–4 entries, blocked from the 5th entry onward.”

※This post is a development log of the indicators I developed and automated trading software (EA) researched and validated with an AI agent on real accounts. We accumulate both successes and failures as evaluation data and continuously refine the logic. The validation process is also shown in a 24/7 weekday live trading stream.

Why we introduced a limit on entry attempts

Even if a trend appears to continue, there can be little remaining price movement toward the end. Therefore, the AI agent counts how many re-entry candidates have appeared within the same flow, and is designed to stop late candidates at the entry point.

This time's definitive rules

  • Targets are fresh/REALTIME/warn_level=0/same symbol/same direction.
  • The 1st–4th entries of the same episode are allowed for new-entry evaluation.
  • From the 5th entry onward, both SOUL1 and NORMAL are blocked.
  • warn1/2, LEGACY_HISTORY, stale, and opposite directions are not counted.
  • BUY and SELL are counted separately, and duplicate records on the same bar are treated as one.
  • If the interval of same-direction warn0 is 6 bars or more, reset as a new episode.

Blocked candidates areguard_rejected / skip_warn0_late_clusterpersisted so that after restart the same candidates are not re-evaluated.

Validation before and after deployment

We compiled and performed integration tests on local, VM candidate, and live production modules, and behavior tests passed on 7/7. Confirmed are: permission of the 4th entry, blocking of the 5th entry, reset at 6 bars, exclusion of warn1/2, history, stale, independence of BUY/SELL, non-accumulation of duplicates on the same bar, and not progressing to Gemini/signal/order after blocking.

In the September 17, 2026 replay, the 1st entry corresponding to ticket 29721052 was allowed, and the 8th entry corresponding to ticket 29721491 was blocked. The 21:15 cron alsono_breakout / exit=0ended normally, and there were no exceptions after loading the new source.

What has not changed

EA5, NORMAL, and SOUL1 entry functionality remains enabled. There have been no changes to the management of existing positions, SL, settlement, lot sizes, SL/TP, Quality Gate, AutoTrading, OBS, or YouTube. The deployment heartbeat had one unsettled position, but there were no policy changes or settlements for that position.

Things not yet proven

Natural trades that trigger the 5th entry or more in the same direction and actuallyskip_warn0_late_clusterstop at that point are not confirmed at deployment.We will verify through code tests and replays, without conflating with real trades, by waiting for the next relevant episode.

Video verification:The focus this time is the deployment record of entry rules; there are no videos yet of the natural trades blocking the 5th entry. We will publish evidence when actual blocks are observed, without embedding irrelevant recordings.

GOLD13 Sales Page
Weekday 24-Hour Trade Live

This article is a development and verification log and does not guarantee profit.

FinTech AI / AI Agent Research and Development

Financial AI / AI Agent Research and Development: Deploying the “4 entries max” rule to stop slow re-entries

To curb the pattern of repeatedly re-entering in the same direction toward the end of a trend, we added an upper limit to GOLD13's new-entry evaluation: “same-direction warn0 is allowed for the 1st–4th entries; the 5th entry onward is blocked.”

※This post is a development log of indicators I developed and automated trading software (EA) that are researched and validated with AI agents on real accounts. We accumulate both successes and failures as evaluation data and continuously improve the logic. Validation is also shown in a 24/7 weekday live trading stream.

Why we introduced a limit on entry attempts

Even if the same trend seems to continue, there may be little price movement left at the end. Therefore, we count how many re-entry candidates appear within the same flow on the AI agent side and design to stop late candidates at the entry point.

This time's definitive rules

  • Targets are fresh/REALTIME/warn_level=0/same symbol/same direction.
  • The 1st–4th entries of the same episode are allowed for new-entry evaluation.
  • From the 5th entry onward, both SOUL1 and NORMAL are blocked.
  • warn1/2, LEGACY_HISTORY, stale, and opposite directions are not counted.
  • BUY and SELL are counted separately, and duplicate records on the same bar are treated as one.
  • If the interval of same-direction warn0 is 6 bars or more, reset as a new episode.

Blocked candidates areguard_rejected / skip_warn0_late_clusterpersisted so that after restart the same candidates are not re-evaluated.

Validation before and after deployment

We compiled and performed integration tests on local, VM candidate, and live production modules, and behavior tests passed on 7/7. Confirmed are: permission of the 4th entry, blocking of the 5th entry, reset at 6 bars, exclusion of warn1/2, history, stale, independence of BUY/SELL, non-accumulation of duplicates on the same bar, and not progressing to Gemini/signal/order after blocking.

In the September 17, 2026 replay, the 1st entry corresponding to ticket 29721052 was allowed, and the 8th entry corresponding to ticket 29721491 was blocked. The 21:15 cron alsono_breakout / exit=0ended normally, and there were no exceptions after loading the new source.

What has not changed

EA5, NORMAL, and SOUL1 entry functionality remains enabled. There have been no changes to the management of existing positions, SL, settlements, lot sizes, SL/TP, Quality Gate, AutoTrading, OBS, or YouTube. The deployment heartbeat had one unsettled position, but there were no policy changes or settlements for that position.

Things not yet proven

Natural trades that trigger the 5th entry or more in the same direction and actuallyskip_warn0_late_clusterstop at that point are not confirmed at deployment.We will verify through code tests and replays, without conflating with natural trading, by waiting for the next relevant episode.

Video verification:The focus this time is the deployment record of entry rules, and there are no videos yet of natural trades blocking the 5th entry. We will publish evidence when actual blocks are observed, without embedding unrelated recordings.

GOLD13 Sales Page
Weekday 24-Hour Trade Live

This article is a development and verification log and does not guarantee profit.

Why implement the entry-limit on the AI side

If you fix a fixed count only in the EA, there is a risk of mechanically stopping without considering market context. Conversely, relying solely on the model's textual judgment can yield variations in expression for the same input, making it impossible to reproduce the blocking reasons. Therefore, we implemented deterministic guard for episode counting and upper-limit evaluation, with AI agents handling the preceding market assessment in a two-layer structure.

What we seek in financial AI is not to make the AI think of everything. We separate where to handle ambiguity from where to guarantee identical results, and make the constraints stronger as orders approach. This change adds that boundary of responsibility at the entry point.

Expected failure patterns

  • Count the same-bar duplicate records as separate candidates and reach the limit quickly
  • Episode state disappears on restart and the 5th is treated as the 1st
  • Mixing warn1/2 and history data may wrongly block a valid candidate
  • BUY counts flow into SELL and blocks candidates right after a reversal
  • Only the block log remains while signals and orders progress in the background

To avoid these, we verify counting conditions, persistence, and late non-execution separately. We decided not just to say tests passed, but to specify what evidence would appear when something breaks.

Operational data to collect

Going forward, we will store, in the same format, the occurrence time, price, spread, ATR, and immediately subsequent maximum favorable and maximum adverse moves for the allowed 1–4 entries and the blocked 5th entry onward. If blocked candidates expand significantly, we will treat it as false blocking and reassess the time-limit and 6-bar reset conditions.

Conversely, if retracements after the 5th entry are frequent, we can evaluate not only win rate but also expected value, maximum drawdown, consecutive losses, and opportunity loss as evaluation targets.

What will be published in the next update

If a relevant episode appears in natural trades, we will combine the telemetry candidate ID, guard_rejected log, no signal generation, no order transmission, and corresponding chart recording into one timeline. If no qualifying recording exists, we will not publish separate trades just to give the impression of footage.

Summary

This implementation of the “4 entries max” is not a guaranteed optimal profit value, but the initial constraint to make slow re-entries measurable. By establishing fixed rules, collecting evidence, and evaluating both false blocking and loss avoidance before updating, we will move AI and EA improvements from intuition-based to verifiable development.