Why did the stock appear to crash right after the trade was filled?

The stock did a 3-for-1 split on September 29. yfinance adjusts past daily candles down to one-third after a split, while the ledger kept the pre-split prices, so the comparison looked like an instant one-third drop.

How many records changed after the fix?

Of the 17 rewritten settlements, 9 changed in value. The cumulative figures also moved: the 1-business-day-after count went from 12 to 13 out of 23, and the 3-business-day-after count from 9 to 10 out of 15, as of October 5.

Did the fix change the prices stored in the ledger?

No. The ledger keeps the values from the moment of extraction. The fix only changes the comparison, dividing the ledger prices by the split ratio just before judging.

AI Coding AssistantsAI Stocks & MarketsZenn AI/ML1d agoOct 27, 2026, 13:00 JST

ALPHA FORGE fixes 17 wrong calls from split-adjusted price mismatch

Send on LINE

3 Key Points

What happened
On October 5, Claude Code spotted that one stock had 17 judgments where the invalidation level was hit on the very day of execution; that stock had done a 3-for-1 split on September 29.
Why it matters
yfinance returns past daily candles adjusted down to one-third after a split, while the prediction ledger keeps the original pre-split prices, so comparing them made the stock look like it crashed one-third instantly and triggered the invalidation judgment, according to the developer.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, every morning.

Get these free every morning

Context & Analysis

ALPHA FORGE is a personal project: a Japanese stock AI screening and continuous-learning terminal developed by one person together with Claude Code. Each morning the AI extracts candidate stocks and records them in a prediction ledger, storing only what was known at extraction time — the simulated entry price, the scenario invalidation level, and the expected range ceiling. Every evening, a settlement process compares those ledger prices with subsequent daily candles to see whether the entry filled and which of the invalidation level or the range ceiling was reached first.

The trouble surfaced on the morning of October 5, when Claude Code was checking the settlement numbers and noticed one stock behaving oddly: judgments showed the invalidation level being reached on the very day of execution. That stock had split one share into three on September 29. Because yfinance returns historical daily candles adjusted retroactively to one-third after a split — a common convenience so charts look continuous — the ledger's pre-split prices were being measured against post-split candles. The developer checked 24 other already-settled stocks and found no other discrepancies.

The repair keeps the ledger's recorded values untouched so the original morning's reasoning stays readable, and instead aligns the units just before comparison. Splits effective on or after the extraction day are multiplied into a cumulative ratio, and the three prices are divided by it; candles without a Stock Splits column, such as those from a backup data source, are left uncorrected. The 17 existing settlements were rewritten only after a full backup of the production database, with the resolved_at timestamps kept at their original times. The developer notes that because the numbers moved in a favorable direction, recording the changed count and the before-and-after values mattered all the more — a discipline the AI's instructions formalize as confirmation, backup, and a report of changed counts and values.

FAQ

Why did the stock appear to crash right after the trade was filled?: The stock did a 3-for-1 split on September 29. yfinance adjusts past daily candles down to one-third after a split, while the ledger kept the pre-split prices, so the comparison looked like an instant one-third drop.
How many records changed after the fix?: Of the 17 rewritten settlements, 9 changed in value. The cumulative figures also moved: the 1-business-day-after count went from 12 to 13 out of 23, and the 3-business-day-after count from 9 to 10 out of 15, as of October 5.
Did the fix change the prices stored in the ledger?: No. The ledger keeps the values from the moment of extraction. The fix only changes the comparison, dividing the ledger prices by the split ratio just before judging.

Zenn AI/MLRead Original Article

Send on LINE

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Get it free

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.