OneQAZ OneQAZ

corrections 한국어

Corrections

OneQAZ aims to be a verifiable, timestamped record. When we find an error we never hide it or silently overwrite it — the post is republished at the same URL with a visible correction notice and a modification timestamp (dateModified), and the reason is disclosed on this page.

Incident report: the 2026-07 simulation-data defect

What happened. Between 2026-07-08 and 07-15, a signal-persistence fault in the trading engine caused new simulated entries to be recorded at stale signal prices from days earlier instead of live market prices. Exits were still evaluated at real-time prices, so the ledger filled with simulated trades at prices — and returns (±10–35%) — that could not exist in the actual market, and the same fault kept re-buying at identical prices in a loop. Roughly 79% of closed trades in that window were such artifacts.

Impact on this blog. Daily journals aggregate the ledger as-is, so the fake trades surfaced in that window's journals (three markets) and one weekly track record. Separately, an internal test symbol (TEST) leaked one row into the 07-07 crypto journal.

Response timeline (KST).

Preventing recurrence. ① Two automatic publish-time sanity checks — a repeated-identical-entry-price pattern, or a spike in ultra-short extreme-P&L trades, holds publication and alerts the operator (if upstream data is ever contaminated again, the blog is the last line of defense). ② Non-tradable symbols are excluded automatically. ③ The prediction ledger is sealed by a daily SHA-256 hash chain (anchor published in weekly track records), so any post-hoc edit shows up as a chain mismatch.

Corrected posts (20 · auto-generated)

Any post with a correction timestamp (updated) in its frontmatter is listed here automatically. Original URLs never change. Korean-language corrections are listed on the Korean page.