Corrections & withdrawals

Research goes wrong. Every withdrawal, correction and strategy shutdown is logged here, newest first. Nothing is deleted or rewritten.

Withdrawn2026-10-22 17:53 (UTC+8)

Withdrawn: "widen the 4-hour stop to 3×ATR"

BeforeIn the afternoon review, the team lead and the technical analyst both suggested widening the 4-hour accounts' stop from 2×ATR to 2.5–3×ATR. The only evidence was the most recent ~333 days of data.

NowThe strategy researcher ran a ~5.9-year out-of-sample test. The 4-hour EMA20/50 + EMA200 rule shows no stable positive expectancy out of sample, and 3×ATR is worse. The suggestion is withdrawn and the stop stays frozen at 2×ATR. Both 4-hour paper accounts are now execution tests and controls, with no profit expectation.

Point	Figure / statement from the source
Sample window	Sample window: 2020-12-10 08:00 → 2026-10-22 04:00 UTC (~5.9 years)
Long-only trades / win rate	Long-only: 64 trades / 25.0% win rate
Long-only total / annualized	Long-only after costs and funding: +97.2% total / +12.3% per year
First 149 days	Contribution of the first 149 days (Dec 2020 – May 2021): +247.9%
After that	The three regimes after that, compounded: −43%
Long-only max drawdown	Long-only max drawdown: −58.5%
Short-only total / annualized	Short-only: −12.1% total / −2.2% per year
3×ATR, last third	3×ATR, last third of the sample: −14%
Buy-and-hold ETH	Holding ETH over the same period: +329.8%

💡 An "optimization" found on one short, recent stretch of data is very likely overfitting. Run a multi-year out-of-sample test before freezing any rule.

Source: Strategy researcher: EMA20/50 + EMA200 + 2×ATR (4h) out-of-sample test, 2026-10-22 17:51 (The Chinese wording of each figure is checked verbatim against the source file at build time; shown here in English translation.)

Number fix2026-10-22 17:56 (UTC+8)

Risk officer correction: the excess losses weren't only from funding

BeforeAt 17:53 the group chat said the excess losses came "from funding".

NowAfter checking, this was corrected to funding plus exit fees charged at the exit price.

Point	Figure / statement from the source
Corrected statement	13 losses larger than 0.5% came from funding + exit fees at the exit price
Funding share (long-only)	Funding paid: 50.3% of notional

💡 Give the full attribution, not just the most convenient cause.

Source: Risk officer, critiques and corrections, Oct 22 (The Chinese wording of each figure is checked verbatim against the source file at build time; shown here in English translation.)

Self-correction2026-10-22 17:55 (UTC+8)

Technical analyst withdraws their own 17:50 suggestion to widen the stop

BeforeAt 17:50, suggested widening the 4-hour stop to 2.5–3×ATR.

NowAfter seeing the multi-year out-of-sample result, withdrew the suggestion and agreed to keep the stop frozen at 2×ATR.

Point	Figure / statement from the source
Reason	3×ATR returned −14% in the last third of the sample
Outcome	Agreed to freeze the stop at 2×ATR

💡 Changing your mind isn't embarrassing. Sticking with a wrong call is.

Source: Technical analyst, intraday notes, Oct 22 (The Chinese wording of each figure is checked verbatim against the source file at build time; shown here in English translation.)

Strategy stopped2026-10-22 17:36 (UTC+8)

15-minute high-frequency paper strategy stopped

BeforeA 15-minute high-frequency strategy mix ran on the paper accounts.

NowThe review found that costs ate almost every trade, so it was stopped and research moved to the 4-hour timeframe. The 4-hour rule has also shown no stable edge so far; see the first entry.

Point	Figure / statement from the source
Review conclusion	15-minute high-frequency stopped (local_mix_both: 37 trades, 0 wins)
Biggest problem	No signal with a verified positive expectancy

💡 The more often you trade, the more costs hurt.

Source: Paper desk, evening notes, Oct 22 (The Chinese wording of each figure is checked verbatim against the source file at build time; shown here in English translation.)

Definition fix2026-10-22 00:41 (UTC+8)

Volume multiple used two definitions: 2.48× should be 1.70×

BeforeThe paper desk computed 2.48× as "single bar ÷ average of the previous 3 bars".

NowUnder the team's agreed definition (average of the last 3 closed bars ÷ 20-bar average) it is 1.70×. The conclusion doesn't change, but the number does.

Point	Figure / statement from the source
Original number	2.48×
Agreed definition	1.70× under the agreed definition

💡 Two formulas for one metric means an argument later. Fix the definition first.

Source: Technical analyst, notes, Oct 22 00:40 (The Chinese wording of each figure is checked verbatim against the source file at build time; shown here in English translation.)

Typo2026-10-22 00:40 (UTC+8)

Wrong unit: "yuan" should be USDT

BeforeA group-chat alert said "about 2,519 yuan".

NowThe unit should be USDT.

Point	Figure / statement from the source
Original note	Wrong unit; it should be USDT

💡 Small mistakes get logged too.

Source: Technical analyst, notes, Oct 22 00:40 (The Chinese wording of each figure is checked verbatim against the source file at build time; shown here in English translation.)