Upgrade log
Every real change to Anchor, newest first: bug fixes, new safeguards, rule changes, better test methods and site fixes. Each entry says what changed, why, who raised it, and the numbers before and after. Only changes that actually happened and are in effect are listed; failures are listed together with their fix.
Oct 31, 2026
Our public Telegram channel (t.me/ProjectAnchorHQ) launched and is now linked from every page footer, the home page and the Chinese home page.
Why. Test results should be easy to follow outside X. The channel only posts our own test records, pass or fail; it has no discussion and no trading calls.
Before and after. Before: X was the only public channel. After: every test result is also posted to the Telegram channel, and a weekly digest follows on Sundays.
Raised by: Our own idea
Oct 31, 2026
This public upgrade log page went live, linked from the footer of every page.
Why. Fixes, new safeguards and rule changes should be visible in one place, together with who raised them and the numbers before and after.
Before and after. Before: changes were only mentioned in scattered posts. After: one dated page in English (/changelog/) and Chinese (/zh/changelog/), generated from the same source file as each change is made.
Raised by: Our own idea
Oct 30, 2026
Strategy test reports now list out-of-sample results for each rolling sub-period, with a stability summary, instead of only one overall pass/fail.
Why. A strategy can pass on the total while making all its money in one stretch and losing in the rest. Per-period results show whether the edge is steady.
Before and after. Before: one aggregate verdict per test. After: from the week of Nov 2, 2026, every report lists each pre-registered segment (trades, win rate, net R per trade, drawdown). The segmenting is locked before the test; for LIQ_CASCADE_REVERSAL_V1 it is six half-year windows (2024-01 to 2026-09), locked Oct 31, 2026.
Raised by: An industry practitioner's feedback on LinkedIn
Oct 30, 2026
Fixed a "phantom short" in the signal watcher, and set how a missed signal is handled.
Why. The MF4 short signal at 13:39 (UTC+8) on Oct 30, 2026 (entry ref 77,834, stop 78,913.8) was sent, but no order was placed because the assistant that places orders had run out of usage. On restart, the watcher replayed history and believed it was holding that short, so it would have ignored the next real signal.
Before and after. Before: watcher state "short" while the Kraken account was flat. After: the account was confirmed flat, the missed trade is recorded as missed (not chased later at a worse price), and a one-off flat override keyed to that entry time makes the watcher flat without blocking any new position. Watcher restarted Oct 30, 2026, 6:58 AM ET (10:58 UTC) and has shown "flat" since.
Raised by: Found by us
Oct 28, 2026
Position sizing on the evaluation account now also reserves about two days of perpetual funding, and an open position is closed early if funding could push a stop-out to within $10 of a loss limit.
Why. Kraken charges funding every 4 hours and counts it toward the daily loss limit, so a long hold can breach the limit even if the stop is hit exactly.
Before and after. Cost factor k = 1 + c × entry / stop distance. Before: c = 0.0008 (commission only). After: c = 0.0014 (commission + funding). Example with entry 77,834 and stop distance 1,079.8: k goes from 1.058 to 1.101, so the dollar risk per trade shrinks by about 4%.
Raised by: Follow-up to @breachless_dev's point on X
Oct 28, 2026
Position sizing on the evaluation account now counts round-trip fees, so a full stop-out including fees still leaves a $30 buffer above both the daily loss line and the overall drawdown floor.
Why. @breachless_dev pointed out on X that Kraken counts commission (and funding) toward the daily loss limit. Our formula only looked at the price distance to the stop.
Before and after. Before: the room left above each loss line was not reduced for fees. After: risk = min(200, (equity − daily line − 30)/k, (equity − 9,400 − 30)/k) where k = 1 + 0.0008 × entry / stop distance; no trade is taken if the result is under $50. Example with entry 77,834 and stop distance 1,079.8: fees add about 5.8% to the loss of a full stop-out, which the old formula ignored.
Raised by: @breachless_dev on X