Do not swing the written bias to London close
Direction is 20/42 = 48% at 16:00Z (55% at 4h). Hard LONG is 0/3. The thesis is conditional; scoring it as a day-hold overstates it.
Freeze 2026-09-14 16:00 UTC · London close · n = 42
Daily CoinGlass theses are a map of levels, not a directional forecast. Unconditional bias is a coin-flip by London close. The usable edges are a green A2 that keeps running through 02:00–08:00, a London fade after a green Asia on SHORT theses, and a first-touch support scalp.
Rules ranked by evidence. Each one names the trade, the window, and the sample that backs it.
Mark support, trigger, targets. Note SHORT vs LONG family and funding. Do not enter yet.
Green 00:00–02:00 is not a leftover close. 02:00–08:00 kept going in 16/17, adding +5.8% on average.
SHORT thesis + green Asia → London down 81%. This is the flagship session trade.
Excluded. NIULAI’s +60% print was a 20:26Z spike. Do not score it.
Direction is 20/42 = 48% at 16:00Z (55% at 4h). Hard LONG is 0/3. The thesis is conditional; scoring it as a day-hold overstates it.
Not tautological. After a green 00:00–02:00, rest of Asia (02:00–08:00) was also up in 16/17 = 94%. Mean rest +5.8% on top of mean A2 +4.6%. Zero days closed green only because A2 was already large. A2 down is the leftover side: remainder bounced 60%.
Best leaf: London down 81% (n=16). Unconditional London is already down 69%. The extra edge is the SHORT filter, not “Asia up” alone.
LONG + Asia down → London up is 50% (n=6). Too small, no edge. Asia-down days still finish London down 69%.
18 first touches ≤6h from dev: 50% win, +0.25R. 0.75% target / 0.5% stop / 30 min. Median win is 1 minute. 83% of tested supports break by London close.
Target-zone scalps are −0.04R. Trigger first-touch is −0.10R. Trade the first support visit; do not hunt the extension.
The 94% is not “Asia finished green because 00:00–02:00 already did the work.” Rest of Asia is the 02:00→08:00 return.
16/17 kept going
After a green A2, 02:00–08:00 was also up. Mean rest +5.8% on top of mean A2 +4.6%, so full Asia averaged +10.8%.
Do not fade 02:00–08:00 after a green A2.
15/25 bounced
After a red A2, the rest of Asia was up 60% (mean +1.3%). Full Asia still finished down 60% because the first two hours did most of the damage.
A red A2 is not a reason to press a short into 08:00.
Mean path after A2 up — rest of Asia added more than the first two hours did.
| Date | Name | A2 00–02 | Rest 02–08 | Full Asia | Path |
|---|
Answer four questions as the day unfolds. The panel on the right is the live recommendation from this sample.
Same window as the playbook: development bar to London close. Direction vs the primary bias family.
| Bias class | n | Correct 4h | Correct London close |
|---|---|---|---|
| CONDITIONAL-LONG | 12 | 8 (67%) | 6 (50%) |
| CONDITIONAL-SHORT | 13 | 6 (46%) | 6 (46%) |
| SHORT-LEAN | 11 | 7 (64%) | 8 (73%) |
| LONG | 3 | 0 (0%) | 0 (0%) |
| SHORT | 1 | 0 (0%) | 0 (0%) |
| RANGE | 2 | 2 (100%) | 0 (0%) |
Post-anchor touches only. +0.75% / −0.5% / 30 min. TIME marked to market. Break-even win rate at 1.5R is 40%.
First touch within 6h: 18 trades, 50%, +0.25R. This is the only scalp with a clean story: fade the level once, flatten fast.
All-touch EV is barely positive. First touch is −0.10R. Do not systematically short every “below X–Y” kiss.
The extension is where the move is spent. Take the first reaction; leave the far liquidity alone.
83% break by London close. A support scalp is not a support investment. Invalidation is a close through the zone.
Click a row for the bias line and session path. A2 is 00:00–02:00; Rest is 02:00–08:00. Two 09-11 USELESS write-ups are the same coin with opposite labels; both are kept.
| Date | Name | Family | A2 | Rest 02–08 | Asia | London | Verdict |
|---|
The pipeline is in this repo. These constraints are load-bearing; ignore them and the old 53% / +60% figures come back.
Window. Development-time 1m anchor (Sydney, before Tokyo) to D 16:00Z. NY is out.
Sample. 42 theses, 7–14 Sep 2026. 09-15 dropped. B3 dropped (venue anomaly). Flagship leaves are n=16–17.
Direction. SHORT-family = SHORT / SHORT-LEAN / CONDITIONAL-SHORT. “Bearish below X, neutral at Y” is not a RANGE.
Session closes. Asia open = 00:00Z open. Asia close = last 1m before 08:00. London close = last 1m before 16:00.
A2 remainder. Rest of Asia is the 02:00→08:00 return, (1+Asia)/(1+A2)−1. A2-up 94% is the same 16/17 as rest-up. Zero leftover closes. A2-down 60% is weaker: 5 of 15 red Asia closes were leftover from A2 while the rest bounced.
Scalps. Paper fills at the touch close. No fees, no slippage. Microcap 1m spreads will eat part of +0.036R.
Unverified. CVD, taker flow, and net-long quotes inside the theses cannot be checked on public Bybit history.
Not odds. These are in-sample frequencies on eight calendar days. Treat 81% and 94% as hypotheses to validate forward, not as fixed edge.
Reproduce. python scripts/05_reanchor_dev.py then 03 → 04 → 06 → 07. Freeze is hardcoded in scripts/common.py.