← Academy home
Deep Dive · Sentinel · Part B of D

Forensics: Why USDJPY?

Part A found a lookout that might have been asleep, and a specialist whose talent nobody could explain. Part B puts both under the microscope — with the watchman's actual trades as evidence. Both mysteries crack.

B.1The pip_value trial — asleep, but innocent

The charge from Part A: with the default pip_value = 0.0001, USDJPY's Waddah dead zone was ~100× too small, so the "energy must shake" lookout may have been waving everything through in Rounds 1–3. The experiment: run Sentinel over the same five years of USDJPY H1 twice — wrong pip value vs correct — and compare everything.

wrong pip (0.0001)correct pip (0.01)
bars where WAE "shaking"99.9%85.0%
trades (5 years)1,8771,871
profit factor1.161.16
total return+19.0%+18.9%
Verdict

Guilty of sleeping, innocent of fraud. The energy check really was waving nearly everything through (99.9% of bars!) — but correcting it changed almost nothing, because the WAE vote has a second half: "rising", which only passes on 61.5% of bars and was doing the real doorwork all along. Six trades out of ~1,880 differ. The USDJPY edge survives the correction intact — and the roster's pip_value: 0.01 stands as the correct config.

B.2One watchman, two cities

Same strategy, same settings, same five years, same risk stack (ATR×1.5, 1:3, cooldown 3). Only the pair changes:

USDJPY H1EURUSD H1
trades1,8711,884
profit factor1.160.84
total return+18.9%−21.5%
win rate34.6%31.6%
median hold3 h2 h
take-profits hit105106
stop-losses hit202315
bars flagged choppy0.9%3.4%

Read the exit money and the case solves itself:

+6,238 · USDJPY+5,181 · EURUSDTP wins-4,229 · USDJPY-6,272 · EURUSDSL losses-118 · USDJPY-1,056 · EURUSDsignal exits
Profit contribution by exit type, five years. The take-profit harvest is nearly identical on both pairs — the entire difference between +19% and −21% is stop-outs: EURUSD hits 55% more of them.

So Sentinel isn't better at winning on USDJPY. It's better at not being thrown off the boat before the win arrives. Why would the same mooring line snap so much more often on EURUSD?

B.3The river and the lake

The obvious suspect — "USDJPY just trends more" — is false by the standard measure. The hiker test says they're twins: ADX above 25 on 46.0% of USDJPY bars and 46.1% of EURUSD bars. Equal ripples. The real difference only appears when you zoom all the way out:

USDJPY +48% EURUSD +2% start = 100 Sep 2021 → Jul 2026 · weekly closes, normalized
Five years, both starting at 100. USDJPY: a river — local ripples on a persistent one-way current (+48%). EURUSD: a lake — the same ripples, but the water ends up where it started (+2%).

Sentinel is a boat that always sails with the local current. On the river, every pullback is shallow relative to the flow — the mooring line holds, the 1:3 target gets reached. On the lake, every "trend" eventually round-trips — the same line gets snapped again and again on the way back. The buy/sell ledger agrees: USDJPY signals ran 58% long over the period (riding the drift); EURUSD's were an indifferent 49/51.

The honest implication

Sentinel's edge is drift harvesting — and the drift is the era of yen weakness, not a law of nature. The per-year ledger shows it: strong while the river ran hard (PF 1.42 in 2021, 1.28 in 2022), thinner as it calmed, and 0.78 in 2026 so far — the river may be slowing. This goes straight into Part D as the thing the campaign must watch: a drift gauge and a kill criterion, not blind faith.

yearUSDJPY PFEURUSD PF
20211.420.72
20221.280.79
20231.150.8
20241.290.73
20251.071.05
20260.780.93

B.4The clock — when the watchman works

USDJPY entries EURUSD entries 036912151821 hour of day (broker server time)
Entries by hour (broker server time). USDJPY's entries cluster in two humps — the Tokyo morning and the New York afternoon — while EURUSD's pile into the London day. The watchman keeps each pair's local hours, exactly as Lesson 6's session logic predicts.

B.5The rudder finding — a bonus for Part C

One more pattern fell out of the near-miss data. When Sentinel is one vote short of firing (about 15% of all bars), which lookout is the dissenter? On both pairs, overwhelmingly the same one:

dissenting lookout at 6-of-7USDJPYEURUSD
④ 50 SMA slope (the tanker's rudder)2,7823,024
⑤ WAE active+rising1,140807
⑦ MACD bar196161
① Heiken Ashi185191
② SSL Channel97105
③ NonLagDot · ⑥ WAE direction00

The SMA slope is Sentinel's real throttle — it vetoes more near-signals than every other lookout combined. Its 3-bar slope window is therefore the highest-leverage dial in the whole strategy, and the first thing Part C's sensitivity map should sweep. Also notable: NonLagDot and WAE-direction never cast the lone dissent — whenever they disagree, others already do. They may be redundant passengers; Part C can test removing them.

B.6Part B verdict

Case closed — three findings

① The pip_value scare is resolved: the lookout slept, the edge stands, the roster config is correct.

② "Why USDJPY" has an answer: not more trend, but net drift — equal ripples, different rivers. Sentinel is a drift harvester, and its stop-outs, not its wins, are where pair selection matters.

③ Two leads for Part C: the SMA-slope rudder is the strategy's dominant dial (sweep it), and two lookouts may be redundant (test them). Plus the standing question from Part A: verify the WAE formula against your MT5 chart — the trial reduced the urgency, but fidelity still matters.

Homework, optional: if you want Part C to test whether the edge transfers to other rivers, run export_history.bat --symbols EURJPY,GBPJPY,AUDJPY --timeframes H1 next time MT5 is open — other yen pairs shared the same drift.