TL;DR — v4.2.5 stacked five new filters simultaneously, violating our own one-variable-at-a-time discipline. The result: the bot went nearly silent, missing clean 300-point trends that met the spirit of every entry rule but failed on the stacked technicalities. v4.2.6 corrects this by keeping the one filter the data actually proved (2-bar consecutive confirmation) and returning the selectivity gates to the levels where the engine was profitable. One real improvement over the proven baseline — not five untested ones stacked on top of each other.
What Happened — The Lesson
We over-tightened. Analysis of v4.2.4 showed that single-bar spike entries were the primary drag (43% win vs 73% for multi-bar entries). The correct fix was the 2-bar consecutive confirmation gate. But v4.2.5 also raised the power threshold, widened the gap requirement, increased the ADX floor, required rising ADX, and added bar-quality filters — all at once. Each made individual sense; stacked together, they created an entry gate that almost nothing could pass. A clean trending session with 300+ points of movement produced near-zero trades. The cure was worse than the disease.
This is the exact mistake the validation discipline is designed to catch: changing multiple variables simultaneously makes it impossible to know which change helped and which hurt. v4.2.6 resets to known-good levels and adds only the one filter with direct data evidence behind it.
What Changed — v4.2.5 → v4.2.6
Gates returned to proven levelsCalibration
The selectivity gates that were raised in v4.2.4/v4.2.5 are returned to the levels where the Power Engine was demonstrably profitable (v4.2.3).
Setting
v4.2.5
v4.2.6
Rationale
Min Power (dominant)
85
80
The engine was profitable at 80; raising to 85 blocked valid setups
Min Power Gap
55
50
Same — 50 was the proven level
Min ADX
23
20
23 was too strict for 1-min chart; 20 was where it worked
Require Rising ADX
On
Off
ADX oscillates bar-to-bar on 1-min; requiring it to rise every bar killed most entries
Kept from v4.2.5 — the real improvementsRetained
Everything that was data-driven and not part of the gate-stacking problem carries forward unchanged:
2-bar consecutive confirmation — the core filter; directly supported by the 43% vs 73% win-rate split in the data
Bar quality (body ≥ 0.35 ATR, close in strong zone) — lightweight, filters weak dojis
Sizing 3 / 5 / 7 with $400 hard per-trade cap
No C-grade trades (minimum grade B)
5-bar early-out window (loosened from 3 to give confirmed entries room)
Stepped stop ladder (breakeven at +9, lock half at +18, stepped trail ~7pts behind)
All safety systems (latched lockout, blackouts, pre-halt, disaster stops)
Power as default entry type, display defaults baked (Surf Report, Large, Bottom Left)
The Design Principle — Stated for the Record
v4.2.6 = the profitable v4.2.3 engine + the one filter the data demanded.
The Power Engine at 80/50/20 gates was net profitable. The 2-bar consecutive confirmation filters the single-bar fake spikes that were the primary drag. Together, they represent one real improvement over a proven baseline — which is how validation is supposed to work. Every other filter (rising ADX, tighter gates, higher thresholds) remains available as a toggleable setting for future testing, but is off by default until data earns it individually.
What This Version Should Look Like
Compared to what we've seen this week:
Version
Trades/Session
Behavior
v4.2.3 (legacy gates, no consec)
~83
Too many trades, fake spikes, cost drag
v4.2.4 (raised gates, no consec)
~45
Fewer trades, but fakes still got in
v4.2.5 (raised gates + all filters)
~0–3
Near-silent, missed clean trends
v4.2.6 (proven gates + 2-bar consec)
~20–35 (expected)
Fewer than v4.2.3, better quality than v4.2.4, actually trades
Deployment
Load v4.2.6; confirm dashboard banner reads POWER v4.2.6.
Recreate the execution alert on v4.2.6.
TradersPost signal multiplier confirmed at 1× on all accounts.
Confirm the first trade's reason code reads "Power".
No manual configuration needed — all defaults are baked.
What We’re Watching
(1) Does it actually trade? (v4.2.5's failure mode.) (2) Is the win/loss size ratio back above 1.0×? (The 2-bar filter's job.) (3) Are the fake single-bar spikes filtered while real multi-bar moves are caught? (4) Trade count in the 20–35 range? (Between v4.2.3's over-trading and v4.2.5's silence.)