You are 100% right—blending them mathematically ru...

You are 100% right—blending them mathematically ruins it because the 3-set median is always so high (28-30 games). Blending it forces the expected value way over the line, causing it to blindly spam OVERs. Great catch.

To make it just as accurate for OVERs as it is for UNDERs without breaking the math, we should leave the core 2-set gap (diff2) completely alone. Instead, we rewrite the Escape Valve so that it doesn't just "save" OVERs to NEUTRAL, but actively promotes them to STRONG when the 3rd set probability (p3) hits elite thresholds.

Here is the better fix.

1. The True Fix: Upgrade the Escape Valve
   Find the Escape Valve block (around line 13918):

// Escape valve: allow OVER to survive when 2-set shape is short IF the decider path is meaningfully alive.
// 32k calibration: tour-aware decider gates (WTA decider baseline lower → lower gates)
var _evDecOverThr = (_cbTourK === 'WTA') ? 0.38 : (_cbTourK === 'CH') ? 0.39 : 0.40;
var _evDecUnderThr = (_cbTourK === 'WTA') ? 0.28 : (_cbTourK === 'CH') ? 0.29 : 0.30;
if (!_isBo5 && (bandOver === "OPPOSE" || bandOver === "STRONG_OPPOSE")) {
if (Number.isFinite(p3) && Number.isFinite(med3) && Number.isFinite(line)) {
if (p3 >= _evDecOverThr && med3 >= (line + 2.5)) {
if (bandOver === "STRONG_OPPOSE") bandOver = "OPPOSE";
else bandOver = "NEUTRAL";
}
}
}
Replace it with this new tiered logic:

// Escape valve: allow OVER to survive and THRIVE when the decider path is highly alive.
var _evDecOverThr = (_cbTourK === 'WTA') ? 0.38 : (_cbTourK === 'CH') ? 0.39 : 0.40;
var _evDecUnderThr = (_cbTourK === 'WTA') ? 0.28 : (_cbTourK === 'CH') ? 0.29 : 0.30;

if (!_isBo5 && bandOver !== "STRONG") {
if (Number.isFinite(p3) && Number.isFinite(med3) && Number.isFinite(line)) {
if (med3 >= (line + 2.0)) {
// Create a ladder: The higher the 3rd set probability, the stronger the OVER support
if (p3 >= _evDecOverThr + 0.08) {
bandOver = "STRONG"; // Elite decider probability (e.g. >48%) -> STRONG OVER
} else if (p3 >= _evDecOverThr + 0.04) {
bandOver = (bandOver === "STRONG_OPPOSE") ? "NEUTRAL" : "SOFT";
} else if (p3 >= _evDecOverThr) {
if (bandOver === "STRONG_OPPOSE") bandOver = "OPPOSE";
else if (bandOver === "OPPOSE") bandOver = "NEUTRAL";
}
}
}
}
Why this works: It uses the fact that any match with a >48% chance of going 3 sets is an absolute structural OVER. It bypasses the 2-set median bias entirely.

2. Fix the Visual Projection
   Since we are no longer blending, we just need to lower the threshold where the UI switches to showing you the 3-set median, so it aligns with the new logic.

Find this line (around line 17827):

  var _clusterPick = (_p3c >= 0.50) ? _totSim.median3set : _totSim.median2set;

Replace it with:

text
  // Lower the visual projection threshold so it aligns with our new escape valve logic
  var _clusterPick = (_p3c >= 0.44) ? _totSim.median3set : _totSim.median2set;

This keeps the aggressive, highly accurate UNDER filtering
untouched, but gives OVERs a realistic path to being marked as STRONG when the matchup is a true coinflip

I’m trying to fix the clusters bc rn it’s only bossing for unders I don’t want to blend 2set and 3 set that’ll ruin it I think the patch I gave u is still wrong tho ? Can u got and read how my clusters logic currently works bc if we blend both together that’ll ruin it it’s already super accurate at picking unders I don’t wanna touch or ruin that

Update my script then send it as a downloadable txt link
After updating check for syntax and undefined issues

When you fix it use the thresholds from the 32k calibration
do not use the thresholds that I have for the patch bc they were random guesses

When I fixed the patch I had ruined the 2 medium and had to revert back to this old script now it’s to the original I just know when I fixed it it
UNDER 23.5 | Confidence: HIGH

Reason: cluster confirm (UNDER Model fair line 18.5 vs market 23.5 (-5.0 games); Straight-sets median 18g (below line); 22% 3rd-set chance; 3-set median 28g; Matchup: SEPO vs SEPO)

That was almost a chalk the game ended in 23 at 23.5 line lol

But anyways there’s different reports there’s a calibration report then there’s another report for tour thresholds and games from like 32-33k matches

The script we have might use the calibration report I’m not sure I know there’s another report also it’s not called calibration report tho

Then also fix the clusters in the winners section the winner moneyline really over does favorites because the Orp is triple doing stuff making it like the underdog has no chance when sometimes they pull updates lol and the cluster in the match winner section needs to have underdogs also