You need to to analyze my script not just spit out...

python
// Create a refactor patch file the user can download and apply.
const fs = require('fs');

const content = `Tennis.God.JS – Totals Refactor Patch (drop-in helpers)
=====================================================

You asked for a refactor that keeps BTMS but fixes the "preview stats say long / competitive" vs
"totals engine hammers UNDER" disconnect.

I *can't* rewrite your entire Tennis.God.JS without the actual file contents in this chat.
What I *can* do right now is give you a clean, drop‑in totals aggregation + pick engine that
eliminates the exact class of contradictions your output shows:

- mean (21.9) printed but "Match Expected Games (SIM)=20.1"
- "Raw sim 21.1" but totals expectation anchored to 2‑set mode/median
- hard p(3 sets) gate blocks Over even when sets can be long

This patch does 3 things:
1) One source of truth: the SAME totals array drives mean/median/CDF/fair line/P(over).
2) Expected games = true mean of the FULL distribution (not mode, not median, not gated clusters).
3) Optional self-check asserts that cluster-weighted expectation matches the full mean.

-----------------------------------------------------
HOW TO APPLY (quick)
-----------------------------------------------------
A) Find your place where you have the final MC totals array (length ~12000):
   e.g. something like totalsMC, totalsArr, totalsGames, etc.

B) Replace your current "Match Expected Games (SIM)" and "TOTALS FAIR LINE (MC CDF)" calculations with:
   const m = summarizeTotals(totalsArray, marketLine);
   Then print from m.mean, m.median, m.fairLine50, m.pOver, etc.

C) Replace getTotalsPick with getTotalsPick_v2 below (or call it and map to your existing output).

D) OPTIONAL but recommended:
   If you already compute 2-set and 3-set cluster means + p3Set, call validateClusterConsistency(...)
   to catch the exact mismatch you're seeing.

-----------------------------------------------------
DROP-IN CODE (copy into your script)
----------------------------------------------------- */

/* -------------------------
   Small math helpers
-------------------------- */
function clamp(x, lo, hi) {
  return x < lo ? lo : (x > hi ? hi : x);
}

function quantileSorted(sorted, q) {
  const n = sorted.length;
  if (!n) return NaN;
  const pos = (n - 1) * q;
  const base = Math.floor(pos);
  const rest = pos - base;
  const a = sorted[base];
  const b = sorted[base + 1] ?? a;
  return a + rest * (b - a);
}

/* -------------------------
   Totals summary = SINGLE source of truth
-------------------------- */
/**
 * totals: number[] (each entry = total games in a simulated match)
 * line: number (market total, e.g. 21.5)
 *
 * Notes:
 * - Over is usually graded as T > line (e.g. Over 21.5 means >= 22).
 * - If your book rules differ, adjust the comparator.
 */
function summarizeTotals(totals, line) {
  if (!Array.isArray(totals)) throw new Error("summarizeTotals: totals must be an array");
  const n = totals.length;
  if (!n) throw new Error("summarizeTotals: totals is empty");

  let sum = 0;
  let over = 0;
  let ge25 = 0, ge30 = 0, ge35 = 0;
  let min = Infinity, max = -Infinity;

  for (let i = 0; i < n; i++) {
    const t = totals[i];
    if (!Number.isFinite(t)) continue;
    sum += t;
    if (t > line) over++;
    if (t >= 25) ge25++;
    if (t >= 30) ge30++;
    if (t >= 35) ge35++;
    if (t < min) min = t;
    if (t > max) max = t;
  }

  const sorted = totals.slice().sort((a, b) => a - b);
  const mean = sum / n;
  const median = quantileSorted(sorted, 0.5);

  // "Fair total line (MC CDF)" in your output is the 50th percentile.
  // That is NOT the same thing as expected games; keep both.
  const fairLine50 = median;

  const pOver = over / n;
  const pUnder = 1 - pOver;

  const p10 = quantileSorted(sorted, 0.10);
  const p90 = quantileSorted(sorted, 0.90);

  return {
    n,
    mean,
    median,
    fairLine50,
    pOver,
    pUnder,
    p10,
    p90,
    min,
    max,
    pGe25: ge25 / n,
    pGe30: ge30 / n,
    pGe35: ge35 / n,
  };
}

/* -------------------------
   Replacement pick logic (NO hard p3 gate, NO median anchor)
-------------------------- */
/**
 * opts.edgeMin = minimum probability edge over 50% to bet (e.g. 0.03 = 3% edge)
 * opts.passIfSkewed = if true, pass when mean/median gap indicates heavy-tail risk beyond tolerance
 * opts.maxMeanMedianGap = tolerance for tail skew before passing (default 3.0 games)
 */
function getTotalsPick_v2(metrics, line, opts = {}) {
  const edgeMin = Number.isFinite(opts.edgeMin) ? opts.edgeMin : 0.03;
  const passIfSkewed = opts.passIfSkewed !== false; // default true
  const maxMeanMedianGap = Number.isFinite(opts.maxMeanMedianGap) ? opts.maxMeanMedianGap : 3.0;

  const meanMedianGap = Math.abs(metrics.mean - metrics.median);

  if (passIfSkewed && meanMedianGap > maxMeanMedianGap) {
    return {
      pick: "PASS",
      confidence: "LOW",
      reason: \`Heavy-tail skew (|mean-median|=\${meanMedianGap.toFixed(2)} > \${maxMeanMedianGap})\`,
    };
  }

  const edge = metrics.pOver - 0.5;
  if (Math.abs(edge) < edgeMin) {
    return {
      pick: "PASS",
      confidence: "LOW",
      reason: \`No edge (P(Over)=\${(metrics.pOver * 100).toFixed(1)}%)\`,
    };
  }

  // Confidence buckets are optional; tune however you like
  const absEdge = Math.abs(edge);
  const confidence =
    absEdge >= 0.10 ? "STRONG" :
    absEdge >= 0.06 ? "HIGH" :
    absEdge >= 0.03 ? "MED" : "LOW";

  return {
    pick: edge > 0 ? "OVER" : "UNDER",
    confidence,
    reason: \`P(Over)=\${(metrics.pOver * 100).toFixed(1)}% vs line \${line}\`,
  };
}

/* -------------------------
   Sanity check: your cluster math MUST match full mean
-------------------------- */
/**
 * expectedByCluster = (1 - p3) * e2 + p3 * e3
 * If that doesn't match metrics.mean, you are mixing different arrays,
 * gating samples, or overwriting expectation with a mode/median.
 */
function validateClusterConsistency(metrics, e2, e3, p3, toleranceGames = 0.5) {
  if (![e2, e3, p3].every(Number.isFinite)) return { ok: true, note: "cluster stats not provided" };
  const expectedByCluster = (1 - p3) * e2 + p3 * e3;
  const diff = Math.abs(metrics.mean - expectedByCluster);

  if (diff > toleranceGames) {
    return {
      ok: false,
      expectedByCluster,
      diff,
      message:
        \`CLUSTER/MEAN MISMATCH: mean=\${metrics.mean.toFixed(2)} but weighted clusters=\${expectedByCluster.toFixed(2)} (diff=\${diff.toFixed(2)}). You are not aggregating from one distribution.\`,
    };
  }
  return { ok: true, expectedByCluster, diff };
}

/* -----------------------------------------------------
   Example wiring (adapt variable names to your script)
--------------------------------------------------------

const line = 21.5;
const totals = totalsMC; // your 12000 match totals

const m = summarizeTotals(totals, line);

// Print these consistently:
console.log(\`[DEBUG TOTALS] Line=\${line} | Min=\${m.min} Max=\${m.max} | P(T>=25)=\${(m.pGe25*100).toFixed(1)}% | P(T>=30)=\${(m.pGe30*100).toFixed(1)}% | P(T>=35)=\${(m.pGe35*100).toFixed(1)}%\`);
console.log(\`[TOTALS FAIR LINE] \${m.fairLine50.toFixed(1)} (50/50 from CDF)\`);
console.log(\`[TOTALS RANGE] P10=\${m.p10.toFixed(0)} | P50=\${m.median.toFixed(0)} | P90=\${m.p90.toFixed(0)}\`);
console.log(\`[TOTALS EXPECTED] Match Expected Games (SIM) = \${m.mean.toFixed(1)}\`);

// Pick:
const pick = getTotalsPick_v2(m, line, { edgeMin: 0.03, maxMeanMedianGap: 3.0 });

// Optional sanity check if you compute clusters:
const chk = validateClusterConsistency(m, e2Mean, e3Mean, p3Set, 0.5);
if (!chk.ok) console.warn(chk.message);

-----------------------------------------------------
WHY YOUR CURRENT OUTPUT IS SELF-CONTRADICTORY
-----------------------------------------------------
In the snippet you pasted, you simultaneously report:
- mean 21.9
- Raw sim 21.1
- BUT "Match Expected Games (SIM) = 20.1"
- AND you anchor to "Cluster projection: 18.0g (mode-pick)"

That can only happen if:
1) You compute mean from one array (e.g. Markov/Dist or pre-filter),
2) then compute "expected games" from a different array (e.g. gated/cluster-only),
3) or you overwrite expected with a mode/median projection when p3Set < gate.

This patch removes that entire class of bugs.

-----------------------------------------------------
NEXT STEP (so I can deliver the *full* rewritten script)
-----------------------------------------------------
Upload Tennis.God.JS (or paste the getTotalsPick + runMatchSimulation_Exact + whatever builds totalsMC).
Then I can:
- apply this refactor in-place,
- run a syntax check,
- scan for undefined variables / dead branches,
- and give you a single downloadable updated script.

`;

fs.writeFileSync('/mnt/data/Tennis.God.JS_totals_refactor_patch.txt', content, 'utf8');
console.log('Wrote /mnt/data/Tennis.God.JS_totals_refactor_patch.txt');

python
from pathlib import Path

content = """Tennis.God.JS - Totals Refactor Patch (drop-in helpers)
=====================================================

You asked for a refactor that keeps BTMS but fixes the "preview stats say long / competitive" vs
"totals engine hammers UNDER" disconnect.

I *can't* rewrite your entire Tennis.God.JS without the actual file contents in this chat.
What I *can* do right now is give you a clean, drop-in totals aggregation + pick engine that
eliminates the exact class of contradictions your output shows:

- mean (21.9) printed but "Match Expected Games (SIM)=20.1"
- "Raw sim 21.1" but totals expectation anchored to 2-set mode/median
- hard p(3 sets) gate blocks Over even when sets can be long

This patch does 3 things:
1) One source of truth: the SAME totals array drives mean/median/CDF/fair line/P(over).
2) Expected games = true mean of the FULL distribution (not mode, not median, not gated clusters).
3) Optional self-check asserts that cluster-weighted expectation matches the full mean.

-----------------------------------------------------
HOW TO APPLY (quick)
-----------------------------------------------------
A) Find your place where you have the final MC totals array (length ~12000):
   e.g. something like totalsMC, totalsArr, totalsGames, etc.

B) Replace your current "Match Expected Games (SIM)" and "TOTALS FAIR LINE (MC CDF)" calculations with:
   const m = summarizeTotals(totalsArray, marketLine);
   Then print from m.mean, m.median, m.fairLine50, m.pOver, etc.

C) Replace getTotalsPick with getTotalsPick_v2 below (or call it and map to your existing output).

D) OPTIONAL but recommended:
   If you already compute 2-set and 3-set cluster means + p3Set, call validateClusterConsistency(...)
   to catch the exact mismatch you're seeing.

-----------------------------------------------------
DROP-IN CODE (copy into your script)
-----------------------------------------------------

/* -------------------------
   Small math helpers
-------------------------- */
function clamp(x, lo, hi) {
  return x < lo ? lo : (x > hi ? hi : x);
}

function quantileSorted(sorted, q) {
  const n = sorted.length;
  if (!n) return NaN;
  const pos = (n - 1) * q;
  const base = Math.floor(pos);
  const rest = pos - base;
  const a = sorted[base];
  const b = sorted[base + 1] ?? a;
  return a + rest * (b - a);
}

/* -------------------------
   Totals summary = SINGLE source of truth
-------------------------- */
/**
 * totals: number[] (each entry = total games in a simulated match)
 * line: number (market total, e.g. 21.5)
 *
 * Notes:
 * - Over is usually graded as T > line (e.g. Over 21.5 means >= 22).
 * - If your book rules differ, adjust the comparator.
 */
function summarizeTotals(totals, line) {
  if (!Array.isArray(totals)) throw new Error("summarizeTotals: totals must be an array");
  const n = totals.length;
  if (!n) throw new Error("summarizeTotals: totals is empty");

  let sum = 0;
  let over = 0;
  let ge25 = 0, ge30 = 0, ge35 = 0;
  let min = Infinity, max = -Infinity;

  for (let i = 0; i < n; i++) {
    const t = totals[i];
    if (!Number.isFinite(t)) continue;
    sum += t;
    if (t > line) over++;
    if (t >= 25) ge25++;
    if (t >= 30) ge30++;
    if (t >= 35) ge35++;
    if (t < min) min = t;
    if (t > max) max = t;
  }

  const sorted = totals.slice().sort((a, b) => a - b);
  const mean = sum / n;
  const median = quantileSorted(sorted, 0.5);

  // "Fair total line (MC CDF)" in your output is the 50th percentile.
  // That is NOT the same thing as expected games; keep both.
  const fairLine50 = median;

  const pOver = over / n;
  const pUnder = 1 - pOver;

  const p10 = quantileSorted(sorted, 0.10);
  const p90 = quantileSorted(sorted, 0.90);

  return {
    n,
    mean,
    median,
    fairLine50,
    pOver,
    pUnder,
    p10,
    p90,
    min,
    max,
    pGe25: ge25 / n,
    pGe30: ge30 / n,
    pGe35: ge35 / n,
  };
}

/* -------------------------
   Replacement pick logic (NO hard p3 gate, NO median anchor)
-------------------------- */
/**
 * opts.edgeMin = minimum probability edge over 50% to bet (e.g. 0.03 = 3% edge)
 * opts.passIfSkewed = if true, pass when mean/median gap indicates heavy-tail risk beyond tolerance
 * opts.maxMeanMedianGap = tolerance for tail skew before passing (default 3.0 games)
 */
function getTotalsPick_v2(metrics, line, opts = {}) {
  const edgeMin = Number.isFinite(opts.edgeMin) ? opts.edgeMin : 0.03;
  const passIfSkewed = opts.passIfSkewed !== false; // default true
  const maxMeanMedianGap = Number.isFinite(opts.maxMeanMedianGap) ? opts.maxMeanMedianGap : 3.0;

  const meanMedianGap = Math.abs(metrics.mean - metrics.median);

  if (passIfSkewed && meanMedianGap > maxMeanMedianGap) {
    return {
      pick: "PASS",
      confidence: "LOW",
      reason: `Heavy-tail skew (|mean-median|=${meanMedianGap.toFixed(2)} > ${maxMeanMedianGap})`,
    };
  }

  const edge = metrics.pOver - 0.5;
  if (Math.abs(edge) < edgeMin) {
    return {
      pick: "PASS",
      confidence: "LOW",
      reason: `No edge (P(Over)=${(metrics.pOver * 100).toFixed(1)}%)`,
    };
  }

  // Confidence buckets are optional; tune however you like
  const absEdge = Math.abs(edge);
  const confidence =
    absEdge >= 0.10 ? "STRONG" :
    absEdge >= 0.06 ? "HIGH" :
    absEdge >= 0.03 ? "MED" : "LOW";

  return {
    pick: edge > 0 ? "OVER" : "UNDER",
    confidence,
    reason: `P(Over)=${(metrics.pOver * 100).toFixed(1)}% vs line ${line}`,
  };
}

/* -------------------------
   Sanity check: your cluster math MUST match full mean
-------------------------- */
/**
 * expectedByCluster = (1 - p3) * e2 + p3 * e3
 * If that doesn't match metrics.mean, you are mixing different arrays,
 * gating samples, or overwriting expectation with a mode/median.
 */
function validateClusterConsistency(metrics, e2, e3, p3, toleranceGames = 0.5) {
  if (![e2, e3, p3].every(Number.isFinite)) return { ok: true, note: "cluster stats not provided" };
  const expectedByCluster = (1 - p3) * e2 + p3 * e3;
  const diff = Math.abs(metrics.mean - expectedByCluster);

  if (diff > toleranceGames) {
    return {
      ok: false,
      expectedByCluster,
      diff,
      message:
        `CLUSTER/MEAN MISMATCH: mean=${metrics.mean.toFixed(2)} but weighted clusters=${expectedByCluster.toFixed(2)} (diff=${diff.toFixed(2)}). You are not aggregating from one distribution.`,
    };
  }
  return { ok: true, expectedByCluster, diff };
}

/* -----------------------------------------------------
   Example wiring (adapt variable names to your script)
--------------------------------------------------------

const line = 21.5;
const totals = totalsMC; // your 12000 match totals

const m = summarizeTotals(totals, line);

// Print these consistently:
console.log(`[DEBUG TOTALS] Line=${line} | Min=${m.min} Max=${m.max} | P(T>=25)=${(m.pGe25*100).toFixed(1)}% | P(T>=30)=${(m.pGe30*100).toFixed(1)}% | P(T>=35)=${(m.pGe35*100).toFixed(1)}%`);
console.log(`[TOTALS FAIR LINE] ${m.fairLine50.toFixed(1)} (50/50 from CDF)`);
console.log(`[TOTALS RANGE] P10=${m.p10.toFixed(0)} | P50=${m.median.toFixed(0)} | P90=${m.p90.toFixed(0)}`);
console.log(`[TOTALS EXPECTED] Match Expected Games (SIM) = ${m.mean.toFixed(1)}`);

// Pick:
const pick = getTotalsPick_v2(m, line, { edgeMin: 0.03, maxMeanMedianGap: 3.0 });

// Optional sanity check if you compute clusters:
const chk = validateClusterConsistency(m, e2Mean, e3Mean, p3Set, 0.5);
if (!chk.ok) console.warn(chk.message);

-----------------------------------------------------
WHY YOUR CURRENT OUTPUT IS SELF-CONTRADICTORY
-----------------------------------------------------
In the snippet you pasted, you simultaneously report:
- mean 21.9
- Raw sim 21.1
- BUT "Match Expected Games (SIM) = 20.1"
- AND you anchor to "Cluster projection: 18.0g (mode-pick)"

That can only happen if:
1) You compute mean from one array (e.g. Markov/Dist or pre-filter),
2) then compute "expected games" from a different array (e.g. gated/cluster-only),
3) or you overwrite expected with a mode/median projection when p3Set < gate.

This patch removes that entire class of bugs.

-----------------------------------------------------
NEXT STEP (so I can deliver the full rewritten script)
-----------------------------------------------------
Upload Tennis.God.JS (or paste the getTotalsPick + runMatchSimulation_Exact + whatever builds totalsMC).
Then I can:
- apply this refactor in-place,
- run a syntax check,
- scan for undefined variables / dead branches,
- and give you a single downloadable updated script.
"""

out_path = Path("/mnt/data/Tennis.God.JS_totals_refactor_patch.txt")
out_path.write_text(content, encoding="utf-8")
str(out_path)

'/mnt/data/Tennis.God.JS_totals_refactor_patch.txt'