TENNISLOCKS v17.10.x – 3-SET SUPPRESSION FIX (SURGICAL PATCH)

============================================================

READ THIS SECTION FIRST – AI EXECUTION INSTRUCTIONS

---

This is NOT a copy-paste patch file. This is a SURGICAL AUDIT INSTRUCTION SET.

Your AI assistant must:

1. LOAD the full script file (Tennislocks v17.10.0 line_mass_refactor_multichain_test_patched_updated)
2. SEARCH for each function/block by exact function name and comment markers
3. LOCATE the specific lines described in each patch instruction
4. VERIFY the current values match the "FIND" pattern before changing
5. APPLY the "REPLACE WITH" values exactly as specified
6. PRESERVE all surrounding code, comments, and whitespace
7. TEST one known 3-set match (Noskova–Gibson or similar) after each pass
8. FIX any undefined variable errors encountered during execution

DO NOT:

• Copy-paste from this instruction file into the script
• Make edits without verifying the current code first
• Skip verification steps
• Change anything not explicitly listed

---

PART A – ROOT CAUSE DIAGNOSIS

---

The 3-set suppression hitting 55–70% favorites is caused by 5 cascading layers:

1. Winner seed layer – Elo weight can stay too high (0.30 max) even when stats disagree, inflating the mismatch classification seed
2. Mismatch mapping – Floors too low (0.50–0.52), ranges too tight (0.33–0.35), so 65% favorites classified as "meaningful mismatch"
3. Set-2 taper – Soft thresholds all at 0.35, maxPen for CH is 0.120 (3x ATP), boost multipliers too high
4. SET_DELTA_MP sweep pressure – Boost gate starts at mismatch 0.15, ramps too fast
5. Totals-only tanking – 2-game trigger fires in totals mode with full mismatch scaling (1.2x multiplier)

Secondary layers (Markov trim, display blend) only need adjustment AFTER fixing the above.

Fix order: Winner seed → Mismatch floors → Taper knobs → Sweep pressure → Tanking → (Pass 2: Markov, if needed)

---

PART B – PASS 1 SURGICAL PATCHES

---

PATCH 1: Reduce Elo Leverage in Winner Seed

LOCATION: winnerEngineLayers() function, Elo weight calculation block

FIND: The section computing wE (Elo weight) with trust-based boosting

AUDIT STEPS:

1. Search for: var wE = THRESHOLDS.BLENDWELO
2. Locate the trust-boosting block: if (trustMinbl2 < blTrustGate)
3. Find the neutrality band compression block: if (Math.abs(pModelA - 0.50) < blNeutBand)

CURRENT BEHAVIOR:

• Default Elo weight can reach 0.30+ via trust boosting
• Even with disagreement, Elo can still dominate at 0.20–0.25
• This inflates pickTaperSeedA() median, triggering over-classification

CHANGES:

Change 1.1 – Default Elo Weight Cap

text
FIND:
var blEloCap = blIsWta ? 0.20 : (blIsCh ? 0.22 : 0.25)

REPLACE WITH:
var blEloCap = blIsWta ? 0.18 : (blIsCh ? 0.19 : 0.22)

Change 1.2 – Disagreement Penalty Enforcement

text
FIND (inside neutrality compression block):
var targetE = Math.min(wE, 0.05)

REPLACE WITH:
var targetE = Math.min(wE, 0.03)

Change 1.3 – Add Hard Elo Cap After All Branches

text
FIND (after all wE calculations, before final weight normalization):
var wSum = wM + wS + wE + wFmp

INSERT BEFORE wSum line:
// Hard Elo cap – prevent trust branch from inflating taper seed
wE = Math.min(wE, 0.22)

EXPECTED RESULT: Elo weight capped at 0.22 max, drops to 0.10 with one disagreement, 0.03 with two+. Reduces false mismatch classification for 64–68% favorites.

---

PATCH 2: Re-Center Mismatch Mapping

LOCATION: mismatchIdxFromSeeds() function, floor/range calculation block

FIND: Lines defining miFloor and miRange for each tour

CURRENT VALUES:

text
var miFloor = miIsWta ? 0.50 : (miIsCh ? 0.51 : (miIsItf ? 0.51 : 0.52))
var miRange = miIsWta ? 0.33 : (miIsCh ? 0.34 : (miIsItf ? 0.35 : 0.34))

PROBLEM: 65% favorite maps to mismatchIdx ≈ 0.44 (above soft threshold 0.35)

REPLACE WITH:

javascript
var miFloor = miIsWta ? 0.56 : (miIsCh ? 0.59 : (miIsItf ? 0.58 : 0.58))
var miRange = miIsWta ? 0.32 : (miIsCh ? 0.31 : (miIsItf ? 0.32 : 0.30))

EXPECTED RESULT:

• 65% favorite → mismatchIdx ≈ 0.28 (below soft threshold)
• 70% favorite → mismatchIdx ≈ 0.47 (moderate)
• 78%+ favorites → Still classified as true mismatches

---

PATCH 3: Soften Set-2 Taper (Multi-Step)

LOCATION: taperParamsForTour() function return statements for each tour

AUDIT: Find the return blocks for ATP, WTA, CH, ITF tours

Change 3.1 – Raise Soft Thresholds

FIND:

javascript
var TAPERSET2SOFTTHRESHOLDATP = 0.35
var TAPERSET2SOFTTHRESHOLDCH = 0.35
var TAPERSET2SOFTTHRESHOLDITF = 0.35
var TAPERSET2SOFTTHRESHOLDWTA = 0.35

REPLACE WITH:

javascript
var TAPERSET2SOFTTHRESHOLDATP = 0.45
var TAPERSET2SOFTTHRESHOLDCH = 0.48
var TAPERSET2SOFTTHRESHOLDITF = 0.46
var TAPERSET2SOFTTHRESHOLDWTA = 0.43

Change 3.2 – Reduce Max Penalties (CRITICAL for CH/ITF)

FIND:

javascript
var TAPERMAXHOLDPENATP = 0.036
var TAPERMAXHOLDPENCH = 0.120
var TAPERMAXHOLDPENITF = 0.054
var TAPERMAXHOLDPENWTA = 0.039

REPLACE WITH:

javascript
var TAPERMAXHOLDPENATP = 0.032
var TAPERMAXHOLDPENCH = 0.075
var TAPERMAXHOLDPENITF = 0.045
var TAPERMAXHOLDPENWTA = 0.035

JUSTIFICATION: CH penalty was 3.3x ATP (0.120 vs 0.036) – this is the single biggest over-suppression source in Challenger matches.

Change 3.3 – Reduce Boost Multipliers

FIND:

javascript
var TAPERMAXPENBOOSTATP = 0.35
var TAPERMAXPENBOOSTCH = 0.70
var TAPERMAXPENBOOSTITF = 0.70
var TAPERMAXPENBOOSTWTA = 0.45

REPLACE WITH:

javascript
var TAPERMAXPENBOOSTATP = 0.25
var TAPERMAXPENBOOSTCH = 0.35
var TAPERMAXPENBOOSTITF = 0.40
var TAPERMAXPENBOOSTWTA = 0.30

EXPECTED RESULT: Set-2 taper now gentle for 55–72% favorites, only aggressive above 75%+. CH/ITF suppression reduced by ~40%.

---

PATCH 4: Delay SET_DELTA_MP Activation

LOCATION: runMatchSimulationExact() function, SET_DELTA_MP calculation block

FIND: The section computing SET_DELTA_MP with mismatch-based increases

CURRENT LOGIC: Boost starts at mismatchIdx >= 0.15

CHANGES:

Change 4.1 – Raise Boost Gate

text
FIND:
if (mismatchIdx > 0.15) {

REPLACE WITH:
if (mismatchIdx > 0.40) {

Change 4.2 – Reduce Boost Slope

text
FIND (inside the boost calculation):
SET_DELTA_MP += ... * (mismatchIdx - 0.15) * ...

REPLACE WITH:
SET_DELTA_MP += ... * (mismatchIdx - 0.40) * 0.65

NOTE: The exact multiplication factor depends on your current code structure. Apply 0.65x scaling to whatever multiplier currently exists.

EXPECTED RESULT: Sweep pressure only activates for 73%+ favorites, ramps slower. Normal 65% favorites get zero sweep bias.

---

PATCH 5: Totals-Only Tanking Suppression

LOCATION: runMatchSimulationExact() function, tanking/blowout logic block

FIND: The section applying tanking penalties when setDiff >= tankTrigger

CHANGES:

Change 5.1 – Add Totals-Only Mismatch Gate

text
FIND:
if (setDiff >= tankTrigger) {

REPLACE WITH:
// Totals-only mode: require much higher mismatch for 2-game trigger
var tankMismatchGate = (opts && opts.totalsOnly) ? 0.70 : 0.50
var allowTanking = (mIdx2 >= tankMismatchGate) || (setDiff >= 3)

if (allowTanking && setDiff >= tankTrigger) {

Change 5.2 – Reduce Mismatch Scaling in Totals Mode

text
FIND (inside tanking penalty calculation):
basePenalty = basePenalty * (1 + 1.2 * mIdx2)

REPLACE WITH:
var tankMismatchMult = (opts && opts.totalsOnly) ? 0.6 : 1.2
basePenalty = basePenalty * (1 + tankMismatchMult * mIdx2)

EXPECTED RESULT: Totals-only sims don't snowball into blowouts unless mismatch is extreme (70%+) or set deficit is 3+ games.

---

PART C – UNDEFINED VARIABLE FIX

---

INSTRUCTION: During execution, if you encounter ReferenceError: [variable] is not defined:

1. Search for the variable name in the entire script
2. Check if it's:
   • Misspelled (e.g., opts.totalsOnly vs totalsOnly)
   • Defined in wrong scope (function vs global)
   • Missing declaration (var, let, or function parameter)
3. Common culprits in this script:
   • opts parameter not passed to nested functions
   • tourLabel vs tourLabelMaybe inconsistency
   • mismatchIdx vs mIdx vs mIdx2 naming variants

FIX PATTERN:

javascript
// If function uses opts.totalsOnly but opts not passed:
FIND:
function myFunction(hA, hB) {
    if (opts.totalsOnly) { ... }

REPLACE WITH:
function myFunction(hA, hB, opts) {
    if (opts && opts.totalsOnly) { ... }

// Then find all callers and add opts parameter

---

PART D – VALIDATION PROTOCOL

---

After each patch pass, run this test:

Test Match: Noskova (67% favorite) vs Gibson, ATP hard court, BO3

Expected Values After Pass 1:

Metric	Before Fix	After Pass 1	After Pass 2 (if needed)
Mismatch Index	0.44	0.33	0.33
Raw MC P3	35%	41%	42%
Markov Adj P3	28%	37%	38%
Display P3	26%	36%	38%
Set-2 Loser Hold Penalty	4.1%	2.8%	2.8%
SET_DELTA_MP Boost	0.008	0	0

Pass/Fail Criteria:

• ✅ PASS if Display P3 is 35–40% for 67% favorite
• ❌ FAIL if Display P3 still below 33%
• ⚠️ CAUTION if Display P3 above 43% (may have over-corrected)

If Pass 1 achieves 35–40% display P3, STOP. Do not proceed to Markov/display patches.

---

PART E – PASS 2 (ONLY IF NEEDED)

---

TRIGGER: If after Pass 1, Display P3 for 67% favorite is still below 33%

PATCH 6: Soften Markov Compression

LOCATION: Lines computing p3AdjMP or p3setAdj via Markov adjustment

FIND: Mismatch compression coefficients

CURRENT:

javascript
var mpMmC = mpTK === 'WTA' ? 0.20 : (mpTK === 'CH' ? 0.18 : 0.16)
var mpClP = mpTK === 'WTA' ? 0.03 : (mpTK === 'CH' ? 0.028 : 0.025)

REPLACE WITH:

javascript
var mpMmC = mpTK === 'WTA' ? 0.12 : (mpTK === 'CH' ? 0.11 : 0.10)
var mpClP = mpTK === 'WTA' ? 0.015 : (mpTK === 'CH' ? 0.012 : 0.012)

Add Conditional Gate:

javascript
// Only compress if mismatch is high OR Markov raw >> MC raw
var shouldCompress = (mmMP > 0.60) || (pDecRawMP > p3setRaw + 0.03)
if (!shouldCompress) {
    p3AdjMP = pDecRawMP  // Skip compression
}

---

PART F – FINAL CHECKLIST

---

Before marking this patch complete:

• All 5 Pass 1 patches applied and verified
• undefined errors fixed
• Test match run successfully
• Display P3 for 67% favorite is 35–40%
• UNDER picks still intact (no side-effect bias flip)
• Script executes without errors
• Logged output includes: mismatchIdx, SET_DELTA_MP, raw pDecider, Set-2 taper penalty, Markov adj p3, display p3

COMPLETION SIGNATURE:

text
=== TENNISLOCKS v17.10.x 3-SET SUPPRESSION FIX APPLIED ===
Date: [INSERT DATE]
Patches Applied: 1.1-1.3, 2, 3.1-3.3, 4.1-4.2, 5.1-5.2
Pass 2 Needed: [YES/NO]
Test Match P3: [RAW%] / [ADJ%] / [DISPLAY%]
Status: [PASS/FAIL/CAUTION]

---

EMERGENCY ROLLBACK INSTRUCTIONS

---

If the fix breaks the script or over-corrects:

1. Restore from backup before patch application
2. Apply only Patches 2 and 3 (mismatch floors + taper)
3. Skip Patches 1, 4, 5 on first attempt
4. Re-test and report which layer still overfires

DO NOT blindly revert all changes – the CH taper penalty (Patch 3.2) is objectively broken at 0.120 and must be fixed regardless.

---

END OF INSTRUCTION FILE

Sources

Update my script then send it as a downloadable txt link

U must read through my script bc u might find other things not logical that needs patched that I didn’t specify

U can see for an example this prior match it was favoring under 2sets bc if the 26% that affects all down stream this match had went to 3 sets

3/12/2026 16:52:13

🎾 TENNISLOCKS PREDICTION ENGINE v17.10.0 🎾
[VERSION] v17.10.0 LINE MASS REFACTOR — SETS/TOTALS COHERENCE 2026-03-12T23:52:00.811Z
Totals sims: 12,000
Prop sims: 0

🎯 WTA 100 (RIGHT) (Outdoor) Best of 3 | Line: 21.5
Generated: 6:52:08 PM

---

Linda noskova vs Talia Gibson

📊 Win Probability:
Linda noskova: 64.4% (engine raw: 61.6%)
Talia Gibson: 35.6% (engine raw: 38.4%)
[MARKET ANCHOR] Market Anchor: 9% applied (Soft Alignment)
[SCHEDULE] Linda noskova: HARD (ranks 6, trust 1.11) | Talia Gibson: SOLID (ranks 6, trust 1.02)
[RANK] Linda noskova: 14 | Talia Gibson: 112
[AVG OPP RANK] Linda noskova: 43 | Talia Gibson: 98
[OPP-ADJ FORM] delta=-0.2 pts | w=8.2%

🎲 Total Games Projection: 22.2 (weighted EV) | sim mean 23.3
Decider Prob: 32% (MID) | adj 27% | raw 42%
Match clusters: 2-set 19.0 med (19.2 mean) | 3-set 29.0 med (28.9 mean)
Cluster EV: 22.2g | E[T|2 sets] 19.2 | E[T|3 sets] 28.9

========================================
🏆 MODEL WINNER RANGE: Linda noskova 64.4% | Talia Gibson 35.6%
⚠️ MODEL LEAN: PASS Talia Gibson — PASS (no betting value at current odds)
Raw win%: Linda noskova 61.6% | Talia Gibson 38.4%
Value Edge: 12.2% (below threshold)
Reasons: DOG_KILL|edge=12.2 p2=68 lane=false sweep=18 trust=6, DOG_RESCUE|edge=12.2 struct=YES strongEdge, bimodal cluster (10g separation), momentum+, layer conflict, cold start
Cluster: DOG_RESCUE

========================================

🎲 Sets Market:

• 2 Sets: 64.2% | Book est -218

• 3 Sets: 35.8% | Book est +162

Sets Bet: BET 2 SETS | Confidence: MID

Over/Under 2.5 Sets:

• Under 2.5: 64.2% | Book est -225

• Over 2.5: 35.8% | Book est +159

O/U Bet: BET UNDER 2.5 | Confidence: MID

Read: Sweep pressure is strong enough to keep the 2-set side in front. (winner policy: volatility/fragility) (p3 sources spread 15pp) | Confidence: MID
P3 sources: disp 31.9% | adj 26.6% | raw 41.8% | mkt 34.6% | hold 39.7% | spread 15.2pp | hook FRAGILE

📊 Player Stats (Live):

• Hist. Surface (HARD): Linda noskova (73.3%) | Talia Gibson (100.0%)
• Linda noskova: Hold 71.8% (raw: 83.9%, adj: Tier Penalty)
• Talia Gibson: Hold 68.6% (raw: 84.2%, adj: Tier Penalty)
• Style: Linda noskova [Spot Server] | Talia Gibson [Spot Server]
• Fuzzy Momentum: Linda noskova 70.8% Positive (Matches: 4W-2L, Straight Sets: 2W-1L, Non-Straight: 2W-1L, Sets: 9W-6L) ; Talia Gibson 100.0% Elite (Matches: 6W-0L, Straight Sets: 3W-0L, Non-Straight: 3W-0L, Sets: 12W-3L)
• 1st Srv Win: Linda noskova 67.7% | Talia Gibson 73.7%
• 2nd Srv Win: Linda noskova 53.2% | Talia Gibson 43.0%
• 1st Srv In: Linda noskova 68.0% | Talia Gibson 65.3%
• SPW: Linda noskova 63.1% | Talia Gibson 63.0%
• Break Rate: Linda noskova 31.4% | Talia Gibson 28.2%
• Dominance Ratio: Linda noskova 1.18 | Talia Gibson 0.91 [MISMATCH]
• Volatility: Linda noskova 9.8% | Talia Gibson 4.8% | Match: 7.3%
• Elo: Linda noskova favored (84.3%)
  Source: Elo_Lookup sheet (Linda=1885, Talia=1673)
• Blend (Linda noskova): MC 62.5% | Stats 67.3% | Elo 84.3%
• Analytical Hold: Linda noskova 78.3% | Talia Gibson 78.2% (Newton-Keller)

########################################
🏆 FINAL PICKS 🏆
########################################

🎾 Linda noskova vs Talia Gibson

Totals Fair Line: 22.5 (50/50 from CDF)
Totals Range: P10=17 | P50=22 | P90=31
Totals Expected: 23.2 (mean) | Median: 22.0
Most common exact: 19g | Highest-density zone: 18-20g 22.4%
Weighted set-center (EV): 22.2g (E[T|2] 19.2 | E[T|3] 28.9 @ p3(display)=32% | raw 42% | adj 27%)
Totals PMF (top exacts): 19g 7.8% | 18g 7.4% | 20g 7.3% | 17g 7.2% | 22g 6.9% | 21g 5.6% | 16g 4.9% | 26g 4.9%
[TARGET] TOTAL GAMES:

Totals regime STABLE (mc chains 3, avg spread 0.00 g)
PASS 21.5 | Confidence: NONE

Why no bet:
The match shape is too balanced at the current number.

Shape:
2-set center 19g | 3-set center 29g | pDecider 32% | Weighted center 22.2g

Pricing:
Model fair 22.5 vs market 21.5 (+1.0g)

Sets fallback:
Under 2.5 sets @ 64.2% | Confidence: MID

Engine Breakdown:

1. Raw MC Over: 54.8% (includes pushes) [lanes: 2set 22.8% | 3set 99.3%]
2. No-push MC: 52.5% Over
3. Governor Over: 47.2% (blend weight: 30%)
4. Owner (pick): n/a
   Markov (diagnostic): 55.3%
   Fair lines - MC: 22.5 | Markov: 22.5 | Gap: 2.8pts, 0.00g
   P3 authority: used 38.8% support 2 spread 15.2pp laneGap 76.4pp FRAGILE thinSupport
   Half-line 21.5 - MC: Under 45.2% | Over 54.8%
   Half-line 21.5 - Markov: Under 44.7% | Over 55.3%
   Tail - P(>=30): 17.8% | P(>=31): 13.4% | P(>=35): 2.7%
   Buckets: leftTail(L-6) 5.1% | localMass low 15.1% high 7.3%
   Near-line mass @21.5: 12.4% (P(21)+P(22)) | Local 20-22: 19.7% | 19-23: 32.1%
   Adjacent exacts: P(21g) 5.6% | P(22g) 6.9%
   Markov 3-set adj: 35.2% (penalty 0.0%)

📊 UNDERDOG ASSESSMENT:
DOG LEAN: Talia Gibson @ 38.4% (+12.2% edge)
Pricing edge exists, but main winner governance did not fully confirm.

Also there’s an undefined issue from the totals gusto gram