Finally fixed b03 I rebuilt the BO3 section and w...

作成日: 2026年9月15日

質問

TennisLocks_FINAL_v1158_BO3_CROSS_SET_RESPONSE_FIX_20260915.txt

Finally fixed b03

I rebuilt the BO3 section and wired the structural correction ahead of the recent-form blend. The original file explicitly had no systemic P3 correction, while the dead _structuralNoPopulationSetTiltV1105 flag was still being passed into production roots. (TennisLocks_FINAL_v1157_BO3_CONDITIONAL_COVERAGE_FIX_20260915.txt) (TennisLocks_FINAL_v1157_BO3_CONDITIONAL_COVERAGE_FIX_20260915.txt)

The v1158 fix now does four important things:

No corpus P3 target and no fixed P3 anchor/cap. The new structural correction is derived from the point model’s existing opposing-relative predictive-state dispersion, which is point-outcome evidence rather than set-count corpus data.
Structural qA/qB are corrected first. Raw P(steal one set | lose) from the fixed-SPW tree gets the cross-set response adjustment before _tlBo3BlendCoverageV1155_ sees it.
Recent form is optional instead of a gatekeeper. Missing/invalid recent losses no longer throw away the systemic structural correction.
The dead flag is physically removed. _structuralNoPopulationSetTiltV1105 now occurs zero times in the rebuilt script.

I also kept Match Winner immutable and retained the exact-score/Fréchet geometry. Syntax validation passes, and a 5,000-case randomized identity test had maximum numerical error of 2.22e-16. In synthetic checks, raw P3 35.0% became 45.5% from the structural response alone; a raw 42.0% case reached 52.9% naturally—there is no set-count target forcing it there.

Download the corrected v1158 full script

Download the v1157 → v1158 diff

I want to continue correcting

I want to make sure that now it can call under 2.5 on actual matches this finally got p3 to rise up I also want u to look for more issues both ways how p3 can rise and p2 can rise so its not doing it on wrong matches do not hardcode anything this is very deep must do research on the web to further correct and replace the script replace means replace the deleted old parts and wire correctly

Over 2.5 worked here proof on this match we didn’t hardcode either

════════════════════════════════════════
🎾 TENNISLOCKS 🔒
OFFICIAL MATCH MODEL
VERSION 3.0
GENERATED 12:34 AM | September 15, 2026
ENGINE Point • Game • Set Probability Model
════════════════════════════════════════

🎯 WTA 500 (OUTDOOR) | Best of 3 | Line: 20.5
Tour: WTA | Court speed (CPI): 38
Metadata confidence: HIGH

────────────────────────────────────────
Cristina Bucsa vs Panna Udvardy
────────────────────────────────────────

────────────────────────────────────────
💰 MODEL PICKS:

TOP [TOTAL GAMES] OVER 20.5: model pick | settlement 72.6% | status OFFICIAL BET
#2 [PROP] Panna Udvardy win 1+ set: probability 73.7% | model odds -280 | status OFFICIAL BET

📊 LEANS:

Sets 2.5: OVER 56.7% | MEDIUM confidence

🚫 NO BETS:

Match Winner: NO BET | forecast Cristina Bucsa 59.3% | forecast side retained, but betting status is below OFFICIAL BET
────────────────────────────────────────

Match type: Mixed serve and return, close matchup. (MIXED_EVEN)
Risk: 0.00 (LOW)
Pricing data quality: STRONG | opponent-rank samples 7/7 | trust -

PLAYER INTEL
┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄
Cristina Bucsa Panna Udvardy
Rank 44 83
Elo 1723 1685
Avg Opp Rank 75 85
Schedule A: SOLID (trust -, ranks 0) | B: SOLID (trust -, ranks 0)
Serve Style ace 1.6% ace 8.2%
Momentum RECENT_RESULTS RECENT_RESULTS
Hold % 70.3% 66.3%
Recent-row SPW (raw) 54.3% 58.8%
Dominance Ratio 0.74 0.77
Recent Hold SD 25.4% 18.1%
Break Rate 33.7% 29.7%
1st Srv Win % 58.2% 70.5%
2nd Srv Win % 45.6% 45.3%
1st Srv In % 63.0% 55.3%
Recent-row implied hold60.6% (54.3% SPW) 71.1% (58.8% SPW)

════════════════════════════════════════

🎲 SETS OUTLOOK
[SET RESEARCH REF] CANONICAL_POINT_ROOT | WTA/HARD/MAIN/CANONICAL_POINT_STATE_SET_COUNTS_V1144 | read-only, no live blend
[SET INPUTS] SPW A/B 58.5% / 56.7% | Hold A/B 70.3% / 66.3% | route UNIFIED_CURRENT_POINT_ROOT_V1113
[SET TB CAL] not applied | tree P(7-6) 15.0% | raw 15.0% | hist not measured | n null | CANONICAL_POINT_ROOT_NO_HISTORICAL_SET_TB_MUTATOR_V1144 | set-winner margin preserved by construction
[SET AUTHORITY] ACTIVE | BO3_POINT_STATE_RESPONSE_PLUS_LOSS_CONDITIONAL_COVERAGE_V1158 | BO3 length priced from player 1+ set coverage and reconciled to Match Winner
[BO3 COVERAGE MODEL] winner anchored | raw structural q -> point-state cross-set response -> optional recent-loss shrinkage | P3 = P(B wins)*qA + P(A wins)*qB | no raw-coverage blend | no corpus/population P3 target | no fixed P3 cap
[BO3 COVERAGE EFFECT] canonical P3 49.2% | response prior P3 60.6% | final P3 56.7% | response +11.4pp | recent -3.9pp
[BO3 PLAYER COVERAGE] A wins 1+ set 83.1% | B wins 1+ set 73.7% | identity 56.7%
[BO3 CONDITIONAL q] A raw/response/recent/final 52.9% / 64.2% / 0.0% / 58.3% || B raw/response/recent/final 46.7% / 58.2% / 40.0% / 55.6%
[SET LENGTH ROOT] final Sets Won / Both Win a Set / Over 2.5 identity P3 56.7% | one exact-score PMF
[SET WINNER ALIGN] final winner error 0.0e+0 | final set-count margin error 0.0e+0
[SET EXACT PMF] 2-0 26.3% | 2-1 33.0% | 0-2 16.9% | 1-2 23.7% | final P3 56.7%
[SET ACTION] LEAN OVER 2.5 | probability 56.7% | model fair odds -131 | MEDIUM | forecast only
[SET BETTING GATE] final exact-score PMF direction always visible | HIGH >= 60.0% = official PICK | MID 55.0%-<60.0% = LEAN | LOW >50.0%-<55.0% = forecast only | no BO3 data-quality confidence cap
[SET FAIR PRICE] Over 2.5 -131 | Under 2.5 +131
[SET TREE DIAGNOSTIC] canonical P(2) 50.8% | canonical P(3) 49.2% | canonical point/game/set tree

📊 Player Stats (Current Live-Source Audit):

Serve/return diagnostic: ret2 A/B 54.9% / 52.9% | BP save A/B 38.8% / 59.6%
Visible target-surface row coverage: Cristina Bucsa through 2026-09-13 [UNVERIFIED] | Panna Udvardy through 2026-09-13 [UNVERIFIED] | CURRENT POINT INPUTS ELIGIBLE
Live row sources: Cristina Bucsa [CURRENT_EXACT_TML_V939 x7] | Panna Udvardy [CURRENT_EXACT_TML_V939 x7] | date precision A/B TOURNEY_START_DATE x7 / TOURNEY_START_DATE x7
Surface SPW reference (HARD): Cristina Bucsa (No verified same-tour surface SPW rate) | Panna Udvardy (No verified same-tour surface SPW rate) [TA_SURFACE_SPW_RATE_UNAVAILABLE]
Surface serve priors: Cristina Bucsa Ace 2.2% / DF 4.2% / 1stIn 63.7% | Panna Udvardy Ace 5.7% / DF 3.9% / 1stIn 55.2% [AUTOFILL_CURRENT_EXACT_THIS_TOUR_364D_PERSISTED_V1076]
Cristina Bucsa: Hold 70.3% (raw: 61.6%, serve vs this returner) [hold seed]
Panna Udvardy: Hold 66.3% (raw: 72.2%, serve vs this returner) [hold seed]
Style: Cristina Bucsa [ace 1.6% / ace 1.6%] | Panna Udvardy [ace 8.2% / ace 8.2%]
Recent current-source results (audit): Cristina Bucsa W-L 4-3, SS 2-3, Sets 8-8 ; Panna Udvardy W-L 2-5, SS 1-3, Sets 6-11
1st Srv Win: Cristina Bucsa 58.2% | Panna Udvardy 70.5%
2nd Srv Win: Cristina Bucsa 45.6% | Panna Udvardy 45.3%
1st Srv In: Cristina Bucsa 63.0% | Panna Udvardy 55.3%
Raw recent-row SPW: Cristina Bucsa 54.3% | Panna Udvardy 58.8% [diagnostic row aggregate; official pricing uses the exact-point posterior root]
Break Rate (from hold): Cristina Bucsa 33.7% | Panna Udvardy 29.7%
Dominance Ratio: Cristina Bucsa 0.74 | Panna Udvardy 0.77
Recent Hold SD: Cristina Bucsa 25.4% | Panna Udvardy 18.1% | Match: 21.7%
Elo (diagnostic only; not official serve authority): Cristina Bucsa 55.4%
Source: Elo_Lookup sheet (Cristina Bucsa=1723, Panna Udvardy=1685)
Serve vs this returner (Cristina Bucsa): 59.3% | Elo 55.4% (calibrates official serve when induce fires)
Recent-row implied hold (diagnostic): Cristina Bucsa 60.6% (SPW 54.3%) | Panna Udvardy 71.1% (SPW 58.8%) (small sample)

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

🎯 TOTAL GAMES
[OFFICIAL TOTAL GAMES DECISION] PICK OVER 20.5 | 72.6% | OFFICIAL BET
Final pricing direction: OVER 72.6% from the official cumulative full-match Total Games threshold probability.
Pricing method: all legal full-match score paths are summed against your Total Games line. No single exact score controls the pick.
At 20.5: Over 72.6% | Under 27.4%
Total Games probability authority: ONE canonical joint score+games PMF | no second threshold recalibration is applied after the current length root.
Set-count decomposition at 20.5:
2-set lane: 43.3% match mass | P(Over | 2 sets) 36.8% | contributes 16.0pp raw Over mass
3-set lane: 56.7% match mass | P(Over | 3 sets) 99.9% | contributes 56.7pp raw Over mass
Combined no-push P(Over 20.5) = 72.6% from all lanes.
First-server sensitivity (diagnostic only): A serves first -> Over 72.6% | B serves first -> Over 72.6% | mean-total gap 0.02g
Projected total-games distribution: fair line 25.5 | mean 25.2 | median 26 | largest single exact bucket 29g (6.4%, not a majority and not the O/U authority)
Exact-total concentration: dominant 3-game cluster 28-30g = 18.8% | cluster side OVER at 20.5
OVER threshold mass is spread across 19 exact totals | strongest OVER exact 29g = 6.4% unconditional / 8.8% of the OVER side | effective support 21.7 totals.
Unconditional pricing distribution: 80% range 17-32 | SD 5.7 | mode 29g (6.4%) | leaders 29g 6.4% | 30g 6.2% | 28g 6.1% | 19g 6.1% | 31g 5.9%
########################################
🎯 PROP PROJECTIONS 🎯
########################################

📊 Cristina Bucsa - Player Props:
Games Won: mean 13.1 | median 13 | mode 12 | full-match distribution
1st Set Games Won: 5.07 projected
Sets Won: LEAN 2+ SETS | 59.3% | MEDIUM
Serve Games: not requested | enter a service prop line to price
Serve Points Played: not requested | enter a service prop line to price
Serve Points Won: not requested | enter a Serve Points Won line to price
Aces: not requested | enter a Aces line to price
Double Faults: not requested | enter a Double Faults line to price
Breaks Won: not requested | enter a Breaks Won line to price
Break Points Created: not requested | enter a Break Points line to price
BP Conversion: not requested | enter a Break Points line to price
Opp BP Save: not requested | enter a Break Points line to price
Opponent Matchup: opp return 32% | context only; official pricing uses the final match tree
Historical sample: 12.3 service games | projected Games Won CV: 26%

📊 Panna Udvardy - Player Props:
Games Won: mean 12.1 | median 13 | mode 12 | full-match distribution
1st Set Games Won: 4.75 projected
Sets Won: BET 1+ SET | 73.7% | HIGH
Serve Games: not requested | enter a service prop line to price
Serve Points Played: not requested | enter a service prop line to price
Serve Points Won: not requested | enter a Serve Points Won line to price
Aces: not requested | enter a Aces line to price
Double Faults: not requested | enter a Double Faults line to price
Breaks Won: not requested | enter a Breaks Won line to price
Break Points Created: not requested | enter a Break Points line to price
BP Conversion: not requested | enter a Break Points line to price
Opp BP Save: not requested | enter a Break Points line to price
Opponent Matchup: opp return 34% | context only; official pricing uses the final match tree
Historical sample: 12.5 service games | projected Games Won CV: 32%

🎲 Match-Level Context:
Sets Played: LEAN OVER 2.5 | P(2 sets) 43.3% / P(3 sets) 56.7% | expected 2.57 sets
3-set match / both players win a set: YES 56.7% | NO 43.3% | no betting action
Exact match-score paths: Cristina Bucsa 2-0 26.3% | Cristina Bucsa 2-1 33.0% | Panna Udvardy 2-0 16.9% | Panna Udvardy 2-1 23.7%
If forecast winner Cristina Bucsa wins: straight sets 44.4% | Panna Udvardy steals one set 55.6% conditional (33.0% of all match paths)
Individual win 1+ set: Cristina Bucsa 83.1% | Panna Udvardy 73.7% | marginal coverage only; NOT the BO3 3-set probability; includes paths where that player wins the match
Straight-set loss chance: Cristina Bucsa 16.9% | Panna Udvardy 26.3%
Games/aces/breaks use the full match tree, not a single 2-0 or 2-1 scoreline.
Surface: HARD | Tour ace reference 4.4% | double-fault reference 5.1%

Resource uri: /response/turn2 Showing 266 of 266 lines.

undefined

思考

undefined

思考

undefined

思考

Created tennis model validation report and audited version references and syntax

Worked for 12m 36s

回答

I replaced v1158 with a new v1159 directed set-transition build. The main BO3 problem you were worried about was real: in v1158, whenever the structural response fired, both conditional steal probabilities were transformed with logit(q) + shift. That means the structural stage itself was inherently one-directional—it could systematically increase P3, while P2 pressure had to come later from recent-form evidence. (TennisLocks_FINAL_v1158_BO3_CROSS_SET_RESPONSE_FIX_20260915.txt) The coverage target then used that upward-shifted structural prior before the recent evidence was blended. (TennisLocks_FINAL_v1158_BO3_CROSS_SET_RESPONSE_FIX_20260915.txt)

What v1159 changes

For BO3, I removed that entire universal cross-set-response mechanism. The model now treats match length according to the actual causal event:

Set 3 exists if and only if Set 2 reverses the winner of Set 1.

So the model now explicitly prices the two relevant scoreboard states: after A wins Set 1, how likely is B to reverse in Set 2; and after B wins Set 1, how likely is A to reverse in Set 2. Only S1→S2 ordered transitions can alter BO3 length. S2→S3 history has zero length ownership because at that point Set 3 already exists.

That is more consistent with the research. The classical point model is an iid benchmark and even Newton & Keller note evidence for non-iid tennis effects. (Wiley Online Library) More importantly, the 66,262-match BO3 study by Depken, Gandar, and Shapiro finds meaningful strategic and psychological set-level effects after the opening set, so there is no empirical justification for universally interpreting uncertainty as “the loser responds more.” (Sage Journals) Ingram's point-based Bayesian model also treats player serve/return skill as time-varying, using a Gaussian random walk, which supports giving matchup-local/current evidence precedence over stale historical transitions. (De Gruyter Brill)

The resulting BO3 adjustment is now genuinely signed. Reversal-heavy evidence increases P3/Over 2.5. Persistence-heavy evidence increases P2/Under 2.5. Missing ordered evidence leaves the clean structural point-state tree intact instead of inventing a direction. There is still no corpus P3 target, floor, fixed anchor, cap, or iid set-rate bridge.

For BO5, I kept it structurally separate rather than copying BO3 logic. BO5 length uses the directed states S1→S2, S2→S3, and S3→S4. S4→S5 remains excluded from length ownership because once the score reaches 2-2, Set 5 already exists. That distinction is also supported by BO5 research showing that later-set behavior departs from the opening-set theoretical benchmark. (Taylor & Francis Online) I also removed the fixed priorN=12 pseudo-count and the requirement that both player profiles must be present; one verified profile can contribute, with mirrored duplicate matches de-duplicated.

Validation

Test	Result
JavaScript syntax	PASS
Random BO3 reconciliation cases	5,000 PASS
Max Match Winner error	`2.22e-16`
Max set-count margin error	`2.22e-16`
Same structural P3 = 35%, reversal-heavy evidence	P3 → 63.9%
Same structural P3 = 35%, persistence-heavy evidence	P3 → 19.4% / P2 → 80.6%
BO3 S2→S3 length observations used	0
BO5 S4→S5 length observations used	0
Old v1158 response-authority references	0
`_structuralNoPopulationSetTiltV1105` references	0
BO5 fixed `priorN=12`	0
BO5 both-profile-required gate	0

I also traced the production wiring through the final root. BO3 and BO5 ordered profiles are supplied to the official exact-score length root; after reconciliation, Sets Won, Sets Played, Total Games, Player Games, and the downstream game/service props consume the same final PMF. So when v1159 legitimately raises P2, that Under pressure also propagates into Total Games instead of being overwritten by another length model.

One important point about your Bucsa–Udvardy example: I did not force v1159 to preserve the 56.7% Over result. The card you posted contains aggregate recent records but not the actual ordered S1→S2 histories that the new authority deliberately needs. Fabricating those would defeat the entire correction. On the real sheet, v1159 will read the ordered Match Inputs and decide whether that 49.2% canonical P3 should move upward like v1158 did—or downward if the players' actual state evidence indicates persistence. That is exactly what you wanted: Over and Under must both be earned by the match rather than by a built-in direction.

回答

😂 Yes — v1159 overcorrected it.

The mistake was replacing the newly fixed v1158 BO3 baseline instead of building safeguards around it. v1158 establishes the response-corrected structural qA/qB first and only then lets recent evidence shrink around that prior. (TennisLocks_FINAL_v1158_BO3_CROSS_SET_RESPONSE_FIX_20260915.txt) v1159 bypassed that lift, which is why P3 could collapse again.

I fixed that in v1160:

BO3 is restored exactly to v1158. I compared the complete BO3 pricing block byte-for-byte; the SHA-256 hashes are identical.
The v1159 S1→S2 replacement has zero live BO3 references.
Your structural response that finally lifted P3 is back.
Recent-loss evidence remains optional and can moderate the lift, exactly as v1158 did.
Match Winner remains immutable.
No P3 corpus target, fixed anchor, cap, or floor.
_structuralNoPopulationSetTiltV1105 remains 0 occurrences.
Syntax passes.

And importantly, I’m not touching BO3 again with another unvalidated transition theory. Research shows set dependence exists, but it can include both strategic persistence and reversal effects; that supports studying a bidirectional modifier, not replacing the working root. (IDEAS/RePEc)

BO5 stays isolated because it genuinely has different 3/4/5-set state geometry.

Download v1160 full replacement script

Download v1158 → v1160 diff

Download rollback validation

Do not use v1159 for BO3. v1160 is the continuation of the BO3 fix you actually wanted.

質問

So research correctly don’t over corrected u should be knowing what adding this or removing that does to the script that’s how yk if it works or worsens

思考

Worked for 4s

回答

Exactly. The right standard is not “this idea sounds more sophisticated.” It is: what does this exact change do to the existing probability system, and does it improve the behavior we were fixing without damaging behavior that was already correct?

For this script, every future BO3/BO5 change should be treated like a controlled model experiment:

Identify the current behavior first. Record canonical P2/P3, qA/qB, Match Winner, exact-score PMF, Total Games, Sets Won, and downstream props before touching anything.
Predict the direction of the code change before implementing it. If I add a positive logit shift to both q values, I should explicitly state beforehand: “this structurally raises P3.” If I replace that with persistence transitions, I should state: “this can materially lower P3.” If I cannot explain the expected effect, the change should not go live.
Change one owner at a time. No replacing the BO3 root while simultaneously changing evidence weighting, history, and reconciliation. Otherwise we cannot know what helped or hurt.
Run ablations. Baseline → change A only → change B only → A+B. That tells us which piece caused the movement.
Test both directions. High-P3 matches must still rise. Strong sweep/persistence matches must be capable of raising P2. Balanced matches should stay near baseline instead of being pushed artificially.
Test invariants after every modification. Match Winner unchanged, PMF sums to 1, exact-score identities hold, no negative probabilities, no duplicate owner, Total Games uses the same final PMF.
Compare distributions, not one showcase match. A fix is bad if it improves one example while systematically shifting hundreds of normal matches the wrong way.
Only promote a change when it beats the existing version. If research suggests something interesting but the implementation worsens calibration or directional behavior, it stays research-only.

Worked for 13m 52s

回答

Done. I researched it and rebuilt this as v1163, without undoing the v1158 P3 correction.

The important finding was that v1158’s problem was not that its P3 lift needed replacing. The missing piece was on the recent conditional evidence side. v1158 estimates qA from A’s losses and qB from B’s losses only. But once Match Winner is fixed, there is another independent-looking sample of the same conditional exact-score event that does not require changing Match Winner. The original v1158 structure already makes those q values the straight-set-vs-three-set split inside each fixed winner row. (TennisLocks_FINAL_v1158_BO3_CROSS_SET_RESPONSE_FIX_20260915.txt)

So v1163 now works like this:

qA = P(A steals a set | B wins) uses A's losses (1-2 vs 0-2) + B's wins (2-1 vs 2-0).
qB = P(B steals a set | A wins) uses B's losses + A's wins.
Sweep-heavy evidence therefore pushes P2 upward: favorite often wins 2-0 + underdog often loses 0-2.
Competitive evidence pushes P3 upward: winner often needs 2-1 + loser often loses 1-2.
The v1158 structural response, structural weighting, Match Winner anchor, Fréchet geometry, and exact-score reconstruction are unchanged.
If both Match Inputs contain the same recent H2H, v1163 de-duplicates it when the row fingerprints are available.
No population P3 target, fixed P3 cap/floor, corpus anchor, fixed pseudo-count, or raw win-rate blend was added.

This is much closer to what the research supports. A 66,262-match BO3 study finds both strategic and psychological set-level effects, so the evidence does not justify forcing every match toward either comebacks or sweeps. (Sage Journals) Ingram's point-based work models player strength as changing over time, supporting your matchup-specific point root remaining the structural authority rather than replacing it with historical set frequencies. (IDEAS/RePEc)

I also rejected two ideas from the experimental v1162 work instead of blindly promoting them. I did not replace your existing structural weight with a beta “pseudo-sample size,” and I did not let official opponent rank directionally mutate q. Gollub's tennis work supports opponent adjustment using serve/return ability, while research comparing rating systems finds official rankings relatively poor compared with Elo-based measures. Your point model already has opponent-specific serve/return information, so adding a rank-driven direction on top risked double-counting opponent strength. (Sage Journals)

The Bucsa–Udvardy regression check is important. From your displayed v1158 diagnostics, the structural weights infer to about 29.64 and 30.00. Your recent records imply v1163 paired evidence of qA = 1/5 = 20.0% and qB = 4/9 = 44.4%. With the exact same v1158 structural response, that gives an estimated P3 ≈ 56.16% versus your v1158 56.7%—only about -0.54 percentage points. So this does not destroy the match that proved your P3 fix.

At the other extreme, with a response prior q=58% and the same structural weight of 30, a genuinely sweep-heavy 0/12 conditional sample moves q to 41.43%, while a three-set-heavy 12/12 sample moves it to 70.00%. That is the behavior you were asking for: the match evidence decides which way it goes.

Validation passed: JavaScript syntax passes; 5,000 randomized Winner/PMF/conditional-identity tests had max numerical error 2.22e-16; the old dead structural flag remains at zero; there are zero new P3 targets/caps/anchors. The five core v1158 BO3 functions—including the response calibration and structural weighting—were verified byte-for-byte unchanged. BO5's live state-duration functions are also byte-for-byte unchanged from v1160. That separation is intentional: BO5 research shows later-set reversal and fatigue/tactical variation, so it should remain a scoreboard-state model rather than inherit this BO3 estimator. (Taylor & Francis Online)

Download the full v1163 replacement script

Download v1160 → v1163 diff

Download v1158 → v1163 full diff

回答

Finished the Match Winner audit and replacement as v1164.

The research supported a minimal correction rather than changing your winner probability root. Point-based serve/return models benefit from opponent adjustment, surface, time-varying form, and schedule strength; surface Elo adds useful independent signal, while official rankings are generally weaker. (IDEAS/RePEc) Tennis markets also show favorite–longshot bias, so I deliberately did not let betting odds decide who the model should pick. (IDEAS/RePEc)

The real script issue was that v1163 already calculated winner uncertainty—pWinALow/pWinAHigh, whether the posterior crossed 50%, alternate profile-window winner probabilities, and whether those windows crossed 50%—but _winnerDecision_ ignored all of it.

v1164 behavior

Central Match Winner probability is unchanged.
Winner side is never flipped by Elo.
No Elo blend weight.
No rank blend.
No market-odds blend.
Fresh exact-surface Elo is only an independent challenger when the point-root winner is internally unstable.
A stable point-root underdog remains eligible even when Elo favors the other player.
An unstable underdog/favorite forecast remains visible, but it cannot become an OFFICIAL BET unless fresh surface Elo confirms the same side.
BO3 P2/P3 code was untouched.
BO5 length code was untouched.

I specifically tested the underdog cases:

Situation	v1164
Stable point-root underdog, Elo favors favorite	Underdog survives
Unstable underdog, Elo favors favorite	Forecast underdog retained, official bet blocked
Unstable underdog, Elo also favors underdog	Underdog allowed
Stable favorite, Elo disagrees	Favorite survives
Exact profile unavailable	Existing fallback behavior unchanged

Validation:

JavaScript syntax: PASS
5,000 randomized winner tests: PASS
Probability mutations introduced: 0
Winner-side flips introduced: 0
New winner-stability helper: 1 definition / 1 live caller
New surface-Elo challenger: 1 definition / 1 live caller
Core point predictor: byte-for-byte unchanged
Winner probability owner: byte-for-byte unchanged
Posterior winner integrator: byte-for-byte unchanged
Exact winner profile model: byte-for-byte unchanged
All five main BO3 coverage functions: byte-for-byte unchanged

That means I did not repeat what happened with P3: the working model was preserved, and the correction was added only where the audit showed missing wiring.

Download the full v1164 replacement script

Download the v1163 → v1164 diff

Download the Match Winner research + validation report

質問

思考

思考

思考

思考

思考

思考

思考

思考

思考

思考

思考

思考

思考

回答

What v1159 changes

Validation

質問

思考

思考

思考

思考

回答

質問

思考

回答

質問

思考

思考

思考

思考

思考

思考

思考

思考

思考

思考

思考

思考

思考

思考

思考

思考

思考

思考

思考

回答

質問

思考

思考

思考

思考

思考

思考

回答

v1164 behavior

このQ&Aを共有