Our Locks Went a Perfect 0-for-Wrong — So Why Did 55 High-Confidence Picks Miss This Week?

By A. | SaltyTrack Team — July 2026

Over the last seven days, SaltyTrack missed on 55 high-confidence calls — and zero lock-tier calls.^[1] That is the whole story in one sentence, so let us say the good part loud: our single highest-confidence tier, the LOCK, went a perfect 0-for-wrong. Not one lock busted. Every miss lived one rung down, in the “high” bucket — and 55 misses across a week’s worth of high-confidence picks is roughly 8%, which is exactly, precisely, textbook what a 90%-plus model is supposed to produce. This is not a bad week. This is a Tuesday. :)

Now, we love us some transparency around here, so we are not going to bury the number or spin it — we are going to explain it. “55 wrong predictions” sounds like a five-alarm fire until you see the denominator, and the denominator changes everything.

The Good News First: The Locks Held the Line

SaltyTrack sorts every call it makes into confidence tiers. LOCK is the top of the mountain — “bet your mama’s Salty Bucks” confident. HIGH is the next rung down: a strong favorite we like a lot, but not a near-certainty.

Here is what those two tiers did this week:

Tier	What it means	Wrong this week
Lock	Highest confidence. Near-certainty.	0^[1]
High	Strong favorite, not a lock.	55^[1]

Every single miss was a HIGH, not a LOCK.^[1] That distinction matters way more than the raw 55, because LOCK is the tier people actually go all-in on — and it held. When we stamp a fight as a lock, the implicit promise is “we are very sure,” and this week the model kept that promise 100% of the time. For the flip side of that story — the rare week our locks actually did crack — we wrote a whole honest post-mortem in When Our Lock Predictions Missed.

Wait — 55 Misses. Is the Model Broken?

Short answer: no. Longer answer: let’s do the math, honestly.

Here’s the honesty part first. We did not capture the exact denominator for this specific seven-day window — the total number of high-confidence calls we made — so we are not going to invent a shiny accuracy percentage for this week. Making up a number would be the exact opposite of the point.

What we can do is scale the 55 against a recent, comparable week. From June 25 to July 2, 2026, SaltyTrack’s high-confidence accuracy was 91.7% — 618 correct out of 674 calls — and lock accuracy was a perfect 100% (1 for 1).^[2]

Window (June 25 – July 2)	Correct	Total	Accuracy
High-confidence calls	618	674	91.7%^[2]
Lock calls	1	1	100%^[2]

See it now? At roughly 600 to 700 high-confidence calls a week, 55 misses is about 8% — a hit rate somewhere in the low 90s.^[1][2] Fifty-five sounds enormous in isolation. Sitting on top of a stack of 600-plus correct calls, it is the completely ordinary tax a 90%-plus model pays every week. A model that never missed a high-confidence pick would not be confident — it would be lying to you.

Where Do the Upsets Actually Live?

Glad you asked, because upsets are not random static — they cluster. When we pulled a handful of recent high-confidence misses (these are examples from July 1 and 2, not the specific 55 above), a pattern jumped right out of the ring: the chaos lives in S-tier.^[3]

We favored	Our confidence	Lost to	Winner’s tier
Akane inuwaka ex	~83%	Chaos hiryu	S
Thundermiss.shermie	~86%	Real mikudayo-	S
Saber_alter	~82%	Dragon claw EX3	S
Ikusan	~84%	Es	S
Angelg	~92%	Ultimate aizen	A
Cecil_damon	~83%	Straight cougar	A

Four of those six upsets were pulled off by S-tier fighters.^[3] That is not a coincidence, and it is the single most useful nugget in this whole article: S-tier is where SaltyBet’s chaos concentrates. The top of the roster is stuffed with absurd, glitchy, one-punch-kill joke characters that can flatten a “cleaner” favorite before the round clock finishes its first tick. We built an entire metric around measuring exactly this kind of unpredictability — go see SaltyBet’s Chaos Score for how we grade the wildest days on the platform.

(One caveat, stated plainly: the betting-odds field came back empty on this pull, so we are deliberately not quoting any payout numbers here. No made-up odds on our watch.)

Why Does a 90%-Plus Model Still Miss?

Because SaltyBet is not chess. It is two AI-piloted MUGEN characters throwing hands with real, structural randomness baked in. A few honest reasons even a great model eats the occasional upset:

MUGEN AI variance. The characters are computer-controlled, and their decision-making has genuine run-to-run randomness. The “better” fighter does not always execute its best game.
Exhibition-mode randomness. Off-hours exhibition matches lean on weirder, wackier characters and modes where anything goes. More coinflips, more chaos.
Thin-sample fighters. When a fighter has only a handful of matches on record, our read on it is a best guess, not a verdict. New and rare fighters carry more uncertainty by definition.
S-tier chaos. As the table above shows, the top tier is home to the one-hit-kill lunatics. A 92% favorite can still walk into an instant-death setup and lose in two seconds flat.

None of that means the fights are fixed, by the way — upsets are the system working as designed, not evidence of a rig. We dug into that exact question in Is SaltyBet Rigged?, and the short version is: variance is not conspiracy.

So What Should You Do With a “High” Call?

Treat it like what it is: a strong lean, not a guarantee. LOCK is the tier to trust with your whole (virtual) wallet; HIGH is a favorite we like but would not mortgage the house on. And all of it — every tier, every number — is for funzies only. SaltyBet runs on pretend Salty Bucks, nobody is getting rich, and our predictions are a fun edge, not financial advice. New around here? Start with the Beginner’s Guide to SaltyBet.

FAQ

Did SaltyTrack’s lock predictions miss this week?
No. Over the last seven days, zero lock-tier predictions were wrong. All 55 of the week’s high-confidence misses came from the second-highest “high” tier, not the top lock tier.^[1]

Is 55 wrong predictions a bad week for SaltyTrack?
No. At roughly 600–700 high-confidence calls per week — the volume we saw in the comparable June 25 – July 2 window, where accuracy was 91.7% — 55 misses is about 8%, which is exactly what a 90%-plus model produces.^[1][2]

Why do high-confidence SaltyBet predictions still miss?
Because SaltyBet has real randomness: MUGEN AI characters vary run to run, exhibition mode adds chaos, thinly-sampled fighters are hard to read, and S-tier is packed with one-hit-kill characters that can flatten any favorite.^[3]

What’s the difference between a “lock” and a “high” prediction?
Lock is SaltyTrack’s highest confidence tier — near-certainty. High is the next rung down: a strong favorite we like, but not a sure thing. Locks are the calls to trust most; highs are strong leans.^[1]

References

SaltyTrack internal data (queried July 2026) — confidence-tier miss counts over the last 7 days: 55 high-confidence predictions wrong, 0 lock-tier predictions wrong. The exact total high-confidence call count for the window was not captured.
SaltyTrack internal data (queried July 2026) — high-confidence and lock accuracy for the comparable window June 25 – July 2, 2026: high-confidence 91.7% (618 of 674), lock 100% (1 of 1).
SaltyTrack internal data (queried July 2026) — individual high-confidence upset examples recorded July 1–2, 2026, with our pre-match confidence and the winning fighter’s tier. Betting-odds field unavailable on this pull; no odds cited.