Price Lab: APS vs ShopSavvy — Softballs, Cleats & a Refurbished iPhone 16

The Final Scoreboard

| Search | AI Price Search (APS) | ShopSavvy | Result | |---|---|---|---| | Bucket of softballs | No reliable winner. A $49.97 Walmart bucket set appeared but wasn't sufficiently verified. APS also surfaced a $10.49 single softball as its lowest price. | $79.99 Hit Run Steal bucket with 12 softballs, listed in stock. Shipping and tax weren't confirmed. | ShopSavvy | | Women's size 8 softball cleats | $35.54 at Amazon with free shipping, but the listing did not establish size 8 in the result. Tax excluded. | $49.99 at Scheels for Nike Lunar Hyperdiamond 3 cleats in size 8, listed in stock. Shipping and tax weren't confirmed. | ShopSavvy | | Refurbished iPhone 16 | $560 at Amazon for a renewed, unlocked 128GB iPhone 16 with free shipping. Tax excluded. | $599 at Walmart for a refurbished, unlocked 128GB iPhone 16, listed in stock. Shipping and tax weren't confirmed. | APS |

Score: ShopSavvy 2 — APS 1

But that simple score hides the most interesting part of this test.

Test 1: Bucket of Softballs

This sounds like an easy shopping query.

It isn't.

A shopper searching for a "bucket of softballs" probably doesn't want one softball. They also probably don't want an empty bucket marketed as softball equipment.

They want the bundle.

APS

APS struggled with that distinction.

One potentially relevant result was a $49.97 Walmart bucket set, but APS couldn't verify it strongly enough for us to call it the winner.

More importantly, APS's apparent lowest price was $10.49 for a single softball.

That's cheaper, but it doesn't satisfy the query.

This is exactly why comparing prices without comparing product identity can produce nonsense.

ShopSavvy

ShopSavvy produced a much clearer match:

Hit Run Steal bucket with 12 softballs — $79.99

The item was listed in stock.

We couldn't confirm shipping or tax, so $79.99 should not be interpreted as a confirmed delivered total.

ShopSavvy also surfaced a roughly $35 Zappos Rawlings offer, but that price looked inconsistent with the surrounding offers and could not be verified well enough for us to treat it as the winner.

Winner: ShopSavvy

Not because $79.99 was necessarily the absolute cheapest deal on the internet.

It won because ShopSavvy gave us the clearest current result that actually represented what we searched for.

Price Lab finding for APS: bundle quantity and bundle identity still need work on searches where individual components are dramatically cheaper than the requested set.

Test 2: Women's Size 8 Softball Cleats

This test attacked a different problem.

A pair of women's softball cleats isn't enough.

They need to be size 8.

APS

APS found an attractive price:

$35.54 at Amazon — New Balance softball cleats

APS also reported free shipping.

That's considerably cheaper than ShopSavvy's strongest result.

The problem is that the APS result did not establish that size 8 was actually available for that offer.

So we couldn't award APS the win.

A $35 shoe in the wrong size has effectively zero value to the shopper who needs size 8.

ShopSavvy

ShopSavvy produced a more specific result:

Nike Lunar Hyperdiamond 3 — women's size 8 — $49.99 at Scheels

The offer was listed in stock.

ShopSavvy also showed a cheaper $46.84 Amazon offer, but availability couldn't be established strongly enough to use it as the winner.

Shipping and tax on the Scheels offer remained unresolved.

Winner: ShopSavvy

APS may actually have found the cheaper pair.

But Price Lab isn't awarding victories based on "maybe."

ShopSavvy gets this round because its result gave stronger evidence that the product actually met the shopper's size requirement.

Price Lab finding for APS: size-specific searches need stronger variant confirmation before an offer can be presented as the lowest relevant price.

Test 3: Refurbished iPhone 16

And then things got interesting.

The original APS search failed.

Not by a few dollars.

It failed product identity.

The First APS Result

Our original search for:

refurbished iPhone 16

returned older iPhones among the qualifying results, including an iPhone 6 and an iPhone XR.

APS had also found actual iPhone 16 listings.

That meant this wasn't simply a product-coverage problem.

The correct products were present.

The identity system was allowing incorrect generations to compete with them.

For a price comparison engine, that's a serious problem.

A $159 refurbished iPhone XR isn't a fantastic price on an iPhone 16.

It isn't an iPhone 16.

So we stopped the benchmark.

Instead of quietly discarding the result, we investigated it.

Price Lab Found a Bug

The first correction tightened generation identity.

After that change, older generations such as the iPhone 6, XR, 13, 14 and 15 no longer leaked into the result.

Problem solved?

Not quite.

We reran the search.

APS then selected a renewed iPhone 16e for $378.39 as the lowest match for "iPhone 16."

That's much closer semantically.

But it still isn't the same model.

An iPhone 16e is part of the iPhone 16 generation, but a shopper explicitly asking for the standard iPhone 16 shouldn't have a cheaper 16e silently substituted and crowned Best Price.

That second failure led us deeper into the identity logic.

The Root Cause

The eventual problem wasn't an Apple-specific rule.

It was structural.

APS's modelConflicts logic contained a phone-model identity check that occurred after an early return involving numeric model tokens.

That created a broader weakness for model names that don't behave like standalone numbers.

The fix moved phone-model identity checking ahead of that early return rather than adding a special rule saying, essentially, "don't confuse iPhone 16 with iPhone 16e."

That distinction matters.

A good fix should improve the engine's understanding of product identity — not memorize the test it just failed.

The corrected logic went through 42 ICE regression tests, plus the broader identity, accessory, search-quality, search-contract and APS regression suites.

They passed.

Then we went back to production.

The Final iPhone 16 Retest

We ran a fresh, uncached search for:

refurbished iPhone 16

at ZIP 72104.

This time APS returned three exact standard iPhone 16 matches.

No iPhone 16e.

No iPhone XR.

No iPhone 6.

No older generations.

The lowest result was:

Amazon Renewed — Apple iPhone 16, unlocked, 128GB, Ultramarine — $560

APS reported free shipping.

Other qualifying results included:

• 128GB Black — $569.95
• 128GB Ultramarine, Renewed Premium — $586.18

All three were verified and Best Price eligible.

APS correctly kept the $560 result provisional, however, because tax had not been resolved.

ShopSavvy

ShopSavvy's strongest comparable result was:

Walmart — refurbished unlocked iPhone 16, 128GB — $599

It was listed in stock, but shipping and tax weren't sufficiently confirmed for us to calculate a final delivered total.

Winner: APS

APS had the lower relevant listed price:

$560 vs $599 — a $39 difference before tax.

It also provided clearer shipping information on the winning offer.

But we're not calling $560 the final delivered price because tax wasn't resolved.

That distinction is intentional.

What This Price Lab Actually Showed

If we only published the final results, APS would look pretty good.

It found a cheaper refurbished iPhone 16 than ShopSavvy.

But that isn't what actually happened.

The first APS search was wrong.

Price Lab caught it.

We fixed one identity problem.

The retest uncovered another.

We investigated again, found a broader weakness in the model-conflict logic, fixed that, ran the regression suites and returned to the original search.

Only then did APS produce the result we expected.

Meanwhile, the other two searches remain unresolved weaknesses.

APS still needs better bundle identity for searches like "bucket of softballs."

It still needs stronger size/variant verification before a cheap pair of shoes can win a size-specific search.

We're leaving those results in this report.

Why?

Because Price Lab isn't supposed to prove that AI Price Search always wins.

It's supposed to find out when it doesn't.

A comparison engine that only tests searches it already handles well doesn't learn very much.

Failures are useful.

If another engine beats APS, we want to know why.

Sometimes the competitor simply has the better deal.

Sometimes APS lacks enough data to verify the result.

And sometimes the test exposes a flaw in the engine itself.

This test did all three.

Final Result

ShopSavvy: 2 wins

• Bucket of softballs
• Women's size 8 softball cleats

AI Price Search: 1 win

• Refurbished iPhone 16

But perhaps the more important Price Lab result is this:

One benchmark produced two identity discoveries and a stronger APS engine than the one that entered the test.

That's exactly what Price Lab is for.

---

Prices, inventory, shipping and availability can change after testing. Prices shown here reflect the offers observed during our October 7, 2026 test. Unless specifically stated otherwise, totals in this test do not include unresolved sales tax or unconfirmed shipping charges.