What a Steal Score is — and what it refuses to say

A Steal Score is a compact comparison between a current reviewed price and the recent reviewed history of the same product. It is not a copy of the retailer’s advertised discount, a prediction that an item will appreciate, or a guarantee that the price will remain available.

The most important part of the method is not the number. It is the set of conditions that can prevent a number from appearing.

When the evidence is thin, ambiguous, stale, or unreviewed, StealIQ returns UNKNOWN. Silence is a feature of the scoring system because an unsupported answer would be more misleading than no answer.

The question the score answers

The score asks: how favorable is the current obtainable price relative to this exact product’s reviewed recent typical price?

That wording contains several constraints.

Current means a fresh source observation, not a price carried forward indefinitely. Obtainable means the item was marked available. Exact product means the model, capacity, bundle, and other material variant details match. Reviewed means a human-confirmed source capture, not a crowd submission or unverified crawl. Recent typical price means a median derived from enough elapsed-time evidence to serve as a baseline.

Remove any of those constraints and the number may answer a different, weaker question.

What enters the rollup

An eligible observation must include a positive price, a dated HTTPS source URL, a retailer identity, an available offer, crawl context, Verified evidence status, and explicit validation. Observations older than the active history window do not drive the current rollup.

A crawl result begins as Unverified. StealIQ stores its captured product title, amount, availability, URL, extraction method, and timestamp. An operator can verify it only after a fresh source check confirms the same product and price. Rejected or failed observations do not enter the model.

Crowd-submitted prices can help identify what needs review, but they do not become score inputs merely because a user typed them.

The thin-data rule

Fewer than three eligible observations cannot produce a meaningful recorded low in the current method. With one observation, that price is simultaneously the highest and lowest only as a mathematical accident. With two, a single error or temporary condition can dominate the range.

The public verdict gate is stricter. A typical-price baseline requires at least five distinct Chicago calendar days spanning at least fourteen days. A score also requires at least two retailers and a fresh canonical-source quote no older than forty-eight hours.

Until every prerequisite passes:

• Steal Score remains empty;
• Verdict remains UNKNOWN;
• confidence remains Low or Unknown;
• no Product page can enter the public projection merely because someone filled publication fields manually.

Two retailers agreeing today are useful confirmation of today’s price. They are not fourteen days of history.

Why the median is the baseline

Once the time gate passes, StealIQ uses the median of eligible recent observations as the typical price. The median is the middle value after sorting the observations. It is less sensitive than an average to one extreme quote.

That protection is limited. A median cannot repair mixed products, duplicate refreshes, or bad evidence. StealIQ therefore keeps one quote per source per Chicago day for aggregation. A worker that checks the same page ten times does not receive ten votes.

Product identity and review happen before statistics. The Bureau of Labor Statistics uses a much more sophisticated methodology for national price indexes, but one relevant principle is shared: price comparisons should follow the same or a comparable defined item, and quality differences must not be confused with price changes. StealIQ does not calculate a consumer price index; it applies the same discipline at product scale.

The deterministic score

When the evidence gate passes, the current deterministic score uses the percentage distance between typical and current price:

discount = (typical price − current price) ÷ typical price

score = clamp(round(50 + 250 × discount), 0, 100)

A price equal to the observed typical price centers near 50. A price below typical moves the score upward; a price above typical moves it downward. Clamping prevents values outside the 0-to-100 display range.

The verdict is also deterministic. A current price at least 15 percent below typical produces BUY NOW. A current price at least 15 percent above typical produces SKIP. Prices between those boundaries produce WAIT. These are product-timing labels, not financial advice.

The formula is published so it can be tested. Affiliate availability, sponsorship, and retailer relationships do not enter it.

Lowest price and durability

The 90-day lowest observed price is contextual. It shows the minimum among eligible observations after at least three exist. It does not automatically become the current recommendation.

A brief low that vanished may show what has happened before, but not what can be purchased now. A durable low observed across dates or retailers is more actionable. The score uses the typical-to-current relationship, while the low helps a reader understand the range and judge whether waiting for a repeat is plausible.

StealIQ does not call the first observed price an “all-time low.” The system has no evidence about time before monitoring began.

Fake-discount claims are separate

A displayed list price is captured as a retailer claim. It is not used as the typical price. The FTC’s deceptive-pricing guidance explains why former-price comparisons require a genuine basis. StealIQ tests a reference-price claim against observed history only after enough evidence exists.

A large gap between the printed list price and the observed median can justify a review flag. It does not automatically justify a public accusation. Fake Deal reports require their own evidence summary, rights clearance, public approval, and publication decision. Retailers may respond; payment cannot remove a report.

What the score refuses to say

A Steal Score does not say:

• that the item is a good fit for the buyer;
• that the seller is trustworthy beyond the observed page;
• that taxes, shipping, membership, or financing costs are included unless stated;
• that a resale profit exists;
• that the product will be cheaper or more expensive on a particular future date;
• that an advertised reference price is deceptive without reviewed evidence;
• that historical availability guarantees checkout availability;
• that a score from one variant applies to another.

Flip economics require sold comps, fees, fulfillment, returns, and holding cost. Those are deliberately outside the consumer Steal Score.

Publication is another gate

A calculated verdict is not automatically public. The Product must separately be marked Published, Public Approved, and Rights Cleared. At read time, the public site recomputes the deterministic rollup from eligible observations and compares it with the stored score, prices, verdict, confidence, and check timestamp. If they do not match, the Product does not render and does not enter the sitemap.

This prevents a manually edited score or stale publication record from bypassing the evidence rules.

Why refusal builds trust

Price tools are under pressure to answer instantly. Retail marketing is designed to make delay feel costly. A system can appear useful by turning one price and one list-price claim into a confident recommendation.

StealIQ chooses a slower standard. UNKNOWN means the product is being observed but the evidence cannot yet support a public verdict. It invites more collection and review without pretending that collection has already happened.

A trustworthy score is not the one that always speaks. It is the one whose silence has rules.

Informational only. A Steal Score describes reviewed price evidence, not a guarantee of savings, availability, resale value, or future pricing.