How the price ranking works

Cost per TB

uglyHD defines total cost as the item price plus shipping. Cost per TB is that total divided by the catalogue capacity in decimal terabytes. A $200 10 TB drive with $20 shipping is therefore $22/TB. Sales tax is not included because no tax amount is present in the offer data and the actual charge depends on the buyer.

Multi-drive lots are reduced to a per-drive item price and shipping amount before screening. For each product, the query prefers in-stock offers and selects the lowest surviving total. The default ranking then sorts products by the resulting cost per TB.

Product matching and price plausibility

Source results first pass a product matcher so accessories, empty boxes, incompatible models and misleading lot titles do not become offers for the wrong SKU. Surviving candidates then pass two price checks before they can set a headline number.

• A deliberately wide floor and ceiling rejects order-of-magnitude errors: HDD $1.50–$220/TB, with an $8 floor for SATA; SSD $8–$900/TB, with a $6 floor for QLC; and RAM $200–$60,000/TB. RAM capacity is stored as GB divided by 1,000, which accounts for its larger per-TB numbers.
• When at least four candidates remain, an offer below 40% of their median total is rejected as an implausible low outlier. With fewer candidates, only the fixed bounds apply because a small-sample median is not reliable.

These checks are broad enough to preserve a real used-market bargain. They are aimed at bait listings, decimal mistakes and mismatched items that would otherwise win solely because the ranking chooses the cheapest total. Rejections are recorded with a reason so the filter can be audited rather than silently erasing data.

Freshness and stale offers

Every offer receives a source-specific expiration time when it is ingested. eBay offers remain quotable for 12 hours. Newegg and Best Buy offers remain quotable for 24 hours. The local sample fixture lasts one year and is always labelled as sample data. Any other source uses the conservative 12-hour default until it has an explicit cadence.

Once that time passes, read queries hide the offer instead of keeping its last price on a ranking page. If any fresh real offer exists for a product, sample offers are also excluded, so placeholder data cannot outrank or blend with market observations.

Condition by source

All source vocabularies are normalised to three values: new, renewed and used. The source determines how the value is established.

• eBay: the Browse API condition ID. New IDs map to new, the 2000-series maps to renewed, and used IDs map to used.
• Newegg: listing-title terms such as refurbished, recertified, used or open box, combined with whether the result came from Newegg’s refurbished facet. An explicit title term wins over the facet default.
• Best Buy: the API condition field. Refurbished, certified, excellent and open-box values map to renewed; pre-owned and used values map to used; empty or new values map to new.

Condition is applied before the cheapest offer is selected. A buyer filtering for new or renewed stock therefore gets the cheapest offer inside that condition set, not a product removed because its overall cheapest listing happened to be used. See every indexed merchant and its typical condition mix on the source list.

Pumpkin deal score

Each drive with a live offer gets a deal score from one to six pumpkins. Six means an amazing, off-the-chart deal, and it is deliberately hard to earn.

The score judges three separate things about today’s best offer: how it compares with the same drive’s own price history in the same condition over the last 90 days; how its cost per TB compares with comparable drives of the same type and interface; and whether the price looks believable, because a used drive priced far below anything plausible, with no warranty, is more likely an error or a scam than a bargain. An AI judge (TypeSafe’s Jev) rates each of those from the numbers; plain code then combines them, so the arithmetic is auditable. A believability ceiling, enforced in code as well as by the judge, keeps a suspiciously cheap listing from earning many pumpkins, and a drive with under two weeks of history cannot score above four. If the judge is unavailable a rule-based fallback applies the same steps. Scores refresh every two hours. A score describes one offer, so it is shown only beside that offer: if the offer goes out of stock or its price moves, the pumpkins disappear until the next scoring run, and they are never shown beside a sample price.

A deal verdict is only as good as the match behind it, so pumpkins appear only when the listing is well evidenced to be this exact drive: its model line or part number is named alongside the brand, capacity and interface (for memory, the module’s rated speed), not just the brand and capacity. A cheap listing that might be a different drive, an OEM relabel matched on specs alone, or a memory kit priced as one module still shows its price, but no pumpkins. An independent audit with the same AI judge found weakly evidenced matches were a different drive far more often, which is why the bar exists. Where a real, in-stock listing falls short of that bar, the ranking tables mark it “Listing identity unconfirmed” and its drive page says “Not rated”, so a missing rating is explained rather than blank. The mark belongs to that listing, not to the drive: once a better-evidenced listing is the one shown, it can be rated.

NAS appliances

The appliance catalog has 29 models from ASUSTOR, QNAP, Synology, TerraMaster, UGREEN. Every specification on an appliance page (drive bays, M.2 slots, CPU, RAM as shipped and at most, built-in network ports) is copied from the manufacturer’s own pages, and each group of rows on a model page links the pages it came from. Nothing about speed or performance is claimed, because the catalog does not measure it.

How listings are matched. Only eBay and Newegg are searched, one query per model. A listing becomes a price for a model only when its title names that model number as a whole word, with the right tier (an F4-424 is not an F4-424 Pro), RAM variant and bay count, and it is a single unit sold without drives. Titles that point to a populated unit, parts, accessories, a bundle, a kit, a multi-pack, a rackmount or an expansion unit are refused. Where those rules cannot settle a title on their own, an AI judge (TypeSafe’s Jev) reads it as well and has to pass it; when the judge is unavailable that listing is refused and no price is shown. A separate audit re-checks accepted listings with an independent judge, and is how we find out whether the rules are working.

Price checks and freshness. A total (price plus shipping) below $35 or above $1,200 per drive bay is refused, as is an offer below 40% of the median of the same model’s same-condition listings once there are enough of them, and an eBay new listing priced under 60% of a live Newegg new price for the same model. Offers expire exactly as drive offers do (12 hours for eBay, 24 for Newegg), and a sold-out listing is not shown as a price. A sample (placeholder) price can appear only for a model that has no real listing, and it is always labelled as a sample.

Ranking, and what $/bay means. Models are grouped by drive bays (1, 2, 4, 5 to 6, 8) and, inside a group, sorted by the cheapest total. $/bay is that total divided by the number of bays. It ignores CPU, RAM and network, so it is only fair between models in the same bay class, and it is a secondary column and a sort of its own, not a score. New and refurbished listings are shown by default and used ones only on request; discontinued models are hidden unless you ask for them. A model with no listing that passed the checks is listed with its specifications and no price.

No deal ratings for appliances. Drives can carry a deal score, which compares today’s price with the same product’s own price history and its peers. Appliances have no such history yet and each model has few listings to compare with, so they get no rating. Each appliance page shows the lowest price seen in the last 30 days instead, once at least 2 observations exist: it is the lowest total we recorded for that condition, a fact and not a verdict. We will revisit ratings when there is history to support them.

What we do not cover. Amazon and Best Buy are not searched for appliances, so a price there will not appear here, and this index is not a statement about which NAS to buy.