How our Deal Score and Flip Score work

The anchor: what comparable lots actually sold for

Almost every signal on the site starts from one number we call the comp value: the median final price of recent, comparable lots that have already sold. We keep a rolling snapshot of roughly 164,721 completed U.S. sales from the last 365 days and match each live lot against it:

• Same category, similar title. A comp has to share a meaningful share of the lot's significant title words (we drop filler words and stock numbers first).
• Vehicles stay in their lane. Vehicle comps are constrained to within four model years, and to the same make when the title names one - so a 2013 Porsche isn't priced off 2013 Jeeps.
• Per-unit math. A "lot of 14 laptops" is divided down to a per-unit price before we take the median, then scaled back up - so bulk lots and singles compare fairly.
• Enough agreement, or nothing. A lot only earns a comp value when we find at least five solid matches that agree closely enough. If the matches are too few or too scattered, we don't publish a comp anchor - we fall back to a conservative estimate and rank the lot lower.

Real estate is handled separately, because a title tells you little about a house: HUD and Fannie Mae homes are anchored to their own published appraised price, and residential auction lots are anchored to the Zillow Home Value Index for that ZIP - but only when the listing clearly describes an actual house (beds, baths, square footage or a street address), so a vacant lot never inherits a six-figure neighborhood median.

Canadian comps also include sold records published by the Government of Canada in its GCSurplus "What has Sold" dataset. Contains information licensed under the Open Government Licence - Canada.

The Deal Score (free): how the Best Deals feed is ranked

The Deal Score compares the current bid - including the buyer's premium - against the lot's value, preferring the comp value above, and only falling back to a conservative category estimate when no comp anchor exists. It blends two things:

• How much you save in dollars (value minus bid), and
• How big the discount is in percent.

Both are square-root-compressed so a single enormous lot can't dominate the feed, then weighted by confidence: a deal is more certain when an auction is closing soon and has real bidding activity, and less certain when it has days left and no bids. The single most important rule:

We also refuse to score some lots, on purpose:

• Lots already bid at or above their estimated value (no headline "deal").
• Lots flagged as heavily damaged or salvage in the listing text.
• Suspiciously cheap non-running wrecks, and accessory/parts lots that happen to word-match a whole machine - both are held back rather than shown as fake mega-deals.

The GovAuctions Flip Score: a free grade, with the numbers behind it on Pro

The Flip Score is a 0-100 grade for one lot, built from up to 25 of its closest completed sales (recency-weighted over roughly the last 90 days). It filters those comps hard - by title similarity, by model-year and make for vehicles, by configuration for electronics (a stripped laptop only comps against other stripped laptops) - and discards comp sets that are too scattered to trust. The grade itself is free on the listing page for any lot that scores 80 or higher; Pro unlocks the figures behind it. Here's how it works:

• At the comp median, the score is 50. Bid well below the median and it climbs toward 100; bid above it and it falls toward 0. (Precisely: 100 / (1 + (bid / median)²).)
• Weak evidence stays hedged. When there are few comps or low similarity, the score is clamped into a narrower band so a thin match can never read as a slam-dunk.
• A true resale margin. Pro also shows resale value minus your bid minus the real costs of flipping - buyer's premium, an approximate 12% resale fee, and category-based transport. The conservative end of the range is the headline number.
• A projected closing range. Government auctions hard-close at a fixed time, and bidder count is the strongest driver of the final price, so we project where the lot is likely to close and how hot demand is.

Damage overrides everything: a lot the listing describes as salvage, non-running, or missing major parts has its margin zeroed and its score capped, because the comps are working units and the comparison no longer holds.

How accurate the Flip Score is

In September 2026 we went back over 35,910 auctions that had finished, scored each one from a reading taken while it was still open, and compared that score against what the lot really sold for. No outcome information went into the score.

Badged lots sold for a median of 43.8% of what comparable items fetch, so under half price, and 64.8% of them sold below even the cheapest quarter of that range. Lots scoring under 35 sold for 298.4% of comparable, so buyers paid close to three times the going rate.

Every band in between lands where it should, in order, with no exceptions. A score of 50 means the bid is around the going rate, and those lots sold at 102.4% of it. That gradient holding across 35,910 auctions is what tells you the number is reading something real in the price rather than sorting lots at random.

The test was built so we could not flatter ourselves. Each score was fixed from a reading taken while the auction was still running, the outcomes are confirmed winning prices pulled from the auction site after the close, and lots that failed to sell are counted in the results rather than quietly dropped. The comparable figure everything is measured against is checked the same way: used on its own as a guess at what a lot will sell for, it came within 2% of the real price across 27,828 finished auctions.

A bargain that visible attracts other bidders. Between our reading and the close, badged lots rise about 1.47x, against 1.09x for the lowest-scoring lots, so part of the discount gets bid away before the end. It survives anyway: 86.8% of badged lots still sold below the going rate.

Most of that bidding happens in the middle of the window, which is why the discount holds up best on lots you decide about in the last couple of hours.

Measured in September 2026 and fixed at that reading, so the figures above describe a specific study rather than a number that moves under you. Two limits worth stating. We do not know what buyers resold these lots for - nobody records it, so no auction site can honestly tell you what people earned, and what is measurable is the gap between the price you pay and what comparable government lots sell for. And the outcomes above are confirmed winning bids recovered from the source after the close, which restricts the test to GovDeals US and Canada plus Municibid, the platforms that publish a confirmed final price.

Where the trust comes from

We don't run any of these auctions or take a cut of a bid, we never add markup or hide fees, and we don't pad the catalog with private-marketplace or general real-estate filler. You can see every platform we index, with live per-source counts, on the sources page.

FAQ

Is the Deal Score based on AI or a guess?

Neither. Every score is deterministic - the same lot always produces the same number - and the strongest scores are anchored to a median of what comparable lots actually sold for in the last year. No language model decides whether a lot is a good deal. The logic is rules plus real sold-price data, and this page describes it in full.

What is the comp anchor?

The comp anchor (we call it the comp value) is the median final price of recent, comparable sold lots in the same category. We match on title-word overlap, and for vehicles we constrain comps to within four model years. A lot only gets a comp anchor when we find at least five solid matches at medium-or-better confidence - otherwise we fall back to a conservative category estimate and rank it below comp-verified lots. Two different bars are at work and it is worth being precise: the comp anchor that ranks the Best Deals feed needs five matches, while the Pro Lot Analyst panel shows its numbers from three, with the confidence it reports lowered accordingly.

Is the Flip Score free, and what does Pro add?

The GovAuctions Flip Score - a 0-100 grade for a single lot, computed from up to 25 of its closest completed sales - is free for everyone on the listing page whenever a lot scores 80 or higher (a genuinely strong buy). What Pro unlocks is the Lot Analyst behind that grade: a recommended max bid (the highest bid that still clears your target resale margin after the buyer's premium, sales tax, transport and resale fees), anchored on the comparable sold-price range (P25 / median / P75), an estimated resale margin, and a projected closing-price range. So the grade tells you a lot is worth a look; Pro tells you exactly what to bid and what it should sell for. (Separately, the free Deal Score is what ranks the Best Deals feed.)

How accurate is the Flip Score?

We measured it. Across 35,910 government auctions that had finished as of September 2026, scored from a reading taken while bidding was still open and graded against the confirmed winning price, lots badged 80 or above sold for a median of 43.8% of what comparable lots fetch, and 64.8% of them sold below even the cheapest quarter of that comparable range. Lots scoring under 35 sold for 298.4% of comparable. Every band in between lands in order. The comparable figure used as the yardstick is itself accurate to within 2% of real sold prices across 27,828 finished auctions. Full method and tables are in the "How accurate the Flip Score is" section of this page.

How fresh is the data?

New listings are added every 30 minutes. The full catalog is rebuilt twice a day, around 06:00 and 18:00 UTC. Ended auctions drop off automatically, so what you see is what is open right now. The exact last-refreshed timestamp is shown on this page and on the sources page.

Do you de-duplicate listings cross-posted to several platforms?

Yes. An item listed on three platforms is merged into one listing, not counted three times. We normalize each title - stripping plate numbers, years, model codes, quantities and condition words - and collapse near-identical runs from the same seller. Aggregators that re-list the same lot under each source inflate their totals; we don't.

Does a high score mean I should bid?

No - it's a starting point, not advice. Scores assume the comps are working units, so heavy damage, missing parts, or a salvage title can make a 'great' score wrong. We flag damage we detect from the listing text, but you should always read the full description and inspect before bidding. Every listing links straight to the official source so you can verify it yourself.