Seven steps from raw listings and auction logs to a single "Should I act now?" verdict — with every supporting signal shown, scored, and traceable. No black boxes, no anchoring on staged sales.
Every two hours we snapshot active retail listings across the trims we track on CarGurus — price, mileage, dealer, color, days-on-market, MSRP when published.
Snapshots are diffed nightly so the platform remembers every price move and every relist, even if the dealer pulled and re-posted the same VIN.
21,448 active listings tracked tonightBring a Trailer and Cars & Bids completed auctions are pulled daily — sold price, RNM, withdrawn, plus bid logs when available. Auction IDs are unified into a single auction_results table so cross-source memory stacks cleanly.
Bidder histories live in a separate auction_bids table; we keep username, sequence, winner flag, and amount per bid for the auctions whose detail pages we've enriched.
1,122 BaT/CnB clearing sales in memoryWhen a VIN appears on an auction and again on a dealer lot within ~12 months, we record the (auction, retail listing) pair with gap-days and markup percentage — that's the dealer-flip signature in plain SQL.
The relist_linkages table is the substrate everything else builds on: dealer scoring, trust adjustment, retail-relist factor on auction comps, and the post-auction outcome badge on each listing.
29 auction → retail linkages trackedEvery BaT outcome gets a 0–100 trust score: penalties for RNM, repeat-listed VINs, post-auction retail relist at a premium, low bidder diversity, top-bidder concentration, dealer-as-winner, and seller-as-winner (self-dealt).
Factor breakdown is stored as JSON so the /bat tooltip shows exactly which signals fired and how many points each subtracted. Low-authenticity comps get downweighted before the comp engine uses them.
1,286 auctions scored · 86 flagged below 60Each tracked dealer gets a scorecard: market trust, archetype, auction-flip rate, median relist premium, time-to-relist, and stale-share. Trust flags surface anchoring, volatile pricing, and aggressive cutting independently of the archetype label.
Dealer pages also show their full active inventory with a per-listing post-auction trail when applicable — so you can spot which units they grabbed from BaT and at what premium.
2,493 dealers analyzed · 23 actively flipping auction-sourced inventoryPer model, we synthesize a single buyer-facing verdict — Buy now / Negotiate / Wait / Patient buyer / Hold off — driven by market state, ask vs auction comps, expected haircut, sell-through, inventory delta, and confidence depth.
The verdict carries a confidence chip (low/medium/high) so you can see when the data depth is thin. Every supporting metric has a tooltip explaining what it measures and why the threshold matters.
120 trims with a live verdictEvery morning at 04:15 PT we post a digest covering likely price-cut listings, fresh auction → retail flips, rare-spec new arrivals, suspicious comps, and dealer softening. Per-key dedup prevents the same alert firing twice.
Suppression is 60-day default with material-change override (a relist whose markup percentage shifts ≥3pp re-alerts). Real-time triggers and per-VIN watchlists land when user accounts are wired up.
1,126 live signals across watched modelsReceipts, not vibes. Every score breaks into named factors with their point contributions visible on hover. If a comp gets downweighted, you can see exactly which signals fired and how the math went.
Confidence is a chip, not a gate. Thin-data models still get a verdict — they just carry a "low confidence" label so you know to weight it less. No "wait for more data" black holes.
Cross-source memory. A car that RNM'd on BaT then resurfaced at a dealer 26 days later is one entity in our data, not two unrelated rows. That's the only way to call out the "test the auction market, then dealer-anchor" pattern that's the single biggest noise source in exotic-car pricing.
A 0–100 number we attach to every BaT sale measuring how likely the outcome reflects a real arm's-length transaction. It starts at 100 and subtracts for known noise patterns: Reserve Not Met (-30), retail relist at a premium (-25 to -35), self-dealt outcomes where the winner is the seller (-40), low bidder diversity (-20), and a few smaller signals. The breakdown is shown as a tooltip on every listing so you see exactly what's flagged.
RNM means the high bid on a BaT or Cars & Bids auction didn't meet the seller's hidden reserve, so no sale occurred. The price shown is what the market refused to pay — not a clearing transaction. Treating an RNM "price" as a comp inflates your reference; we always downweight RNMs and tag them visibly.
VIN. When the same VIN appears on a BaT or Cars & Bids auction and on a CarGurus dealer listing within a year, we record the pair as a relist linkage with gap-days and markup percentage. The post-auction outcome badge under a listing's title and the dealer auction-flip scorecards both read from this single table.
Per model, we anchor on BaT 6-month median sold price when available (real transactions), falling back to a confidence-adjusted dealer median (ask discounted by expected reduction share × median cut size). Each listing additionally carries a hedonic residual that prices the spec — mileage, color, options — against the model cohort.
2,493 dealers across 120 trims of Porsche GT3 (991.1/991.2/992.1/992.2 in coupe and touring), Ferrari SF90, 296, F8, Lamborghini Huracán/Aventador/Revuelto, McLaren 750S/765LT, Nissan GT-R, and several other sports & exotic categories. Each trim has its own page with live market state and active inventory.
Retail listings snapshot every 2 hours. BaT auction ingest runs daily at 03:30 PT; Cars & Bids closed-auction ingest at 03:45 PT; bidder-anomaly + authenticity scoring at 03:55 PT; dealer + market analytics rebuild at 04:00 PT; daily alert digest at 04:15 PT. The carstats build that publishes the static site runs on top of all of that.
No signup needed to browse. Free accounts add VIN watchlists, daily alerts, and saved searches.