How Starling works
Starling is the buy-side face of lectr. lectr spent a year building the sell-side truth — a 1.1M+ sold-lot corpus with exact-identity keys and conformal price bands. Starling points that knowledge at eBay Buy It Now and surfaces the listings priced deep below it.
The thesis
eBay Buy It Now is the largest fixed-price collectibles surface on earth, and it is priced by humans without lectr's knowledge. Sellers misprice constantly — wrong comps, stale comps, no comps. For every listing whose identity we can pin to an exact corpus key, we look up what that exact thing actually trades for, and surface the ones listed deep below it. The moat isn't the eBay integration — anyone can call the Browse API. The moat is the value behind every depth call: 761K+ settled prices, conformal bands, and a published backtest record.
The honesty rules
Starling manufactures no numbers. A deal exists only when an eBay listing's identity pins to an exact corpus key. If the identity is unclear, the matcher abstains — it never fuzzy-matches. Every number on a card is a corpus fact, one tap from the comps on lectr.
The risk grade is never a black box. Every A/B/C/D grade lists its reasons on the card — authenticity anchor, seller signals, listing quality, per-vertical fake-rate prior. Nothing is silently hidden; a D-grade deal is shown at the bottom with its reasons, not removed.
One hard suppression, stated. A listing more than 90% under book reads as fake/broken and is suppressed with the rule named — the only hard gate in the system. Everything else is ranked, never gated.
Cards get no special prominence. Cards are first-class and fully polled, but the board defaults to the mixed view and ranking has no cards penalty and no cards bonus. Rank is depth × confidence × risk — a great watch deal and a great card deal compete on equal footing.
The method
- Depth = 1 − allIn / book median, where all-in includes the cheapest shipping option. A deal must sit between 25% and 90% under book to qualify.
- Confidence is lectr's tier for the corpus row — high or medium; low never ships. It weights the rank and is shown on every card.
- Rank = depth × confidence-weight × risk-weight × a small freshness boost for listings under 24h old (they get sniped fast, so we surface them first).
- Receipts. Every surfaced deal is logged with its call frozen, then graded against what happened — see the tape. Did the market agree the deal was real?
Per-vertical metrics — never blended
If watches surface 3 deals a week and cards surface 300, that gap is a visible roadmap item, not a hidden fact. Coverage and yield are tracked per vertical, from day one.
| Vertical | Polled | Matched | Surfaced |
|---|---|---|---|
| Sports Cards | 7 | 6 | 2 |
| Autographs | 9 | 6 | 4 |
| Pokémon | 10 | 8 | 5 |
| Art Editions | 8 | 7 | 4 |
Figures from the board built Aug 15, 2026, against the lectr corpus dated Aug 14, 2026.
Powered by lectr
Starling holds no corpus of its own — it consumes lectr's public price products and deep-links every call back to the proof. lectr.bid is the sell-side truth engine: 1.1M lots, 761K settled prices, every call replayed against history. Starling is the buy-side lens.
Affiliate disclosure: some links to eBay are affiliate links and Starling may earn a commission if you buy, at no additional cost to you. eBay is a trademark of eBay Inc.; Starling is an independent affiliate and is not endorsed by eBay.