First Pass

First Pass is a photo-to-verdict research assistant for secondhand resellers. A reseller photographs a messy pile of finds — the way items actually sit at a thrift store or estate sale. First Pass segments the frame so each physical item is judged on its own, examines each item the way an experienced picker does, and returns a verdict grounded in verified sold prices — never asking prices, never guesses. When the evidence is thin, it says so and names the one additional photo that would settle it. When an item fails legitimacy or economics, the verdict is PASS, not a price. Built on the Meta Model API (Responses API): SAM 3.1 segments the photo; Muse Spark 1.3 examines each item — form, materials, maker's marks, era markers — and states what it cannot establish. In the live demo run on a real pile photo (Nov 3, 2026), SAM 3.1 returned 9 segmentation masks and Muse Spark isolated 24 items, each with frame location, visible evidence, an explicit cannot-establish list, and a research verdict (EXAMINE 10, UNCERTAIN 12, SKIP 2) — and stated zero prices, because at that step there is no sold evidence and inventing one is defined as failure. Final verdicts are priced only from verified sold evidence in the operator's audited records: LIST (Coach Ergo hobo — 3 exact verified solds $360–$439, list $425), RESEARCH (Polo set, size M — no verified sale of the exact set, no call made), PASS (brown LV-embossed wallet — flagged counterfeit at the research gate, $5.94 sunk cost recorded). The method's track record is inspectable: a shirt called at $20 before listing sold for exactly $20.00; a jacket called at $20–$30 sold for $35 — the miss stays in the record on purpose. I'm Jessica Orlic, a self-taught developer in Painesville, Ohio. I run Rejected Treasures, a vintage resale business, and have processed 300+ items through this method with every decision logged. This is my world; the tool is built from it.