Channel3 · 100M+ products, 25,000+ stores

Scammers win when nobody knows what something is worth. Handshake asks Channel3 for live prices across 25,000 retailers, and uses them for three different jobs across a deal.

Try it live

01 · Scam check

Too cheap to be real

A listed price is compared with the median of live retail prices. Used goods normally sell 20–50% under retail, so only a deeper gap is flagged.

iPhone 13 128GB

Listed $90

63% under the $242 retail median → flagged as bait

02 · Fair range

What it's really worth used

Retail prices plus the item's condition give a fair used range, computed in code. Both negotiating agents argue inside it.

Nintendo Switch OLED

Retail $442

Used, small scuff → fair range $220–$300

03 · Damage pricing

Refunds from real part prices

When an item arrives damaged, the broken part is priced from live listings, so a partial refund is evidence, not a guess.

Cracked left Joy-Con

Part $49.99

Refund ≈ the part's price, capped by severity in code

How it works

The same Channel3 search powers each step. What changes is the question we ask it, and how carefully we clean the answer before money depends on it.

  1. 01

    Read the deal

    AI reads the pasted chat or screenshot and writes a precise store query for the exact item, model and storage.

    AI · Vision
  2. 02

    Search 25,000 stores

    Channel3 returns normalized products with prices, images and buy links in one call.

    POST /v1/search
  3. 03

    Clean the results

    Cases, chargers, docks and protectors are filtered out of an item search, so a $12 case can't drag down an iPhone's price.

    Accessory filter
  4. 04

    Take the median

    The middle price of what's left is the retail benchmark. Code, not AI, computes every percentage.

    Deterministic
  5. 05

    Use it three ways

    Flag bait prices, set the agents' fair range, and price a damaged part for a partial refund.

    Scan · Agents · Claims

Clean data in, fair money out

In testing, a search for a cracked “Switch OLED left Joy-Con” returned whole consoles next to the part, and the refund was priced from a $304 median. A $50 part nearly cost a seller $138.

Name the part

Part prices only count listings whose title actually names the broken part, so a Joy-Con is priced from Joy-Cons.

Nothing priced like the item

Any listing that costs most of the deal itself is dropped from a part search. A console can't be a “replacement Joy-Con”.

Locked in by tests

The exact Channel3 results from that bug are saved as test data. If the filter ever regresses, the tests fail and the deploy stops.

$304

$49.99

The same cracked Joy-Con, priced before and after the part filter, from the same Channel3 results.

Proof

3

jobs for one API: the scam check on every deal, the agents' fair range, and damage pricing on claims.

Uses

63%

under retail: the $90 “iPhone 13” that Handshake flagged as bait against a $242 median.

Scam check

$49.99

for a cracked Joy-Con, from real single Joy-Con listings, not consoles.

Damage pricing

What we call

Item search

POST /v1/search

The exact item from the deal. Prices become the retail median for the scam check and the agents' fair range.

Part search

POST /v1/search

Just the damaged part, filtered to listings that name it, to price a partial refund.

Product data

title · price · image · buy link

Shown to the buyer as the comparison list, so every number links to a real product.

Docs & source

Check a deal Sell with Handshake