What we learned by buying our own remnant

We bought remnant display so we could describe the supply path to customers who advertise to ordinary people, not so we could publish a victory CTR.

Supply Bench is that buy. The campaigns ran on the open exchange with ordinary 300x250 creatives and interaction pages we control. The intent was characterization: what placements show up, which domains take the clicks, whether ads.txt authorization changes the mix, and whether a web-only environment still buys browser games. The intent is operator-judged. The grain in the later pull is seat-measured and lives in Lab.

What "ordinary people" forced us to notice

A customer who sells to people needs a person to arrive. Remnant display is good at producing billed clicks and bad at promising that arrival. If we only reported CTR, we would be describing how clicky leftover inventory is. CTR is a trap is the short version of that sentence.

The 2026-09-02 pull of inventory_agg and domain_agg for campaigns 4791 through 4810 covered the top spend slice of both Supply Bench waves, about $316.64 of a roughly $432 combined total. This seat ignores report offset and 50,000-row limits 400, so the slice is biased toward larger placements. Inside it: 1,149 placements, 1,535 domains, 1,073 placements with zero clicks taking 64 percent of observed spend, and seven domains producing 41.3 percent of clicks from 4.7 percent of impressions.

That shape is the finding we owed the customer. Most placements did nothing. A handful of hosts produced the clicks. Reseller clicked 2.3 times more than Authorized-Direct. The same creative's CTR ran from OPENX at 0.111 percent to SMARTADSERVER at 3.074 percent. msn.com had already shown up on a July learning flight.

What we will not say about the buy

We did not have an IVT vendor attached. invalid_impressions was zero because we were measuring nothing. We will not call the remnant "clean" or "dirty" from that zero. We will not treat billed clicks as people. We will not turn the consumer probe pages into a signup funnel. Those pages exist to see whether a stranger interacts. The product pitch belongs on Guides.

ads.txt authorization is available on this seat and we have never used it on the live portfolio. Unauthorized inventory was $0.48 of that pull. Authorization would not have saved the seven-domain concentration. Authorized inventory can still be junk. That is a Lab sentence, and it is why "buy remnant" is not a strategy by itself.

Why we keep buying it

Customers who advertise to ordinary people will still meet remnant if they buy open exchange. Pretending the path is a brand marketplace helps no one. The honest product job is to let an agent launch a prepaid test, then read the grain without lying: placements that spent and got nothing, domains that took the clicks, freshness of the report, and measurement that is actually attached.

Buying our own remnant was how we stopped arguing from vendor slides. The next judged flight updates the scorecards. This note stays about the reason we flew the first ones.