BidTune
Log in
Publisher case study

Inside JSONLint’s
Prebid setup.

A live site. A clearer view of the auction. Three diagnostic findings that show where the next investigation should start.

Explore the findings
JSONLint
AUCTION DATA REVIEW

September 17–30, 2026

Historical diagnostics from a snapshot collected October 1. A verified revenue-lift result is not yet published.

What the data revealed

Three signals.
Three places to look closer.

These observations shaped the work. They describe the historical data, not revenue recovered or the setup’s current state.

01 · Bidder health

Requests don’t guarantee competition.

Medianet, PubMatic, and TTD recorded no bid responses in this window. Other adapters did respond.

The next questionIs this adapter contributing demand on this traffic?
Bid response share Selected adapters · Sep 17–30
criteo39.2%
openx29.9%
medianet0%
pubmatic0%
ttd0%

Share of recorded response, no-bid, and timeout outcomes. Zero means no observed bid responses in this snapshot.

02 · Auction depth

One bidder changes the floor question.

396,463 of 1,405,812 auction/ad-unit groups with positive USD bids had just one responding bidder.

The next questionHow much downside could a higher floor create when there is no competing bid?
Depth of responding demand Groups with positive USD bids

One bidder396,463 groups

Multiple bidders1,009,349 groups

This is a subset of auction/ad-unit groups with positive USD bids, not all auctions.

03 · Currency handling

Count value in the currency it arrives in.

The snapshot contained 6,407 EUR-denominated bid-win events from smilewanted. A query limited to USD would exclude those events.

The next questionAre currencies handled consistently before we compare value?
A measurement requirement Currency normalization
EUR bid wins6,407recorded events
Comparable valueNormalize before comparison

Bid-win events are not rendered revenue. This finding does not establish revenue recovered.

From finding to decision

Follow the question.
Then follow the evidence.

A timeout idea illustrates the investigation path. First inspect the data; use a limited test where the answer depends on bidder behavior.

  1. Find the signal

    Inspect the auction data.

  2. Frame the question

    Estimate the value of waiting.

  3. Limit the change

    Test what the data cannot settle.

  4. Verify and evaluate

    Check application, delivery, and value.

Keep the evidence. Revisit the question as demand changes.
What would prove improvement?

A holdback comparison.
And a ledger.

A permanent 10% holdback keeps the current configuration untouched and compares it with the optimized setup over the same period.

A published result needs dated samples, an uncertainty interval, and the experiment ledger: every change, with its measured lift. This page does not show that result.

Explore the measurement approach
Measurement design · no lift result shown
Base configurationUntouched comparison
Optimized setupChanges under evaluation
Rendered Prebid revenue
per exposed session

Same period. Consistent valuation.
Uncertainty included.

About these observations

Source: JSONLint’s auction-event snapshot collected October 1, 2026, covering September 17–30, 2026. This is a historical diagnostic analysis, not a verified optimization outcome.

Response shares divide bid responses by the sum of response, no-bid, and timeout events, rounded to one decimal place. Auction depth counts distinct responding bidders per auction/ad-unit group, using positive USD bids and a recorded auction identifier. Currency counts refer to bid-win events, not rendered impressions or publisher profit.

A published revenue result would need the holdback comparison. This page does not include one.

Start from your own auction data.

14 days free after activation. No credit card to get started.