A practical research agenda
The questions we work on
AI work creates long stretches where a person is waiting, reviewing, or deciding what to do next. That time is not automatically valuable. It becomes useful only when the product can distinguish an open surface from a focused person who is actually there.
Our work concentrates on verified attention rather than raw delivery, share-of-voice allocation rather than highest-bid-wins, quality-adjusted rewards rather than simple counters, and privacy-safe signals that make coordinated abuse harder without turning product use into surveillance.
A rewardable attention window
= a real AI wait + a focused surface + recent activity + server-side eligibility
September 2026
Research notes
- 01Attention is not an impressionA card being delivered is not evidence that it was seen. The useful unit is a short, focused, measurable moment while a person is present in their work.Read the product context →
- 02A fair market cannot be bid-onlyPrice matters, but relevance, quality, and share of voice prevent a single bidder from turning a shared discovery surface into a private billboard.Read the product context →
- 03Open methodology, private thresholdsWe can explain the rules and publish settled outcomes without publishing the thresholds that would make abuse easier to tune.Read the product context →
An intentional boundary
What we publish
BackSpin publishes the parts of its methodology that customers, builders, and researchers need to understand the market. That includes the product model, the public algorithm, aggregated ledger outcomes, and the reasoning behind quality-adjusted attention.
We do not publish live fraud thresholds, challenge details, or account-specific risk signals. That is not a contradiction. A rule can be inspectable at the policy level while its operational thresholds remain protected from people trying to game it.
Continue reading