Research behind the work
Compelle’s historical Bittensor experiments put competing arguments through hundreds of thousands of debates. The useful result was not a machine that could never be wrong. It was a clearer view of where reasoning breaks—and how to examine it.
Historical research · not client outcomes
Recorded in Compelle’s aggregate snapshot, captured September 26, 2026.
These are historical network totals, not a count of customer engagements or a live activity feed. The current public selection does not contain every underlying debate.
What the experiments taught us
01 / Challenge certainty
A June 21 report compared a single-model answer about college with adversarial exchanges on the same motion. The revealing issue was the difference between “most students” and graduates.
That was a finding about argument and framing—not proof that college is or is not worthwhile.
Read the experiment02 / Test the setup
Early experiments appeared to show a persistent advantage for the negative side. Episode 14 revisited that interpretation after later results changed sharply.
Publishing a correction matters more than protecting an attractive explanation.
Read or hear the correction03 / Inspect the exchange
An April 11 report recorded 56 concessions in the first tournament after a debate-model change, following 16 tournaments with none.
This was a setup-specific observation. A concession alone does not establish factual accuracy.
Read the historical reportFrom experiments to your decision
Your decision is not another number in a network.
Compelle is a white-glove service: we work with you to frame the question, examine opposing cases, and make the assumptions and remaining uncertainty legible.
Work with usWorking papers · June 2026
Two companion manuscripts grew out of the historical network research. These are working drafts, with author identities withheld in the linked versions for anonymous review—not claims of journal acceptance or independent validation of our service.
Mechanism design
Costly Registration as Sybil-Resistance in Graded-Reward Networks.
The paper studies the tension between open entry, resistance to duplicate identities, and performance-based rewards within a specified class of fixed-pool mechanisms. Its conclusions depend on those assumptions.
Read the working paper (PDF)Governance & falsifiability
A Falsification Protocol for Parameter Disputes in Permissionless Networks.
A proposed framework for turning disagreement into a testable prediction and a pre-registered experiment. A theoretical mechanism, not a demonstrated guarantee that disputes will resolve correctly.
Read the working paper (PDF)Make it concrete
Start with a curated sample. Then bring us the question that matters to you.