Completion-rate distribution.
How task outcomes vary across comparable cohorts.
Awaiting publication-gated evidenceHow agents actually use software.
A research program for task completion, failure surfaces, controlled treatments, and the gap between Eval behavior and production outcomes.
Publication standard
Research begins after the evidence gate.
Research will publish findings only when the task, sample, provenance, time window, method, and limitations are explicit.
Research agenda
Ten questions, published only when the evidence is ready.
These are the chapters GrowthOS is building toward. They are an agenda, not findings, charts, or implied benchmark results.
Where agent tasks fail.
The product surfaces and barriers that stop completion.
Awaiting publication-gated evidenceProduction versus controlled behavior.
Where live journeys and managed runs agree or diverge.
Awaiting publication-gated evidenceCompletion by product category.
Category patterns only after task definitions are comparable.
Awaiting publication-gated evidencePerformance across agents and models.
Agent and model context without universal rankings.
Awaiting publication-gated evidenceDocumentation, authentication, API, and MCP barriers.
The structural friction found across core surfaces.
Awaiting publication-gated evidenceEffect of product treatments.
What changes under controlled docs, access, and interface variants.
Awaiting publication-gated evidenceConfidence and unresolved activity.
How uncertainty changes the interpretation of production evidence.
Awaiting publication-gated evidenceBefore/after completion lift.
Production change only after a verified time window exists.
Awaiting publication-gated evidenceResearch limitations.
Coverage, evaluator fallibility, samples, and unobserved behavior.
Awaiting publication-gated evidenceEvidence available now
Inspect task-level evidence before aggregate claims.
Illustrative entries stay explicitly separate from future production studies.