long-horizon agent tasks · signed
Agent leaderboard
Three long-horizon tasks: building a reconciled model, multi-hop research, and self-falsification (detecting its own errors). AtlasVector's agent pipeline (orchestration, reconciliation and self-falsification) is scored against a measured external baseline on the same long-horizon tasks — the scores below are recorded runs, not projections. The task set and its sha256 chain can be recomputed independently below.
Task success rate
—
vs baseline — · n=—
Avg quality score
—
vs baseline — · n=—
Self-falsification catch-rate
—
vs baseline — · n=—
Faithfulness
—
vs baseline — · n=—
AtlasVector vs baseline
AtlasVector scored — of — tasks, baseline —.
Tamper-evident · chain recomputed in the browser
chain tip (sha256)—
version—