Fetching from the wire…
01
02
03
04
05
06
07
08
Research2026-08-11 · source-backed
SpecPath (arXiv 2608.09799) holds repository, final contract, verifier, agent system, and execution budget constant, varying only the revision history by which a spec became final. Same end state, different path there. Thirty-five percent failed on at least one contract-equivalent history, and aggregate accuracy across paths looked nearly identical, which means path sensitivity is completely invisible to standard benchmarking. Agents are partly following surface cues from how a request evolved. Evaluate against messy multi-turn requirement histories, not the tidy ticket you'd write in hindsight.
Each link below shares sources, entities, or timing with this story.
Shared entity: SpecPath / Same source / Shared topic / Tension
Both cover SpecPath; cite the same source (SpecPath (arXiv 2608.09799)); overlapping topics (accuracy, against, agent, aggregate, benchmarking).
Shared entity: Same / Same source domain / Shared topic / Earlier coverage
Both cover Same; reported by the same outlet (arxiv.org); overlapping topics (agent, became, only).
Both cover Same; reported by the same outlet (arxiv.org); overlapping topics (against, agent).
Both cover Same; reported by the same outlet (arxiv.org); overlapping topics (accuracy, agent).
Same source domain / Shared topic / Tension
Reported by the same outlet (arxiv.org); overlapping topics (against, agent, budget, change); pushes against this story (against).
Reported by the same outlet (arxiv.org); overlapping topics (against, agent, budget, change); pushes against this story (but).
Reported by the same outlet (arxiv.org); overlapping topics (budget, change, history, only); pushes against this story (but).
Reported by the same outlet (arxiv.org); overlapping topics (against, agent, budget, change); pushes against this story (but).