End-to-end evaluation scores conflate three sources of improvement: base capability, harness-updating quality, and harness-benefit, leaving it unclear which models produce useful updates or benefit most from them

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

• harness-updating is flat in base capability: models across capability tiers produce updates that yield similar gains, and even the Qwen3.5-9B evolver induces gains comparable to Claude Opus 4.6quote0.801
  Verbatim summary of first major finding from conclusion
• Harness-updating capability is flat in base capability: models from different capability tiers produce harness updates that lead to surprisingly similar gainsclaim0.800
  First major claim of the paper, supported by narrow spread across evolvers and case study
• Harness-benefit is non-monotonic in base capability: weak-tier models benefit little, mid-tier models benefit most, and strong-tier models benefit less than mid-tierclaim0.792
  Second major claim of the paper, supported by Δbenefit measurements across six models on three benchmarks
• Strong-tier models benefit less from harness evolution because they already solve many tasks under the initial harness, leaving less room for improvement (ceiling effect)hypothesis0.760
  Explanation offered for why high-base-capability models show lower Δbenefit
• For feedback to be meaningful, the end-result must be unpredictable; a predetermined end-state shuts off the possibility of adaptation.claim0.759
  Unpredictability is a necessary condition for genuine adaptation.
• Baseline scores blend together at least three different things: latent reflective capacity, default accessibility, and stability of access.claim0.759
  Conceptual decomposition arising from the data showing different models dissociate these traits
• what explains why weak-tier models with the most performance headroom benefit least from harness evolution?question0.750
  In-depth diagnostic question addressed by the two failure mode analysis
• Capability budget should be allocated to the task-solving agent rather than the evolver, since harness-updating varies by at most 3.1 pp across evolversclaim0.749
  Primary design recommendation derived from harness-updating flatness finding