Add TL;DR, update banner

README.md

@@ -121,11 +121,15 @@ Per-version compatibility matrices split APIs into *stable* ones, whose signatur
 
 <p align="center"><img src="https://huggingface.co/datasets/JetBrains/LibEvoBench/resolve/main/assets/pipeline.png" alt="Pipeline: mine GitHub repos that pin a version, resolve call sites to versioned APIs, build per-release compatibility matrices, generate the three tasks, classify every error by matrix lookup and rank by SEUS." width="100%"></p>
 
-## Results at a glance
+## TL;DR
+
+- **Even SOTA models lag on version-specific knowledge**: a persistent 5–10 point gap between stable and evolving APIs that neither scale nor training recency closes.
+- **Models don't store the version as a dimension of library knowledge**: giving the target version yields no gain, while giving documentation adds +10–20 points.
+- **Failures are temporal confusion, not fabrication:** models fall back on version-agnostic, high-frequency symbols that hide the gap.
 
 <p align="center"><img src="https://huggingface.co/datasets/JetBrains/LibEvoBench/resolve/main/assets/results.png" alt="Stable versus evolving API scores for 13 models, ranked by SEUS. Every model scores lower on evolving APIs." width="100%"></p>
 
-All 13 models score lower on evolving APIs than on stable ones, and naming the target version in the prompt gives no measurable gain. The interactive leaderboard is on the [project page](https://jetbrains-research.github.io/LibEvoBench/#leaderboard).
+The interactive leaderboard is on the [project page](https://jetbrains-research.github.io/LibEvoBench/#leaderboard).
 
 ## Loading the data