Release CanRun FPS Model v1.0

README.md

@@ -133,6 +133,19 @@ On its own, the model turns hardware scores and Steam store specs into an FPS nu
 
 Where the app costs accuracy: its store-spec fallback gives the model fewer inputs than the `inference.py` examples below (no VRAM, release year or genre). On F1 24, the app is 54% off while the bare model is 29% off. The measured curves also carry the bias of the hardware they were measured on. They overestimate older cards in OpenML (Rainbow Six Siege +139%, ARK +116%) and underestimate new ones in TechSpot (Forza Horizon 6 at High, -56%).
 
+### Compared with other FPS estimators
+
+| Tool | Published accuracy | Test data |
+| :--- | :--- | :--- |
+| **CanRun (this model plus app)** | 15.2% to 30.7% median error | TechPowerUp, TechSpot and OpenML, listed above |
+| [FPSGauge](https://fpsgauge.com/methodology) | 26% median error | 1,990 independent results |
+| [FPSHQ](https://fpshq.com/en/methodology/) | 41.6% median error | 1,341 independent review results |
+| [PCBottleneck](https://www.pcbottleneck.com/methodology) | 21.6% mean error | Its own estimate path |
+| [MyRigCheck](https://myrigcheck.games/accuracy), TechSavant | about 15% (claimed) | No public test set |
+| Ramkumar et al. 2025 | R² 0.99 | OpenML 42737, in-sample |
+
+Each tool was tested on different data, so this compares the size of the error, not a head-to-head result. Among the tools that publish error on independent data, CanRun is in line with or better than the others, and it is the only one that runs fully on the user's PC with an open model.
+
 ---
 
 ## Mathematical Architecture