card: 265->268 MB (hub-displayed size)

README.md

@@ -64,7 +64,7 @@ never on this benchmark.
 deliberately adversarial: every positive uses phrasing templates the model never saw,
 gold is multi-label, and 24% of cases are near-miss negatives (third-party probes,
 mechanism/definition traps) built to punish keyword matching. For scale on the *same*
-benchmark: **mistral:7b, a 4.4 GB LLM, scores 0.495 macro-F1** — this 265 MB model beats
+benchmark: **mistral:7b, a 4.4 GB LLM, scores 0.495 macro-F1** — this 268 MB model beats
 it while running every turn on-device.
 
 | metric | value | mistral:7b, same benchmark |
@@ -90,7 +90,7 @@ LLM's actual answer on those exact cases.
 | **hybrid: this model + llama3.2** | **0.500** | **0.508** | **0.976** | 0.146 | 16.4% |
 
 The result that matters for low-RAM machines: **a 2 GB-LLM machine running the hybrid
-(0.500) outperforms a 4.4 GB-LLM machine running LLM-only (0.495)** — the head is 265 MB
+(0.500) outperforms a 4.4 GB-LLM machine running LLM-only (0.495)** — the head is 268 MB
 and identical on every hardware tier, so only ~1/6th of traffic degrades with weaker
 hardware.