docs: C engine exists now — retire the no-engine limitation, add torch-free inference section

README.md

@@ -351,14 +351,23 @@ Two of these reversed conclusions that would otherwise have shipped on intuition
   `live_parallel` 0.0 on BFCL — repeated instantiations of one schema, as
   opposed to calls the query motivates in sequence.
 - **768-token context.** 151 of 3,641 BFCL rows (4.1%) do not fit and score as misses.
-- **Deployment.** 48.12M parameters is ~11.5MB at 2-bit and ~92MB at bf16, but
-  what ships here is the 184MB fp32 checkpoint and there is no on-device
-  inference engine. The size figure is a property of the parameter count, not of
-  a runnable microcontroller artifact.
+- **Microcontrollers.** ~11.5MB at 2-bit is a property of the parameter count,
+  not a shippable artifact: 2-bit would need quantization-aware retraining this
+  model never had. The smallest thing that actually runs is the 48MB int8
+  build of the C engine below — Pi-class and up, not Cortex-M.
 
 Scale is the honest explanation for most of this: ~1B unique tokens, no
 pretraining phase, a corpus deliberately spent on depth instead of breadth.
 
+## Torch-free inference (C engine)
+
+The repo ships [cengine/](https://github.com/nikshepsvn/thimble/tree/master/cengine),
+the full decoder — tokenizer, trunk, grammar walk, name head — in one
+dependency-free C file. 48MB int8 weights, ~20ms load, ~350ms per full request
+on an Apple M3, and fp32 output is byte-identical to this checkpoint run
+through the Python stack on 100/100 checked eval rows. Export the weights from
+this repo's checkpoint with `cengine/export.py`.
+
 ## Model details
 
 | | |