Make README benchmark summary clearer

Author: CrazyAngelm

README.md

@@ -1,38 +1,49 @@
 # MegaQuant KV-Cache
 
-MegaQuant KV-Cache is a **CPU/Python research proof-of-concept** for metadata-aware low-bit KV-cache quantization.
+[![Research PoC](https://img.shields.io/badge/status-research%20PoC-blue)](#scope)
+[![CPU/Python](https://img.shields.io/badge/benchmark-CPU%2FPython-lightgrey)](#scope)
+[![Metadata aware](https://img.shields.io/badge/accounting-metadata--aware-green)](#what-effective_bits_per_dim-means)
 
-It compares MegaQuant methods against locally tested TurboQuant / RotorQuant / IsoQuant / PlanarQuant-style Python baseline implementations in a narrow GPT-2 KV-cache quality benchmark.
+**Low-bit KV-cache compression experiments with honest metadata accounting.**
 
-> Scope warning: this repository does **not** claim production speed, real VRAM savings, CUDA kernel quality, or general superiority across LLMs. The headline is an observed result in the benchmark described below.
+## At a glance
 
-## Current headline
+In a local GPT-2 CPU/Python fake-quant benchmark, MegaQuant's best current point gives:
 
-Current strongest method in this repo's benchmark:
+| What you care about | Result | Compared with |
+|---|---:|---|
+| Modeled KV payload size | **19.6% of FP16** | **5.11x compression / 80.4% saving** vs FP16 |
+| Attention-output quality | **+11.0% higher** | vs local `RotorQuant-3b` baseline (`0.942270` vs `0.848665`) |
+| Memory cost vs RotorQuant-3b | **+4.35% more** | `3.130399` vs `3.000000` effective bits/dim |
+
+Main method:
 
 ```text
 affine_seven_level_3bit_g64_meta4
+3.130399 effective bits/dim
+0.942270 attention-output cosine
 ```
 
-Full GPT-2 CPU/Python fake-quant benchmark result:
+Need lower memory? The 2-bit Hadamard variant uses **24.8% less modeled memory than local RotorQuant-3b** (`2.255399` vs `3.000000` bits/dim) while landing in the same attention-output-cosine range in this benchmark (`0.851023` vs `0.848665`).
 
-```text
-effective_bits_per_dim = 3.130399
-attn_out_cos_mean      = 0.942270
-score_cos_mean         = 0.997257
-```
+## Scope
+
+This repository is a **research proof-of-concept**, not a production inference engine.
 
-Within this repository's current benchmark setup, this is the best observed quality/compression tradeoff among the tested methods.
+The numbers above are:
 
-## Plain-language comparison
+- from a narrow GPT-2 KV-cache quality benchmark,
+- CPU/Python fake-quant results,
+- based on modeled `effective_bits_per_dim` including declared metadata,
+- comparisons against local Python baseline implementations.
 
-For readers who just want the headline numbers:
+They are **not** claims about CUDA kernels, real VRAM, decode throughput, or general superiority across LLMs.
+
+## Related repository
 
-- vs **FP16 KV-cache payload**: `affine_seven_level_3bit_g64_meta4` uses about **19.6%** of the modeled payload, i.e. about **80.4% memory saving** and about **5.11x compression**.
-- vs local **RotorQuant-3b** baseline in this benchmark: it uses about **4.35% more modeled memory** (`3.130399` vs `3.000000` effective bits/dim), but gives about **11.03% higher attention-output cosine** (`0.942270` vs `0.848665`).
-- if you want a lower-memory point instead of the main quality point, `hadamard_affine_four_level_2bit_g64_meta8` uses about **24.8% less modeled memory** than local `RotorQuant-3b` (`2.255399` vs `3.000000` bits/dim) while showing slightly higher attention-output cosine in this benchmark (`0.851023` vs `0.848665`).
+RAG/vector-index companion project:
 
-These are benchmark-local modeled payload comparisons, not production VRAM or kernel-throughput claims.
+- https://github.com/CrazyAngelm/megaquant-rag-compress
 
 ## What `effective_bits_per_dim` means
 
@@ -44,12 +55,6 @@ Small implementation overheads such as padding, headers, estimator state, and ru
 
 For simulated `meta4`/`meta8` methods, the public tables add a small conservative term for shared metadata-range parameters. This is still a modeled storage budget, not a packed-kernel measurement.
 
-## Related repository
-
-RAG/vector-index companion project:
-
-- https://github.com/CrazyAngelm/megaquant-rag-compress
-
 ## Benchmark setup
 
 ```text