diff --git a/README.md b/README.md
index 4c9d97f..140f19c 100644
--- a/README.md
+++ b/README.md
@@ -1,33 +1,66 @@
-# EVA: A Long-Context Generative Foundation Model for Versatile RNA Design
-
-
-
-
-
-[Paper](https://www.biorxiv.org/content/10.64898/2026.03.17.712398v1) ·
-[Model weights](https://huggingface.co/GENTEL-Lab/EVA) ·
-[OpenRNA data](https://huggingface.co/datasets/GENTEL-Lab/OpenRNA-v1-114M) ·
-[Website](https://evabio.net/) · [Apache-2.0 license](LICENSE)
-
-EVA is a generative RNA foundation model for sequence scoring and conditional
-sequence modeling. The released model family includes a 1.4B-parameter
-mixture-of-experts model with an 8,192-token context window, trained on OpenRNA v1.
-This repository contains model code, command-line tools and paper reproduction
+
+
+
+
+
+
EVA
+
A Long-Context Generative Foundation Model for Versatile RNA Design
+
+EVA brings RNA sequence scoring and conditional sequence modeling into one
+framework, with public model code, checkpoints and selected paper reproduction
workflows. Model weights and large datasets are downloaded separately.
-## Start here
-
-| I want to… | Start with | Requirements |
-|---|---|---|
-| Check a bundled result without a GPU | [CPU result recalculation](docs/REPRODUCTION.md#cpu-recalculate-stored-predictions) | Python 3.10/3.11 only |
-| Run a benchmark from model weights | [135-sequence Milena example](docs/REPRODUCTION.md#gpu-run-the-complete-milena-benchmark) | Linux/NVIDIA GPU; Docker recommended |
-| Use scoring or generation | [CLI guide](docs/USAGE.md) | Compatible GPU runtime and a checkpoint |
-| Train or fine-tune | [Training workflow](docs/REPRODUCTION.md#training-and-fine-tuning) | Complete source checkout and GPU runtime |
-| Find a paper result or comparison model | [Paper workflows](docs/PAPER_WORKFLOWS.md) | Requirements vary by workflow |
-
-## Quick Start
-
-Clone the repository and recalculate the bundled Milena benchmark metric:
+
+
+
+
+
+
+
+## Get started
+
+
+
+
+
01 Explore on CPU
+
Recalculate a bundled benchmark result from saved predictions.
+
+### Quick start
+
+A first result with Python's standard library:
```bash
git clone https://github.com/GENTEL-Lab/EVA.git
@@ -36,15 +69,25 @@ python3 scripts/reproduce_historical_benchmark.py \
--artifact-only --output results/milena_stored
```
-This standard-library-only example writes `report.json` and
-`sequence_label_audit.csv` for all 135 stored predictions. The expected Spearman
-correlation is **0.8360456283218484**, and the command returns 0. It recalculates
-an archived result; the [GPU example](docs/REPRODUCTION.md#gpu-run-the-complete-milena-benchmark)
-performs new model inference. Use a new output directory when rerunning.
+**Expected:** 135 stored predictions, Spearman **0.8360456283218484**, exit code `0`.
+The output directory contains `report.json` and `sequence_label_audit.csv`.
+This recalculates an archived result; the GPU path below runs new inference.
+Use a new output directory when rerunning.
+
+
+
+## Reproduce the paper
-## Installation
+The representative workflow connects **pinned checkpoint → 135 input sequences
+→ new predictions → metrics & plots**. Inputs, model files and the scoring
+protocol are bound to version and checksum records.
-For model inference, build and enter the runtime from the repository root:
+
+
+
+Run the GPU example · Environment setup and complete inference command
+
+From the repository root, build and enter the runtime **on the host**:
```bash
mkdir -p checkpoint results
@@ -53,14 +96,7 @@ docker run --rm -it --gpus device=0 --name eva-repro \
-v "$PWD":/eva -w /eva eva:local bash
```
-Then follow the GPU example below **inside the container**. Full setup, source
-installation and HPC instructions are in [Installation](docs/INSTALLATION.md).
-
-## Reproducing the paper
-
-The complete representative example downloads a pinned public checkpoint,
-checks its files, scores all 135 Milena sequences, calculates the metric and
-writes diagnostic plots:
+Then run **inside the container**:
```bash
python scripts/download_reproduction_checkpoint.py \
@@ -69,150 +105,100 @@ python scripts/reproduce_milena.py \
--checkpoint checkpoint/EVA_1.4B_CLM --output results/milena_fresh
```
-Observed new-inference Spearman: **0.8394237924835843**; archived reference:
-**0.8360456283218484**. The difference remains visible in the output and is not
-an execution failure. Add `--strict-reference` only to request a nonzero exit
-when comparison at the stored reference precision fails. Input, dependency,
-scoring and plotting failures always remain errors.
+See [Installation](docs/INSTALLATION.md) for source installs, runtime versions
+and Singularity / Apptainer instructions.
-The recorded A100 run scored the assay in about **12 seconds**, with **2.94 GiB**
-peak PyTorch-allocated GPU memory. These are measurements for this example,
-not minimum hardware requirements or estimates for other workloads; downloads,
-loading, hashing and plotting add overhead.
+
-- [Reproduction guide](docs/REPRODUCTION.md): commands, inputs, outputs and validation scope.
-- [Reviewer requirements](docs/REVIEWER_REQUIREMENTS.md): six software requirements and supporting evidence.
-- [Paper workflow coverage](docs/PAPER_WORKFLOWS.md): supported entry points and missing resources.
-- [Validation and release status](docs/RELEASE_STATUS.md): dated evidence and publication state.
+
+Expected results & measured runtime · Reference comparison, exit codes and hardware
-## Model and data resources
-
-The [resource index](docs/RESOURCES.md) links checkpoints, OpenRNA and deposited
-analysis data. The representative example fixes the model revision and SHA256
-checksums in its manifest. Historical competitor resources and available
-execution entry points are listed in [comparison models](docs/OFFICIAL_MODEL_RESOURCES.md).
-
-## Repository layout
+| Measurement | Recorded result |
+|---|---|
+| New-inference Spearman | **0.8394237924835843** |
+| Archived reference | **0.8360456283218484** |
+| Scoring time | About **12 seconds** for all 135 sequences |
+| Peak PyTorch-allocated GPU memory | **2.94 GiB** |
+| Environment | One NVIDIA A100-SXM4-80GB; batch size 1; pinned Docker runtime |
+
+Default mode returns `0` when inference, metrics and plots complete successfully.
+The reference difference remains visible. Add `--strict-reference` to return `2`
+when comparison at the stored reference precision fails. Input, dependency,
+scoring and plotting failures remain errors in either mode.
+
+These are measurements for this example, not minimum hardware requirements
+or estimates for other workloads. Downloads, loading, hashing and plotting
+add overhead. A successful example does not establish reproduction of every
+paper experiment. See [dated validation records](docs/RELEASE_STATUS.md).
+
+
+
+[Protocol & outputs](docs/REPRODUCTION.md) ·
+[Paper workflow coverage](docs/PAPER_WORKFLOWS.md) ·
+[Reviewer requirements](docs/REVIEWER_REQUIREMENTS.md)
+
+
+
+## Documentation
+
+
-
+ 
+