v1.0 fixes: consistent risk reporting, input handling, solvers, benchmarks and Docker

.github/workflows/docker-publish.yml

@@ -22,6 +22,10 @@ jobs:
       - name: Checkout
         uses: actions/checkout@v4
 
+      # QEMU emulates arm64 so the image can also be built for it.
+      - name: Set up QEMU
+        uses: docker/setup-qemu-action@v3
+
       # Buildx is needed for multi-arch builds and layer caching.
       - name: Set up Docker Buildx
         uses: docker/setup-buildx-action@v3
@@ -54,7 +58,6 @@ jobs:
           labels: ${{ steps.meta.outputs.labels }}
           cache-from: type=gha
           cache-to: type=gha,mode=max
-          # Publish for x86_64 Linux hosts. Add ,linux/arm64 here if you
-          # also want ARM (e.g. M-series Mac / Pi 4+) — it roughly doubles
-          # build time because there's no native arm64 runner on public GH.
-          platforms: linux/amd64
+          # x86_64 hosts and ARM hosts (e.g. Apple silicon Macs). The arm64
+          # build runs under QEMU, which roughly doubles the build time.
+          platforms: linux/amd64,linux/arm64

DOCKER.md

@@ -14,6 +14,8 @@ Run the pre-built image from the GitHub Container Registry (use a release tag su
 docker run -p 5000:5000 ghcr.io/kiguli/scen-opt:latest
 ```
 
+The image is built for both x86_64 and ARM (e.g. Apple silicon) machines.
+
 Or build the image yourself:
 
 ```bash

Dockerfile

@@ -30,4 +30,7 @@ USER appuser
 
 EXPOSE 5000
 
-CMD ["gunicorn", "--bind", "0.0.0.0:5000", "--workers", "2", "--timeout", "7200", "app:app"]
+# Default worker count. Override with -e GUNICORN_CMD_ARGS="--workers 4".
+ENV GUNICORN_CMD_ARGS="--workers 2"
+
+CMD ["gunicorn", "--bind", "0.0.0.0:5000", "--timeout", "7200", "app:app"]

README.md

@@ -10,7 +10,7 @@ The tool is implemented as a Python Flask web application with a modern JavaScri
 
 ## Installation
 
-**Prerequisites:** Python 3.12+ and pip.
+**Prerequisites:** Python 3.10 or newer (3.12 recommended) and pip.
 
 ```bash
 # Clone the repository
@@ -26,7 +26,7 @@ python3 app.py
 
 The server starts at `http://127.0.0.1:5000`. Open this URL in a browser to access the web interface.
 
-**Optional:** For best performance on SDP problems, install [MOSEK](https://www.mosek.com/) (free academic license available). See [MOSEK Support](#mosek-support) for license configuration.
+**MOSEK:** [MOSEK](https://www.mosek.com/) is the recommended solver: it is the most stable for these problems and was used for all published results. Its Python package is installed by `requirements.txt`, but it needs a licence (free for academics). See [MOSEK Support](#mosek-support) for license configuration.
 
 ## Quick Start
 
@@ -94,7 +94,7 @@ where the scenario constraints are Linear Matrix Inequalities (LMIs) parameteriz
 | Scenarios | delta_i | Sampled uncertainty realizations (N samples) |
 | Slack penalty | rho | Penalty on constraint relaxation (rho = 0 means hard scenario constraints) |
 | Regularization | tau | Regularization strength toward reference point x_ref |
-| Norm type | p | Norm for regularization (1, 2, inf, fro, nuc) |
+| Norm type | p | Norm for regularization (any p ≥ 1, inf, or fro) |
 | Confidence | beta | Confidence parameter for risk bounds (e.g. beta = 0.01 for 99% confidence) |
 
 ### Risk Quantification
@@ -170,7 +170,6 @@ Each benchmark follows a standardized layout:
 │   ├── program_numeric.json   Upload-Program definition (numeric mode)
 │   ├── scenarios.csv   Sampled uncertainty realizations
 │   ├── scenarios_numeric.csv  Per-row-flattened matrices for numeric mode
-│   ├── c.csv           Objective vector
 │   └── ...             Additional problem-specific data
 └── results/
     ├── metrics.json    Solver output (cost, risk bounds, complexity)
@@ -183,13 +182,13 @@ Each benchmark follows a standardized layout:
 ```bash
 cd benchmarks/<benchmark_name>
 
-# Regenerate scenarios (optional, data already provided)
+# Regenerate scenarios (optional, data already provided; some generators need extra packages, listed at the top of the script)
 python generate.py
 
-# Solve the program
+# Solve the program (MOSEK by default; e.g. --solver CLARABEL without a MOSEK licence)
 python run.py
 
-# Generate the paper figure
+# Generate the paper figure (needs matplotlib: pip install matplotlib)
 python plot.py
 ```
 
@@ -269,7 +268,7 @@ Identifies the support list of a solved scenario program: the constraints whose
 
 - **License upload:** When MOSEK is selected as the solver, a modal prompts you to upload your `mosek.lic` file. The license is **not** stored on the server.
 - **Browser caching:** After the first upload, the license is cached in your browser's session storage. Subsequent solves skip the modal automatically. A cache indicator and "Clear" button appear below the solver dropdown.
-- **Local license:** If MOSEK is installed locally with a system license (e.g., at `~/mosek/mosek.lic`), click "Skip" in the modal to use it directly.
+- **Local license:** If MOSEK is installed locally with a system license (e.g., at `~/mosek/mosek.lic`), click "I have a local license" in the modal to use it directly.
 - **Concurrent users:** Each MOSEK solve with an uploaded license runs in an isolated subprocess with the license written to an ephemeral `/tmp` directory (RAM-backed, auto-purged). Multiple users can solve simultaneously without conflicts.
 
 ## Docker