Develop to main

.dockerignore

@@ -0,0 +1,13 @@
+# The image is built from pom.xml and src/ only; keep everything else out of the build context
+.git
+.idea
+target
+docs
+vectikz
+evaluation
+tools
+data
+*.log
+*.md
+.env
+.env.*

.env.example

@@ -1,5 +1,6 @@
 # UDAV Environment Configuration Example
-# Copy this file to .env and modify values as needed
+# Optional: docker-compose.yml has defaults for everything. Copy this file to .env only to
+# override them; the values below are the ones used for large corpus imports.
 
 # ============================================
 # Database Configuration

.github/workflows/ci.yml

@@ -0,0 +1,19 @@
+name: CI
+
+on:
+  push:
+    branches: [main, develop]
+  pull_request:
+
+jobs:
+  test:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/setup-java@v4
+        with:
+          distribution: temurin
+          java-version: '21'
+          cache: maven
+      # The browser tests are opt-in (profile browser-tests) and stay out of CI
+      - run: mvn -B test

Dockerfile

@@ -30,7 +30,7 @@ RUN addgroup -g 1000 appuser && \
 
 # Copy JAR from builder stage. --chown here instead of a later "chown -R /app": a chown after
 # COPY rewrites the whole 1.2 GB jar into a second image layer.
-COPY --from=builder --chown=appuser:appuser /build/target/UDAV-1.0.jar app.jar
+COPY --from=builder --chown=appuser:appuser /build/target/UDAV-2.0.0.jar app.jar
 
 # Bundled pipelines and JSON sources (the demos and the evaluation workload). The importers read
 # them from disk, not from the classpath, so the container needs its own copy. Point
@@ -73,7 +73,7 @@ ENV DUUI_IMPORTER_CAS_POOL_SIZE="8"
 
 # Application Configuration
 ENV APP_INPUT_DIR="/app/data/input"
-ENV SROUCE_BUILDER="false"
+ENV SOURCE_BUILDER="false"
 ENV PIPELINE_IMPORTER="true"
 ENV PIPELINE_IMPORTER_FOLDER="/app/pipelines"
 ENV JSON_IMPORTER_FOLDER="/app/sourcefilesJSON"

README.md

@@ -1,5 +1,5 @@
 <div align="center">
-  <a href="/LICENSE"> <img src="https://img.shields.io/github/license/Texttechnologylab/Unified-Dynamic-Annotation-Visualizer"></a>
+  <a href="/LICENSE.txt"> <img src="https://img.shields.io/github/license/Texttechnologylab/Unified-Dynamic-Annotation-Visualizer"></a>
   <a href="https://github.com/texttechnologylab/Unified-Dynamic-Annotation-Visualizer/releases"> <img src="https://img.shields.io/github/v/release/Texttechnologylab/Unified-Dynamic-Annotation-Visualizer"></a>
   <a href="https://lrec2026.info/"> <img src="https://img.shields.io/badge/conference-LREC--2026-4b44ce.svg"></a>
   <a href="https://aclanthology.org/2026.lrec-1.533.pdf"> <img src="https://img.shields.io/badge/paper-LREC--2026-fb44ce.svg"></a>
@@ -79,7 +79,7 @@ pipeline are exported concurrently inside a single page. Both are tunable:
 | `TOMCAT_MAX_THREADS` | `200` | Servlet worker pool that the budget in the note below is expressed against. |
 
 > [!NOTE]
-> The headless page calls back into this same application for `/api/data` and `/api/convertions/*`.
+> The headless page calls back into this same application for `/api/data` and `/api/conversions/*`.
 > Batch-attributable worker usage is bounded by roughly
 > `(max-sessions + max-waiters) + 6 * max-sessions` (Chromium caps connections per origin at ~6),
 > so keep that under `server.tomcat.threads.max` (pinned to `200` by default).
@@ -104,7 +104,7 @@ pipeline are exported concurrently inside a single page. Both are tunable:
 
 #### TeX export: VecTikZ
 
-The `tex` format of the toolbar and of the batch API is produced by `POST /api/convertions/tikz`.
+The `tex` format of the toolbar and of the batch API is produced by `POST /api/conversions/tikz`.
 Widgets with a native LaTeX representation (Table, Highlight Text) generate it directly; every
 SVG-drawing widget is converted by **VecTikZ** (`org.texttechnologylab.udav.widgets.svgtolatex`),
 an SVG-to-TikZ converter that turns shapes, paths, text, gradients, patterns, markers, clip paths
@@ -129,6 +129,10 @@ the view shows a pager, the data API reports the group as `meta.total` / `meta.i
 "Export all" toolbar entry and the batch API's `bulk=true` export every page. Only JSON/XML
 sources can be grouped.
 
+The corpus filter of a view selects documents by file, by tag and by import date: tags are the
+`collectionId` of the `DocumentMetaData` the DUUI importer stores with each document, the date is
+the import time of the document's sofa.
+
 `CategoryNumber` and `TextFormatting` evaluate the filter lists `filesWhitelist` / `filesBlacklist`
 and `categoriesWhitelist` / `categoriesBlacklist`, set on the generator or on its source.
 
@@ -161,11 +165,17 @@ publishes the app on port 18080 with measurements switched on) and run
 `mvn test -Dtest=BatchExportEvaluationIT -DUDAV_BASE_URL=http://localhost:18080`, adding
 `-Dudav.eval.bugHunt=true` for the smoke variant.
 
-The report of the campaign the paper is based on is `evaluation/evaluation-report-2026-08-26.txt`.
+The paper is based on the three JVM runs of 2026-10-31, `evaluation/evaluation-report-2026-10-31-run1.txt`
+to `-run3.txt`; `evaluation/evaluation-report-2026-08-26.txt` is the earlier single-run campaign.
 The script writes `evaluation-report.txt`, whose first section states whether the numbers are usable;
 section 5 holds the LaTeX rows of the paper's tables and section 7 the raw per-run data. Harness
 knobs (`-Dudav.eval.analysisRuns=15`, `-Dudav.eval.pipelines=P1,P5`, ...) are passed through.
-`tools/make_eval_pipelines.py` regenerates the pipelines and their JSON sources.
+`tools/make_eval_pipelines.py` regenerates the pipelines and their JSON sources, and
+`tools/cost_model.py evaluation/evaluation-report-2026-10-31-run*.txt` recomputes the workload and
+cost-model tables of the paper, the bootstrap intervals, the marginal resource costs and the other
+numbers of the evaluation section from section 7 of the reports. `tools/run-evaluation.sh --docker --jvm-runs 3` runs the campaign three
+times with a restart of UDAV in between and writes one report per run; `tools/cost_model.py` pools
+any number of reports.
 
 `mvn test` runs the unit tests. The tests tagged `browser` launch Playwright's own Chromium (downloaded
 on the first run, about 1 GB together with Firefox and WebKit) and are opt-in: `mvn test -Pbrowser-tests`.
@@ -201,26 +211,42 @@ startup, so the pipeline list is populated right away.
 
 ### Quick Start (Docker Compose)
 
-1. **Clone the repository:**
+Nothing to configure: `docker-compose.yml` carries defaults for everything.
+
+1. **Clone and start:**
 
    ```bash
    git clone https://github.com/texttechnologylab/Unified-Dynamic-Annotation-Visualizer.git
    cd Unified-Dynamic-Annotation-Visualizer
+   docker compose up -d
    ```
 
-2. **Create your `.env` file** by copying the provided example:
+   The first start builds the image (a Maven build inside Docker, several minutes depending on the
+   connection); later starts take seconds. The web UI is available at
+   [http://localhost:8080](http://localhost:8080) once the container is healthy. The bundled demo
+   and evaluation pipelines with their JSON sources are imported and built in the first minute
+   after that; pipelines whose sources are UIMA annotation types stay empty until a corpus is
+   imported (see below).
+
+2. **Check it:**
 
    ```bash
-   cp .env.example .env
+   tools/smoke-test.sh
    ```
 
-3. **Start the application:**
+   Needs only bash and curl: it waits for the application and its pipelines, reads widget data
+   through the data API, exports a whole pipeline through the batch API and converts an SVG to
+   TikZ. On Windows run it from Git Bash or WSL, or open the UI and use the export menu of a
+   widget instead.
 
-   ```bash
-   docker compose up -d
-   ```
+3. **Stop it** with `docker compose down`, or `docker compose down -v` to drop the database as well.
 
-   This starts PostgreSQL and the UDAV application. The web UI is available at [http://localhost:8080](http://localhost:8080) once the container is healthy (usually within 30-60 seconds). The image ships the demo and evaluation pipelines with their JSON sources and imports them at startup; set `PIPELINE_IMPORTER=false` or point `PIPELINE_IMPORTER_FOLDER` / `JSON_IMPORTER_FOLDER` at mounted folders to import your own instead. The default memory settings suit a laptop; `.env.example` lists the values for large corpus imports.
+The defaults cap the Java heap at 4 GB and give PostgreSQL 1 GB of shared buffers; with the bundled
+pipelines built, the two containers use about 1.5 GB of memory together, which suits a laptop. To change anything, create a `.env` next to `docker-compose.yml` with the variables to
+override; `.env.example` documents them with the values used for large corpus imports. Set
+`PIPELINE_IMPORTER=false`, or point `PIPELINE_IMPORTER_FOLDER` / `JSON_IMPORTER_FOLDER` at mounted
+folders, to import your own pipelines instead of the bundled ones. After a code change,
+`docker compose up -d --build` rebuilds the image.
 
 > [!NOTE]
 > If you're looking for a small demo without any setup, check our [open demo](https://demo.udav.texttechnologylab.org/).
@@ -343,7 +369,7 @@ JAVA_OPTS=-Xmx10G -Xms1024m
 
 ## License
 
-This project is published under the AGPL-3.0 [license](/LICENSE).
+This project is published under the AGPL-3.0 [license](/LICENSE.txt).
 
 # Cite
 If you want to use the project please quote this as follows:

docker-compose.yml

@@ -26,8 +26,6 @@ services:
       interval: 10s
       timeout: 5s
       retries: 10
-    ports:
-      - "5432:5432"
     volumes:
       - postgres_data:/var/lib/postgresql/data
 
@@ -61,7 +59,7 @@ services:
       PIPELINE_IMPORTER_FOLDER: ${PIPELINE_IMPORTER_FOLDER:-/app/pipelines}
       JSON_IMPORTER_FOLDER: ${JSON_IMPORTER_FOLDER:-/app/sourcefilesJSON}
       PIPELINE_IMPORTER_REPLACE_IF_DIFFERENT: ${PIPELINE_IMPORTER_REPLACE_IF_DIFFERENT:-false}
-      SROUCE_BUILDER: ${SROUCE_BUILDER:-false}
+      SOURCE_BUILDER: ${SOURCE_BUILDER:-false}
       LLM_BASE_URL: ${LLM_BASE_URL:-}
       LLM_API_TOKEN: ${LLM_API_TOKEN:-}
       # Batch export API (headless Chromium is shipped in the image; see README)