Pin the task images to a public base image: quay.io/jupyter/scipy-notebook (dated tag) instead of the <base-image>:latest placeholder

README.md

@@ -213,9 +213,11 @@ Earlier reported results used a `[-1, 1]` normalization; convert with
 
 Install `docker`, then for each task:
 
-1. build the image from `environment/Dockerfile` — it `COPY data/ /app/data/`,
+1. build the image from `environment/Dockerfile` — it builds on the public
+   `quay.io/jupyter/scipy-notebook` image (see below), `COPY data/ /app/data/`,
    creates the `agent` and `verifier` users declared in `task.toml`, and asserts
    that no ground-truth material is present in the image;
+   the container is meant to be started and then driven with `docker exec`;
 2. run the agent as the `agent` user; it must leave its report at
    `/app/answer.txt`;
 3. run `tests/test.sh` as the `verifier` user — it sources `tests/judge.env`,
@@ -233,8 +235,15 @@ ANTHROPIC_MODEL      # judge model id
 
 The file ships with placeholders (`<your-judge-token>`, `<judge-gateway>`,
 `<judge-model-path-or-id>`) — fill in your own endpoint and token before running.
-The same applies to `environment/Dockerfile`, which builds from a
-`<base-image>:latest` placeholder and public package mirrors.
+
+`environment/Dockerfile` needs no such placeholders. It builds from
+`quay.io/jupyter/scipy-notebook`, pinned to a dated tag: a public Python
+(conda-forge) environment that already ships the data-science stack these tasks
+use — pandas, numpy, scipy, statsmodels, scikit-learn, matplotlib, h5py,
+openpyxl, pytables, dask — and pulls the remaining analysis packages (pyarrow,
+pypdf, pillow) from public PyPI on top. Any image carrying `python3`/`pip` can be
+substituted: the apt and pip steps probe for existence first and skip whatever is
+already installed.
 
 `judge.py` prefers `JUDGE_*` over `ANTHROPIC_*` when both are set, streams the
 request (so a long judgment is never killed by a total-time timeout) and falls

dataworld-p000014-cause-1/environment/Dockerfile

@@ -3,10 +3,33 @@
 # required.
 # Leak prevention: the whitelist COPYs only data/ -- WORLD.md, .build and dag.json must
 # never make it into the image.
-# [RELEASE] Replace the base image below with your own generic image (it needs
-# python3/pip, otherwise the apt step below installs them); the apt/pip steps all probe
-# for existence first, so anything the base image already provides is not reinstalled.
-FROM <base-image>:latest
+# Base image: a public, batteries-included Python environment -- the Jupyter scipy stack
+# (conda-forge). It already ships the data-science packages these tasks use (pandas, numpy,
+# scipy, statsmodels, scikit-learn, matplotlib, h5py, openpyxl, pytables, dask, numexpr);
+# pyarrow/pypdf/pillow are covered by the pip step below. The Docker Hub mirror of this
+# image is stale (2023-10), so the maintained quay.io registry is used, and the tag is a
+# dated release (index digest
+# sha256:8f8a5763d9a01e7f488eff8d9379654788e9cc08286dbcbc6bcda129c1b0edcd) so that builds
+# are reproducible -- `latest` tracks the same image. Any image with python3/pip works just
+# as well; the apt/pip steps probe for existence first, so nothing the base image already
+# provides is reinstalled. Two properties of this base image are reset below: it ends on a
+# non-root user (this build needs root to create the agent/verifier users and to open up
+# /app), and it carries a Jupyter entrypoint (the container is started by the harness with
+# its own command and has to stay up for `docker exec`).
+FROM quay.io/jupyter/scipy-notebook:2026-10-12
+
+USER root
+# Drop the inherited Jupyter entrypoint/CMD: exec whatever command the harness passes, and
+# stay alive when it passes none.
+ENTRYPOINT ["/bin/sh", "-c", "exec \"$@\"", "-"]
+CMD ["sleep", "infinity"]
+
+# The conda prefix is the only Python in this image; link it into the standard /usr/local/bin
+# so that python3/pip3 still resolve when the harness switches to the agent/verifier user
+# with a login shell, which would otherwise drop /opt/conda/bin from PATH.
+RUN set -eux; \
+    ln -sf /opt/conda/bin/python3 /usr/local/bin/python3; \
+    ln -sf /opt/conda/bin/pip3 /usr/local/bin/pip3
 
 # Package mirrors -- the release build uses public mirrors (the original build
 # environment had no internet access and went through internal mirrors).

dataworld-p000014-consequence-2/environment/Dockerfile

@@ -3,10 +3,33 @@
 # required.
 # Leak prevention: the whitelist COPYs only data/ -- WORLD.md, .build and dag.json must
 # never make it into the image.
-# [RELEASE] Replace the base image below with your own generic image (it needs
-# python3/pip, otherwise the apt step below installs them); the apt/pip steps all probe
-# for existence first, so anything the base image already provides is not reinstalled.
-FROM <base-image>:latest
+# Base image: a public, batteries-included Python environment -- the Jupyter scipy stack
+# (conda-forge). It already ships the data-science packages these tasks use (pandas, numpy,
+# scipy, statsmodels, scikit-learn, matplotlib, h5py, openpyxl, pytables, dask, numexpr);
+# pyarrow/pypdf/pillow are covered by the pip step below. The Docker Hub mirror of this
+# image is stale (2023-10), so the maintained quay.io registry is used, and the tag is a
+# dated release (index digest
+# sha256:8f8a5763d9a01e7f488eff8d9379654788e9cc08286dbcbc6bcda129c1b0edcd) so that builds
+# are reproducible -- `latest` tracks the same image. Any image with python3/pip works just
+# as well; the apt/pip steps probe for existence first, so nothing the base image already
+# provides is reinstalled. Two properties of this base image are reset below: it ends on a
+# non-root user (this build needs root to create the agent/verifier users and to open up
+# /app), and it carries a Jupyter entrypoint (the container is started by the harness with
+# its own command and has to stay up for `docker exec`).
+FROM quay.io/jupyter/scipy-notebook:2026-10-12
+
+USER root
+# Drop the inherited Jupyter entrypoint/CMD: exec whatever command the harness passes, and
+# stay alive when it passes none.
+ENTRYPOINT ["/bin/sh", "-c", "exec \"$@\"", "-"]
+CMD ["sleep", "infinity"]
+
+# The conda prefix is the only Python in this image; link it into the standard /usr/local/bin
+# so that python3/pip3 still resolve when the harness switches to the agent/verifier user
+# with a login shell, which would otherwise drop /opt/conda/bin from PATH.
+RUN set -eux; \
+    ln -sf /opt/conda/bin/python3 /usr/local/bin/python3; \
+    ln -sf /opt/conda/bin/pip3 /usr/local/bin/pip3
 
 # Package mirrors -- the release build uses public mirrors (the original build
 # environment had no internet access and went through internal mirrors).

dataworld-p000049-cause-2/environment/Dockerfile

@@ -3,10 +3,33 @@
 # required.
 # Leak prevention: the whitelist COPYs only data/ -- WORLD.md, .build and dag.json must
 # never make it into the image.
-# [RELEASE] Replace the base image below with your own generic image (it needs
-# python3/pip, otherwise the apt step below installs them); the apt/pip steps all probe
-# for existence first, so anything the base image already provides is not reinstalled.
-FROM <base-image>:latest
+# Base image: a public, batteries-included Python environment -- the Jupyter scipy stack
+# (conda-forge). It already ships the data-science packages these tasks use (pandas, numpy,
+# scipy, statsmodels, scikit-learn, matplotlib, h5py, openpyxl, pytables, dask, numexpr);
+# pyarrow/pypdf/pillow are covered by the pip step below. The Docker Hub mirror of this
+# image is stale (2023-10), so the maintained quay.io registry is used, and the tag is a
+# dated release (index digest
+# sha256:8f8a5763d9a01e7f488eff8d9379654788e9cc08286dbcbc6bcda129c1b0edcd) so that builds
+# are reproducible -- `latest` tracks the same image. Any image with python3/pip works just
+# as well; the apt/pip steps probe for existence first, so nothing the base image already
+# provides is reinstalled. Two properties of this base image are reset below: it ends on a
+# non-root user (this build needs root to create the agent/verifier users and to open up
+# /app), and it carries a Jupyter entrypoint (the container is started by the harness with
+# its own command and has to stay up for `docker exec`).
+FROM quay.io/jupyter/scipy-notebook:2026-10-12
+
+USER root
+# Drop the inherited Jupyter entrypoint/CMD: exec whatever command the harness passes, and
+# stay alive when it passes none.
+ENTRYPOINT ["/bin/sh", "-c", "exec \"$@\"", "-"]
+CMD ["sleep", "infinity"]
+
+# The conda prefix is the only Python in this image; link it into the standard /usr/local/bin
+# so that python3/pip3 still resolve when the harness switches to the agent/verifier user
+# with a login shell, which would otherwise drop /opt/conda/bin from PATH.
+RUN set -eux; \
+    ln -sf /opt/conda/bin/python3 /usr/local/bin/python3; \
+    ln -sf /opt/conda/bin/pip3 /usr/local/bin/pip3
 
 # Package mirrors -- the release build uses public mirrors (the original build
 # environment had no internet access and went through internal mirrors).

dataworld-p000049-consequence-2/environment/Dockerfile

@@ -3,10 +3,33 @@
 # required.
 # Leak prevention: the whitelist COPYs only data/ -- WORLD.md, .build and dag.json must
 # never make it into the image.
-# [RELEASE] Replace the base image below with your own generic image (it needs
-# python3/pip, otherwise the apt step below installs them); the apt/pip steps all probe
-# for existence first, so anything the base image already provides is not reinstalled.
-FROM <base-image>:latest
+# Base image: a public, batteries-included Python environment -- the Jupyter scipy stack
+# (conda-forge). It already ships the data-science packages these tasks use (pandas, numpy,
+# scipy, statsmodels, scikit-learn, matplotlib, h5py, openpyxl, pytables, dask, numexpr);
+# pyarrow/pypdf/pillow are covered by the pip step below. The Docker Hub mirror of this
+# image is stale (2023-10), so the maintained quay.io registry is used, and the tag is a
+# dated release (index digest
+# sha256:8f8a5763d9a01e7f488eff8d9379654788e9cc08286dbcbc6bcda129c1b0edcd) so that builds
+# are reproducible -- `latest` tracks the same image. Any image with python3/pip works just
+# as well; the apt/pip steps probe for existence first, so nothing the base image already
+# provides is reinstalled. Two properties of this base image are reset below: it ends on a
+# non-root user (this build needs root to create the agent/verifier users and to open up
+# /app), and it carries a Jupyter entrypoint (the container is started by the harness with
+# its own command and has to stay up for `docker exec`).
+FROM quay.io/jupyter/scipy-notebook:2026-10-12
+
+USER root
+# Drop the inherited Jupyter entrypoint/CMD: exec whatever command the harness passes, and
+# stay alive when it passes none.
+ENTRYPOINT ["/bin/sh", "-c", "exec \"$@\"", "-"]
+CMD ["sleep", "infinity"]
+
+# The conda prefix is the only Python in this image; link it into the standard /usr/local/bin
+# so that python3/pip3 still resolve when the harness switches to the agent/verifier user
+# with a login shell, which would otherwise drop /opt/conda/bin from PATH.
+RUN set -eux; \
+    ln -sf /opt/conda/bin/python3 /usr/local/bin/python3; \
+    ln -sf /opt/conda/bin/pip3 /usr/local/bin/pip3
 
 # Package mirrors -- the release build uses public mirrors (the original build
 # environment had no internet access and went through internal mirrors).

dataworld-p000072-cause-2/environment/Dockerfile

@@ -3,10 +3,33 @@
 # required.
 # Leak prevention: the whitelist COPYs only data/ -- WORLD.md, .build and dag.json must
 # never make it into the image.
-# [RELEASE] Replace the base image below with your own generic image (it needs
-# python3/pip, otherwise the apt step below installs them); the apt/pip steps all probe
-# for existence first, so anything the base image already provides is not reinstalled.
-FROM <base-image>:latest
+# Base image: a public, batteries-included Python environment -- the Jupyter scipy stack
+# (conda-forge). It already ships the data-science packages these tasks use (pandas, numpy,
+# scipy, statsmodels, scikit-learn, matplotlib, h5py, openpyxl, pytables, dask, numexpr);
+# pyarrow/pypdf/pillow are covered by the pip step below. The Docker Hub mirror of this
+# image is stale (2023-10), so the maintained quay.io registry is used, and the tag is a
+# dated release (index digest
+# sha256:8f8a5763d9a01e7f488eff8d9379654788e9cc08286dbcbc6bcda129c1b0edcd) so that builds
+# are reproducible -- `latest` tracks the same image. Any image with python3/pip works just
+# as well; the apt/pip steps probe for existence first, so nothing the base image already
+# provides is reinstalled. Two properties of this base image are reset below: it ends on a
+# non-root user (this build needs root to create the agent/verifier users and to open up
+# /app), and it carries a Jupyter entrypoint (the container is started by the harness with
+# its own command and has to stay up for `docker exec`).
+FROM quay.io/jupyter/scipy-notebook:2026-10-12
+
+USER root
+# Drop the inherited Jupyter entrypoint/CMD: exec whatever command the harness passes, and
+# stay alive when it passes none.
+ENTRYPOINT ["/bin/sh", "-c", "exec \"$@\"", "-"]
+CMD ["sleep", "infinity"]
+
+# The conda prefix is the only Python in this image; link it into the standard /usr/local/bin
+# so that python3/pip3 still resolve when the harness switches to the agent/verifier user
+# with a login shell, which would otherwise drop /opt/conda/bin from PATH.
+RUN set -eux; \
+    ln -sf /opt/conda/bin/python3 /usr/local/bin/python3; \
+    ln -sf /opt/conda/bin/pip3 /usr/local/bin/pip3
 
 # Package mirrors -- the release build uses public mirrors (the original build
 # environment had no internet access and went through internal mirrors).

dataworld-p000072-consequence-1/environment/Dockerfile

@@ -3,10 +3,33 @@
 # required.
 # Leak prevention: the whitelist COPYs only data/ -- WORLD.md, .build and dag.json must
 # never make it into the image.
-# [RELEASE] Replace the base image below with your own generic image (it needs
-# python3/pip, otherwise the apt step below installs them); the apt/pip steps all probe
-# for existence first, so anything the base image already provides is not reinstalled.
-FROM <base-image>:latest
+# Base image: a public, batteries-included Python environment -- the Jupyter scipy stack
+# (conda-forge). It already ships the data-science packages these tasks use (pandas, numpy,
+# scipy, statsmodels, scikit-learn, matplotlib, h5py, openpyxl, pytables, dask, numexpr);
+# pyarrow/pypdf/pillow are covered by the pip step below. The Docker Hub mirror of this
+# image is stale (2023-10), so the maintained quay.io registry is used, and the tag is a
+# dated release (index digest
+# sha256:8f8a5763d9a01e7f488eff8d9379654788e9cc08286dbcbc6bcda129c1b0edcd) so that builds
+# are reproducible -- `latest` tracks the same image. Any image with python3/pip works just
+# as well; the apt/pip steps probe for existence first, so nothing the base image already
+# provides is reinstalled. Two properties of this base image are reset below: it ends on a
+# non-root user (this build needs root to create the agent/verifier users and to open up
+# /app), and it carries a Jupyter entrypoint (the container is started by the harness with
+# its own command and has to stay up for `docker exec`).
+FROM quay.io/jupyter/scipy-notebook:2026-10-12
+
+USER root
+# Drop the inherited Jupyter entrypoint/CMD: exec whatever command the harness passes, and
+# stay alive when it passes none.
+ENTRYPOINT ["/bin/sh", "-c", "exec \"$@\"", "-"]
+CMD ["sleep", "infinity"]
+
+# The conda prefix is the only Python in this image; link it into the standard /usr/local/bin
+# so that python3/pip3 still resolve when the harness switches to the agent/verifier user
+# with a login shell, which would otherwise drop /opt/conda/bin from PATH.
+RUN set -eux; \
+    ln -sf /opt/conda/bin/python3 /usr/local/bin/python3; \
+    ln -sf /opt/conda/bin/pip3 /usr/local/bin/pip3
 
 # Package mirrors -- the release build uses public mirrors (the original build
 # environment had no internet access and went through internal mirrors).

dataworld-p000096-cause-2/environment/Dockerfile

@@ -3,10 +3,33 @@
 # required.
 # Leak prevention: the whitelist COPYs only data/ -- WORLD.md, .build and dag.json must
 # never make it into the image.
-# [RELEASE] Replace the base image below with your own generic image (it needs
-# python3/pip, otherwise the apt step below installs them); the apt/pip steps all probe
-# for existence first, so anything the base image already provides is not reinstalled.
-FROM <base-image>:latest
+# Base image: a public, batteries-included Python environment -- the Jupyter scipy stack
+# (conda-forge). It already ships the data-science packages these tasks use (pandas, numpy,
+# scipy, statsmodels, scikit-learn, matplotlib, h5py, openpyxl, pytables, dask, numexpr);
+# pyarrow/pypdf/pillow are covered by the pip step below. The Docker Hub mirror of this
+# image is stale (2023-10), so the maintained quay.io registry is used, and the tag is a
+# dated release (index digest
+# sha256:8f8a5763d9a01e7f488eff8d9379654788e9cc08286dbcbc6bcda129c1b0edcd) so that builds
+# are reproducible -- `latest` tracks the same image. Any image with python3/pip works just
+# as well; the apt/pip steps probe for existence first, so nothing the base image already
+# provides is reinstalled. Two properties of this base image are reset below: it ends on a
+# non-root user (this build needs root to create the agent/verifier users and to open up
+# /app), and it carries a Jupyter entrypoint (the container is started by the harness with
+# its own command and has to stay up for `docker exec`).
+FROM quay.io/jupyter/scipy-notebook:2026-10-12
+
+USER root
+# Drop the inherited Jupyter entrypoint/CMD: exec whatever command the harness passes, and
+# stay alive when it passes none.
+ENTRYPOINT ["/bin/sh", "-c", "exec \"$@\"", "-"]
+CMD ["sleep", "infinity"]
+
+# The conda prefix is the only Python in this image; link it into the standard /usr/local/bin
+# so that python3/pip3 still resolve when the harness switches to the agent/verifier user
+# with a login shell, which would otherwise drop /opt/conda/bin from PATH.
+RUN set -eux; \
+    ln -sf /opt/conda/bin/python3 /usr/local/bin/python3; \
+    ln -sf /opt/conda/bin/pip3 /usr/local/bin/pip3
 
 # Package mirrors -- the release build uses public mirrors (the original build
 # environment had no internet access and went through internal mirrors).

dataworld-p000096-consequence-1/environment/Dockerfile

@@ -3,10 +3,33 @@
 # required.
 # Leak prevention: the whitelist COPYs only data/ -- WORLD.md, .build and dag.json must
 # never make it into the image.
-# [RELEASE] Replace the base image below with your own generic image (it needs
-# python3/pip, otherwise the apt step below installs them); the apt/pip steps all probe
-# for existence first, so anything the base image already provides is not reinstalled.
-FROM <base-image>:latest
+# Base image: a public, batteries-included Python environment -- the Jupyter scipy stack
+# (conda-forge). It already ships the data-science packages these tasks use (pandas, numpy,
+# scipy, statsmodels, scikit-learn, matplotlib, h5py, openpyxl, pytables, dask, numexpr);
+# pyarrow/pypdf/pillow are covered by the pip step below. The Docker Hub mirror of this
+# image is stale (2023-10), so the maintained quay.io registry is used, and the tag is a
+# dated release (index digest
+# sha256:8f8a5763d9a01e7f488eff8d9379654788e9cc08286dbcbc6bcda129c1b0edcd) so that builds
+# are reproducible -- `latest` tracks the same image. Any image with python3/pip works just
+# as well; the apt/pip steps probe for existence first, so nothing the base image already
+# provides is reinstalled. Two properties of this base image are reset below: it ends on a
+# non-root user (this build needs root to create the agent/verifier users and to open up
+# /app), and it carries a Jupyter entrypoint (the container is started by the harness with
+# its own command and has to stay up for `docker exec`).
+FROM quay.io/jupyter/scipy-notebook:2026-10-12
+
+USER root
+# Drop the inherited Jupyter entrypoint/CMD: exec whatever command the harness passes, and
+# stay alive when it passes none.
+ENTRYPOINT ["/bin/sh", "-c", "exec \"$@\"", "-"]
+CMD ["sleep", "infinity"]
+
+# The conda prefix is the only Python in this image; link it into the standard /usr/local/bin
+# so that python3/pip3 still resolve when the harness switches to the agent/verifier user
+# with a login shell, which would otherwise drop /opt/conda/bin from PATH.
+RUN set -eux; \
+    ln -sf /opt/conda/bin/python3 /usr/local/bin/python3; \
+    ln -sf /opt/conda/bin/pip3 /usr/local/bin/pip3
 
 # Package mirrors -- the release build uses public mirrors (the original build
 # environment had no internet access and went through internal mirrors).

dataworld-p000129-cause-2/environment/Dockerfile

@@ -3,10 +3,33 @@
 # required.
 # Leak prevention: the whitelist COPYs only data/ -- WORLD.md, .build and dag.json must
 # never make it into the image.
-# [RELEASE] Replace the base image below with your own generic image (it needs
-# python3/pip, otherwise the apt step below installs them); the apt/pip steps all probe
-# for existence first, so anything the base image already provides is not reinstalled.
-FROM <base-image>:latest
+# Base image: a public, batteries-included Python environment -- the Jupyter scipy stack
+# (conda-forge). It already ships the data-science packages these tasks use (pandas, numpy,
+# scipy, statsmodels, scikit-learn, matplotlib, h5py, openpyxl, pytables, dask, numexpr);
+# pyarrow/pypdf/pillow are covered by the pip step below. The Docker Hub mirror of this
+# image is stale (2023-10), so the maintained quay.io registry is used, and the tag is a
+# dated release (index digest
+# sha256:8f8a5763d9a01e7f488eff8d9379654788e9cc08286dbcbc6bcda129c1b0edcd) so that builds
+# are reproducible -- `latest` tracks the same image. Any image with python3/pip works just
+# as well; the apt/pip steps probe for existence first, so nothing the base image already
+# provides is reinstalled. Two properties of this base image are reset below: it ends on a
+# non-root user (this build needs root to create the agent/verifier users and to open up
+# /app), and it carries a Jupyter entrypoint (the container is started by the harness with
+# its own command and has to stay up for `docker exec`).
+FROM quay.io/jupyter/scipy-notebook:2026-10-12
+
+USER root
+# Drop the inherited Jupyter entrypoint/CMD: exec whatever command the harness passes, and
+# stay alive when it passes none.
+ENTRYPOINT ["/bin/sh", "-c", "exec \"$@\"", "-"]
+CMD ["sleep", "infinity"]
+
+# The conda prefix is the only Python in this image; link it into the standard /usr/local/bin
+# so that python3/pip3 still resolve when the harness switches to the agent/verifier user
+# with a login shell, which would otherwise drop /opt/conda/bin from PATH.
+RUN set -eux; \
+    ln -sf /opt/conda/bin/python3 /usr/local/bin/python3; \
+    ln -sf /opt/conda/bin/pip3 /usr/local/bin/pip3
 
 # Package mirrors -- the release build uses public mirrors (the original build
 # environment had no internet access and went through internal mirrors).

dataworld-p000129-consequence-1/environment/Dockerfile

@@ -3,10 +3,33 @@
 # required.
 # Leak prevention: the whitelist COPYs only data/ -- WORLD.md, .build and dag.json must
 # never make it into the image.
-# [RELEASE] Replace the base image below with your own generic image (it needs
-# python3/pip, otherwise the apt step below installs them); the apt/pip steps all probe
-# for existence first, so anything the base image already provides is not reinstalled.
-FROM <base-image>:latest
+# Base image: a public, batteries-included Python environment -- the Jupyter scipy stack
+# (conda-forge). It already ships the data-science packages these tasks use (pandas, numpy,
+# scipy, statsmodels, scikit-learn, matplotlib, h5py, openpyxl, pytables, dask, numexpr);
+# pyarrow/pypdf/pillow are covered by the pip step below. The Docker Hub mirror of this
+# image is stale (2023-10), so the maintained quay.io registry is used, and the tag is a
+# dated release (index digest
+# sha256:8f8a5763d9a01e7f488eff8d9379654788e9cc08286dbcbc6bcda129c1b0edcd) so that builds
+# are reproducible -- `latest` tracks the same image. Any image with python3/pip works just
+# as well; the apt/pip steps probe for existence first, so nothing the base image already
+# provides is reinstalled. Two properties of this base image are reset below: it ends on a
+# non-root user (this build needs root to create the agent/verifier users and to open up
+# /app), and it carries a Jupyter entrypoint (the container is started by the harness with
+# its own command and has to stay up for `docker exec`).
+FROM quay.io/jupyter/scipy-notebook:2026-10-12
+
+USER root
+# Drop the inherited Jupyter entrypoint/CMD: exec whatever command the harness passes, and
+# stay alive when it passes none.
+ENTRYPOINT ["/bin/sh", "-c", "exec \"$@\"", "-"]
+CMD ["sleep", "infinity"]
+
+# The conda prefix is the only Python in this image; link it into the standard /usr/local/bin
+# so that python3/pip3 still resolve when the harness switches to the agent/verifier user
+# with a login shell, which would otherwise drop /opt/conda/bin from PATH.
+RUN set -eux; \
+    ln -sf /opt/conda/bin/python3 /usr/local/bin/python3; \
+    ln -sf /opt/conda/bin/pip3 /usr/local/bin/pip3
 
 # Package mirrors -- the release build uses public mirrors (the original build
 # environment had no internet access and went through internal mirrors).

dataworld-p000151-cause-2/environment/Dockerfile

@@ -3,10 +3,33 @@
 # required.
 # Leak prevention: the whitelist COPYs only data/ -- WORLD.md, .build and dag.json must
 # never make it into the image.
-# [RELEASE] Replace the base image below with your own generic image (it needs
-# python3/pip, otherwise the apt step below installs them); the apt/pip steps all probe
-# for existence first, so anything the base image already provides is not reinstalled.
-FROM <base-image>:latest
+# Base image: a public, batteries-included Python environment -- the Jupyter scipy stack
+# (conda-forge). It already ships the data-science packages these tasks use (pandas, numpy,
+# scipy, statsmodels, scikit-learn, matplotlib, h5py, openpyxl, pytables, dask, numexpr);
+# pyarrow/pypdf/pillow are covered by the pip step below. The Docker Hub mirror of this
+# image is stale (2023-10), so the maintained quay.io registry is used, and the tag is a
+# dated release (index digest
+# sha256:8f8a5763d9a01e7f488eff8d9379654788e9cc08286dbcbc6bcda129c1b0edcd) so that builds
+# are reproducible -- `latest` tracks the same image. Any image with python3/pip works just
+# as well; the apt/pip steps probe for existence first, so nothing the base image already
+# provides is reinstalled. Two properties of this base image are reset below: it ends on a
+# non-root user (this build needs root to create the agent/verifier users and to open up
+# /app), and it carries a Jupyter entrypoint (the container is started by the harness with
+# its own command and has to stay up for `docker exec`).
+FROM quay.io/jupyter/scipy-notebook:2026-10-12
+
+USER root
+# Drop the inherited Jupyter entrypoint/CMD: exec whatever command the harness passes, and
+# stay alive when it passes none.
+ENTRYPOINT ["/bin/sh", "-c", "exec \"$@\"", "-"]
+CMD ["sleep", "infinity"]
+
+# The conda prefix is the only Python in this image; link it into the standard /usr/local/bin
+# so that python3/pip3 still resolve when the harness switches to the agent/verifier user
+# with a login shell, which would otherwise drop /opt/conda/bin from PATH.
+RUN set -eux; \
+    ln -sf /opt/conda/bin/python3 /usr/local/bin/python3; \
+    ln -sf /opt/conda/bin/pip3 /usr/local/bin/pip3
 
 # Package mirrors -- the release build uses public mirrors (the original build
 # environment had no internet access and went through internal mirrors).

dataworld-p000151-consequence-1/environment/Dockerfile

@@ -3,10 +3,33 @@
 # required.
 # Leak prevention: the whitelist COPYs only data/ -- WORLD.md, .build and dag.json must
 # never make it into the image.
-# [RELEASE] Replace the base image below with your own generic image (it needs
-# python3/pip, otherwise the apt step below installs them); the apt/pip steps all probe
-# for existence first, so anything the base image already provides is not reinstalled.
-FROM <base-image>:latest
+# Base image: a public, batteries-included Python environment -- the Jupyter scipy stack
+# (conda-forge). It already ships the data-science packages these tasks use (pandas, numpy,
+# scipy, statsmodels, scikit-learn, matplotlib, h5py, openpyxl, pytables, dask, numexpr);
+# pyarrow/pypdf/pillow are covered by the pip step below. The Docker Hub mirror of this
+# image is stale (2023-10), so the maintained quay.io registry is used, and the tag is a
+# dated release (index digest
+# sha256:8f8a5763d9a01e7f488eff8d9379654788e9cc08286dbcbc6bcda129c1b0edcd) so that builds
+# are reproducible -- `latest` tracks the same image. Any image with python3/pip works just
+# as well; the apt/pip steps probe for existence first, so nothing the base image already
+# provides is reinstalled. Two properties of this base image are reset below: it ends on a
+# non-root user (this build needs root to create the agent/verifier users and to open up
+# /app), and it carries a Jupyter entrypoint (the container is started by the harness with
+# its own command and has to stay up for `docker exec`).
+FROM quay.io/jupyter/scipy-notebook:2026-10-12
+
+USER root
+# Drop the inherited Jupyter entrypoint/CMD: exec whatever command the harness passes, and
+# stay alive when it passes none.
+ENTRYPOINT ["/bin/sh", "-c", "exec \"$@\"", "-"]
+CMD ["sleep", "infinity"]
+
+# The conda prefix is the only Python in this image; link it into the standard /usr/local/bin
+# so that python3/pip3 still resolve when the harness switches to the agent/verifier user
+# with a login shell, which would otherwise drop /opt/conda/bin from PATH.
+RUN set -eux; \
+    ln -sf /opt/conda/bin/python3 /usr/local/bin/python3; \
+    ln -sf /opt/conda/bin/pip3 /usr/local/bin/pip3
 
 # Package mirrors -- the release build uses public mirrors (the original build
 # environment had no internet access and went through internal mirrors).

dataworld-p000183-cause-2/environment/Dockerfile

@@ -3,10 +3,33 @@
 # required.
 # Leak prevention: the whitelist COPYs only data/ -- WORLD.md, .build and dag.json must
 # never make it into the image.
-# [RELEASE] Replace the base image below with your own generic image (it needs
-# python3/pip, otherwise the apt step below installs them); the apt/pip steps all probe
-# for existence first, so anything the base image already provides is not reinstalled.
-FROM <base-image>:latest
+# Base image: a public, batteries-included Python environment -- the Jupyter scipy stack
+# (conda-forge). It already ships the data-science packages these tasks use (pandas, numpy,
+# scipy, statsmodels, scikit-learn, matplotlib, h5py, openpyxl, pytables, dask, numexpr);
+# pyarrow/pypdf/pillow are covered by the pip step below. The Docker Hub mirror of this
+# image is stale (2023-10), so the maintained quay.io registry is used, and the tag is a
+# dated release (index digest
+# sha256:8f8a5763d9a01e7f488eff8d9379654788e9cc08286dbcbc6bcda129c1b0edcd) so that builds
+# are reproducible -- `latest` tracks the same image. Any image with python3/pip works just
+# as well; the apt/pip steps probe for existence first, so nothing the base image already
+# provides is reinstalled. Two properties of this base image are reset below: it ends on a
+# non-root user (this build needs root to create the agent/verifier users and to open up
+# /app), and it carries a Jupyter entrypoint (the container is started by the harness with
+# its own command and has to stay up for `docker exec`).
+FROM quay.io/jupyter/scipy-notebook:2026-10-12
+
+USER root
+# Drop the inherited Jupyter entrypoint/CMD: exec whatever command the harness passes, and
+# stay alive when it passes none.
+ENTRYPOINT ["/bin/sh", "-c", "exec \"$@\"", "-"]
+CMD ["sleep", "infinity"]
+
+# The conda prefix is the only Python in this image; link it into the standard /usr/local/bin
+# so that python3/pip3 still resolve when the harness switches to the agent/verifier user
+# with a login shell, which would otherwise drop /opt/conda/bin from PATH.
+RUN set -eux; \
+    ln -sf /opt/conda/bin/python3 /usr/local/bin/python3; \
+    ln -sf /opt/conda/bin/pip3 /usr/local/bin/pip3
 
 # Package mirrors -- the release build uses public mirrors (the original build
 # environment had no internet access and went through internal mirrors).

dataworld-p000183-consequence-1/environment/Dockerfile

@@ -3,10 +3,33 @@
 # required.
 # Leak prevention: the whitelist COPYs only data/ -- WORLD.md, .build and dag.json must
 # never make it into the image.
-# [RELEASE] Replace the base image below with your own generic image (it needs
-# python3/pip, otherwise the apt step below installs them); the apt/pip steps all probe
-# for existence first, so anything the base image already provides is not reinstalled.
-FROM <base-image>:latest
+# Base image: a public, batteries-included Python environment -- the Jupyter scipy stack
+# (conda-forge). It already ships the data-science packages these tasks use (pandas, numpy,
+# scipy, statsmodels, scikit-learn, matplotlib, h5py, openpyxl, pytables, dask, numexpr);
+# pyarrow/pypdf/pillow are covered by the pip step below. The Docker Hub mirror of this
+# image is stale (2023-10), so the maintained quay.io registry is used, and the tag is a
+# dated release (index digest
+# sha256:8f8a5763d9a01e7f488eff8d9379654788e9cc08286dbcbc6bcda129c1b0edcd) so that builds
+# are reproducible -- `latest` tracks the same image. Any image with python3/pip works just
+# as well; the apt/pip steps probe for existence first, so nothing the base image already
+# provides is reinstalled. Two properties of this base image are reset below: it ends on a
+# non-root user (this build needs root to create the agent/verifier users and to open up
+# /app), and it carries a Jupyter entrypoint (the container is started by the harness with
+# its own command and has to stay up for `docker exec`).
+FROM quay.io/jupyter/scipy-notebook:2026-10-12
+
+USER root
+# Drop the inherited Jupyter entrypoint/CMD: exec whatever command the harness passes, and
+# stay alive when it passes none.
+ENTRYPOINT ["/bin/sh", "-c", "exec \"$@\"", "-"]
+CMD ["sleep", "infinity"]
+
+# The conda prefix is the only Python in this image; link it into the standard /usr/local/bin
+# so that python3/pip3 still resolve when the harness switches to the agent/verifier user
+# with a login shell, which would otherwise drop /opt/conda/bin from PATH.
+RUN set -eux; \
+    ln -sf /opt/conda/bin/python3 /usr/local/bin/python3; \
+    ln -sf /opt/conda/bin/pip3 /usr/local/bin/pip3
 
 # Package mirrors -- the release build uses public mirrors (the original build
 # environment had no internet access and went through internal mirrors).

dataworld-p000211-cause-2/environment/Dockerfile

@@ -3,10 +3,33 @@
 # required.
 # Leak prevention: the whitelist COPYs only data/ -- WORLD.md, .build and dag.json must
 # never make it into the image.
-# [RELEASE] Replace the base image below with your own generic image (it needs
-# python3/pip, otherwise the apt step below installs them); the apt/pip steps all probe
-# for existence first, so anything the base image already provides is not reinstalled.
-FROM <base-image>:latest
+# Base image: a public, batteries-included Python environment -- the Jupyter scipy stack
+# (conda-forge). It already ships the data-science packages these tasks use (pandas, numpy,
+# scipy, statsmodels, scikit-learn, matplotlib, h5py, openpyxl, pytables, dask, numexpr);
+# pyarrow/pypdf/pillow are covered by the pip step below. The Docker Hub mirror of this
+# image is stale (2023-10), so the maintained quay.io registry is used, and the tag is a
+# dated release (index digest
+# sha256:8f8a5763d9a01e7f488eff8d9379654788e9cc08286dbcbc6bcda129c1b0edcd) so that builds
+# are reproducible -- `latest` tracks the same image. Any image with python3/pip works just
+# as well; the apt/pip steps probe for existence first, so nothing the base image already
+# provides is reinstalled. Two properties of this base image are reset below: it ends on a
+# non-root user (this build needs root to create the agent/verifier users and to open up
+# /app), and it carries a Jupyter entrypoint (the container is started by the harness with
+# its own command and has to stay up for `docker exec`).
+FROM quay.io/jupyter/scipy-notebook:2026-10-12
+
+USER root
+# Drop the inherited Jupyter entrypoint/CMD: exec whatever command the harness passes, and
+# stay alive when it passes none.
+ENTRYPOINT ["/bin/sh", "-c", "exec \"$@\"", "-"]
+CMD ["sleep", "infinity"]
+
+# The conda prefix is the only Python in this image; link it into the standard /usr/local/bin
+# so that python3/pip3 still resolve when the harness switches to the agent/verifier user
+# with a login shell, which would otherwise drop /opt/conda/bin from PATH.
+RUN set -eux; \
+    ln -sf /opt/conda/bin/python3 /usr/local/bin/python3; \
+    ln -sf /opt/conda/bin/pip3 /usr/local/bin/pip3
 
 # Package mirrors -- the release build uses public mirrors (the original build
 # environment had no internet access and went through internal mirrors).

dataworld-p000211-consequence-1/environment/Dockerfile

@@ -3,10 +3,33 @@
 # required.
 # Leak prevention: the whitelist COPYs only data/ -- WORLD.md, .build and dag.json must
 # never make it into the image.
-# [RELEASE] Replace the base image below with your own generic image (it needs
-# python3/pip, otherwise the apt step below installs them); the apt/pip steps all probe
-# for existence first, so anything the base image already provides is not reinstalled.
-FROM <base-image>:latest
+# Base image: a public, batteries-included Python environment -- the Jupyter scipy stack
+# (conda-forge). It already ships the data-science packages these tasks use (pandas, numpy,
+# scipy, statsmodels, scikit-learn, matplotlib, h5py, openpyxl, pytables, dask, numexpr);
+# pyarrow/pypdf/pillow are covered by the pip step below. The Docker Hub mirror of this
+# image is stale (2023-10), so the maintained quay.io registry is used, and the tag is a
+# dated release (index digest
+# sha256:8f8a5763d9a01e7f488eff8d9379654788e9cc08286dbcbc6bcda129c1b0edcd) so that builds
+# are reproducible -- `latest` tracks the same image. Any image with python3/pip works just
+# as well; the apt/pip steps probe for existence first, so nothing the base image already
+# provides is reinstalled. Two properties of this base image are reset below: it ends on a
+# non-root user (this build needs root to create the agent/verifier users and to open up
+# /app), and it carries a Jupyter entrypoint (the container is started by the harness with
+# its own command and has to stay up for `docker exec`).
+FROM quay.io/jupyter/scipy-notebook:2026-10-12
+
+USER root
+# Drop the inherited Jupyter entrypoint/CMD: exec whatever command the harness passes, and
+# stay alive when it passes none.
+ENTRYPOINT ["/bin/sh", "-c", "exec \"$@\"", "-"]
+CMD ["sleep", "infinity"]
+
+# The conda prefix is the only Python in this image; link it into the standard /usr/local/bin
+# so that python3/pip3 still resolve when the harness switches to the agent/verifier user
+# with a login shell, which would otherwise drop /opt/conda/bin from PATH.
+RUN set -eux; \
+    ln -sf /opt/conda/bin/python3 /usr/local/bin/python3; \
+    ln -sf /opt/conda/bin/pip3 /usr/local/bin/pip3
 
 # Package mirrors -- the release build uses public mirrors (the original build
 # environment had no internet access and went through internal mirrors).

dataworld-p000213-cause-2/environment/Dockerfile

@@ -3,10 +3,33 @@
 # required.
 # Leak prevention: the whitelist COPYs only data/ -- WORLD.md, .build and dag.json must
 # never make it into the image.
-# [RELEASE] Replace the base image below with your own generic image (it needs
-# python3/pip, otherwise the apt step below installs them); the apt/pip steps all probe
-# for existence first, so anything the base image already provides is not reinstalled.
-FROM <base-image>:latest
+# Base image: a public, batteries-included Python environment -- the Jupyter scipy stack
+# (conda-forge). It already ships the data-science packages these tasks use (pandas, numpy,
+# scipy, statsmodels, scikit-learn, matplotlib, h5py, openpyxl, pytables, dask, numexpr);
+# pyarrow/pypdf/pillow are covered by the pip step below. The Docker Hub mirror of this
+# image is stale (2023-10), so the maintained quay.io registry is used, and the tag is a
+# dated release (index digest
+# sha256:8f8a5763d9a01e7f488eff8d9379654788e9cc08286dbcbc6bcda129c1b0edcd) so that builds
+# are reproducible -- `latest` tracks the same image. Any image with python3/pip works just
+# as well; the apt/pip steps probe for existence first, so nothing the base image already
+# provides is reinstalled. Two properties of this base image are reset below: it ends on a
+# non-root user (this build needs root to create the agent/verifier users and to open up
+# /app), and it carries a Jupyter entrypoint (the container is started by the harness with
+# its own command and has to stay up for `docker exec`).
+FROM quay.io/jupyter/scipy-notebook:2026-10-12
+
+USER root
+# Drop the inherited Jupyter entrypoint/CMD: exec whatever command the harness passes, and
+# stay alive when it passes none.
+ENTRYPOINT ["/bin/sh", "-c", "exec \"$@\"", "-"]
+CMD ["sleep", "infinity"]
+
+# The conda prefix is the only Python in this image; link it into the standard /usr/local/bin
+# so that python3/pip3 still resolve when the harness switches to the agent/verifier user
+# with a login shell, which would otherwise drop /opt/conda/bin from PATH.
+RUN set -eux; \
+    ln -sf /opt/conda/bin/python3 /usr/local/bin/python3; \
+    ln -sf /opt/conda/bin/pip3 /usr/local/bin/pip3
 
 # Package mirrors -- the release build uses public mirrors (the original build
 # environment had no internet access and went through internal mirrors).

dataworld-p000213-consequence-2/environment/Dockerfile

@@ -3,10 +3,33 @@
 # required.
 # Leak prevention: the whitelist COPYs only data/ -- WORLD.md, .build and dag.json must
 # never make it into the image.
-# [RELEASE] Replace the base image below with your own generic image (it needs
-# python3/pip, otherwise the apt step below installs them); the apt/pip steps all probe
-# for existence first, so anything the base image already provides is not reinstalled.
-FROM <base-image>:latest
+# Base image: a public, batteries-included Python environment -- the Jupyter scipy stack
+# (conda-forge). It already ships the data-science packages these tasks use (pandas, numpy,
+# scipy, statsmodels, scikit-learn, matplotlib, h5py, openpyxl, pytables, dask, numexpr);
+# pyarrow/pypdf/pillow are covered by the pip step below. The Docker Hub mirror of this
+# image is stale (2023-10), so the maintained quay.io registry is used, and the tag is a
+# dated release (index digest
+# sha256:8f8a5763d9a01e7f488eff8d9379654788e9cc08286dbcbc6bcda129c1b0edcd) so that builds
+# are reproducible -- `latest` tracks the same image. Any image with python3/pip works just
+# as well; the apt/pip steps probe for existence first, so nothing the base image already
+# provides is reinstalled. Two properties of this base image are reset below: it ends on a
+# non-root user (this build needs root to create the agent/verifier users and to open up
+# /app), and it carries a Jupyter entrypoint (the container is started by the harness with
+# its own command and has to stay up for `docker exec`).
+FROM quay.io/jupyter/scipy-notebook:2026-10-12
+
+USER root
+# Drop the inherited Jupyter entrypoint/CMD: exec whatever command the harness passes, and
+# stay alive when it passes none.
+ENTRYPOINT ["/bin/sh", "-c", "exec \"$@\"", "-"]
+CMD ["sleep", "infinity"]
+
+# The conda prefix is the only Python in this image; link it into the standard /usr/local/bin
+# so that python3/pip3 still resolve when the harness switches to the agent/verifier user
+# with a login shell, which would otherwise drop /opt/conda/bin from PATH.
+RUN set -eux; \
+    ln -sf /opt/conda/bin/python3 /usr/local/bin/python3; \
+    ln -sf /opt/conda/bin/pip3 /usr/local/bin/pip3
 
 # Package mirrors -- the release build uses public mirrors (the original build
 # environment had no internet access and went through internal mirrors).

dataworld-p000221-cause-2/environment/Dockerfile

@@ -3,10 +3,33 @@
 # required.
 # Leak prevention: the whitelist COPYs only data/ -- WORLD.md, .build and dag.json must
 # never make it into the image.
-# [RELEASE] Replace the base image below with your own generic image (it needs
-# python3/pip, otherwise the apt step below installs them); the apt/pip steps all probe
-# for existence first, so anything the base image already provides is not reinstalled.
-FROM <base-image>:latest
+# Base image: a public, batteries-included Python environment -- the Jupyter scipy stack
+# (conda-forge). It already ships the data-science packages these tasks use (pandas, numpy,
+# scipy, statsmodels, scikit-learn, matplotlib, h5py, openpyxl, pytables, dask, numexpr);
+# pyarrow/pypdf/pillow are covered by the pip step below. The Docker Hub mirror of this
+# image is stale (2023-10), so the maintained quay.io registry is used, and the tag is a
+# dated release (index digest
+# sha256:8f8a5763d9a01e7f488eff8d9379654788e9cc08286dbcbc6bcda129c1b0edcd) so that builds
+# are reproducible -- `latest` tracks the same image. Any image with python3/pip works just
+# as well; the apt/pip steps probe for existence first, so nothing the base image already
+# provides is reinstalled. Two properties of this base image are reset below: it ends on a
+# non-root user (this build needs root to create the agent/verifier users and to open up
+# /app), and it carries a Jupyter entrypoint (the container is started by the harness with
+# its own command and has to stay up for `docker exec`).
+FROM quay.io/jupyter/scipy-notebook:2026-10-12
+
+USER root
+# Drop the inherited Jupyter entrypoint/CMD: exec whatever command the harness passes, and
+# stay alive when it passes none.
+ENTRYPOINT ["/bin/sh", "-c", "exec \"$@\"", "-"]
+CMD ["sleep", "infinity"]
+
+# The conda prefix is the only Python in this image; link it into the standard /usr/local/bin
+# so that python3/pip3 still resolve when the harness switches to the agent/verifier user
+# with a login shell, which would otherwise drop /opt/conda/bin from PATH.
+RUN set -eux; \
+    ln -sf /opt/conda/bin/python3 /usr/local/bin/python3; \
+    ln -sf /opt/conda/bin/pip3 /usr/local/bin/pip3
 
 # Package mirrors -- the release build uses public mirrors (the original build
 # environment had no internet access and went through internal mirrors).

dataworld-p000221-consequence-2/environment/Dockerfile

@@ -3,10 +3,33 @@
 # required.
 # Leak prevention: the whitelist COPYs only data/ -- WORLD.md, .build and dag.json must
 # never make it into the image.
-# [RELEASE] Replace the base image below with your own generic image (it needs
-# python3/pip, otherwise the apt step below installs them); the apt/pip steps all probe
-# for existence first, so anything the base image already provides is not reinstalled.
-FROM <base-image>:latest
+# Base image: a public, batteries-included Python environment -- the Jupyter scipy stack
+# (conda-forge). It already ships the data-science packages these tasks use (pandas, numpy,
+# scipy, statsmodels, scikit-learn, matplotlib, h5py, openpyxl, pytables, dask, numexpr);
+# pyarrow/pypdf/pillow are covered by the pip step below. The Docker Hub mirror of this
+# image is stale (2023-10), so the maintained quay.io registry is used, and the tag is a
+# dated release (index digest
+# sha256:8f8a5763d9a01e7f488eff8d9379654788e9cc08286dbcbc6bcda129c1b0edcd) so that builds
+# are reproducible -- `latest` tracks the same image. Any image with python3/pip works just
+# as well; the apt/pip steps probe for existence first, so nothing the base image already
+# provides is reinstalled. Two properties of this base image are reset below: it ends on a
+# non-root user (this build needs root to create the agent/verifier users and to open up
+# /app), and it carries a Jupyter entrypoint (the container is started by the harness with
+# its own command and has to stay up for `docker exec`).
+FROM quay.io/jupyter/scipy-notebook:2026-10-12
+
+USER root
+# Drop the inherited Jupyter entrypoint/CMD: exec whatever command the harness passes, and
+# stay alive when it passes none.
+ENTRYPOINT ["/bin/sh", "-c", "exec \"$@\"", "-"]
+CMD ["sleep", "infinity"]
+
+# The conda prefix is the only Python in this image; link it into the standard /usr/local/bin
+# so that python3/pip3 still resolve when the harness switches to the agent/verifier user
+# with a login shell, which would otherwise drop /opt/conda/bin from PATH.
+RUN set -eux; \
+    ln -sf /opt/conda/bin/python3 /usr/local/bin/python3; \
+    ln -sf /opt/conda/bin/pip3 /usr/local/bin/pip3
 
 # Package mirrors -- the release build uses public mirrors (the original build
 # environment had no internet access and went through internal mirrors).

dataworld-p000231-cause-1/environment/Dockerfile

@@ -3,10 +3,33 @@
 # required.
 # Leak prevention: the whitelist COPYs only data/ -- WORLD.md, .build and dag.json must
 # never make it into the image.
-# [RELEASE] Replace the base image below with your own generic image (it needs
-# python3/pip, otherwise the apt step below installs them); the apt/pip steps all probe
-# for existence first, so anything the base image already provides is not reinstalled.
-FROM <base-image>:latest
+# Base image: a public, batteries-included Python environment -- the Jupyter scipy stack
+# (conda-forge). It already ships the data-science packages these tasks use (pandas, numpy,
+# scipy, statsmodels, scikit-learn, matplotlib, h5py, openpyxl, pytables, dask, numexpr);
+# pyarrow/pypdf/pillow are covered by the pip step below. The Docker Hub mirror of this
+# image is stale (2023-10), so the maintained quay.io registry is used, and the tag is a
+# dated release (index digest
+# sha256:8f8a5763d9a01e7f488eff8d9379654788e9cc08286dbcbc6bcda129c1b0edcd) so that builds
+# are reproducible -- `latest` tracks the same image. Any image with python3/pip works just
+# as well; the apt/pip steps probe for existence first, so nothing the base image already
+# provides is reinstalled. Two properties of this base image are reset below: it ends on a
+# non-root user (this build needs root to create the agent/verifier users and to open up
+# /app), and it carries a Jupyter entrypoint (the container is started by the harness with
+# its own command and has to stay up for `docker exec`).
+FROM quay.io/jupyter/scipy-notebook:2026-10-12
+
+USER root
+# Drop the inherited Jupyter entrypoint/CMD: exec whatever command the harness passes, and
+# stay alive when it passes none.
+ENTRYPOINT ["/bin/sh", "-c", "exec \"$@\"", "-"]
+CMD ["sleep", "infinity"]
+
+# The conda prefix is the only Python in this image; link it into the standard /usr/local/bin
+# so that python3/pip3 still resolve when the harness switches to the agent/verifier user
+# with a login shell, which would otherwise drop /opt/conda/bin from PATH.
+RUN set -eux; \
+    ln -sf /opt/conda/bin/python3 /usr/local/bin/python3; \
+    ln -sf /opt/conda/bin/pip3 /usr/local/bin/pip3
 
 # Package mirrors -- the release build uses public mirrors (the original build
 # environment had no internet access and went through internal mirrors).

dataworld-p000231-consequence-2/environment/Dockerfile

@@ -3,10 +3,33 @@
 # required.
 # Leak prevention: the whitelist COPYs only data/ -- WORLD.md, .build and dag.json must
 # never make it into the image.
-# [RELEASE] Replace the base image below with your own generic image (it needs
-# python3/pip, otherwise the apt step below installs them); the apt/pip steps all probe
-# for existence first, so anything the base image already provides is not reinstalled.
-FROM <base-image>:latest
+# Base image: a public, batteries-included Python environment -- the Jupyter scipy stack
+# (conda-forge). It already ships the data-science packages these tasks use (pandas, numpy,
+# scipy, statsmodels, scikit-learn, matplotlib, h5py, openpyxl, pytables, dask, numexpr);
+# pyarrow/pypdf/pillow are covered by the pip step below. The Docker Hub mirror of this
+# image is stale (2023-10), so the maintained quay.io registry is used, and the tag is a
+# dated release (index digest
+# sha256:8f8a5763d9a01e7f488eff8d9379654788e9cc08286dbcbc6bcda129c1b0edcd) so that builds
+# are reproducible -- `latest` tracks the same image. Any image with python3/pip works just
+# as well; the apt/pip steps probe for existence first, so nothing the base image already
+# provides is reinstalled. Two properties of this base image are reset below: it ends on a
+# non-root user (this build needs root to create the agent/verifier users and to open up
+# /app), and it carries a Jupyter entrypoint (the container is started by the harness with
+# its own command and has to stay up for `docker exec`).
+FROM quay.io/jupyter/scipy-notebook:2026-10-12
+
+USER root
+# Drop the inherited Jupyter entrypoint/CMD: exec whatever command the harness passes, and
+# stay alive when it passes none.
+ENTRYPOINT ["/bin/sh", "-c", "exec \"$@\"", "-"]
+CMD ["sleep", "infinity"]
+
+# The conda prefix is the only Python in this image; link it into the standard /usr/local/bin
+# so that python3/pip3 still resolve when the harness switches to the agent/verifier user
+# with a login shell, which would otherwise drop /opt/conda/bin from PATH.
+RUN set -eux; \
+    ln -sf /opt/conda/bin/python3 /usr/local/bin/python3; \
+    ln -sf /opt/conda/bin/pip3 /usr/local/bin/pip3
 
 # Package mirrors -- the release build uses public mirrors (the original build
 # environment had no internet access and went through internal mirrors).

dataworld-p000233-cause-1/environment/Dockerfile

@@ -3,10 +3,33 @@
 # required.
 # Leak prevention: the whitelist COPYs only data/ -- WORLD.md, .build and dag.json must
 # never make it into the image.
-# [RELEASE] Replace the base image below with your own generic image (it needs
-# python3/pip, otherwise the apt step below installs them); the apt/pip steps all probe
-# for existence first, so anything the base image already provides is not reinstalled.
-FROM <base-image>:latest
+# Base image: a public, batteries-included Python environment -- the Jupyter scipy stack
+# (conda-forge). It already ships the data-science packages these tasks use (pandas, numpy,
+# scipy, statsmodels, scikit-learn, matplotlib, h5py, openpyxl, pytables, dask, numexpr);
+# pyarrow/pypdf/pillow are covered by the pip step below. The Docker Hub mirror of this
+# image is stale (2023-10), so the maintained quay.io registry is used, and the tag is a
+# dated release (index digest
+# sha256:8f8a5763d9a01e7f488eff8d9379654788e9cc08286dbcbc6bcda129c1b0edcd) so that builds
+# are reproducible -- `latest` tracks the same image. Any image with python3/pip works just
+# as well; the apt/pip steps probe for existence first, so nothing the base image already
+# provides is reinstalled. Two properties of this base image are reset below: it ends on a
+# non-root user (this build needs root to create the agent/verifier users and to open up
+# /app), and it carries a Jupyter entrypoint (the container is started by the harness with
+# its own command and has to stay up for `docker exec`).
+FROM quay.io/jupyter/scipy-notebook:2026-10-12
+
+USER root
+# Drop the inherited Jupyter entrypoint/CMD: exec whatever command the harness passes, and
+# stay alive when it passes none.
+ENTRYPOINT ["/bin/sh", "-c", "exec \"$@\"", "-"]
+CMD ["sleep", "infinity"]
+
+# The conda prefix is the only Python in this image; link it into the standard /usr/local/bin
+# so that python3/pip3 still resolve when the harness switches to the agent/verifier user
+# with a login shell, which would otherwise drop /opt/conda/bin from PATH.
+RUN set -eux; \
+    ln -sf /opt/conda/bin/python3 /usr/local/bin/python3; \
+    ln -sf /opt/conda/bin/pip3 /usr/local/bin/pip3
 
 # Package mirrors -- the release build uses public mirrors (the original build
 # environment had no internet access and went through internal mirrors).

dataworld-p000233-consequence-1/environment/Dockerfile

@@ -3,10 +3,33 @@
 # required.
 # Leak prevention: the whitelist COPYs only data/ -- WORLD.md, .build and dag.json must
 # never make it into the image.
-# [RELEASE] Replace the base image below with your own generic image (it needs
-# python3/pip, otherwise the apt step below installs them); the apt/pip steps all probe
-# for existence first, so anything the base image already provides is not reinstalled.
-FROM <base-image>:latest
+# Base image: a public, batteries-included Python environment -- the Jupyter scipy stack
+# (conda-forge). It already ships the data-science packages these tasks use (pandas, numpy,
+# scipy, statsmodels, scikit-learn, matplotlib, h5py, openpyxl, pytables, dask, numexpr);
+# pyarrow/pypdf/pillow are covered by the pip step below. The Docker Hub mirror of this
+# image is stale (2023-10), so the maintained quay.io registry is used, and the tag is a
+# dated release (index digest
+# sha256:8f8a5763d9a01e7f488eff8d9379654788e9cc08286dbcbc6bcda129c1b0edcd) so that builds
+# are reproducible -- `latest` tracks the same image. Any image with python3/pip works just
+# as well; the apt/pip steps probe for existence first, so nothing the base image already
+# provides is reinstalled. Two properties of this base image are reset below: it ends on a
+# non-root user (this build needs root to create the agent/verifier users and to open up
+# /app), and it carries a Jupyter entrypoint (the container is started by the harness with
+# its own command and has to stay up for `docker exec`).
+FROM quay.io/jupyter/scipy-notebook:2026-10-12
+
+USER root
+# Drop the inherited Jupyter entrypoint/CMD: exec whatever command the harness passes, and
+# stay alive when it passes none.
+ENTRYPOINT ["/bin/sh", "-c", "exec \"$@\"", "-"]
+CMD ["sleep", "infinity"]
+
+# The conda prefix is the only Python in this image; link it into the standard /usr/local/bin
+# so that python3/pip3 still resolve when the harness switches to the agent/verifier user
+# with a login shell, which would otherwise drop /opt/conda/bin from PATH.
+RUN set -eux; \
+    ln -sf /opt/conda/bin/python3 /usr/local/bin/python3; \
+    ln -sf /opt/conda/bin/pip3 /usr/local/bin/pip3
 
 # Package mirrors -- the release build uses public mirrors (the original build
 # environment had no internet access and went through internal mirrors).

dataworld-p000236-cause-2/environment/Dockerfile

@@ -3,10 +3,33 @@
 # required.
 # Leak prevention: the whitelist COPYs only data/ -- WORLD.md, .build and dag.json must
 # never make it into the image.
-# [RELEASE] Replace the base image below with your own generic image (it needs
-# python3/pip, otherwise the apt step below installs them); the apt/pip steps all probe
-# for existence first, so anything the base image already provides is not reinstalled.
-FROM <base-image>:latest
+# Base image: a public, batteries-included Python environment -- the Jupyter scipy stack
+# (conda-forge). It already ships the data-science packages these tasks use (pandas, numpy,
+# scipy, statsmodels, scikit-learn, matplotlib, h5py, openpyxl, pytables, dask, numexpr);
+# pyarrow/pypdf/pillow are covered by the pip step below. The Docker Hub mirror of this
+# image is stale (2023-10), so the maintained quay.io registry is used, and the tag is a
+# dated release (index digest
+# sha256:8f8a5763d9a01e7f488eff8d9379654788e9cc08286dbcbc6bcda129c1b0edcd) so that builds
+# are reproducible -- `latest` tracks the same image. Any image with python3/pip works just
+# as well; the apt/pip steps probe for existence first, so nothing the base image already
+# provides is reinstalled. Two properties of this base image are reset below: it ends on a
+# non-root user (this build needs root to create the agent/verifier users and to open up
+# /app), and it carries a Jupyter entrypoint (the container is started by the harness with
+# its own command and has to stay up for `docker exec`).
+FROM quay.io/jupyter/scipy-notebook:2026-10-12
+
+USER root
+# Drop the inherited Jupyter entrypoint/CMD: exec whatever command the harness passes, and
+# stay alive when it passes none.
+ENTRYPOINT ["/bin/sh", "-c", "exec \"$@\"", "-"]
+CMD ["sleep", "infinity"]
+
+# The conda prefix is the only Python in this image; link it into the standard /usr/local/bin
+# so that python3/pip3 still resolve when the harness switches to the agent/verifier user
+# with a login shell, which would otherwise drop /opt/conda/bin from PATH.
+RUN set -eux; \
+    ln -sf /opt/conda/bin/python3 /usr/local/bin/python3; \
+    ln -sf /opt/conda/bin/pip3 /usr/local/bin/pip3
 
 # Package mirrors -- the release build uses public mirrors (the original build
 # environment had no internet access and went through internal mirrors).

dataworld-p000236-consequence-1/environment/Dockerfile

@@ -3,10 +3,33 @@
 # required.
 # Leak prevention: the whitelist COPYs only data/ -- WORLD.md, .build and dag.json must
 # never make it into the image.
-# [RELEASE] Replace the base image below with your own generic image (it needs
-# python3/pip, otherwise the apt step below installs them); the apt/pip steps all probe
-# for existence first, so anything the base image already provides is not reinstalled.
-FROM <base-image>:latest
+# Base image: a public, batteries-included Python environment -- the Jupyter scipy stack
+# (conda-forge). It already ships the data-science packages these tasks use (pandas, numpy,
+# scipy, statsmodels, scikit-learn, matplotlib, h5py, openpyxl, pytables, dask, numexpr);
+# pyarrow/pypdf/pillow are covered by the pip step below. The Docker Hub mirror of this
+# image is stale (2023-10), so the maintained quay.io registry is used, and the tag is a
+# dated release (index digest
+# sha256:8f8a5763d9a01e7f488eff8d9379654788e9cc08286dbcbc6bcda129c1b0edcd) so that builds
+# are reproducible -- `latest` tracks the same image. Any image with python3/pip works just
+# as well; the apt/pip steps probe for existence first, so nothing the base image already
+# provides is reinstalled. Two properties of this base image are reset below: it ends on a
+# non-root user (this build needs root to create the agent/verifier users and to open up
+# /app), and it carries a Jupyter entrypoint (the container is started by the harness with
+# its own command and has to stay up for `docker exec`).
+FROM quay.io/jupyter/scipy-notebook:2026-10-12
+
+USER root
+# Drop the inherited Jupyter entrypoint/CMD: exec whatever command the harness passes, and
+# stay alive when it passes none.
+ENTRYPOINT ["/bin/sh", "-c", "exec \"$@\"", "-"]
+CMD ["sleep", "infinity"]
+
+# The conda prefix is the only Python in this image; link it into the standard /usr/local/bin
+# so that python3/pip3 still resolve when the harness switches to the agent/verifier user
+# with a login shell, which would otherwise drop /opt/conda/bin from PATH.
+RUN set -eux; \
+    ln -sf /opt/conda/bin/python3 /usr/local/bin/python3; \
+    ln -sf /opt/conda/bin/pip3 /usr/local/bin/pip3
 
 # Package mirrors -- the release build uses public mirrors (the original build
 # environment had no internet access and went through internal mirrors).

dataworld-p000249-cause-1/environment/Dockerfile

@@ -3,10 +3,33 @@
 # required.
 # Leak prevention: the whitelist COPYs only data/ -- WORLD.md, .build and dag.json must
 # never make it into the image.
-# [RELEASE] Replace the base image below with your own generic image (it needs
-# python3/pip, otherwise the apt step below installs them); the apt/pip steps all probe
-# for existence first, so anything the base image already provides is not reinstalled.
-FROM <base-image>:latest
+# Base image: a public, batteries-included Python environment -- the Jupyter scipy stack
+# (conda-forge). It already ships the data-science packages these tasks use (pandas, numpy,
+# scipy, statsmodels, scikit-learn, matplotlib, h5py, openpyxl, pytables, dask, numexpr);
+# pyarrow/pypdf/pillow are covered by the pip step below. The Docker Hub mirror of this
+# image is stale (2023-10), so the maintained quay.io registry is used, and the tag is a
+# dated release (index digest
+# sha256:8f8a5763d9a01e7f488eff8d9379654788e9cc08286dbcbc6bcda129c1b0edcd) so that builds
+# are reproducible -- `latest` tracks the same image. Any image with python3/pip works just
+# as well; the apt/pip steps probe for existence first, so nothing the base image already
+# provides is reinstalled. Two properties of this base image are reset below: it ends on a
+# non-root user (this build needs root to create the agent/verifier users and to open up
+# /app), and it carries a Jupyter entrypoint (the container is started by the harness with
+# its own command and has to stay up for `docker exec`).
+FROM quay.io/jupyter/scipy-notebook:2026-10-12
+
+USER root
+# Drop the inherited Jupyter entrypoint/CMD: exec whatever command the harness passes, and
+# stay alive when it passes none.
+ENTRYPOINT ["/bin/sh", "-c", "exec \"$@\"", "-"]
+CMD ["sleep", "infinity"]
+
+# The conda prefix is the only Python in this image; link it into the standard /usr/local/bin
+# so that python3/pip3 still resolve when the harness switches to the agent/verifier user
+# with a login shell, which would otherwise drop /opt/conda/bin from PATH.
+RUN set -eux; \
+    ln -sf /opt/conda/bin/python3 /usr/local/bin/python3; \
+    ln -sf /opt/conda/bin/pip3 /usr/local/bin/pip3
 
 # Package mirrors -- the release build uses public mirrors (the original build
 # environment had no internet access and went through internal mirrors).

dataworld-p000249-consequence-1/environment/Dockerfile

@@ -3,10 +3,33 @@
 # required.
 # Leak prevention: the whitelist COPYs only data/ -- WORLD.md, .build and dag.json must
 # never make it into the image.
-# [RELEASE] Replace the base image below with your own generic image (it needs
-# python3/pip, otherwise the apt step below installs them); the apt/pip steps all probe
-# for existence first, so anything the base image already provides is not reinstalled.
-FROM <base-image>:latest
+# Base image: a public, batteries-included Python environment -- the Jupyter scipy stack
+# (conda-forge). It already ships the data-science packages these tasks use (pandas, numpy,
+# scipy, statsmodels, scikit-learn, matplotlib, h5py, openpyxl, pytables, dask, numexpr);
+# pyarrow/pypdf/pillow are covered by the pip step below. The Docker Hub mirror of this
+# image is stale (2023-10), so the maintained quay.io registry is used, and the tag is a
+# dated release (index digest
+# sha256:8f8a5763d9a01e7f488eff8d9379654788e9cc08286dbcbc6bcda129c1b0edcd) so that builds
+# are reproducible -- `latest` tracks the same image. Any image with python3/pip works just
+# as well; the apt/pip steps probe for existence first, so nothing the base image already
+# provides is reinstalled. Two properties of this base image are reset below: it ends on a
+# non-root user (this build needs root to create the agent/verifier users and to open up
+# /app), and it carries a Jupyter entrypoint (the container is started by the harness with
+# its own command and has to stay up for `docker exec`).
+FROM quay.io/jupyter/scipy-notebook:2026-10-12
+
+USER root
+# Drop the inherited Jupyter entrypoint/CMD: exec whatever command the harness passes, and
+# stay alive when it passes none.
+ENTRYPOINT ["/bin/sh", "-c", "exec \"$@\"", "-"]
+CMD ["sleep", "infinity"]
+
+# The conda prefix is the only Python in this image; link it into the standard /usr/local/bin
+# so that python3/pip3 still resolve when the harness switches to the agent/verifier user
+# with a login shell, which would otherwise drop /opt/conda/bin from PATH.
+RUN set -eux; \
+    ln -sf /opt/conda/bin/python3 /usr/local/bin/python3; \
+    ln -sf /opt/conda/bin/pip3 /usr/local/bin/pip3
 
 # Package mirrors -- the release build uses public mirrors (the original build
 # environment had no internet access and went through internal mirrors).

dataworld-p000341-cause-2/environment/Dockerfile

@@ -3,10 +3,33 @@
 # required.
 # Leak prevention: the whitelist COPYs only data/ -- WORLD.md, .build and dag.json must
 # never make it into the image.
-# [RELEASE] Replace the base image below with your own generic image (it needs
-# python3/pip, otherwise the apt step below installs them); the apt/pip steps all probe
-# for existence first, so anything the base image already provides is not reinstalled.
-FROM <base-image>:latest
+# Base image: a public, batteries-included Python environment -- the Jupyter scipy stack
+# (conda-forge). It already ships the data-science packages these tasks use (pandas, numpy,
+# scipy, statsmodels, scikit-learn, matplotlib, h5py, openpyxl, pytables, dask, numexpr);
+# pyarrow/pypdf/pillow are covered by the pip step below. The Docker Hub mirror of this
+# image is stale (2023-10), so the maintained quay.io registry is used, and the tag is a
+# dated release (index digest
+# sha256:8f8a5763d9a01e7f488eff8d9379654788e9cc08286dbcbc6bcda129c1b0edcd) so that builds
+# are reproducible -- `latest` tracks the same image. Any image with python3/pip works just
+# as well; the apt/pip steps probe for existence first, so nothing the base image already
+# provides is reinstalled. Two properties of this base image are reset below: it ends on a
+# non-root user (this build needs root to create the agent/verifier users and to open up
+# /app), and it carries a Jupyter entrypoint (the container is started by the harness with
+# its own command and has to stay up for `docker exec`).
+FROM quay.io/jupyter/scipy-notebook:2026-10-12
+
+USER root
+# Drop the inherited Jupyter entrypoint/CMD: exec whatever command the harness passes, and
+# stay alive when it passes none.
+ENTRYPOINT ["/bin/sh", "-c", "exec \"$@\"", "-"]
+CMD ["sleep", "infinity"]
+
+# The conda prefix is the only Python in this image; link it into the standard /usr/local/bin
+# so that python3/pip3 still resolve when the harness switches to the agent/verifier user
+# with a login shell, which would otherwise drop /opt/conda/bin from PATH.
+RUN set -eux; \
+    ln -sf /opt/conda/bin/python3 /usr/local/bin/python3; \
+    ln -sf /opt/conda/bin/pip3 /usr/local/bin/pip3
 
 # Package mirrors -- the release build uses public mirrors (the original build
 # environment had no internet access and went through internal mirrors).

dataworld-p000341-consequence-2/environment/Dockerfile

@@ -3,10 +3,33 @@
 # required.
 # Leak prevention: the whitelist COPYs only data/ -- WORLD.md, .build and dag.json must
 # never make it into the image.
-# [RELEASE] Replace the base image below with your own generic image (it needs
-# python3/pip, otherwise the apt step below installs them); the apt/pip steps all probe
-# for existence first, so anything the base image already provides is not reinstalled.
-FROM <base-image>:latest
+# Base image: a public, batteries-included Python environment -- the Jupyter scipy stack
+# (conda-forge). It already ships the data-science packages these tasks use (pandas, numpy,
+# scipy, statsmodels, scikit-learn, matplotlib, h5py, openpyxl, pytables, dask, numexpr);
+# pyarrow/pypdf/pillow are covered by the pip step below. The Docker Hub mirror of this
+# image is stale (2023-10), so the maintained quay.io registry is used, and the tag is a
+# dated release (index digest
+# sha256:8f8a5763d9a01e7f488eff8d9379654788e9cc08286dbcbc6bcda129c1b0edcd) so that builds
+# are reproducible -- `latest` tracks the same image. Any image with python3/pip works just
+# as well; the apt/pip steps probe for existence first, so nothing the base image already
+# provides is reinstalled. Two properties of this base image are reset below: it ends on a
+# non-root user (this build needs root to create the agent/verifier users and to open up
+# /app), and it carries a Jupyter entrypoint (the container is started by the harness with
+# its own command and has to stay up for `docker exec`).
+FROM quay.io/jupyter/scipy-notebook:2026-10-12
+
+USER root
+# Drop the inherited Jupyter entrypoint/CMD: exec whatever command the harness passes, and
+# stay alive when it passes none.
+ENTRYPOINT ["/bin/sh", "-c", "exec \"$@\"", "-"]
+CMD ["sleep", "infinity"]
+
+# The conda prefix is the only Python in this image; link it into the standard /usr/local/bin
+# so that python3/pip3 still resolve when the harness switches to the agent/verifier user
+# with a login shell, which would otherwise drop /opt/conda/bin from PATH.
+RUN set -eux; \
+    ln -sf /opt/conda/bin/python3 /usr/local/bin/python3; \
+    ln -sf /opt/conda/bin/pip3 /usr/local/bin/pip3
 
 # Package mirrors -- the release build uses public mirrors (the original build
 # environment had no internet access and went through internal mirrors).

dataworld-p000348-cause-2/environment/Dockerfile

@@ -3,10 +3,33 @@
 # required.
 # Leak prevention: the whitelist COPYs only data/ -- WORLD.md, .build and dag.json must
 # never make it into the image.
-# [RELEASE] Replace the base image below with your own generic image (it needs
-# python3/pip, otherwise the apt step below installs them); the apt/pip steps all probe
-# for existence first, so anything the base image already provides is not reinstalled.
-FROM <base-image>:latest
+# Base image: a public, batteries-included Python environment -- the Jupyter scipy stack
+# (conda-forge). It already ships the data-science packages these tasks use (pandas, numpy,
+# scipy, statsmodels, scikit-learn, matplotlib, h5py, openpyxl, pytables, dask, numexpr);
+# pyarrow/pypdf/pillow are covered by the pip step below. The Docker Hub mirror of this
+# image is stale (2023-10), so the maintained quay.io registry is used, and the tag is a
+# dated release (index digest
+# sha256:8f8a5763d9a01e7f488eff8d9379654788e9cc08286dbcbc6bcda129c1b0edcd) so that builds
+# are reproducible -- `latest` tracks the same image. Any image with python3/pip works just
+# as well; the apt/pip steps probe for existence first, so nothing the base image already
+# provides is reinstalled. Two properties of this base image are reset below: it ends on a
+# non-root user (this build needs root to create the agent/verifier users and to open up
+# /app), and it carries a Jupyter entrypoint (the container is started by the harness with
+# its own command and has to stay up for `docker exec`).
+FROM quay.io/jupyter/scipy-notebook:2026-10-12
+
+USER root
+# Drop the inherited Jupyter entrypoint/CMD: exec whatever command the harness passes, and
+# stay alive when it passes none.
+ENTRYPOINT ["/bin/sh", "-c", "exec \"$@\"", "-"]
+CMD ["sleep", "infinity"]
+
+# The conda prefix is the only Python in this image; link it into the standard /usr/local/bin
+# so that python3/pip3 still resolve when the harness switches to the agent/verifier user
+# with a login shell, which would otherwise drop /opt/conda/bin from PATH.
+RUN set -eux; \
+    ln -sf /opt/conda/bin/python3 /usr/local/bin/python3; \
+    ln -sf /opt/conda/bin/pip3 /usr/local/bin/pip3
 
 # Package mirrors -- the release build uses public mirrors (the original build
 # environment had no internet access and went through internal mirrors).

dataworld-p000348-consequence-1/environment/Dockerfile

@@ -3,10 +3,33 @@
 # required.
 # Leak prevention: the whitelist COPYs only data/ -- WORLD.md, .build and dag.json must
 # never make it into the image.
-# [RELEASE] Replace the base image below with your own generic image (it needs
-# python3/pip, otherwise the apt step below installs them); the apt/pip steps all probe
-# for existence first, so anything the base image already provides is not reinstalled.
-FROM <base-image>:latest
+# Base image: a public, batteries-included Python environment -- the Jupyter scipy stack
+# (conda-forge). It already ships the data-science packages these tasks use (pandas, numpy,
+# scipy, statsmodels, scikit-learn, matplotlib, h5py, openpyxl, pytables, dask, numexpr);
+# pyarrow/pypdf/pillow are covered by the pip step below. The Docker Hub mirror of this
+# image is stale (2023-10), so the maintained quay.io registry is used, and the tag is a
+# dated release (index digest
+# sha256:8f8a5763d9a01e7f488eff8d9379654788e9cc08286dbcbc6bcda129c1b0edcd) so that builds
+# are reproducible -- `latest` tracks the same image. Any image with python3/pip works just
+# as well; the apt/pip steps probe for existence first, so nothing the base image already
+# provides is reinstalled. Two properties of this base image are reset below: it ends on a
+# non-root user (this build needs root to create the agent/verifier users and to open up
+# /app), and it carries a Jupyter entrypoint (the container is started by the harness with
+# its own command and has to stay up for `docker exec`).
+FROM quay.io/jupyter/scipy-notebook:2026-10-12
+
+USER root
+# Drop the inherited Jupyter entrypoint/CMD: exec whatever command the harness passes, and
+# stay alive when it passes none.
+ENTRYPOINT ["/bin/sh", "-c", "exec \"$@\"", "-"]
+CMD ["sleep", "infinity"]
+
+# The conda prefix is the only Python in this image; link it into the standard /usr/local/bin
+# so that python3/pip3 still resolve when the harness switches to the agent/verifier user
+# with a login shell, which would otherwise drop /opt/conda/bin from PATH.
+RUN set -eux; \
+    ln -sf /opt/conda/bin/python3 /usr/local/bin/python3; \
+    ln -sf /opt/conda/bin/pip3 /usr/local/bin/pip3
 
 # Package mirrors -- the release build uses public mirrors (the original build
 # environment had no internet access and went through internal mirrors).

dataworld-p000350-cause-2/environment/Dockerfile

@@ -3,10 +3,33 @@
 # required.
 # Leak prevention: the whitelist COPYs only data/ -- WORLD.md, .build and dag.json must
 # never make it into the image.
-# [RELEASE] Replace the base image below with your own generic image (it needs
-# python3/pip, otherwise the apt step below installs them); the apt/pip steps all probe
-# for existence first, so anything the base image already provides is not reinstalled.
-FROM <base-image>:latest
+# Base image: a public, batteries-included Python environment -- the Jupyter scipy stack
+# (conda-forge). It already ships the data-science packages these tasks use (pandas, numpy,
+# scipy, statsmodels, scikit-learn, matplotlib, h5py, openpyxl, pytables, dask, numexpr);
+# pyarrow/pypdf/pillow are covered by the pip step below. The Docker Hub mirror of this
+# image is stale (2023-10), so the maintained quay.io registry is used, and the tag is a
+# dated release (index digest
+# sha256:8f8a5763d9a01e7f488eff8d9379654788e9cc08286dbcbc6bcda129c1b0edcd) so that builds
+# are reproducible -- `latest` tracks the same image. Any image with python3/pip works just
+# as well; the apt/pip steps probe for existence first, so nothing the base image already
+# provides is reinstalled. Two properties of this base image are reset below: it ends on a
+# non-root user (this build needs root to create the agent/verifier users and to open up
+# /app), and it carries a Jupyter entrypoint (the container is started by the harness with
+# its own command and has to stay up for `docker exec`).
+FROM quay.io/jupyter/scipy-notebook:2026-10-12
+
+USER root
+# Drop the inherited Jupyter entrypoint/CMD: exec whatever command the harness passes, and
+# stay alive when it passes none.
+ENTRYPOINT ["/bin/sh", "-c", "exec \"$@\"", "-"]
+CMD ["sleep", "infinity"]
+
+# The conda prefix is the only Python in this image; link it into the standard /usr/local/bin
+# so that python3/pip3 still resolve when the harness switches to the agent/verifier user
+# with a login shell, which would otherwise drop /opt/conda/bin from PATH.
+RUN set -eux; \
+    ln -sf /opt/conda/bin/python3 /usr/local/bin/python3; \
+    ln -sf /opt/conda/bin/pip3 /usr/local/bin/pip3
 
 # Package mirrors -- the release build uses public mirrors (the original build
 # environment had no internet access and went through internal mirrors).

dataworld-p000350-consequence-2/environment/Dockerfile

@@ -3,10 +3,33 @@
 # required.
 # Leak prevention: the whitelist COPYs only data/ -- WORLD.md, .build and dag.json must
 # never make it into the image.
-# [RELEASE] Replace the base image below with your own generic image (it needs
-# python3/pip, otherwise the apt step below installs them); the apt/pip steps all probe
-# for existence first, so anything the base image already provides is not reinstalled.
-FROM <base-image>:latest
+# Base image: a public, batteries-included Python environment -- the Jupyter scipy stack
+# (conda-forge). It already ships the data-science packages these tasks use (pandas, numpy,
+# scipy, statsmodels, scikit-learn, matplotlib, h5py, openpyxl, pytables, dask, numexpr);
+# pyarrow/pypdf/pillow are covered by the pip step below. The Docker Hub mirror of this
+# image is stale (2023-10), so the maintained quay.io registry is used, and the tag is a
+# dated release (index digest
+# sha256:8f8a5763d9a01e7f488eff8d9379654788e9cc08286dbcbc6bcda129c1b0edcd) so that builds
+# are reproducible -- `latest` tracks the same image. Any image with python3/pip works just
+# as well; the apt/pip steps probe for existence first, so nothing the base image already
+# provides is reinstalled. Two properties of this base image are reset below: it ends on a
+# non-root user (this build needs root to create the agent/verifier users and to open up
+# /app), and it carries a Jupyter entrypoint (the container is started by the harness with
+# its own command and has to stay up for `docker exec`).
+FROM quay.io/jupyter/scipy-notebook:2026-10-12
+
+USER root
+# Drop the inherited Jupyter entrypoint/CMD: exec whatever command the harness passes, and
+# stay alive when it passes none.
+ENTRYPOINT ["/bin/sh", "-c", "exec \"$@\"", "-"]
+CMD ["sleep", "infinity"]
+
+# The conda prefix is the only Python in this image; link it into the standard /usr/local/bin
+# so that python3/pip3 still resolve when the harness switches to the agent/verifier user
+# with a login shell, which would otherwise drop /opt/conda/bin from PATH.
+RUN set -eux; \
+    ln -sf /opt/conda/bin/python3 /usr/local/bin/python3; \
+    ln -sf /opt/conda/bin/pip3 /usr/local/bin/pip3
 
 # Package mirrors -- the release build uses public mirrors (the original build
 # environment had no internet access and went through internal mirrors).

dataworld-p000372-cause-2/environment/Dockerfile

@@ -3,10 +3,33 @@
 # required.
 # Leak prevention: the whitelist COPYs only data/ -- WORLD.md, .build and dag.json must
 # never make it into the image.
-# [RELEASE] Replace the base image below with your own generic image (it needs
-# python3/pip, otherwise the apt step below installs them); the apt/pip steps all probe
-# for existence first, so anything the base image already provides is not reinstalled.
-FROM <base-image>:latest
+# Base image: a public, batteries-included Python environment -- the Jupyter scipy stack
+# (conda-forge). It already ships the data-science packages these tasks use (pandas, numpy,
+# scipy, statsmodels, scikit-learn, matplotlib, h5py, openpyxl, pytables, dask, numexpr);
+# pyarrow/pypdf/pillow are covered by the pip step below. The Docker Hub mirror of this
+# image is stale (2023-10), so the maintained quay.io registry is used, and the tag is a
+# dated release (index digest
+# sha256:8f8a5763d9a01e7f488eff8d9379654788e9cc08286dbcbc6bcda129c1b0edcd) so that builds
+# are reproducible -- `latest` tracks the same image. Any image with python3/pip works just
+# as well; the apt/pip steps probe for existence first, so nothing the base image already
+# provides is reinstalled. Two properties of this base image are reset below: it ends on a
+# non-root user (this build needs root to create the agent/verifier users and to open up
+# /app), and it carries a Jupyter entrypoint (the container is started by the harness with
+# its own command and has to stay up for `docker exec`).
+FROM quay.io/jupyter/scipy-notebook:2026-10-12
+
+USER root
+# Drop the inherited Jupyter entrypoint/CMD: exec whatever command the harness passes, and
+# stay alive when it passes none.
+ENTRYPOINT ["/bin/sh", "-c", "exec \"$@\"", "-"]
+CMD ["sleep", "infinity"]
+
+# The conda prefix is the only Python in this image; link it into the standard /usr/local/bin
+# so that python3/pip3 still resolve when the harness switches to the agent/verifier user
+# with a login shell, which would otherwise drop /opt/conda/bin from PATH.
+RUN set -eux; \
+    ln -sf /opt/conda/bin/python3 /usr/local/bin/python3; \
+    ln -sf /opt/conda/bin/pip3 /usr/local/bin/pip3
 
 # Package mirrors -- the release build uses public mirrors (the original build
 # environment had no internet access and went through internal mirrors).

dataworld-p000372-consequence-1/environment/Dockerfile

@@ -3,10 +3,33 @@
 # required.
 # Leak prevention: the whitelist COPYs only data/ -- WORLD.md, .build and dag.json must
 # never make it into the image.
-# [RELEASE] Replace the base image below with your own generic image (it needs
-# python3/pip, otherwise the apt step below installs them); the apt/pip steps all probe
-# for existence first, so anything the base image already provides is not reinstalled.
-FROM <base-image>:latest
+# Base image: a public, batteries-included Python environment -- the Jupyter scipy stack
+# (conda-forge). It already ships the data-science packages these tasks use (pandas, numpy,
+# scipy, statsmodels, scikit-learn, matplotlib, h5py, openpyxl, pytables, dask, numexpr);
+# pyarrow/pypdf/pillow are covered by the pip step below. The Docker Hub mirror of this
+# image is stale (2023-10), so the maintained quay.io registry is used, and the tag is a
+# dated release (index digest
+# sha256:8f8a5763d9a01e7f488eff8d9379654788e9cc08286dbcbc6bcda129c1b0edcd) so that builds
+# are reproducible -- `latest` tracks the same image. Any image with python3/pip works just
+# as well; the apt/pip steps probe for existence first, so nothing the base image already
+# provides is reinstalled. Two properties of this base image are reset below: it ends on a
+# non-root user (this build needs root to create the agent/verifier users and to open up
+# /app), and it carries a Jupyter entrypoint (the container is started by the harness with
+# its own command and has to stay up for `docker exec`).
+FROM quay.io/jupyter/scipy-notebook:2026-10-12
+
+USER root
+# Drop the inherited Jupyter entrypoint/CMD: exec whatever command the harness passes, and
+# stay alive when it passes none.
+ENTRYPOINT ["/bin/sh", "-c", "exec \"$@\"", "-"]
+CMD ["sleep", "infinity"]
+
+# The conda prefix is the only Python in this image; link it into the standard /usr/local/bin
+# so that python3/pip3 still resolve when the harness switches to the agent/verifier user
+# with a login shell, which would otherwise drop /opt/conda/bin from PATH.
+RUN set -eux; \
+    ln -sf /opt/conda/bin/python3 /usr/local/bin/python3; \
+    ln -sf /opt/conda/bin/pip3 /usr/local/bin/pip3
 
 # Package mirrors -- the release build uses public mirrors (the original build
 # environment had no internet access and went through internal mirrors).

dataworld-p000384-cause-1/environment/Dockerfile

@@ -3,10 +3,33 @@
 # required.
 # Leak prevention: the whitelist COPYs only data/ -- WORLD.md, .build and dag.json must
 # never make it into the image.
-# [RELEASE] Replace the base image below with your own generic image (it needs
-# python3/pip, otherwise the apt step below installs them); the apt/pip steps all probe
-# for existence first, so anything the base image already provides is not reinstalled.
-FROM <base-image>:latest
+# Base image: a public, batteries-included Python environment -- the Jupyter scipy stack
+# (conda-forge). It already ships the data-science packages these tasks use (pandas, numpy,
+# scipy, statsmodels, scikit-learn, matplotlib, h5py, openpyxl, pytables, dask, numexpr);
+# pyarrow/pypdf/pillow are covered by the pip step below. The Docker Hub mirror of this
+# image is stale (2023-10), so the maintained quay.io registry is used, and the tag is a
+# dated release (index digest
+# sha256:8f8a5763d9a01e7f488eff8d9379654788e9cc08286dbcbc6bcda129c1b0edcd) so that builds
+# are reproducible -- `latest` tracks the same image. Any image with python3/pip works just
+# as well; the apt/pip steps probe for existence first, so nothing the base image already
+# provides is reinstalled. Two properties of this base image are reset below: it ends on a
+# non-root user (this build needs root to create the agent/verifier users and to open up
+# /app), and it carries a Jupyter entrypoint (the container is started by the harness with
+# its own command and has to stay up for `docker exec`).
+FROM quay.io/jupyter/scipy-notebook:2026-10-12
+
+USER root
+# Drop the inherited Jupyter entrypoint/CMD: exec whatever command the harness passes, and
+# stay alive when it passes none.
+ENTRYPOINT ["/bin/sh", "-c", "exec \"$@\"", "-"]
+CMD ["sleep", "infinity"]
+
+# The conda prefix is the only Python in this image; link it into the standard /usr/local/bin
+# so that python3/pip3 still resolve when the harness switches to the agent/verifier user
+# with a login shell, which would otherwise drop /opt/conda/bin from PATH.
+RUN set -eux; \
+    ln -sf /opt/conda/bin/python3 /usr/local/bin/python3; \
+    ln -sf /opt/conda/bin/pip3 /usr/local/bin/pip3
 
 # Package mirrors -- the release build uses public mirrors (the original build
 # environment had no internet access and went through internal mirrors).

dataworld-p000384-consequence-1/environment/Dockerfile

@@ -3,10 +3,33 @@
 # required.
 # Leak prevention: the whitelist COPYs only data/ -- WORLD.md, .build and dag.json must
 # never make it into the image.
-# [RELEASE] Replace the base image below with your own generic image (it needs
-# python3/pip, otherwise the apt step below installs them); the apt/pip steps all probe
-# for existence first, so anything the base image already provides is not reinstalled.
-FROM <base-image>:latest
+# Base image: a public, batteries-included Python environment -- the Jupyter scipy stack
+# (conda-forge). It already ships the data-science packages these tasks use (pandas, numpy,
+# scipy, statsmodels, scikit-learn, matplotlib, h5py, openpyxl, pytables, dask, numexpr);
+# pyarrow/pypdf/pillow are covered by the pip step below. The Docker Hub mirror of this
+# image is stale (2023-10), so the maintained quay.io registry is used, and the tag is a
+# dated release (index digest
+# sha256:8f8a5763d9a01e7f488eff8d9379654788e9cc08286dbcbc6bcda129c1b0edcd) so that builds
+# are reproducible -- `latest` tracks the same image. Any image with python3/pip works just
+# as well; the apt/pip steps probe for existence first, so nothing the base image already
+# provides is reinstalled. Two properties of this base image are reset below: it ends on a
+# non-root user (this build needs root to create the agent/verifier users and to open up
+# /app), and it carries a Jupyter entrypoint (the container is started by the harness with
+# its own command and has to stay up for `docker exec`).
+FROM quay.io/jupyter/scipy-notebook:2026-10-12
+
+USER root
+# Drop the inherited Jupyter entrypoint/CMD: exec whatever command the harness passes, and
+# stay alive when it passes none.
+ENTRYPOINT ["/bin/sh", "-c", "exec \"$@\"", "-"]
+CMD ["sleep", "infinity"]
+
+# The conda prefix is the only Python in this image; link it into the standard /usr/local/bin
+# so that python3/pip3 still resolve when the harness switches to the agent/verifier user
+# with a login shell, which would otherwise drop /opt/conda/bin from PATH.
+RUN set -eux; \
+    ln -sf /opt/conda/bin/python3 /usr/local/bin/python3; \
+    ln -sf /opt/conda/bin/pip3 /usr/local/bin/pip3
 
 # Package mirrors -- the release build uses public mirrors (the original build
 # environment had no internet access and went through internal mirrors).

dataworld-p000419-cause-2/environment/Dockerfile

@@ -3,10 +3,33 @@
 # required.
 # Leak prevention: the whitelist COPYs only data/ -- WORLD.md, .build and dag.json must
 # never make it into the image.
-# [RELEASE] Replace the base image below with your own generic image (it needs
-# python3/pip, otherwise the apt step below installs them); the apt/pip steps all probe
-# for existence first, so anything the base image already provides is not reinstalled.
-FROM <base-image>:latest
+# Base image: a public, batteries-included Python environment -- the Jupyter scipy stack
+# (conda-forge). It already ships the data-science packages these tasks use (pandas, numpy,
+# scipy, statsmodels, scikit-learn, matplotlib, h5py, openpyxl, pytables, dask, numexpr);
+# pyarrow/pypdf/pillow are covered by the pip step below. The Docker Hub mirror of this
+# image is stale (2023-10), so the maintained quay.io registry is used, and the tag is a
+# dated release (index digest
+# sha256:8f8a5763d9a01e7f488eff8d9379654788e9cc08286dbcbc6bcda129c1b0edcd) so that builds
+# are reproducible -- `latest` tracks the same image. Any image with python3/pip works just
+# as well; the apt/pip steps probe for existence first, so nothing the base image already
+# provides is reinstalled. Two properties of this base image are reset below: it ends on a
+# non-root user (this build needs root to create the agent/verifier users and to open up
+# /app), and it carries a Jupyter entrypoint (the container is started by the harness with
+# its own command and has to stay up for `docker exec`).
+FROM quay.io/jupyter/scipy-notebook:2026-10-12
+
+USER root
+# Drop the inherited Jupyter entrypoint/CMD: exec whatever command the harness passes, and
+# stay alive when it passes none.
+ENTRYPOINT ["/bin/sh", "-c", "exec \"$@\"", "-"]
+CMD ["sleep", "infinity"]
+
+# The conda prefix is the only Python in this image; link it into the standard /usr/local/bin
+# so that python3/pip3 still resolve when the harness switches to the agent/verifier user
+# with a login shell, which would otherwise drop /opt/conda/bin from PATH.
+RUN set -eux; \
+    ln -sf /opt/conda/bin/python3 /usr/local/bin/python3; \
+    ln -sf /opt/conda/bin/pip3 /usr/local/bin/pip3
 
 # Package mirrors -- the release build uses public mirrors (the original build
 # environment had no internet access and went through internal mirrors).

dataworld-p000419-consequence-1/environment/Dockerfile

@@ -3,10 +3,33 @@
 # required.
 # Leak prevention: the whitelist COPYs only data/ -- WORLD.md, .build and dag.json must
 # never make it into the image.
-# [RELEASE] Replace the base image below with your own generic image (it needs
-# python3/pip, otherwise the apt step below installs them); the apt/pip steps all probe
-# for existence first, so anything the base image already provides is not reinstalled.
-FROM <base-image>:latest
+# Base image: a public, batteries-included Python environment -- the Jupyter scipy stack
+# (conda-forge). It already ships the data-science packages these tasks use (pandas, numpy,
+# scipy, statsmodels, scikit-learn, matplotlib, h5py, openpyxl, pytables, dask, numexpr);
+# pyarrow/pypdf/pillow are covered by the pip step below. The Docker Hub mirror of this
+# image is stale (2023-10), so the maintained quay.io registry is used, and the tag is a
+# dated release (index digest
+# sha256:8f8a5763d9a01e7f488eff8d9379654788e9cc08286dbcbc6bcda129c1b0edcd) so that builds
+# are reproducible -- `latest` tracks the same image. Any image with python3/pip works just
+# as well; the apt/pip steps probe for existence first, so nothing the base image already
+# provides is reinstalled. Two properties of this base image are reset below: it ends on a
+# non-root user (this build needs root to create the agent/verifier users and to open up
+# /app), and it carries a Jupyter entrypoint (the container is started by the harness with
+# its own command and has to stay up for `docker exec`).
+FROM quay.io/jupyter/scipy-notebook:2026-10-12
+
+USER root
+# Drop the inherited Jupyter entrypoint/CMD: exec whatever command the harness passes, and
+# stay alive when it passes none.
+ENTRYPOINT ["/bin/sh", "-c", "exec \"$@\"", "-"]
+CMD ["sleep", "infinity"]
+
+# The conda prefix is the only Python in this image; link it into the standard /usr/local/bin
+# so that python3/pip3 still resolve when the harness switches to the agent/verifier user
+# with a login shell, which would otherwise drop /opt/conda/bin from PATH.
+RUN set -eux; \
+    ln -sf /opt/conda/bin/python3 /usr/local/bin/python3; \
+    ln -sf /opt/conda/bin/pip3 /usr/local/bin/pip3
 
 # Package mirrors -- the release build uses public mirrors (the original build
 # environment had no internet access and went through internal mirrors).

dataworld-p000478-cause-1/environment/Dockerfile

@@ -3,10 +3,33 @@
 # required.
 # Leak prevention: the whitelist COPYs only data/ -- WORLD.md, .build and dag.json must
 # never make it into the image.
-# [RELEASE] Replace the base image below with your own generic image (it needs
-# python3/pip, otherwise the apt step below installs them); the apt/pip steps all probe
-# for existence first, so anything the base image already provides is not reinstalled.
-FROM <base-image>:latest
+# Base image: a public, batteries-included Python environment -- the Jupyter scipy stack
+# (conda-forge). It already ships the data-science packages these tasks use (pandas, numpy,
+# scipy, statsmodels, scikit-learn, matplotlib, h5py, openpyxl, pytables, dask, numexpr);
+# pyarrow/pypdf/pillow are covered by the pip step below. The Docker Hub mirror of this
+# image is stale (2023-10), so the maintained quay.io registry is used, and the tag is a
+# dated release (index digest
+# sha256:8f8a5763d9a01e7f488eff8d9379654788e9cc08286dbcbc6bcda129c1b0edcd) so that builds
+# are reproducible -- `latest` tracks the same image. Any image with python3/pip works just
+# as well; the apt/pip steps probe for existence first, so nothing the base image already
+# provides is reinstalled. Two properties of this base image are reset below: it ends on a
+# non-root user (this build needs root to create the agent/verifier users and to open up
+# /app), and it carries a Jupyter entrypoint (the container is started by the harness with
+# its own command and has to stay up for `docker exec`).
+FROM quay.io/jupyter/scipy-notebook:2026-10-12
+
+USER root
+# Drop the inherited Jupyter entrypoint/CMD: exec whatever command the harness passes, and
+# stay alive when it passes none.
+ENTRYPOINT ["/bin/sh", "-c", "exec \"$@\"", "-"]
+CMD ["sleep", "infinity"]
+
+# The conda prefix is the only Python in this image; link it into the standard /usr/local/bin
+# so that python3/pip3 still resolve when the harness switches to the agent/verifier user
+# with a login shell, which would otherwise drop /opt/conda/bin from PATH.
+RUN set -eux; \
+    ln -sf /opt/conda/bin/python3 /usr/local/bin/python3; \
+    ln -sf /opt/conda/bin/pip3 /usr/local/bin/pip3
 
 # Package mirrors -- the release build uses public mirrors (the original build
 # environment had no internet access and went through internal mirrors).

dataworld-p000478-consequence-1/environment/Dockerfile

@@ -3,10 +3,33 @@
 # required.
 # Leak prevention: the whitelist COPYs only data/ -- WORLD.md, .build and dag.json must
 # never make it into the image.
-# [RELEASE] Replace the base image below with your own generic image (it needs
-# python3/pip, otherwise the apt step below installs them); the apt/pip steps all probe
-# for existence first, so anything the base image already provides is not reinstalled.
-FROM <base-image>:latest
+# Base image: a public, batteries-included Python environment -- the Jupyter scipy stack
+# (conda-forge). It already ships the data-science packages these tasks use (pandas, numpy,
+# scipy, statsmodels, scikit-learn, matplotlib, h5py, openpyxl, pytables, dask, numexpr);
+# pyarrow/pypdf/pillow are covered by the pip step below. The Docker Hub mirror of this
+# image is stale (2023-10), so the maintained quay.io registry is used, and the tag is a
+# dated release (index digest
+# sha256:8f8a5763d9a01e7f488eff8d9379654788e9cc08286dbcbc6bcda129c1b0edcd) so that builds
+# are reproducible -- `latest` tracks the same image. Any image with python3/pip works just
+# as well; the apt/pip steps probe for existence first, so nothing the base image already
+# provides is reinstalled. Two properties of this base image are reset below: it ends on a
+# non-root user (this build needs root to create the agent/verifier users and to open up
+# /app), and it carries a Jupyter entrypoint (the container is started by the harness with
+# its own command and has to stay up for `docker exec`).
+FROM quay.io/jupyter/scipy-notebook:2026-10-12
+
+USER root
+# Drop the inherited Jupyter entrypoint/CMD: exec whatever command the harness passes, and
+# stay alive when it passes none.
+ENTRYPOINT ["/bin/sh", "-c", "exec \"$@\"", "-"]
+CMD ["sleep", "infinity"]
+
+# The conda prefix is the only Python in this image; link it into the standard /usr/local/bin
+# so that python3/pip3 still resolve when the harness switches to the agent/verifier user
+# with a login shell, which would otherwise drop /opt/conda/bin from PATH.
+RUN set -eux; \
+    ln -sf /opt/conda/bin/python3 /usr/local/bin/python3; \
+    ln -sf /opt/conda/bin/pip3 /usr/local/bin/pip3
 
 # Package mirrors -- the release build uses public mirrors (the original build
 # environment had no internet access and went through internal mirrors).

dataworld-p000565-cause-1/environment/Dockerfile

@@ -3,10 +3,33 @@
 # required.
 # Leak prevention: the whitelist COPYs only data/ -- WORLD.md, .build and dag.json must
 # never make it into the image.
-# [RELEASE] Replace the base image below with your own generic image (it needs
-# python3/pip, otherwise the apt step below installs them); the apt/pip steps all probe
-# for existence first, so anything the base image already provides is not reinstalled.
-FROM <base-image>:latest
+# Base image: a public, batteries-included Python environment -- the Jupyter scipy stack
+# (conda-forge). It already ships the data-science packages these tasks use (pandas, numpy,
+# scipy, statsmodels, scikit-learn, matplotlib, h5py, openpyxl, pytables, dask, numexpr);
+# pyarrow/pypdf/pillow are covered by the pip step below. The Docker Hub mirror of this
+# image is stale (2023-10), so the maintained quay.io registry is used, and the tag is a
+# dated release (index digest
+# sha256:8f8a5763d9a01e7f488eff8d9379654788e9cc08286dbcbc6bcda129c1b0edcd) so that builds
+# are reproducible -- `latest` tracks the same image. Any image with python3/pip works just
+# as well; the apt/pip steps probe for existence first, so nothing the base image already
+# provides is reinstalled. Two properties of this base image are reset below: it ends on a
+# non-root user (this build needs root to create the agent/verifier users and to open up
+# /app), and it carries a Jupyter entrypoint (the container is started by the harness with
+# its own command and has to stay up for `docker exec`).
+FROM quay.io/jupyter/scipy-notebook:2026-10-12
+
+USER root
+# Drop the inherited Jupyter entrypoint/CMD: exec whatever command the harness passes, and
+# stay alive when it passes none.
+ENTRYPOINT ["/bin/sh", "-c", "exec \"$@\"", "-"]
+CMD ["sleep", "infinity"]
+
+# The conda prefix is the only Python in this image; link it into the standard /usr/local/bin
+# so that python3/pip3 still resolve when the harness switches to the agent/verifier user
+# with a login shell, which would otherwise drop /opt/conda/bin from PATH.
+RUN set -eux; \
+    ln -sf /opt/conda/bin/python3 /usr/local/bin/python3; \
+    ln -sf /opt/conda/bin/pip3 /usr/local/bin/pip3
 
 # Package mirrors -- the release build uses public mirrors (the original build
 # environment had no internet access and went through internal mirrors).

dataworld-p000565-consequence-1/environment/Dockerfile

@@ -3,10 +3,33 @@
 # required.
 # Leak prevention: the whitelist COPYs only data/ -- WORLD.md, .build and dag.json must
 # never make it into the image.
-# [RELEASE] Replace the base image below with your own generic image (it needs
-# python3/pip, otherwise the apt step below installs them); the apt/pip steps all probe
-# for existence first, so anything the base image already provides is not reinstalled.
-FROM <base-image>:latest
+# Base image: a public, batteries-included Python environment -- the Jupyter scipy stack
+# (conda-forge). It already ships the data-science packages these tasks use (pandas, numpy,
+# scipy, statsmodels, scikit-learn, matplotlib, h5py, openpyxl, pytables, dask, numexpr);
+# pyarrow/pypdf/pillow are covered by the pip step below. The Docker Hub mirror of this
+# image is stale (2023-10), so the maintained quay.io registry is used, and the tag is a
+# dated release (index digest
+# sha256:8f8a5763d9a01e7f488eff8d9379654788e9cc08286dbcbc6bcda129c1b0edcd) so that builds
+# are reproducible -- `latest` tracks the same image. Any image with python3/pip works just
+# as well; the apt/pip steps probe for existence first, so nothing the base image already
+# provides is reinstalled. Two properties of this base image are reset below: it ends on a
+# non-root user (this build needs root to create the agent/verifier users and to open up
+# /app), and it carries a Jupyter entrypoint (the container is started by the harness with
+# its own command and has to stay up for `docker exec`).
+FROM quay.io/jupyter/scipy-notebook:2026-10-12
+
+USER root
+# Drop the inherited Jupyter entrypoint/CMD: exec whatever command the harness passes, and
+# stay alive when it passes none.
+ENTRYPOINT ["/bin/sh", "-c", "exec \"$@\"", "-"]
+CMD ["sleep", "infinity"]
+
+# The conda prefix is the only Python in this image; link it into the standard /usr/local/bin
+# so that python3/pip3 still resolve when the harness switches to the agent/verifier user
+# with a login shell, which would otherwise drop /opt/conda/bin from PATH.
+RUN set -eux; \
+    ln -sf /opt/conda/bin/python3 /usr/local/bin/python3; \
+    ln -sf /opt/conda/bin/pip3 /usr/local/bin/pip3
 
 # Package mirrors -- the release build uses public mirrors (the original build
 # environment had no internet access and went through internal mirrors).

dataworld-p000619-cause-2/environment/Dockerfile

@@ -3,10 +3,33 @@
 # required.
 # Leak prevention: the whitelist COPYs only data/ -- WORLD.md, .build and dag.json must
 # never make it into the image.
-# [RELEASE] Replace the base image below with your own generic image (it needs
-# python3/pip, otherwise the apt step below installs them); the apt/pip steps all probe
-# for existence first, so anything the base image already provides is not reinstalled.
-FROM <base-image>:latest
+# Base image: a public, batteries-included Python environment -- the Jupyter scipy stack
+# (conda-forge). It already ships the data-science packages these tasks use (pandas, numpy,
+# scipy, statsmodels, scikit-learn, matplotlib, h5py, openpyxl, pytables, dask, numexpr);
+# pyarrow/pypdf/pillow are covered by the pip step below. The Docker Hub mirror of this
+# image is stale (2023-10), so the maintained quay.io registry is used, and the tag is a
+# dated release (index digest
+# sha256:8f8a5763d9a01e7f488eff8d9379654788e9cc08286dbcbc6bcda129c1b0edcd) so that builds
+# are reproducible -- `latest` tracks the same image. Any image with python3/pip works just
+# as well; the apt/pip steps probe for existence first, so nothing the base image already
+# provides is reinstalled. Two properties of this base image are reset below: it ends on a
+# non-root user (this build needs root to create the agent/verifier users and to open up
+# /app), and it carries a Jupyter entrypoint (the container is started by the harness with
+# its own command and has to stay up for `docker exec`).
+FROM quay.io/jupyter/scipy-notebook:2026-10-12
+
+USER root
+# Drop the inherited Jupyter entrypoint/CMD: exec whatever command the harness passes, and
+# stay alive when it passes none.
+ENTRYPOINT ["/bin/sh", "-c", "exec \"$@\"", "-"]
+CMD ["sleep", "infinity"]
+
+# The conda prefix is the only Python in this image; link it into the standard /usr/local/bin
+# so that python3/pip3 still resolve when the harness switches to the agent/verifier user
+# with a login shell, which would otherwise drop /opt/conda/bin from PATH.
+RUN set -eux; \
+    ln -sf /opt/conda/bin/python3 /usr/local/bin/python3; \
+    ln -sf /opt/conda/bin/pip3 /usr/local/bin/pip3
 
 # Package mirrors -- the release build uses public mirrors (the original build
 # environment had no internet access and went through internal mirrors).

dataworld-p000619-consequence-1/environment/Dockerfile

@@ -3,10 +3,33 @@
 # required.
 # Leak prevention: the whitelist COPYs only data/ -- WORLD.md, .build and dag.json must
 # never make it into the image.
-# [RELEASE] Replace the base image below with your own generic image (it needs
-# python3/pip, otherwise the apt step below installs them); the apt/pip steps all probe
-# for existence first, so anything the base image already provides is not reinstalled.
-FROM <base-image>:latest
+# Base image: a public, batteries-included Python environment -- the Jupyter scipy stack
+# (conda-forge). It already ships the data-science packages these tasks use (pandas, numpy,
+# scipy, statsmodels, scikit-learn, matplotlib, h5py, openpyxl, pytables, dask, numexpr);
+# pyarrow/pypdf/pillow are covered by the pip step below. The Docker Hub mirror of this
+# image is stale (2023-10), so the maintained quay.io registry is used, and the tag is a
+# dated release (index digest
+# sha256:8f8a5763d9a01e7f488eff8d9379654788e9cc08286dbcbc6bcda129c1b0edcd) so that builds
+# are reproducible -- `latest` tracks the same image. Any image with python3/pip works just
+# as well; the apt/pip steps probe for existence first, so nothing the base image already
+# provides is reinstalled. Two properties of this base image are reset below: it ends on a
+# non-root user (this build needs root to create the agent/verifier users and to open up
+# /app), and it carries a Jupyter entrypoint (the container is started by the harness with
+# its own command and has to stay up for `docker exec`).
+FROM quay.io/jupyter/scipy-notebook:2026-10-12
+
+USER root
+# Drop the inherited Jupyter entrypoint/CMD: exec whatever command the harness passes, and
+# stay alive when it passes none.
+ENTRYPOINT ["/bin/sh", "-c", "exec \"$@\"", "-"]
+CMD ["sleep", "infinity"]
+
+# The conda prefix is the only Python in this image; link it into the standard /usr/local/bin
+# so that python3/pip3 still resolve when the harness switches to the agent/verifier user
+# with a login shell, which would otherwise drop /opt/conda/bin from PATH.
+RUN set -eux; \
+    ln -sf /opt/conda/bin/python3 /usr/local/bin/python3; \
+    ln -sf /opt/conda/bin/pip3 /usr/local/bin/pip3
 
 # Package mirrors -- the release build uses public mirrors (the original build
 # environment had no internet access and went through internal mirrors).

dataworld-p000631-cause-1/environment/Dockerfile

@@ -3,10 +3,33 @@
 # required.
 # Leak prevention: the whitelist COPYs only data/ -- WORLD.md, .build and dag.json must
 # never make it into the image.
-# [RELEASE] Replace the base image below with your own generic image (it needs
-# python3/pip, otherwise the apt step below installs them); the apt/pip steps all probe
-# for existence first, so anything the base image already provides is not reinstalled.
-FROM <base-image>:latest
+# Base image: a public, batteries-included Python environment -- the Jupyter scipy stack
+# (conda-forge). It already ships the data-science packages these tasks use (pandas, numpy,
+# scipy, statsmodels, scikit-learn, matplotlib, h5py, openpyxl, pytables, dask, numexpr);
+# pyarrow/pypdf/pillow are covered by the pip step below. The Docker Hub mirror of this
+# image is stale (2023-10), so the maintained quay.io registry is used, and the tag is a
+# dated release (index digest
+# sha256:8f8a5763d9a01e7f488eff8d9379654788e9cc08286dbcbc6bcda129c1b0edcd) so that builds
+# are reproducible -- `latest` tracks the same image. Any image with python3/pip works just
+# as well; the apt/pip steps probe for existence first, so nothing the base image already
+# provides is reinstalled. Two properties of this base image are reset below: it ends on a
+# non-root user (this build needs root to create the agent/verifier users and to open up
+# /app), and it carries a Jupyter entrypoint (the container is started by the harness with
+# its own command and has to stay up for `docker exec`).
+FROM quay.io/jupyter/scipy-notebook:2026-10-12
+
+USER root
+# Drop the inherited Jupyter entrypoint/CMD: exec whatever command the harness passes, and
+# stay alive when it passes none.
+ENTRYPOINT ["/bin/sh", "-c", "exec \"$@\"", "-"]
+CMD ["sleep", "infinity"]
+
+# The conda prefix is the only Python in this image; link it into the standard /usr/local/bin
+# so that python3/pip3 still resolve when the harness switches to the agent/verifier user
+# with a login shell, which would otherwise drop /opt/conda/bin from PATH.
+RUN set -eux; \
+    ln -sf /opt/conda/bin/python3 /usr/local/bin/python3; \
+    ln -sf /opt/conda/bin/pip3 /usr/local/bin/pip3
 
 # Package mirrors -- the release build uses public mirrors (the original build
 # environment had no internet access and went through internal mirrors).

dataworld-p000631-consequence-2/environment/Dockerfile

@@ -3,10 +3,33 @@
 # required.
 # Leak prevention: the whitelist COPYs only data/ -- WORLD.md, .build and dag.json must
 # never make it into the image.
-# [RELEASE] Replace the base image below with your own generic image (it needs
-# python3/pip, otherwise the apt step below installs them); the apt/pip steps all probe
-# for existence first, so anything the base image already provides is not reinstalled.
-FROM <base-image>:latest
+# Base image: a public, batteries-included Python environment -- the Jupyter scipy stack
+# (conda-forge). It already ships the data-science packages these tasks use (pandas, numpy,
+# scipy, statsmodels, scikit-learn, matplotlib, h5py, openpyxl, pytables, dask, numexpr);
+# pyarrow/pypdf/pillow are covered by the pip step below. The Docker Hub mirror of this
+# image is stale (2023-10), so the maintained quay.io registry is used, and the tag is a
+# dated release (index digest
+# sha256:8f8a5763d9a01e7f488eff8d9379654788e9cc08286dbcbc6bcda129c1b0edcd) so that builds
+# are reproducible -- `latest` tracks the same image. Any image with python3/pip works just
+# as well; the apt/pip steps probe for existence first, so nothing the base image already
+# provides is reinstalled. Two properties of this base image are reset below: it ends on a
+# non-root user (this build needs root to create the agent/verifier users and to open up
+# /app), and it carries a Jupyter entrypoint (the container is started by the harness with
+# its own command and has to stay up for `docker exec`).
+FROM quay.io/jupyter/scipy-notebook:2026-10-12
+
+USER root
+# Drop the inherited Jupyter entrypoint/CMD: exec whatever command the harness passes, and
+# stay alive when it passes none.
+ENTRYPOINT ["/bin/sh", "-c", "exec \"$@\"", "-"]
+CMD ["sleep", "infinity"]
+
+# The conda prefix is the only Python in this image; link it into the standard /usr/local/bin
+# so that python3/pip3 still resolve when the harness switches to the agent/verifier user
+# with a login shell, which would otherwise drop /opt/conda/bin from PATH.
+RUN set -eux; \
+    ln -sf /opt/conda/bin/python3 /usr/local/bin/python3; \
+    ln -sf /opt/conda/bin/pip3 /usr/local/bin/pip3
 
 # Package mirrors -- the release build uses public mirrors (the original build
 # environment had no internet access and went through internal mirrors).

dataworld-p000917-cause-1/environment/Dockerfile

@@ -3,10 +3,33 @@
 # required.
 # Leak prevention: the whitelist COPYs only data/ -- WORLD.md, .build and dag.json must
 # never make it into the image.
-# [RELEASE] Replace the base image below with your own generic image (it needs
-# python3/pip, otherwise the apt step below installs them); the apt/pip steps all probe
-# for existence first, so anything the base image already provides is not reinstalled.
-FROM <base-image>:latest
+# Base image: a public, batteries-included Python environment -- the Jupyter scipy stack
+# (conda-forge). It already ships the data-science packages these tasks use (pandas, numpy,
+# scipy, statsmodels, scikit-learn, matplotlib, h5py, openpyxl, pytables, dask, numexpr);
+# pyarrow/pypdf/pillow are covered by the pip step below. The Docker Hub mirror of this
+# image is stale (2023-10), so the maintained quay.io registry is used, and the tag is a
+# dated release (index digest
+# sha256:8f8a5763d9a01e7f488eff8d9379654788e9cc08286dbcbc6bcda129c1b0edcd) so that builds
+# are reproducible -- `latest` tracks the same image. Any image with python3/pip works just
+# as well; the apt/pip steps probe for existence first, so nothing the base image already
+# provides is reinstalled. Two properties of this base image are reset below: it ends on a
+# non-root user (this build needs root to create the agent/verifier users and to open up
+# /app), and it carries a Jupyter entrypoint (the container is started by the harness with
+# its own command and has to stay up for `docker exec`).
+FROM quay.io/jupyter/scipy-notebook:2026-10-12
+
+USER root
+# Drop the inherited Jupyter entrypoint/CMD: exec whatever command the harness passes, and
+# stay alive when it passes none.
+ENTRYPOINT ["/bin/sh", "-c", "exec \"$@\"", "-"]
+CMD ["sleep", "infinity"]
+
+# The conda prefix is the only Python in this image; link it into the standard /usr/local/bin
+# so that python3/pip3 still resolve when the harness switches to the agent/verifier user
+# with a login shell, which would otherwise drop /opt/conda/bin from PATH.
+RUN set -eux; \
+    ln -sf /opt/conda/bin/python3 /usr/local/bin/python3; \
+    ln -sf /opt/conda/bin/pip3 /usr/local/bin/pip3
 
 # Package mirrors -- the release build uses public mirrors (the original build
 # environment had no internet access and went through internal mirrors).