publish SWE-bench Science release (part 3)

tasks/task_083/instruction.md

@@ -0,0 +1,22 @@
+# Repair bond-order species reporting
+
+You are working with a local snapshot of ReacNetGenerator, a tool for building
+reaction networks from reactive molecular-dynamics trajectories.
+
+Run the public workbench:
+
+```bash
+python reproduce.py
+```
+
+The bundled bond-file fixture exercises a small ring-like structure under
+several bond-order assignments. Inspect the report and repair the source so the
+molecule, species, and reaction outputs are chemically consistent for the
+fixture.
+
+Keep changes inside `source/`. Do not edit the public fixtures or the
+reproduction script.
+
+Do not hard-code the supplied fixture or report values. Your fix should work for
+other bond-order variants that preserve atom membership and bonding
+connectivity.

tasks/task_083/metadata.json

@@ -0,0 +1,12 @@
+{
+  "task_id": "083",
+  "title": "Repair bond-order species reporting",
+  "domain": "computational-chemistry",
+  "language": "python",
+  "source_repository": "https://github.com/deepmodeling/reacnetgenerator",
+  "source_commit": "37b0bef6fd9d5945f1b752ac82d4f8564728cd8b",
+  "source_license": "LGPL-3.0-family",
+  "license_source": "LICENSE",
+  "gpl_family": true,
+  "public_payload_sha256": "48f2557e6046a003e567fb8b9055d9a6f9a93401f664313f9c04cedadd30780c"
+}

tasks/task_083/task.toml

@@ -0,0 +1,44 @@
+schema_version = "1.3"
+artifacts = ["/logs/artifacts/model.patch"]
+
+[task]
+name = "openmoss/swe-bench-science-083"
+description = "Repair bond-order species reporting"
+authors = []
+keywords = ["scientific-computing", "software-engineering"]
+
+[metadata]
+task_id = "083"
+language = "python"
+base_commit_hash = "37b0bef6fd9d5945f1b752ac82d4f8564728cd8b"
+
+[agent]
+network_mode = "no-network"
+timeout_sec = 5400.0
+
+[environment]
+docker_image = "docker.io/kevinxulearning/swe-bench-science-environment-python-task-083:v0.1.0@sha256:9483e6962b892e928f5f1e1c608d646d2d35c5899da15f57fa7bc8c23bd2e14f"
+os = "linux"
+allow_internet = false
+cpus = 2
+memory_mb = 8192
+storage_mb = 20480
+workdir = "/app/task_083"
+
+[verifier]
+network_mode = "no-network"
+environment_mode = "separate"
+timeout_sec = 1800.0
+
+[verifier.environment]
+docker_image = "docker.io/kevinxulearning/swe-bench-science-verifier-python-task-083:v0.1.0@sha256:a1c0f1dabd6d4c334c815cb398eda399fbb8b7739b7b5ec60ea8abb1882b7727"
+os = "linux"
+allow_internet = false
+cpus = 2
+memory_mb = 8192
+storage_mb = 20480
+workdir = "/app/task_083"
+
+[[verifier.collect]]
+command = "cd /app/task_083 && git config --global --add safe.directory /app/task_083 && baseline=$(git rev-list --max-parents=0 --reverse HEAD | head -n 1) && git add -A && git diff --cached --binary \"$baseline\" > /logs/artifacts/model.patch"
+timeout_sec = 300.0

tasks/task_083/tests/Dockerfile

@@ -0,0 +1,2 @@
+ARG VERIFIER_IMAGE=docker.io/kevinxulearning/swe-bench-science-verifier-python-task-083:v0.1.0@sha256:a1c0f1dabd6d4c334c815cb398eda399fbb8b7739b7b5ec60ea8abb1882b7727
+FROM ${VERIFIER_IMAGE}

tasks/task_083/tests/test.sh

@@ -0,0 +1,6 @@
+#!/bin/sh
+set -eu
+
+# Pier requires a public tests entrypoint when loading a local task. In
+# separate-verifier mode the executable grader remains inside the verifier image.
+exec python /tests/grader.py

tasks/task_084/environment/Dockerfile

@@ -0,0 +1,60 @@
+ARG BASE_IMAGE=python:3.11-slim
+FROM ${BASE_IMAGE}
+
+ARG DEBIAN_MIRROR=https://mirrors.tuna.tsinghua.edu.cn/debian
+ARG DEBIAN_SECURITY_MIRROR=https://mirrors.tuna.tsinghua.edu.cn/debian-security
+ARG PIP_INDEX_URL=https://pypi.tuna.tsinghua.edu.cn/simple
+ARG INSTALL_FORTRAN=0
+ARG INSTALL_OCTAVE=0
+ENV DEBIAN_FRONTEND=noninteractive \
+    PIP_DISABLE_PIP_VERSION_CHECK=1 \
+    PIP_INDEX_URL=${PIP_INDEX_URL} \
+    PIP_NO_CACHE_DIR=1 \
+    PYTHONDONTWRITEBYTECODE=1
+
+RUN sed -i \
+        -e "s|http://deb.debian.org/debian-security|${DEBIAN_SECURITY_MIRROR}|g" \
+        -e "s|http://deb.debian.org/debian|${DEBIAN_MIRROR}|g" \
+        /etc/apt/sources.list.d/debian.sources
+
+RUN apt-get -o Acquire::Retries=10 update \
+    && apt-get -o Acquire::Retries=10 install -y --no-install-recommends \
+        build-essential \
+        ca-certificates \
+        curl \
+        git \
+        ripgrep \
+    && rm -rf /var/lib/apt/lists/*
+
+RUN if [ "${INSTALL_FORTRAN}" = "1" ]; then \
+        apt-get -o Acquire::Retries=10 update \
+        && apt-get -o Acquire::Retries=10 install -y --no-install-recommends gfortran \
+        && rm -rf /var/lib/apt/lists/*; \
+    fi \
+    && if [ "${INSTALL_OCTAVE}" = "1" ]; then \
+        apt-get -o Acquire::Retries=10 update \
+        && apt-get -o Acquire::Retries=10 install -y --no-install-recommends octave \
+        && rm -rf /var/lib/apt/lists/*; \
+    fi
+
+WORKDIR /app/task_084
+COPY repo/ /app/task_084/source/
+COPY public/ /app/task_084/
+COPY requirements.lock /app/task_084/requirements.lock
+
+RUN python -m pip install --no-cache-dir --upgrade pip \
+    && python -m pip install --no-cache-dir -r requirements.lock
+ENV PYTHONPATH=/app/task_084:/app/task_084/source
+
+RUN git init \
+    && git checkout -b main \
+    && git config user.email science-bench@example.invalid \
+    && git config user.name "SWE-bench Science baseline" \
+    && git add -A \
+    && GIT_AUTHOR_DATE=2025-01-01T00:00:00Z \
+       GIT_COMMITTER_DATE=2025-01-01T00:00:00Z \
+       git commit -m baseline \
+    && git reflog expire --expire=now --all \
+    && git gc --prune=now
+
+CMD ["bash"]

tasks/task_084/instruction.md

@@ -0,0 +1,27 @@
+# Repair ABACUS relaxation archive/reload semantics
+
+The supplied ABACUS relaxation examples can be loaded, but the supported
+archive/reload workflow does not preserve the trajectory in a scientifically
+usable form consistently across the provided examples.
+
+Study the supplied paper and method notes, reproduce the issue, inspect the
+source, and repair the implementation so that valid ABACUS relaxation
+catalogues remain scientifically coherent through the supported
+archive/reload workflow. The repair must generalize beyond the supplied
+examples. Do not replace scientific objects with fixed outputs,
+special-case the supplied data, or change the public reproduction to hide the
+failure.
+
+Work only under this task directory. The evaluation environment is Linux and
+has no network access. You may modify files under `source/`; do not modify the
+task statement, reproduction, fixtures, workflow, or evaluation
+infrastructure.
+
+Run the public smoke test with:
+
+```bash
+python reproduce.py
+```
+
+The command returns `1` when the untouched source reproduces the issue, `0`
+after a successful repair, and `2` for a runner or environment failure.

tasks/task_084/metadata.json

@@ -0,0 +1,12 @@
+{
+  "task_id": "084",
+  "title": "Repair ABACUS relaxation archive/reload semantics",
+  "domain": "materials-science-scientific-data-semantics",
+  "language": "python",
+  "source_repository": "https://github.com/deepmodeling/dpdata",
+  "source_commit": "80abfc5037876c0e883dcb58cd96521397e3fe36",
+  "source_license": "LGPL-3.0-family",
+  "license_source": "LICENSE",
+  "gpl_family": true,
+  "public_payload_sha256": "0b14a30b4b7ceb203ee7c08d2196b2c8b6a4efa48f85b12627a87a49f5f8b83d"
+}

tasks/task_084/task.toml

@@ -0,0 +1,44 @@
+schema_version = "1.3"
+artifacts = ["/logs/artifacts/model.patch"]
+
+[task]
+name = "openmoss/swe-bench-science-084"
+description = "Repair ABACUS relaxation archive/reload semantics"
+authors = []
+keywords = ["scientific-computing", "software-engineering"]
+
+[metadata]
+task_id = "084"
+language = "python"
+base_commit_hash = "80abfc5037876c0e883dcb58cd96521397e3fe36"
+
+[agent]
+network_mode = "no-network"
+timeout_sec = 5400.0
+
+[environment]
+docker_image = "docker.io/kevinxulearning/swe-bench-science-environment-python-task-084:v0.1.0@sha256:735e7ece88fa49a77257cfb2a59b435ed6ef511e99cc1739e24b366ea129836b"
+os = "linux"
+allow_internet = false
+cpus = 2
+memory_mb = 8192
+storage_mb = 20480
+workdir = "/app/task_084"
+
+[verifier]
+network_mode = "no-network"
+environment_mode = "separate"
+timeout_sec = 1800.0
+
+[verifier.environment]
+docker_image = "docker.io/kevinxulearning/swe-bench-science-verifier-python-task-084:v0.1.0@sha256:8cbe4554c727f25c1584f79c48f574a3d08e1fcf6d3c4cb32df36a7d8482216a"
+os = "linux"
+allow_internet = false
+cpus = 2
+memory_mb = 8192
+storage_mb = 20480
+workdir = "/app/task_084"
+
+[[verifier.collect]]
+command = "cd /app/task_084 && git config --global --add safe.directory /app/task_084 && baseline=$(git rev-list --max-parents=0 --reverse HEAD | head -n 1) && git add -A && git diff --cached --binary \"$baseline\" > /logs/artifacts/model.patch"
+timeout_sec = 300.0

tasks/task_084/tests/Dockerfile

@@ -0,0 +1,2 @@
+ARG VERIFIER_IMAGE=docker.io/kevinxulearning/swe-bench-science-verifier-python-task-084:v0.1.0@sha256:8cbe4554c727f25c1584f79c48f574a3d08e1fcf6d3c4cb32df36a7d8482216a
+FROM ${VERIFIER_IMAGE}

tasks/task_084/tests/test.sh

@@ -0,0 +1,6 @@
+#!/bin/sh
+set -eu
+
+# Pier requires a public tests entrypoint when loading a local task. In
+# separate-verifier mode the executable grader remains inside the verifier image.
+exec python /tests/grader.py

tasks/task_085/environment/Dockerfile

@@ -0,0 +1,60 @@
+ARG BASE_IMAGE=python:3.11-slim
+FROM ${BASE_IMAGE}
+
+ARG DEBIAN_MIRROR=https://mirrors.tuna.tsinghua.edu.cn/debian
+ARG DEBIAN_SECURITY_MIRROR=https://mirrors.tuna.tsinghua.edu.cn/debian-security
+ARG PIP_INDEX_URL=https://pypi.tuna.tsinghua.edu.cn/simple
+ARG INSTALL_FORTRAN=1
+ARG INSTALL_OCTAVE=0
+ENV DEBIAN_FRONTEND=noninteractive \
+    PIP_DISABLE_PIP_VERSION_CHECK=1 \
+    PIP_INDEX_URL=${PIP_INDEX_URL} \
+    PIP_NO_CACHE_DIR=1 \
+    PYTHONDONTWRITEBYTECODE=1
+
+RUN sed -i \
+        -e "s|http://deb.debian.org/debian-security|${DEBIAN_SECURITY_MIRROR}|g" \
+        -e "s|http://deb.debian.org/debian|${DEBIAN_MIRROR}|g" \
+        /etc/apt/sources.list.d/debian.sources
+
+RUN apt-get -o Acquire::Retries=10 update \
+    && apt-get -o Acquire::Retries=10 install -y --no-install-recommends \
+        build-essential \
+        ca-certificates \
+        curl \
+        git \
+        ripgrep \
+    && rm -rf /var/lib/apt/lists/*
+
+RUN if [ "${INSTALL_FORTRAN}" = "1" ]; then \
+        apt-get -o Acquire::Retries=10 update \
+        && apt-get -o Acquire::Retries=10 install -y --no-install-recommends gfortran \
+        && rm -rf /var/lib/apt/lists/*; \
+    fi \
+    && if [ "${INSTALL_OCTAVE}" = "1" ]; then \
+        apt-get -o Acquire::Retries=10 update \
+        && apt-get -o Acquire::Retries=10 install -y --no-install-recommends octave \
+        && rm -rf /var/lib/apt/lists/*; \
+    fi
+
+WORKDIR /app/task_085
+COPY repo/ /app/task_085/source/
+COPY public/ /app/task_085/
+COPY requirements.lock /app/task_085/requirements.lock
+
+RUN python -m pip install --no-cache-dir --upgrade pip \
+    && python -m pip install --no-cache-dir -r requirements.lock
+ENV PYTHONPATH=/app/task_085:/app/task_085/source
+
+RUN git init \
+    && git checkout -b main \
+    && git config user.email science-bench@example.invalid \
+    && git config user.name "SWE-bench Science baseline" \
+    && git add -A \
+    && GIT_AUTHOR_DATE=2025-01-01T00:00:00Z \
+       GIT_COMMITTER_DATE=2025-01-01T00:00:00Z \
+       git commit -m baseline \
+    && git reflog expire --expire=now --all \
+    && git gc --prune=now
+
+CMD ["bash"]

tasks/task_085/instruction.md

@@ -0,0 +1,13 @@
+# Inconsistent archived observables
+
+A molecular-modelling workflow is replaying archived DFTB+ calculations with
+this historical source snapshot. Several valid archived records complete
+successfully, yet some reported observables do not stay consistent across the
+documented input conventions.
+
+Run the supplied offline workflow, inspect the raw reports, the complete source
+snapshot, the official DFTB+ materials, and the documented input conventions.
+Repair the historical implementation so supported calculations behave
+consistently for valid inputs. The repair must generalize beyond the supplied
+archives. Do not hard-code the fixtures or reports, and do not modify the
+public workflow or scientific inputs.

tasks/task_085/metadata.json

@@ -0,0 +1,12 @@
+{
+  "task_id": "085",
+  "title": "Inconsistent archived observables",
+  "domain": "computational-chemistry-and-solvation",
+  "language": "fortran",
+  "source_repository": "https://github.com/dftbplus/dftbplus",
+  "source_commit": "d4da8c5027b643291f3a6295efd4340ce79d6a36",
+  "source_license": "AGPL-family",
+  "license_source": "COPYING",
+  "gpl_family": true,
+  "public_payload_sha256": "2db37177603a05c62b276b94c738f8592f40853359df72a3f25205c1d5bb5748"
+}

tasks/task_085/task.toml

@@ -0,0 +1,44 @@
+schema_version = "1.3"
+artifacts = ["/logs/artifacts/model.patch"]
+
+[task]
+name = "openmoss/swe-bench-science-085"
+description = "Inconsistent archived observables"
+authors = []
+keywords = ["scientific-computing", "software-engineering"]
+
+[metadata]
+task_id = "085"
+language = "fortran"
+base_commit_hash = "d4da8c5027b643291f3a6295efd4340ce79d6a36"
+
+[agent]
+network_mode = "no-network"
+timeout_sec = 5400.0
+
+[environment]
+docker_image = "docker.io/kevinxulearning/swe-bench-science-environment-fortran-task-085:v0.1.0@sha256:c4613c3e94fefc6cd721075d314d4fc463269cdd92cf7f6d0b656078a838a7cd"
+os = "linux"
+allow_internet = false
+cpus = 2
+memory_mb = 8192
+storage_mb = 20480
+workdir = "/app/task_085"
+
+[verifier]
+network_mode = "no-network"
+environment_mode = "separate"
+timeout_sec = 1800.0
+
+[verifier.environment]
+docker_image = "docker.io/kevinxulearning/swe-bench-science-verifier-fortran-task-085:v0.1.0@sha256:e68ac55c7cfb257756103351c305b97789d72f1e79203305d2aa7fa39e657eb1"
+os = "linux"
+allow_internet = false
+cpus = 2
+memory_mb = 8192
+storage_mb = 20480
+workdir = "/app/task_085"
+
+[[verifier.collect]]
+command = "cd /app/task_085 && git config --global --add safe.directory /app/task_085 && baseline=$(git rev-list --max-parents=0 --reverse HEAD | head -n 1) && git add -A && git diff --cached --binary \"$baseline\" > /logs/artifacts/model.patch"
+timeout_sec = 300.0

tasks/task_085/tests/Dockerfile

@@ -0,0 +1,2 @@
+ARG VERIFIER_IMAGE=docker.io/kevinxulearning/swe-bench-science-verifier-fortran-task-085:v0.1.0@sha256:e68ac55c7cfb257756103351c305b97789d72f1e79203305d2aa7fa39e657eb1
+FROM ${VERIFIER_IMAGE}

tasks/task_085/tests/test.sh

@@ -0,0 +1,6 @@
+#!/bin/sh
+set -eu
+
+# Pier requires a public tests entrypoint when loading a local task. In
+# separate-verifier mode the executable grader remains inside the verifier image.
+exec python /tests/grader.py

tasks/task_086/environment/Dockerfile

@@ -0,0 +1,63 @@
+ARG BASE_IMAGE=python:3.11-slim
+FROM ${BASE_IMAGE}
+
+ARG DEBIAN_MIRROR=https://mirrors.tuna.tsinghua.edu.cn/debian
+ARG DEBIAN_SECURITY_MIRROR=https://mirrors.tuna.tsinghua.edu.cn/debian-security
+ARG PIP_INDEX_URL=https://pypi.tuna.tsinghua.edu.cn/simple
+ARG INSTALL_FORTRAN=0
+ARG INSTALL_OCTAVE=0
+ENV DEBIAN_FRONTEND=noninteractive \
+    PIP_DISABLE_PIP_VERSION_CHECK=1 \
+    PIP_INDEX_URL=${PIP_INDEX_URL} \
+    PIP_NO_CACHE_DIR=1 \
+    PYTHONDONTWRITEBYTECODE=1
+
+RUN sed -i \
+        -e "s|http://deb.debian.org/debian-security|${DEBIAN_SECURITY_MIRROR}|g" \
+        -e "s|http://deb.debian.org/debian|${DEBIAN_MIRROR}|g" \
+        /etc/apt/sources.list.d/debian.sources
+
+RUN apt-get -o Acquire::Retries=10 update \
+    && apt-get -o Acquire::Retries=10 install -y --no-install-recommends \
+        build-essential \
+        ca-certificates \
+        curl \
+        git \
+        ripgrep \
+    && rm -rf /var/lib/apt/lists/*
+
+RUN if [ "${INSTALL_FORTRAN}" = "1" ]; then \
+        apt-get -o Acquire::Retries=10 update \
+        && apt-get -o Acquire::Retries=10 install -y --no-install-recommends gfortran \
+        && rm -rf /var/lib/apt/lists/*; \
+    fi \
+    && if [ "${INSTALL_OCTAVE}" = "1" ]; then \
+        apt-get -o Acquire::Retries=10 update \
+        && apt-get -o Acquire::Retries=10 install -y --no-install-recommends octave \
+        && rm -rf /var/lib/apt/lists/*; \
+    fi
+
+WORKDIR /app/task_086
+COPY repo/ /app/task_086/source/
+COPY public/ /app/task_086/
+COPY requirements.lock /app/task_086/requirements.lock
+COPY wheels/torch-2.6.0+cpu-cp311-cp311-linux_x86_64.whl /app/task_086/torch-2.6.0+cpu-cp311-cp311-linux_x86_64.whl
+
+RUN python -m pip install --no-cache-dir --upgrade pip \
+    && python -m pip install --no-cache-dir /app/task_086/torch-2.6.0+cpu-cp311-cp311-linux_x86_64.whl \
+    && rm -f /app/task_086/torch-2.6.0+cpu-cp311-cp311-linux_x86_64.whl \
+    && python -m pip install --no-cache-dir -r requirements.lock
+ENV PYTHONPATH=/app/task_086:/app/task_086/source
+
+RUN git init \
+    && git checkout -b main \
+    && git config user.email science-bench@example.invalid \
+    && git config user.name "SWE-bench Science baseline" \
+    && git add -A \
+    && GIT_AUTHOR_DATE=2025-01-01T00:00:00Z \
+       GIT_COMMITTER_DATE=2025-01-01T00:00:00Z \
+       git commit -m baseline \
+    && git reflog expire --expire=now --all \
+    && git gc --prune=now
+
+CMD ["bash"]

tasks/task_086/instruction.md

@@ -0,0 +1,10 @@
+Run `python reproduce.py` from this task directory and inspect the bundled
+source snapshot and offline workbench.
+
+The current preflight is small on purpose, but it exercises a boundary case
+that should not be treated as valid. Repair the package code in `source/` so
+the workbench reaches the expected post-fix status without changing the
+fixture, the public runner, or the task metadata.
+
+Keep the repair general. The corrected code should remain sensible for other
+valid materials inputs, not only for the supplied local case.

tasks/task_086/metadata.json

@@ -0,0 +1,12 @@
+{
+  "task_id": "086",
+  "title": "Repair a silent MatGL fine-tuning incompatibility",
+  "domain": "atomistic-machine-learning-potentials",
+  "language": "python",
+  "source_repository": "https://github.com/materialyzeai/matgl",
+  "source_commit": "02b415ba8bb25244bc5e3ad970491f0fbbea0292",
+  "source_license": "BSD-3-Clause",
+  "license_source": "LICENSE",
+  "gpl_family": false,
+  "public_payload_sha256": "dd83643eed59e22c0cf5fa9d29e4de990944b8add3edd5058391df75a585f537"
+}

tasks/task_086/task.toml

@@ -0,0 +1,44 @@
+schema_version = "1.3"
+artifacts = ["/logs/artifacts/model.patch"]
+
+[task]
+name = "openmoss/swe-bench-science-086"
+description = "Repair a silent MatGL fine-tuning incompatibility"
+authors = []
+keywords = ["scientific-computing", "software-engineering"]
+
+[metadata]
+task_id = "086"
+language = "python"
+base_commit_hash = "02b415ba8bb25244bc5e3ad970491f0fbbea0292"
+
+[agent]
+network_mode = "no-network"
+timeout_sec = 5400.0
+
+[environment]
+docker_image = "docker.io/kevinxulearning/swe-bench-science-environment-python-task-086:v0.1.0@sha256:7a54bd7e811a51abdd7571abbc03213eece0310016b174bfa52dcbe4ad171476"
+os = "linux"
+allow_internet = false
+cpus = 2
+memory_mb = 8192
+storage_mb = 20480
+workdir = "/app/task_086"
+
+[verifier]
+network_mode = "no-network"
+environment_mode = "separate"
+timeout_sec = 1800.0
+
+[verifier.environment]
+docker_image = "docker.io/kevinxulearning/swe-bench-science-verifier-python-task-086:v0.1.0@sha256:fb47312ac7443db59115807334c5d2345e3a2fb7a8f83d89bc8bfb48f7ce4bac"
+os = "linux"
+allow_internet = false
+cpus = 2
+memory_mb = 8192
+storage_mb = 20480
+workdir = "/app/task_086"
+
+[[verifier.collect]]
+command = "cd /app/task_086 && git config --global --add safe.directory /app/task_086 && baseline=$(git rev-list --max-parents=0 --reverse HEAD | head -n 1) && git add -A && git diff --cached --binary \"$baseline\" > /logs/artifacts/model.patch"
+timeout_sec = 300.0

tasks/task_086/tests/Dockerfile

@@ -0,0 +1,2 @@
+ARG VERIFIER_IMAGE=docker.io/kevinxulearning/swe-bench-science-verifier-python-task-086:v0.1.0@sha256:fb47312ac7443db59115807334c5d2345e3a2fb7a8f83d89bc8bfb48f7ce4bac
+FROM ${VERIFIER_IMAGE}

tasks/task_086/tests/test.sh

@@ -0,0 +1,6 @@
+#!/bin/sh
+set -eu
+
+# Pier requires a public tests entrypoint when loading a local task. In
+# separate-verifier mode the executable grader remains inside the verifier image.
+exec python /tests/grader.py

tasks/task_087/environment/Dockerfile

@@ -0,0 +1,60 @@
+ARG BASE_IMAGE=python:3.11-slim
+FROM ${BASE_IMAGE}
+
+ARG DEBIAN_MIRROR=https://mirrors.tuna.tsinghua.edu.cn/debian
+ARG DEBIAN_SECURITY_MIRROR=https://mirrors.tuna.tsinghua.edu.cn/debian-security
+ARG PIP_INDEX_URL=https://pypi.tuna.tsinghua.edu.cn/simple
+ARG INSTALL_FORTRAN=0
+ARG INSTALL_OCTAVE=0
+ENV DEBIAN_FRONTEND=noninteractive \
+    PIP_DISABLE_PIP_VERSION_CHECK=1 \
+    PIP_INDEX_URL=${PIP_INDEX_URL} \
+    PIP_NO_CACHE_DIR=1 \
+    PYTHONDONTWRITEBYTECODE=1
+
+RUN sed -i \
+        -e "s|http://deb.debian.org/debian-security|${DEBIAN_SECURITY_MIRROR}|g" \
+        -e "s|http://deb.debian.org/debian|${DEBIAN_MIRROR}|g" \
+        /etc/apt/sources.list.d/debian.sources
+
+RUN apt-get -o Acquire::Retries=10 update \
+    && apt-get -o Acquire::Retries=10 install -y --no-install-recommends \
+        build-essential \
+        ca-certificates \
+        curl \
+        git \
+        ripgrep \
+    && rm -rf /var/lib/apt/lists/*
+
+RUN if [ "${INSTALL_FORTRAN}" = "1" ]; then \
+        apt-get -o Acquire::Retries=10 update \
+        && apt-get -o Acquire::Retries=10 install -y --no-install-recommends gfortran \
+        && rm -rf /var/lib/apt/lists/*; \
+    fi \
+    && if [ "${INSTALL_OCTAVE}" = "1" ]; then \
+        apt-get -o Acquire::Retries=10 update \
+        && apt-get -o Acquire::Retries=10 install -y --no-install-recommends octave \
+        && rm -rf /var/lib/apt/lists/*; \
+    fi
+
+WORKDIR /app/task_087
+COPY repo/ /app/task_087/source/
+COPY public/ /app/task_087/
+COPY requirements.lock /app/task_087/requirements.lock
+
+RUN python -m pip install --no-cache-dir --upgrade pip \
+    && python -m pip install --no-cache-dir -r requirements.lock
+ENV PYTHONPATH=/app/task_087:/app/task_087/source
+
+RUN git init \
+    && git checkout -b main \
+    && git config user.email science-bench@example.invalid \
+    && git config user.name "SWE-bench Science baseline" \
+    && git add -A \
+    && GIT_AUTHOR_DATE=2025-01-01T00:00:00Z \
+       GIT_COMMITTER_DATE=2025-01-01T00:00:00Z \
+       git commit -m baseline \
+    && git reflog expire --expire=now --all \
+    && git gc --prune=now
+
+CMD ["bash"]

tasks/task_087/instruction.md

@@ -0,0 +1,10 @@
+Run `python reproduce.py` from this task directory and inspect the bundled
+source snapshot and offline workbench.
+
+The public workflow should expose a finite, inspectable diffusion-MRI
+observation on the supplied local case, but the reported output is irregular
+before repair. Repair the implementation in `source/` so the same workflow
+produces a coherent result on the public case and on broader valid inputs.
+
+Keep the repair general. Do not hard-code the supplied observation or change
+the public runner, public metadata, or task directory layout.

tasks/task_087/metadata.json

@@ -0,0 +1,12 @@
+{
+  "task_id": "087",
+  "title": "Repair a silent diffusion-MRI denoising inconsistency",
+  "domain": "diffusion-mri-denoising",
+  "language": "python",
+  "source_repository": "https://github.com/dipy/dipy",
+  "source_commit": "ee6016ab81dd649ab40439d75212f35669302810",
+  "source_license": "BSD-3-Clause",
+  "license_source": "LICENSE",
+  "gpl_family": false,
+  "public_payload_sha256": "442792643e44ca3c36c70b3d63c80708e9f58b5972e5bdf43e90834720834814"
+}

tasks/task_087/task.toml

@@ -0,0 +1,44 @@
+schema_version = "1.3"
+artifacts = ["/logs/artifacts/model.patch"]
+
+[task]
+name = "openmoss/swe-bench-science-087"
+description = "Repair a silent diffusion-MRI denoising inconsistency"
+authors = []
+keywords = ["scientific-computing", "software-engineering"]
+
+[metadata]
+task_id = "087"
+language = "python"
+base_commit_hash = "ee6016ab81dd649ab40439d75212f35669302810"
+
+[agent]
+network_mode = "no-network"
+timeout_sec = 5400.0
+
+[environment]
+docker_image = "docker.io/kevinxulearning/swe-bench-science-environment-python-task-087:v0.1.0@sha256:5a00e4d7a825618ba70259ef5c5e229f7ba2549ab7648ead6fc91fce31256793"
+os = "linux"
+allow_internet = false
+cpus = 2
+memory_mb = 8192
+storage_mb = 20480
+workdir = "/app/task_087"
+
+[verifier]
+network_mode = "no-network"
+environment_mode = "separate"
+timeout_sec = 1800.0
+
+[verifier.environment]
+docker_image = "docker.io/kevinxulearning/swe-bench-science-verifier-python-task-087:v0.1.0@sha256:b1ce39c75913bccef5f09acfd2b989857c071fc64c7fdc0dbff092e14dc5270f"
+os = "linux"
+allow_internet = false
+cpus = 2
+memory_mb = 8192
+storage_mb = 20480
+workdir = "/app/task_087"
+
+[[verifier.collect]]
+command = "cd /app/task_087 && git config --global --add safe.directory /app/task_087 && baseline=$(git rev-list --max-parents=0 --reverse HEAD | head -n 1) && git add -A && git diff --cached --binary \"$baseline\" > /logs/artifacts/model.patch"
+timeout_sec = 300.0

tasks/task_087/tests/Dockerfile

@@ -0,0 +1,2 @@
+ARG VERIFIER_IMAGE=docker.io/kevinxulearning/swe-bench-science-verifier-python-task-087:v0.1.0@sha256:b1ce39c75913bccef5f09acfd2b989857c071fc64c7fdc0dbff092e14dc5270f
+FROM ${VERIFIER_IMAGE}

tasks/task_087/tests/test.sh

@@ -0,0 +1,6 @@
+#!/bin/sh
+set -eu
+
+# Pier requires a public tests entrypoint when loading a local task. In
+# separate-verifier mode the executable grader remains inside the verifier image.
+exec python /tests/grader.py

tasks/task_088/environment/Dockerfile

@@ -0,0 +1,60 @@
+ARG BASE_IMAGE=python:3.11-slim
+FROM ${BASE_IMAGE}
+
+ARG DEBIAN_MIRROR=https://mirrors.tuna.tsinghua.edu.cn/debian
+ARG DEBIAN_SECURITY_MIRROR=https://mirrors.tuna.tsinghua.edu.cn/debian-security
+ARG PIP_INDEX_URL=https://pypi.tuna.tsinghua.edu.cn/simple
+ARG INSTALL_FORTRAN=0
+ARG INSTALL_OCTAVE=0
+ENV DEBIAN_FRONTEND=noninteractive \
+    PIP_DISABLE_PIP_VERSION_CHECK=1 \
+    PIP_INDEX_URL=${PIP_INDEX_URL} \
+    PIP_NO_CACHE_DIR=1 \
+    PYTHONDONTWRITEBYTECODE=1
+
+RUN sed -i \
+        -e "s|http://deb.debian.org/debian-security|${DEBIAN_SECURITY_MIRROR}|g" \
+        -e "s|http://deb.debian.org/debian|${DEBIAN_MIRROR}|g" \
+        /etc/apt/sources.list.d/debian.sources
+
+RUN apt-get -o Acquire::Retries=10 update \
+    && apt-get -o Acquire::Retries=10 install -y --no-install-recommends \
+        build-essential \
+        ca-certificates \
+        curl \
+        git \
+        ripgrep \
+    && rm -rf /var/lib/apt/lists/*
+
+RUN if [ "${INSTALL_FORTRAN}" = "1" ]; then \
+        apt-get -o Acquire::Retries=10 update \
+        && apt-get -o Acquire::Retries=10 install -y --no-install-recommends gfortran \
+        && rm -rf /var/lib/apt/lists/*; \
+    fi \
+    && if [ "${INSTALL_OCTAVE}" = "1" ]; then \
+        apt-get -o Acquire::Retries=10 update \
+        && apt-get -o Acquire::Retries=10 install -y --no-install-recommends octave \
+        && rm -rf /var/lib/apt/lists/*; \
+    fi
+
+WORKDIR /app/task_088
+COPY repo/ /app/task_088/source/
+COPY public/ /app/task_088/
+COPY requirements.lock /app/task_088/requirements.lock
+
+RUN python -m pip install --no-cache-dir --upgrade pip \
+    && python -m pip install --no-cache-dir -r requirements.lock
+ENV PYTHONPATH=/app/task_088:/app/task_088/source
+
+RUN git init \
+    && git checkout -b main \
+    && git config user.email science-bench@example.invalid \
+    && git config user.name "SWE-bench Science baseline" \
+    && git add -A \
+    && GIT_AUTHOR_DATE=2025-01-01T00:00:00Z \
+       GIT_COMMITTER_DATE=2025-01-01T00:00:00Z \
+       git commit -m baseline \
+    && git reflog expire --expire=now --all \
+    && git gc --prune=now
+
+CMD ["bash"]

tasks/task_088/instruction.md

@@ -0,0 +1,22 @@
+# Repair a boundary aggregation failure
+
+A short time series fails when it is aggregated into a coarser bin. The same
+workflow should run to completion and return a finite aggregate for valid
+inputs.
+
+Reproduce the issue from the supplied source snapshot and references, inspect
+the implementation, and repair the behavior so the aggregation workflow works
+for the supported coordinates. The fix must generalize beyond the public
+example and must not hard-code a single span or output value.
+
+Work only in this task directory. The evaluation environment is Linux and has
+no network access. You may modify files under `source/`.
+
+Use:
+
+```bash
+python reproduce.py
+```
+
+The command returns `1` for the reproduced scientific inconsistency, `0` after
+a successful repair, and `2` for a runner or environment failure.

tasks/task_088/metadata.json

@@ -0,0 +1,12 @@
+{
+  "task_id": "088",
+  "title": "Repair a boundary aggregation failure",
+  "domain": "boundary-aggregation",
+  "language": "python",
+  "source_repository": "https://github.com/pydata/xarray",
+  "source_commit": "0b071e369a88f6672accc2a99c40d3b4b334f2f9",
+  "source_license": "Apache-2.0",
+  "license_source": "LICENSE",
+  "gpl_family": false,
+  "public_payload_sha256": "5e7474f8aa8269a9d76e03f048cce31b30e91fadfa8bb5e3401d63b59c63ddbf"
+}

tasks/task_088/task.toml

@@ -0,0 +1,44 @@
+schema_version = "1.3"
+artifacts = ["/logs/artifacts/model.patch"]
+
+[task]
+name = "openmoss/swe-bench-science-088"
+description = "Repair a boundary aggregation failure"
+authors = []
+keywords = ["scientific-computing", "software-engineering"]
+
+[metadata]
+task_id = "088"
+language = "python"
+base_commit_hash = "0b071e369a88f6672accc2a99c40d3b4b334f2f9"
+
+[agent]
+network_mode = "no-network"
+timeout_sec = 5400.0
+
+[environment]
+docker_image = "docker.io/kevinxulearning/swe-bench-science-environment-python-task-088:v0.1.0@sha256:c98a81a135c210d7681b57574c8f39a00f1a92778e458dc5300a7f2775e2ba6f"
+os = "linux"
+allow_internet = false
+cpus = 2
+memory_mb = 8192
+storage_mb = 20480
+workdir = "/app/task_088"
+
+[verifier]
+network_mode = "no-network"
+environment_mode = "separate"
+timeout_sec = 1800.0
+
+[verifier.environment]
+docker_image = "docker.io/kevinxulearning/swe-bench-science-verifier-python-task-088:v0.1.0@sha256:9f8a00b079fb998f0672f57af8f75628dabc8582122a294f2b20cd304503ac6e"
+os = "linux"
+allow_internet = false
+cpus = 2
+memory_mb = 8192
+storage_mb = 20480
+workdir = "/app/task_088"
+
+[[verifier.collect]]
+command = "cd /app/task_088 && git config --global --add safe.directory /app/task_088 && baseline=$(git rev-list --max-parents=0 --reverse HEAD | head -n 1) && git add -A && git diff --cached --binary \"$baseline\" > /logs/artifacts/model.patch"
+timeout_sec = 300.0

tasks/task_088/tests/Dockerfile

@@ -0,0 +1,2 @@
+ARG VERIFIER_IMAGE=docker.io/kevinxulearning/swe-bench-science-verifier-python-task-088:v0.1.0@sha256:9f8a00b079fb998f0672f57af8f75628dabc8582122a294f2b20cd304503ac6e
+FROM ${VERIFIER_IMAGE}

tasks/task_088/tests/test.sh

@@ -0,0 +1,6 @@
+#!/bin/sh
+set -eu
+
+# Pier requires a public tests entrypoint when loading a local task. In
+# separate-verifier mode the executable grader remains inside the verifier image.
+exec python /tests/grader.py

tasks/task_089/environment/Dockerfile

@@ -0,0 +1,60 @@
+ARG BASE_IMAGE=python:3.11-slim
+FROM ${BASE_IMAGE}
+
+ARG DEBIAN_MIRROR=https://mirrors.tuna.tsinghua.edu.cn/debian
+ARG DEBIAN_SECURITY_MIRROR=https://mirrors.tuna.tsinghua.edu.cn/debian-security
+ARG PIP_INDEX_URL=https://pypi.tuna.tsinghua.edu.cn/simple
+ARG INSTALL_FORTRAN=0
+ARG INSTALL_OCTAVE=0
+ENV DEBIAN_FRONTEND=noninteractive \
+    PIP_DISABLE_PIP_VERSION_CHECK=1 \
+    PIP_INDEX_URL=${PIP_INDEX_URL} \
+    PIP_NO_CACHE_DIR=1 \
+    PYTHONDONTWRITEBYTECODE=1
+
+RUN sed -i \
+        -e "s|http://deb.debian.org/debian-security|${DEBIAN_SECURITY_MIRROR}|g" \
+        -e "s|http://deb.debian.org/debian|${DEBIAN_MIRROR}|g" \
+        /etc/apt/sources.list.d/debian.sources
+
+RUN apt-get -o Acquire::Retries=10 update \
+    && apt-get -o Acquire::Retries=10 install -y --no-install-recommends \
+        build-essential \
+        ca-certificates \
+        curl \
+        git \
+        ripgrep \
+    && rm -rf /var/lib/apt/lists/*
+
+RUN if [ "${INSTALL_FORTRAN}" = "1" ]; then \
+        apt-get -o Acquire::Retries=10 update \
+        && apt-get -o Acquire::Retries=10 install -y --no-install-recommends gfortran \
+        && rm -rf /var/lib/apt/lists/*; \
+    fi \
+    && if [ "${INSTALL_OCTAVE}" = "1" ]; then \
+        apt-get -o Acquire::Retries=10 update \
+        && apt-get -o Acquire::Retries=10 install -y --no-install-recommends octave \
+        && rm -rf /var/lib/apt/lists/*; \
+    fi
+
+WORKDIR /app/task_089
+COPY repo/ /app/task_089/source/
+COPY public/ /app/task_089/
+COPY requirements.lock /app/task_089/requirements.lock
+
+RUN python -m pip install --no-cache-dir --upgrade pip \
+    && python -m pip install --no-cache-dir -r requirements.lock
+ENV PYTHONPATH=/app/task_089:/app/task_089/source
+
+RUN git init \
+    && git checkout -b main \
+    && git config user.email science-bench@example.invalid \
+    && git config user.name "SWE-bench Science baseline" \
+    && git add -A \
+    && GIT_AUTHOR_DATE=2025-01-01T00:00:00Z \
+       GIT_COMMITTER_DATE=2025-01-01T00:00:00Z \
+       git commit -m baseline \
+    && git reflog expire --expire=now --all \
+    && git gc --prune=now
+
+CMD ["bash"]

tasks/task_089/instruction.md

@@ -0,0 +1,30 @@
+# Repair inconsistent ORCA excited-state spectrum parsing
+
+A computational-chemistry workflow imports ORCA excited-state spectra from a
+historical source snapshot. The parser completes and returns finite transition
+data, but a downstream spectrum summary made from the parsed attributes is not
+consistent with the supplied ORCA-style record.
+
+Run the public reproduction, inspect the complete source snapshot and supplied
+scientific material, and repair the implementation so valid ORCA excited-state
+spectrum records retain their scientific meaning in the parsed data model. The
+repair must generalize to other valid ORCA spectra, output-version layouts,
+additional transition-property sections, and related spectroscopy sections that
+may appear in the same output.
+
+Do not hard-code the supplied fixture, spectrum names, line numbers, numerical
+values, or generated report. Work only in this task directory. The evaluation
+environment is Linux and has no network access. You may modify files under
+`source/`; do not modify the public reproduction, fixtures, workflow, or
+evaluation infrastructure.
+
+Use:
+
+```bash
+python reproduce.py
+```
+
+The command is a diagnostic workflow. It returns `0` when the parser can produce
+a finite nonempty spectrum observation, `1` when it cannot, and `2` for a runner
+or dependency failure. The independent verifier checks scientific consistency on
+broader spectra.

tasks/task_089/metadata.json

@@ -0,0 +1,12 @@
+{
+  "task_id": "089",
+  "title": "Repair inconsistent ORCA excited-state oscillator strengths",
+  "domain": "computational-chemistry-spectroscopy",
+  "language": "python",
+  "source_repository": "https://github.com/cclib/cclib",
+  "source_commit": "6ba33bdd48bf336a81400147d2870323b241e457",
+  "source_license": "BSD-3-Clause",
+  "license_source": "LICENSE",
+  "gpl_family": false,
+  "public_payload_sha256": "fe78fd81c8f3bde1750d471aec4416de4b93ff45a1490008fe50e529855c6782"
+}

tasks/task_089/task.toml

@@ -0,0 +1,44 @@
+schema_version = "1.3"
+artifacts = ["/logs/artifacts/model.patch"]
+
+[task]
+name = "openmoss/swe-bench-science-089"
+description = "Repair inconsistent ORCA excited-state oscillator strengths"
+authors = []
+keywords = ["scientific-computing", "software-engineering"]
+
+[metadata]
+task_id = "089"
+language = "python"
+base_commit_hash = "6ba33bdd48bf336a81400147d2870323b241e457"
+
+[agent]
+network_mode = "no-network"
+timeout_sec = 5400.0
+
+[environment]
+docker_image = "docker.io/kevinxulearning/swe-bench-science-environment-python-task-089:v0.1.0@sha256:87e8b70f8d192990fe5a0f5d1773dcb304ab1bf5074a187236f2e0bd041ba356"
+os = "linux"
+allow_internet = false
+cpus = 2
+memory_mb = 8192
+storage_mb = 20480
+workdir = "/app/task_089"
+
+[verifier]
+network_mode = "no-network"
+environment_mode = "separate"
+timeout_sec = 1800.0
+
+[verifier.environment]
+docker_image = "docker.io/kevinxulearning/swe-bench-science-verifier-python-task-089:v0.1.0@sha256:66ee7d7df906e3cad4e9d27664f1c6ea9bd6673ef36d5ddd61f5aaa75c88db60"
+os = "linux"
+allow_internet = false
+cpus = 2
+memory_mb = 8192
+storage_mb = 20480
+workdir = "/app/task_089"
+
+[[verifier.collect]]
+command = "cd /app/task_089 && git config --global --add safe.directory /app/task_089 && baseline=$(git rev-list --max-parents=0 --reverse HEAD | head -n 1) && git add -A && git diff --cached --binary \"$baseline\" > /logs/artifacts/model.patch"
+timeout_sec = 300.0

tasks/task_089/tests/Dockerfile

@@ -0,0 +1,2 @@
+ARG VERIFIER_IMAGE=docker.io/kevinxulearning/swe-bench-science-verifier-python-task-089:v0.1.0@sha256:66ee7d7df906e3cad4e9d27664f1c6ea9bd6673ef36d5ddd61f5aaa75c88db60
+FROM ${VERIFIER_IMAGE}

tasks/task_089/tests/test.sh

@@ -0,0 +1,6 @@
+#!/bin/sh
+set -eu
+
+# Pier requires a public tests entrypoint when loading a local task. In
+# separate-verifier mode the executable grader remains inside the verifier image.
+exec python /tests/grader.py

tasks/task_090/environment/Dockerfile

@@ -0,0 +1,60 @@
+ARG BASE_IMAGE=python:3.11-slim
+FROM ${BASE_IMAGE}
+
+ARG DEBIAN_MIRROR=https://mirrors.tuna.tsinghua.edu.cn/debian
+ARG DEBIAN_SECURITY_MIRROR=https://mirrors.tuna.tsinghua.edu.cn/debian-security
+ARG PIP_INDEX_URL=https://pypi.tuna.tsinghua.edu.cn/simple
+ARG INSTALL_FORTRAN=0
+ARG INSTALL_OCTAVE=0
+ENV DEBIAN_FRONTEND=noninteractive \
+    PIP_DISABLE_PIP_VERSION_CHECK=1 \
+    PIP_INDEX_URL=${PIP_INDEX_URL} \
+    PIP_NO_CACHE_DIR=1 \
+    PYTHONDONTWRITEBYTECODE=1
+
+RUN sed -i \
+        -e "s|http://deb.debian.org/debian-security|${DEBIAN_SECURITY_MIRROR}|g" \
+        -e "s|http://deb.debian.org/debian|${DEBIAN_MIRROR}|g" \
+        /etc/apt/sources.list.d/debian.sources
+
+RUN apt-get -o Acquire::Retries=10 update \
+    && apt-get -o Acquire::Retries=10 install -y --no-install-recommends \
+        build-essential \
+        ca-certificates \
+        curl \
+        git \
+        ripgrep \
+    && rm -rf /var/lib/apt/lists/*
+
+RUN if [ "${INSTALL_FORTRAN}" = "1" ]; then \
+        apt-get -o Acquire::Retries=10 update \
+        && apt-get -o Acquire::Retries=10 install -y --no-install-recommends gfortran \
+        && rm -rf /var/lib/apt/lists/*; \
+    fi \
+    && if [ "${INSTALL_OCTAVE}" = "1" ]; then \
+        apt-get -o Acquire::Retries=10 update \
+        && apt-get -o Acquire::Retries=10 install -y --no-install-recommends octave \
+        && rm -rf /var/lib/apt/lists/*; \
+    fi
+
+WORKDIR /app/task_090
+COPY repo/ /app/task_090/source/
+COPY public/ /app/task_090/
+COPY requirements.lock /app/task_090/requirements.lock
+
+RUN python -m pip install --no-cache-dir --upgrade pip \
+    && python -m pip install --no-cache-dir -r requirements.lock
+ENV PYTHONPATH=/app/task_090:/app/task_090/source
+
+RUN git init \
+    && git checkout -b main \
+    && git config user.email science-bench@example.invalid \
+    && git config user.name "SWE-bench Science baseline" \
+    && git add -A \
+    && GIT_AUTHOR_DATE=2025-01-01T00:00:00Z \
+       GIT_COMMITTER_DATE=2025-01-01T00:00:00Z \
+       git commit -m baseline \
+    && git reflog expire --expire=now --all \
+    && git gc --prune=now
+
+CMD ["bash"]

tasks/task_090/instruction.md

@@ -0,0 +1,26 @@
+# Restore multidimensional NMR acquisition grids
+
+A multidimensional NMR acquisition can arrive as a compressed FID stream plus a
+sparse coordinate schedule. Study the source snapshot and the public
+reproduction. Repair the package so that valid acquisition workflows preserve
+that schedule alongside the compressed stream in both eager and low-memory
+reader paths, and so the public processing workflow can restore the compressed
+rows into a full multidimensional acquisition grid without losing the declared
+acquisition and quadrature layout. Preserve ordinary fully-sampled behavior.
+The repair must generalize when valid layouts, dimensions, direct FID lengths,
+dtypes, and sampling patterns change.
+
+Do not hard-code the supplied observations or replace the NMR data model with a
+fixed answer. Work only under this task directory. The evaluation environment is
+offline; all required Python dependencies and source files are provided.
+
+Run the public diagnostic with:
+
+```bash
+python reproduce.py
+```
+
+The command returns `0` when the diagnostic report is produced and `2` for an
+import, dependency, path, or other runner failure. The public report is a
+workbench signal; final evaluation performs independent scientific consistency
+checks on broader cases.

tasks/task_090/metadata.json

@@ -0,0 +1,12 @@
+{
+  "task_id": "090",
+  "title": "Restore multidimensional NMR acquisition grids",
+  "domain": "nuclear-magnetic-resonance-spectroscopy",
+  "language": "python",
+  "source_repository": "",
+  "source_commit": "",
+  "source_license": "BSD-3-Clause",
+  "license_source": "LICENSE.txt",
+  "gpl_family": false,
+  "public_payload_sha256": "dd119886283401127edc350f562371a2bad9d92d79cfe0488afeaa26a8dde9b4"
+}

tasks/task_090/task.toml

@@ -0,0 +1,44 @@
+schema_version = "1.3"
+artifacts = ["/logs/artifacts/model.patch"]
+
+[task]
+name = "openmoss/swe-bench-science-090"
+description = "Restore multidimensional NMR acquisition grids"
+authors = []
+keywords = ["scientific-computing", "software-engineering"]
+
+[metadata]
+task_id = "090"
+language = "python"
+base_commit_hash = ""
+
+[agent]
+network_mode = "no-network"
+timeout_sec = 5400.0
+
+[environment]
+docker_image = "docker.io/kevinxulearning/swe-bench-science-environment-python-task-090:v0.1.0@sha256:a2a7d3de4b603c2e170a1ea679103aeefdd43c1503f5c2210f32fec9f51e3efa"
+os = "linux"
+allow_internet = false
+cpus = 2
+memory_mb = 8192
+storage_mb = 20480
+workdir = "/app/task_090"
+
+[verifier]
+network_mode = "no-network"
+environment_mode = "separate"
+timeout_sec = 1800.0
+
+[verifier.environment]
+docker_image = "docker.io/kevinxulearning/swe-bench-science-verifier-python-task-090:v0.1.0@sha256:8ba7b5dff3e7bd208683b7d457ebfe6277ce4bdd218bea4d2e0ea322f49a5007"
+os = "linux"
+allow_internet = false
+cpus = 2
+memory_mb = 8192
+storage_mb = 20480
+workdir = "/app/task_090"
+
+[[verifier.collect]]
+command = "cd /app/task_090 && git config --global --add safe.directory /app/task_090 && baseline=$(git rev-list --max-parents=0 --reverse HEAD | head -n 1) && git add -A && git diff --cached --binary \"$baseline\" > /logs/artifacts/model.patch"
+timeout_sec = 300.0

tasks/task_090/tests/Dockerfile

@@ -0,0 +1,2 @@
+ARG VERIFIER_IMAGE=docker.io/kevinxulearning/swe-bench-science-verifier-python-task-090:v0.1.0@sha256:8ba7b5dff3e7bd208683b7d457ebfe6277ce4bdd218bea4d2e0ea322f49a5007
+FROM ${VERIFIER_IMAGE}

tasks/task_090/tests/test.sh

@@ -0,0 +1,6 @@
+#!/bin/sh
+set -eu
+
+# Pier requires a public tests entrypoint when loading a local task. In
+# separate-verifier mode the executable grader remains inside the verifier image.
+exec python /tests/grader.py

tasks/task_091/environment/Dockerfile

@@ -0,0 +1,60 @@
+ARG BASE_IMAGE=python:3.11-slim
+FROM ${BASE_IMAGE}
+
+ARG DEBIAN_MIRROR=https://mirrors.tuna.tsinghua.edu.cn/debian
+ARG DEBIAN_SECURITY_MIRROR=https://mirrors.tuna.tsinghua.edu.cn/debian-security
+ARG PIP_INDEX_URL=https://pypi.tuna.tsinghua.edu.cn/simple
+ARG INSTALL_FORTRAN=0
+ARG INSTALL_OCTAVE=0
+ENV DEBIAN_FRONTEND=noninteractive \
+    PIP_DISABLE_PIP_VERSION_CHECK=1 \
+    PIP_INDEX_URL=${PIP_INDEX_URL} \
+    PIP_NO_CACHE_DIR=1 \
+    PYTHONDONTWRITEBYTECODE=1
+
+RUN sed -i \
+        -e "s|http://deb.debian.org/debian-security|${DEBIAN_SECURITY_MIRROR}|g" \
+        -e "s|http://deb.debian.org/debian|${DEBIAN_MIRROR}|g" \
+        /etc/apt/sources.list.d/debian.sources
+
+RUN apt-get -o Acquire::Retries=10 update \
+    && apt-get -o Acquire::Retries=10 install -y --no-install-recommends \
+        build-essential \
+        ca-certificates \
+        curl \
+        git \
+        ripgrep \
+    && rm -rf /var/lib/apt/lists/*
+
+RUN if [ "${INSTALL_FORTRAN}" = "1" ]; then \
+        apt-get -o Acquire::Retries=10 update \
+        && apt-get -o Acquire::Retries=10 install -y --no-install-recommends gfortran \
+        && rm -rf /var/lib/apt/lists/*; \
+    fi \
+    && if [ "${INSTALL_OCTAVE}" = "1" ]; then \
+        apt-get -o Acquire::Retries=10 update \
+        && apt-get -o Acquire::Retries=10 install -y --no-install-recommends octave \
+        && rm -rf /var/lib/apt/lists/*; \
+    fi
+
+WORKDIR /app/task_091
+COPY repo/ /app/task_091/source/
+COPY public/ /app/task_091/
+COPY requirements.lock /app/task_091/requirements.lock
+
+RUN python -m pip install --no-cache-dir --upgrade pip \
+    && python -m pip install --no-cache-dir -r requirements.lock
+ENV PYTHONPATH=/app/task_091:/app/task_091/source
+
+RUN git init \
+    && git checkout -b main \
+    && git config user.email science-bench@example.invalid \
+    && git config user.name "SWE-bench Science baseline" \
+    && git add -A \
+    && GIT_AUTHOR_DATE=2025-01-01T00:00:00Z \
+       GIT_COMMITTER_DATE=2025-01-01T00:00:00Z \
+       git commit -m baseline \
+    && git reflog expire --expire=now --all \
+    && git gc --prune=now
+
+CMD ["bash"]

tasks/task_091/instruction.md

@@ -0,0 +1,21 @@
+# Repair a workflow regression
+
+A bundled input file exercises the public workflow and currently fails the
+smoke test.
+
+Run the public reproduction, inspect the complete source snapshot, and repair
+the implementation so valid inputs are handled consistently across
+representative cases. The repair must generalize to other valid input files and
+similar inputs. Do not hard-code the bundled file or output.
+
+Work only in this task directory. The evaluation environment is Linux and has
+no network access. You may modify files under `source/`.
+
+Use:
+
+```bash
+python reproduce.py
+```
+
+The command returns `1` for the reproduced finite inconsistency, `0` after a
+successful repair, and `2` for a runner or environment failure.

tasks/task_091/metadata.json

@@ -0,0 +1,12 @@
+{
+  "task_id": "091",
+  "title": "Repair a workflow regression",
+  "domain": "materials-science-volumetric-data-semantics",
+  "language": "python",
+  "source_repository": "https://github.com/materialsproject/pymatgen-core",
+  "source_commit": "c8cfee71f3ce5d92c5b5f6ca3d363f8f3a9f4495",
+  "source_license": "MIT",
+  "license_source": "LICENSE",
+  "gpl_family": false,
+  "public_payload_sha256": "ee6d2a8fa5af6c99aded10bacf8ffa2910bcb148ed68fb80707b9cebe7eb0437"
+}