diff --git a/.claude/skills/perturbNMF-runner/scripts/convert_igvf_mudata_to_h5ad.py b/.claude/skills/perturbNMF-runner/scripts/convert_igvf_mudata_to_h5ad.py
index 6ad97ce..cab061a 100644
--- a/.claude/skills/perturbNMF-runner/scripts/convert_igvf_mudata_to_h5ad.py
+++ b/.claude/skills/perturbNMF-runner/scripts/convert_igvf_mudata_to_h5ad.py
@@ -14,7 +14,7 @@
 **binary assignment in layers['guide_assignment']**. This script uses the layer; using
 X would treat ambient guide reads as perturbations.
 
-Generalises the one-off ``Convert_file_adata.py`` written for IGVF_Gersbach_iHep.
+Generalises a one-off ``Convert_file_adata.py`` written for a single IGVF dataset.
 """
 
 import argparse
diff --git a/.claude/skills/perturbNMF-runner/scripts/run_add_umap.sh b/.claude/skills/perturbNMF-runner/scripts/run_add_umap.sh
index 706c09c..6ffa54c 100755
--- a/.claude/skills/perturbNMF-runner/scripts/run_add_umap.sh
+++ b/.claude/skills/perturbNMF-runner/scripts/run_add_umap.sh
@@ -8,40 +8,50 @@
 # Run directly and it measures the h5mu, derives --mem/--time/--cpus-per-task, and
 # re-submits itself with those overrides. Do NOT prefix it with sbatch -- if you do,
 # that job just re-submits a correctly sized one and exits, costing an extra hop.
-# To pin resources yourself, set ADD_UMAP_INJOB to skip the sizing path entirely:
+# To pin resources yourself, set ADD_UMAP_INJOB to skip the sizing path entirely
+# (Slurm does not create the log directory, so make it first):
+#   mkdir -p add_umap_logs
 #   ADD_UMAP_INJOB=1 sbatch --mem=64G --time=06:00:00 run_add_umap.sh <h5mu>
 # The static #SBATCH directives below only apply on that manual path; they are
-# sized for the largest jamboree input.
+# sized for the largest input measured (below) and log to ./add_umap_logs.
+# Edit <partition> (or export SBATCH_PARTITION) and <your_email>, and set
+# PIPELINE_ROOT first:
+#   export PIPELINE_ROOT=/path/to/PerturbNMF
+# On the self-sizing path, logs go to $ADD_UMAP_LOG_DIR (default: ./add_umap_logs
+# in the directory you run it from), which is created automatically.
 #
 # Sizing basis (measured 2026-09-15). Peak RSS is driven by the expression matrix
 # and, in the dense case, by the temporary numpy allocates for X.var(axis=0) inside
-# ensure_umap -- that temporary is the full size of X again, which is why the dense
-# Gersbach matrix dominates everything else:
+# ensure_umap -- that temporary is the full size of X again, which is why a dense
+# matrix dominates everything else:
 #
-#   dataset       rna/X resident                      est. peak
-#   Gersbach      46.95 GB dense f32 1.05M x 11138    ~95-115 GB
-#   Hon_CM         9.91 GB sparse    nnz 1.24e9        ~30-40 GB
-#   Huangfu DE     5.01 GB sparse    nnz 627M          ~15-20 GB
-#   Huangfu ESC    3.37 GB sparse    nnz 422M          ~10-14 GB
+#   rna/X resident                      est. peak
+#   46.95 GB dense f32 1.05M x 11138    ~95-115 GB
+#    9.91 GB sparse    nnz 1.24e9        ~30-40 GB
+#    5.01 GB sparse    nnz 627M          ~15-20 GB
+#    3.37 GB sparse    nnz 422M          ~10-14 GB
 #
 # --in_place additionally needs free disk for a second copy of the h5mu while it
-# writes (48 GB for Gersbach; Oak had ~11 TB free at time of writing).
+# writes (48 GB for the largest input above).
 
 #SBATCH --job-name=add_umap
-#SBATCH --output=/scratch/users/ymo/add_umap_logs/%j.out
-#SBATCH --error=/scratch/users/ymo/add_umap_logs/%j.err
-#SBATCH --partition=engreitz,owners
+#SBATCH --output=add_umap_logs/%j.out   # relative to the submit dir
+#SBATCH --error=add_umap_logs/%j.err
+#SBATCH --partition=<partition>
 #SBATCH --time=16:00:00
 #SBATCH --nodes=1
 #SBATCH --ntasks=1
 #SBATCH --cpus-per-task=16
 #SBATCH --mem=192G
 #SBATCH --mail-type=BEGIN,END,FAIL
-#SBATCH --mail-user=ymo@stanford.edu
+#SBATCH --mail-user=<your_email>
 
 set -o pipefail
 
-SCRIPT_DIR="/oak/stanford/groups/engreitz/Users/ymo/Tools/PerturbNMF/.claude/skills/perturbNMF-runner/scripts"
+# Path to your PerturbNMF checkout (export PIPELINE_ROOT=/path/to/PerturbNMF first)
+: "${PIPELINE_ROOT:?set PIPELINE_ROOT to the PerturbNMF repo root}"
+SCRIPT_DIR="$PIPELINE_ROOT/.claude/skills/perturbNMF-runner/scripts"
+LOG_DIR="${ADD_UMAP_LOG_DIR:-$PWD/add_umap_logs}"
 
 MDATA_PATH="$1"
 if [ -z "$MDATA_PATH" ]; then
@@ -57,8 +67,8 @@ fi
 # Self-sizing submit path.
 #
 # The guard is the ADD_UMAP_INJOB sentinel this script sets on its own sbatch
-# call -- NOT $SLURM_JOB_ID. On Sherlock an OnDemand/code-server session is itself
-# a Slurm allocation that exports SLURM_JOB_ID (and SLURM_CPUS_PER_TASK=2), so a
+# call -- NOT $SLURM_JOB_ID. On clusters where an OnDemand/code-server session is
+# itself a Slurm allocation that exports SLURM_JOB_ID (and SLURM_CPUS_PER_TASK=2), a
 # SLURM_JOB_ID check reads as "already in a batch job" from any interactive shell
 # there and silently runs the whole UMAP in that 2-CPU session instead of queuing.
 # ---------------------------------------------------------------------------
@@ -83,14 +93,16 @@ if [ -z "$ADD_UMAP_INJOB" ]; then
     echo "h5mu:  $MDATA_PATH"
     echo "size:  ${SIZE_GB} GB"
     echo "sized: --mem=${MEM_GB}G --time=${TIME_H}:00:00 --cpus-per-task=${CPUS}"
-    mkdir -p /scratch/users/ymo/add_umap_logs
+    mkdir -p "$LOG_DIR"
     # Absolute self-reference so this works regardless of the caller's cwd.
     # ADD_UMAP_INJOB=1 is what makes the submitted copy run the workload instead of
     # re-submitting again; --export=ALL keeps the rest of the environment.
     exec sbatch --mem="${MEM_GB}G" \
                 --time="${TIME_H}:00:00" \
                 --cpus-per-task="${CPUS}" \
-                --export=ALL,ADD_UMAP_INJOB=1 \
+                --output="$LOG_DIR/%j.out" \
+                --error="$LOG_DIR/%j.err" \
+                --export=ALL,ADD_UMAP_INJOB=1,ADD_UMAP_LOG_DIR="$LOG_DIR" \
                 "$SCRIPT_DIR/run_add_umap.sh" "$@"
 fi
 
@@ -115,8 +127,6 @@ echo "h5mu: $MDATA_PATH"
 echo "Size: $(du -h "$MDATA_PATH" | cut -f1)"
 echo "Extra args: ${EXTRA_ARGS[*]}"
 
-mkdir -p /scratch/users/ymo/add_umap_logs
-
 echo "Activating conda environment: NMF_Benchmarking"
 eval "$(conda shell.bash hook)"
 conda activate NMF_Benchmarking
diff --git a/CHANGELOG.md b/CHANGELOG.md
index aaff40a..5d6c4e0 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -22,6 +22,9 @@ Entries marked ⚠️ change pipeline output — re-run affected analyses.
   repo from their own location instead of a hardcoded checkout path.
 - Annotation defaults are cell-type neutral; the PubMed keyword is optional, and
   Vertex AI project/bucket come from env vars (`VERTEX_PROJECT_ID`, `VERTEX_BUCKET`).
+- Annotation runners locate their scripts via `PIPELINE_ROOT` (fixing paths into a
+  nonexistent `AGeneTic/` folder); `gpe_export.py` reads the GeneProgramExplorer
+  location from `GPE_DIR`.
 
 ### Added
 - `tools/check_no_lab_specific_content.py` guard (CI + pre-commit) that blocks
diff --git a/src/Stage2_Evaluation/A_Metrics/src/association_categorical.py b/src/Stage2_Evaluation/A_Metrics/src/association_categorical.py
index 2dad811..ca7464d 100644
--- a/src/Stage2_Evaluation/A_Metrics/src/association_categorical.py
+++ b/src/Stage2_Evaluation/A_Metrics/src/association_categorical.py
@@ -365,7 +365,7 @@ def compute_categorical_association(
 
     # If test = 'pearsonr' with mode = 'one_vs_all' we are going to do something special until we can more properly integrate this
     if test=='pearsonr' and mode=='one_vs_all':
-        logging.info('Running jamboree specific version of posthoc with pearsonr, this is not yet integrated into the main pipeline')
+        logging.info('Running a one-vs-all pearsonr version of posthoc, this is not yet integrated into the main pipeline')
         res = []        
         # Run posthoc-tests and append results
         Parallel(n_jobs=n_jobs,
diff --git a/src/Stage3_Interpretation/C_Annotation/0.Prepare/Extract_program_information.py b/src/Stage3_Interpretation/C_Annotation/0.Prepare/Extract_program_information.py
index 7847982..228a952 100644
--- a/src/Stage3_Interpretation/C_Annotation/0.Prepare/Extract_program_information.py
+++ b/src/Stage3_Interpretation/C_Annotation/0.Prepare/Extract_program_information.py
@@ -284,7 +284,7 @@ def build_parser():
 
     # context info
     p.add_argument("--programs", type=int, nargs="+", help="Program ids, space separated (e.g. 1 2 3). Default: all.")
-    p.add_argument("--cell_type", default="", help="Cell type, e.g. 'teloHAEC aortic endothelial cell'.")
+    p.add_argument("--cell_type", default="", help="Cell type, e.g. 'aortic endothelial cell'.")
     p.add_argument("--organism", default="human")
 
     # select top items of the program
diff --git a/src/Stage3_Interpretation/C_Annotation/0.Prepare/run_extract_program_information.sh b/src/Stage3_Interpretation/C_Annotation/0.Prepare/run_extract_program_information.sh
index 79b5eb5..364bc5a 100644
--- a/src/Stage3_Interpretation/C_Annotation/0.Prepare/run_extract_program_information.sh
+++ b/src/Stage3_Interpretation/C_Annotation/0.Prepare/run_extract_program_information.sh
@@ -1,6 +1,6 @@
 #!/bin/bash
 #SBATCH --job-name=extract_prog_info
-#SBATCH -p normal
+#SBATCH -p <partition>
 #SBATCH --time=00:30:00
 #SBATCH --cpus-per-task=4
 #SBATCH --mem=16GB
@@ -17,11 +17,13 @@
 # literature steps investigate exactly these pairs.
 
 set -euo pipefail
-source /oak/stanford/groups/engreitz/Users/ymo/miniforge3/etc/profile.d/conda.sh
+eval "$(conda shell.bash hook)"
 conda activate NMF_Benchmarking
 
-SCRIPT_DIR=/oak/stanford/groups/engreitz/Users/ymo/Tools/PerturbNMF/src/Stage3_Interpretation/C_Annotation/AGeneTic/0.Prepare
-RESULT=/oak/stanford/groups/engreitz/Users/ymo/Project/Helen_telohaec/Result/073126_2kG_torch_e8_64
+# Path to your PerturbNMF checkout (export PIPELINE_ROOT=/path/to/PerturbNMF before sbatch)
+: "${PIPELINE_ROOT:?set PIPELINE_ROOT to the PerturbNMF repo root}"
+SCRIPT_DIR="$PIPELINE_ROOT/src/Stage3_Interpretation/C_Annotation/0.Prepare"
+RESULT="/path/to/output_dir/<run_name>"   # edit: run directory with Interpretation/
 OUT=$RESULT/Interpretation/AGeneTic_test
 
 python $SCRIPT_DIR/Extract_program_information.py \
@@ -35,7 +37,7 @@ python $SCRIPT_DIR/Extract_program_information.py \
   --gene_name_key var_names \
   --categorical_key type \
   --log2fc_key log2FC \
-  --cell_type "teloHAEC aortic endothelial cell" \
+  --cell_type "aortic endothelial cell" \
   --organism human \
   --top_gene 15 \
   --top_unique_gene 8 \
diff --git a/src/Stage3_Interpretation/C_Annotation/1.Search/1.1.Search_literature/search_nature.py b/src/Stage3_Interpretation/C_Annotation/1.Search/1.1.Search_literature/search_nature.py
index ee9f40e..2724886 100644
--- a/src/Stage3_Interpretation/C_Annotation/1.Search/1.1.Search_literature/search_nature.py
+++ b/src/Stage3_Interpretation/C_Annotation/1.Search/1.1.Search_literature/search_nature.py
@@ -13,11 +13,11 @@
 
 # Nature portfolio search (Crossref, DOI prefix 10.1038) + PDF download from nature.com.
 #
-# PDFs come from nature.com under Stanford's institutional subscription, which Nature grants
-# by IP address: it works from Sherlock / campus / Stanford VPN (Sherlock egresses from a
-# Stanford 171.67.x address) with no SUNet login. Off the Stanford network, only open-access
-# articles download. Stanford's license covers research use, not bulk harvesting, so results
-# are capped and PDF requests are spaced out.
+# PDFs come from nature.com under an institutional subscription, which Nature grants by IP
+# address: it works from a network covered by your institution's license (campus, VPN, or a
+# cluster that egresses from an institutional address) with no login. Elsewhere, only
+# open-access articles download. Institutional licenses cover research use, not bulk
+# harvesting, so results are capped and PDF requests are spaced out.
 CROSSREF_BASE_URL = "https://api.crossref.org"
 NATURE_BASE_URL = "https://www.nature.com"
 NATURE_DOI_PREFIX = "10.1038"
@@ -122,7 +122,7 @@ def search_Nature(
        bibliographic (title, authors, journal, year) via Crossref, restricted to DOI prefix 10.1038
     2. Adding a journal name or year to the query biases toward it (e.g. "... Nature Genetics 2023")
     3. Only journal articles are returned (no news, editorials split out by Crossref type)
-    4. PDF download uses Stanford's subscription by IP: run on Sherlock / campus / Stanford VPN
+    4. PDF download uses an institutional subscription by IP: run on a licensed network
 
     Env (optional): CROSSREF_MAILTO / PUBMED_EMAIL (Crossref polite pool). No API key needed.
 
@@ -168,7 +168,7 @@ def search_Nature(
             pdf_path, msg = download_paper(out_dir, f"nature_{rec['article_id']}",
                                            doi=rec["doi"], pdf_urls=[rec["pdf_url"]])
             if not pdf_path:
-                msg += "; is this running on the Stanford network (Sherlock / campus / VPN)?"
+                msg += "; is this running on a network with institutional access (campus / VPN)?"
             downloads.append((rec["title"], pdf_path, msg))
             pdf_line = '\n' + 'PDF: ' + (pdf_path or f'Not downloaded ({msg})')
 
@@ -177,7 +177,7 @@ def search_Nature(
     if out_dir:
         report_downloads("Nature", query, downloads, out_dir)
 
-    return "\n".join(data_list) + '\n' + "Data source: Nature (nature.com, Stanford subscription)"
+    return "\n".join(data_list) + '\n' + "Data source: Nature (nature.com, institutional subscription)"
 
 
 if __name__ == "__main__":
diff --git a/src/Stage3_Interpretation/C_Annotation/1.Search/run_MyGene.sh b/src/Stage3_Interpretation/C_Annotation/1.Search/run_MyGene.sh
index c5f2f42..8dc7001 100644
--- a/src/Stage3_Interpretation/C_Annotation/1.Search/run_MyGene.sh
+++ b/src/Stage3_Interpretation/C_Annotation/1.Search/run_MyGene.sh
@@ -1,6 +1,6 @@
 #!/bin/bash
 #SBATCH --job-name=mygene_extend
-#SBATCH -p normal
+#SBATCH -p <partition>
 #SBATCH --time=00:15:00
 #SBATCH --cpus-per-task=1
 #SBATCH --mem=4GB
@@ -15,11 +15,13 @@
 # --programs (required) takes space-separated ids (e.g. 1 2 3); only those P<k>.json are read.
 
 set -euo pipefail
-source /oak/stanford/groups/engreitz/Users/ymo/miniforge3/etc/profile.d/conda.sh
+eval "$(conda shell.bash hook)"
 conda activate NMF_Benchmarking
 
-SCRIPT_DIR=/oak/stanford/groups/engreitz/Users/ymo/Tools/PerturbNMF/src/Stage3_Interpretation/C_Annotation/AGeneTic/1.Search/1.0.Search_database/search_gene
-RESULT=/oak/stanford/groups/engreitz/Users/ymo/Project/Helen_telohaec/Result/073126_2kG_torch_e8_64
+# Path to your PerturbNMF checkout (export PIPELINE_ROOT=/path/to/PerturbNMF before sbatch)
+: "${PIPELINE_ROOT:?set PIPELINE_ROOT to the PerturbNMF repo root}"
+SCRIPT_DIR="$PIPELINE_ROOT/src/Stage3_Interpretation/C_Annotation/1.Search/1.0.Search_database/search_gene"
+RESULT="/path/to/output_dir/<run_name>"   # edit: run directory with Interpretation/
 BUNDLE_DIR=$RESULT/Interpretation/AGeneTic_test/PerturbNMF_Info 
 OVERWRITE=""   # default off; set to --overwrite to re-query genes that already have MyGene info
 OUT_DIR=$RESULT/Interpretation/AGeneTic_test/Gene_info_extended_PerturbNMF_Info
diff --git a/src/Stage3_Interpretation/C_Annotation/1.Search/run_NCBI.sh b/src/Stage3_Interpretation/C_Annotation/1.Search/run_NCBI.sh
index 12ab892..64338d2 100644
--- a/src/Stage3_Interpretation/C_Annotation/1.Search/run_NCBI.sh
+++ b/src/Stage3_Interpretation/C_Annotation/1.Search/run_NCBI.sh
@@ -1,6 +1,6 @@
 #!/bin/bash
 #SBATCH --job-name=ncbi_extend
-#SBATCH -p normal
+#SBATCH -p <partition>
 #SBATCH --time=00:15:00
 #SBATCH --cpus-per-task=1
 #SBATCH --mem=4GB
@@ -16,11 +16,13 @@
 
 # Optional: export NCBI_API_KEY (10 req/s instead of 3) and NCBI_EMAIL.
 set -euo pipefail
-source /oak/stanford/groups/engreitz/Users/ymo/miniforge3/etc/profile.d/conda.sh
+eval "$(conda shell.bash hook)"
 conda activate NMF_Benchmarking
 
-SCRIPT_DIR=/oak/stanford/groups/engreitz/Users/ymo/Tools/PerturbNMF/src/Stage3_Interpretation/C_Annotation/AGeneTic/1.Search/1.0.Search_database/search_gene
-RESULT=/oak/stanford/groups/engreitz/Users/ymo/Project/Helen_telohaec/Result/073126_2kG_torch_e8_64
+# Path to your PerturbNMF checkout (export PIPELINE_ROOT=/path/to/PerturbNMF before sbatch)
+: "${PIPELINE_ROOT:?set PIPELINE_ROOT to the PerturbNMF repo root}"
+SCRIPT_DIR="$PIPELINE_ROOT/src/Stage3_Interpretation/C_Annotation/1.Search/1.0.Search_database/search_gene"
+RESULT="/path/to/output_dir/<run_name>"   # edit: run directory with Interpretation/
 BUNDLE_DIR=$RESULT/Interpretation/AGeneTic_test/PerturbNMF_Info
 OVERWRITE=""   # default off; set to --overwrite to re-query genes that already have NCBI info
 OUT_DIR=$RESULT/Interpretation/AGeneTic_test/Gene_info_extended_PerturbNMF_Info
diff --git a/src/Stage3_Interpretation/C_Annotation/1.Search/run_OmniPath.sh b/src/Stage3_Interpretation/C_Annotation/1.Search/run_OmniPath.sh
index 1804972..58a10f0 100644
--- a/src/Stage3_Interpretation/C_Annotation/1.Search/run_OmniPath.sh
+++ b/src/Stage3_Interpretation/C_Annotation/1.Search/run_OmniPath.sh
@@ -1,6 +1,6 @@
 #!/bin/bash
 #SBATCH --job-name=omnipath_interact
-#SBATCH -p normal
+#SBATCH -p <partition>
 #SBATCH --time=00:30:00
 #SBATCH --cpus-per-task=1
 #SBATCH --mem=4GB
@@ -18,11 +18,13 @@
 # --programs (required) takes space-separated ids (e.g. 1 2 3); only those P<k>.json are read.
 
 set -euo pipefail
-source /oak/stanford/groups/engreitz/Users/ymo/miniforge3/etc/profile.d/conda.sh
+eval "$(conda shell.bash hook)"
 conda activate NMF_Benchmarking
 
-SCRIPT_DIR=/oak/stanford/groups/engreitz/Users/ymo/Tools/PerturbNMF/src/Stage3_Interpretation/C_Annotation/AGeneTic/1.Search/1.0.Search_database/search_gene_interaction
-RESULT=/oak/stanford/groups/engreitz/Users/ymo/Project/Helen_telohaec/Result/073126_2kG_torch_e8_64
+# Path to your PerturbNMF checkout (export PIPELINE_ROOT=/path/to/PerturbNMF before sbatch)
+: "${PIPELINE_ROOT:?set PIPELINE_ROOT to the PerturbNMF repo root}"
+SCRIPT_DIR="$PIPELINE_ROOT/src/Stage3_Interpretation/C_Annotation/1.Search/1.0.Search_database/search_gene_interaction"
+RESULT="/path/to/output_dir/<run_name>"   # edit: run directory with Interpretation/
 BUNDLE_DIR=$RESULT/Interpretation/AGeneTic_test/PerturbNMF_Info
 
 python $SCRIPT_DIR/search_OmniPath.py \
diff --git a/src/Stage3_Interpretation/C_Annotation/1.Search/run_UniProt.sh b/src/Stage3_Interpretation/C_Annotation/1.Search/run_UniProt.sh
index 4e88280..eb7641a 100644
--- a/src/Stage3_Interpretation/C_Annotation/1.Search/run_UniProt.sh
+++ b/src/Stage3_Interpretation/C_Annotation/1.Search/run_UniProt.sh
@@ -1,6 +1,6 @@
 #!/bin/bash
 #SBATCH --job-name=uniprot_extend
-#SBATCH -p normal
+#SBATCH -p <partition>
 #SBATCH --time=00:15:00
 #SBATCH --cpus-per-task=1
 #SBATCH --mem=4GB
@@ -15,11 +15,13 @@
 # --programs (required) takes space-separated ids (e.g. 1 2 3); only those P<k>.json are read.
 
 set -euo pipefail
-source /oak/stanford/groups/engreitz/Users/ymo/miniforge3/etc/profile.d/conda.sh
+eval "$(conda shell.bash hook)"
 conda activate NMF_Benchmarking
 
-SCRIPT_DIR=/oak/stanford/groups/engreitz/Users/ymo/Tools/PerturbNMF/src/Stage3_Interpretation/C_Annotation/AGeneTic/1.Search/1.0.Search_database/search_gene
-RESULT=/oak/stanford/groups/engreitz/Users/ymo/Project/Helen_telohaec/Result/073126_2kG_torch_e8_64
+# Path to your PerturbNMF checkout (export PIPELINE_ROOT=/path/to/PerturbNMF before sbatch)
+: "${PIPELINE_ROOT:?set PIPELINE_ROOT to the PerturbNMF repo root}"
+SCRIPT_DIR="$PIPELINE_ROOT/src/Stage3_Interpretation/C_Annotation/1.Search/1.0.Search_database/search_gene"
+RESULT="/path/to/output_dir/<run_name>"   # edit: run directory with Interpretation/
 BUNDLE_DIR=$RESULT/Interpretation/AGeneTic_test/PerturbNMF_Info
 OVERWRITE=""   # default off; set to --overwrite to re-query genes that already have UniProt info
 OUT_DIR=$RESULT/Interpretation/AGeneTic_test/Gene_info_extended_PerturbNMF_Info
diff --git a/src/Stage3_Interpretation/C_Annotation/1.Search/run_literature_search.sh b/src/Stage3_Interpretation/C_Annotation/1.Search/run_literature_search.sh
index 0ad979c..096b155 100755
--- a/src/Stage3_Interpretation/C_Annotation/1.Search/run_literature_search.sh
+++ b/src/Stage3_Interpretation/C_Annotation/1.Search/run_literature_search.sh
@@ -1,6 +1,6 @@
 #!/bin/bash
 #SBATCH --job-name=lit_search
-#SBATCH -p normal
+#SBATCH -p <partition>
 #SBATCH --time=04:00:00
 #SBATCH --cpus-per-task=1
 #SBATCH --mem=8GB
@@ -14,11 +14,13 @@
 # Tasks with a search_log.json are skipped; add --overwrite to re-search.
 # Uses the Claude Agent SDK (ANTHROPIC_API_KEY or a Claude Code login); spend is capped per task.
 set -euo pipefail
-source /oak/stanford/groups/engreitz/Users/ymo/miniforge3/etc/profile.d/conda.sh
+eval "$(conda shell.bash hook)"
 conda activate geneqa
 
-SCRIPT_DIR=/oak/stanford/groups/engreitz/Users/ymo/Tools/PerturbNMF/src/Stage3_Interpretation/C_Annotation/AGeneTic/1.Search
-RESULT=/oak/stanford/groups/engreitz/Users/ymo/Project/Helen_telohaec/Result/073126_2kG_torch_e8_64
+# Path to your PerturbNMF checkout (export PIPELINE_ROOT=/path/to/PerturbNMF before sbatch)
+: "${PIPELINE_ROOT:?set PIPELINE_ROOT to the PerturbNMF repo root}"
+SCRIPT_DIR="$PIPELINE_ROOT/src/Stage3_Interpretation/C_Annotation/1.Search"
+RESULT="/path/to/output_dir/<run_name>"   # edit: run directory with Interpretation/
 INFO_DIR=$RESULT/Interpretation/AGeneTic_test/Gene_info_extended_PerturbNMF_Info
 LIT_DIR=$RESULT/Interpretation/AGeneTic_test/Literature_info_extended_PerturbNMF_Info
 OVERWRITE=""   # set to --overwrite to re-search
diff --git a/src/Stage3_Interpretation/C_Annotation/1.Search/run_llm_query_agent.sh b/src/Stage3_Interpretation/C_Annotation/1.Search/run_llm_query_agent.sh
index 60b70ab..99e9fc2 100755
--- a/src/Stage3_Interpretation/C_Annotation/1.Search/run_llm_query_agent.sh
+++ b/src/Stage3_Interpretation/C_Annotation/1.Search/run_llm_query_agent.sh
@@ -1,6 +1,6 @@
 #!/bin/bash
 #SBATCH --job-name=lit_query_agent
-#SBATCH -p normal
+#SBATCH -p <partition>
 #SBATCH --time=00:30:00
 #SBATCH --cpus-per-task=1
 #SBATCH --mem=4GB
@@ -14,11 +14,13 @@
 # agent sees which genes lack cell-type evidence (optional). Programs that already have the plan are
 # skipped; add --overwrite to re-plan. Needs ANTHROPIC_API_KEY in AGeneTic/.env.
 set -euo pipefail
-source /oak/stanford/groups/engreitz/Users/ymo/miniforge3/etc/profile.d/conda.sh
+eval "$(conda shell.bash hook)"
 conda activate geneqa
 
-SCRIPT_DIR=/oak/stanford/groups/engreitz/Users/ymo/Tools/PerturbNMF/src/Stage3_Interpretation/C_Annotation/AGeneTic/1.Search
-RESULT=/oak/stanford/groups/engreitz/Users/ymo/Project/Helen_telohaec/Result/073126_2kG_torch_e8_64
+# Path to your PerturbNMF checkout (export PIPELINE_ROOT=/path/to/PerturbNMF before sbatch)
+: "${PIPELINE_ROOT:?set PIPELINE_ROOT to the PerturbNMF repo root}"
+SCRIPT_DIR="$PIPELINE_ROOT/src/Stage3_Interpretation/C_Annotation/1.Search"
+RESULT="/path/to/output_dir/<run_name>"   # edit: run directory with Interpretation/
 INFO_DIR=$RESULT/Interpretation/AGeneTic_test/Gene_info_extended_PerturbNMF_Info
 OUT_DIR=$RESULT/Interpretation/AGeneTic_test/Literature_info_extended_PerturbNMF_Info
 QUESTION=()                       # e.g. QUESTION=(--question "why does VIRMA knockdown reduce translation genes" --plan_name virma)
diff --git a/src/Stage3_Interpretation/C_Annotation/1.Search/run_search_generif.sh b/src/Stage3_Interpretation/C_Annotation/1.Search/run_search_generif.sh
index 2aef8a0..c0706bd 100755
--- a/src/Stage3_Interpretation/C_Annotation/1.Search/run_search_generif.sh
+++ b/src/Stage3_Interpretation/C_Annotation/1.Search/run_search_generif.sh
@@ -1,6 +1,6 @@
 #!/bin/bash
 #SBATCH --job-name=generif_lit
-#SBATCH -p normal
+#SBATCH -p <partition>
 #SBATCH --time=02:00:00
 #SBATCH --cpus-per-task=1
 #SBATCH --mem=8GB
@@ -15,11 +15,13 @@
 # Per-gene results are cached in <OUT_DIR>/gene_cache/; add --overwrite to re-summarize.
 # Needs ANTHROPIC_API_KEY (and optionally NCBI_API_KEY / NCBI_EMAIL / UNPAYWALL_EMAIL) in AGeneTic/.env.
 set -euo pipefail
-source /oak/stanford/groups/engreitz/Users/ymo/miniforge3/etc/profile.d/conda.sh
+eval "$(conda shell.bash hook)"
 conda activate geneqa
 
-SCRIPT_DIR=/oak/stanford/groups/engreitz/Users/ymo/Tools/PerturbNMF/src/Stage3_Interpretation/C_Annotation/AGeneTic/1.Search/1.1.Search_literature
-RESULT=/oak/stanford/groups/engreitz/Users/ymo/Project/Helen_telohaec/Result/073126_2kG_torch_e8_64
+# Path to your PerturbNMF checkout (export PIPELINE_ROOT=/path/to/PerturbNMF before sbatch)
+: "${PIPELINE_ROOT:?set PIPELINE_ROOT to the PerturbNMF repo root}"
+SCRIPT_DIR="$PIPELINE_ROOT/src/Stage3_Interpretation/C_Annotation/1.Search/1.1.Search_literature"
+RESULT="/path/to/output_dir/<run_name>"   # edit: run directory with Interpretation/
 INFO_DIR=$RESULT/Interpretation/AGeneTic_test/Gene_info_extended_PerturbNMF_Info
 OUT_DIR=$RESULT/Interpretation/AGeneTic_test/Literature_info_extended_PerturbNMF_Info
 DOWNLOAD_PDFS="--download_pdfs"   # set to "" to skip PDF downloads
diff --git a/src/Stage3_Interpretation/C_Annotation/2.Evidence_curation/gpe_export.py b/src/Stage3_Interpretation/C_Annotation/2.Evidence_curation/gpe_export.py
index e7b859b..ee62957 100644
--- a/src/Stage3_Interpretation/C_Annotation/2.Evidence_curation/gpe_export.py
+++ b/src/Stage3_Interpretation/C_Annotation/2.Evidence_curation/gpe_export.py
@@ -1,26 +1,33 @@
 """Attach cached paper-qa evidence onto an existing GPE contract, and validate it.
 
-Imports GPE's own contract classes live from Tools/GeneProgramExplorer (``pipeline/src``) so the
-emitted evidence matches the schema exactly. Attach-to-EXISTING-edges only: for each gene-gene
-edge whose pair has a cached result, we append a ``literature`` entry to the edge's ``evidence[]``.
-We never create new edges here.
+Imports GPE's own contract classes live from a GeneProgramExplorer checkout (``pipeline/src``) so
+the emitted evidence matches the schema exactly. Set the ``GPE_DIR`` environment variable to that
+checkout, or pass ``src_dir`` / ``schema_path`` explicitly. Attach-to-EXISTING-edges only: for each
+gene-gene edge whose pair has a cached result, we append a ``literature`` entry to the edge's
+``evidence[]``. We never create new edges here.
 """
 from __future__ import annotations
 
 import json
+import os
 import sys
 from pathlib import Path
 
 sys.path.insert(0, str(Path(__file__).resolve().parent))
 from evidence_cache import read_result  # noqa: E402
 
-GPE_DIR = "/oak/stanford/groups/engreitz/Users/ymo/Tools/GeneProgramExplorer"
+GPE_DIR = os.environ.get("GPE_DIR", "/path/to/GeneProgramExplorer")
 DEFAULT_GPE_SRC = f"{GPE_DIR}/pipeline/src"
 DEFAULT_SCHEMA = f"{GPE_DIR}/contract/schema.json"
 
 
 def _load_gpe(src_dir: str = DEFAULT_GPE_SRC):
     """Put the bundled GPE pipeline on sys.path and return the classes we need."""
+    if not Path(src_dir).is_dir():
+        raise FileNotFoundError(
+            f"GeneProgramExplorer source not found at {src_dir}. Set the GPE_DIR "
+            "environment variable to your GeneProgramExplorer checkout."
+        )
     src_dir = str(Path(src_dir).resolve())
     if src_dir not in sys.path:
         sys.path.insert(0, src_dir)
diff --git a/src/Stage3_Interpretation/C_Annotation/2.Evidence_curation/run_evidence_curation.sh b/src/Stage3_Interpretation/C_Annotation/2.Evidence_curation/run_evidence_curation.sh
index d02b32e..43ebf33 100644
--- a/src/Stage3_Interpretation/C_Annotation/2.Evidence_curation/run_evidence_curation.sh
+++ b/src/Stage3_Interpretation/C_Annotation/2.Evidence_curation/run_evidence_curation.sh
@@ -1,6 +1,6 @@
 #!/bin/bash
 #SBATCH --job-name=evidence_curation
-#SBATCH -p normal
+#SBATCH -p <partition>
 #SBATCH --time=02:00:00
 #SBATCH --cpus-per-task=2
 #SBATCH --mem=16GB
@@ -18,11 +18,13 @@
 # Leave out --cell_type to use each bundle's cell_type.
 
 set -euo pipefail
-source /oak/stanford/groups/engreitz/Users/ymo/miniforge3/etc/profile.d/conda.sh
+eval "$(conda shell.bash hook)"
 conda activate geneqa
 
-SCRIPT_DIR=/oak/stanford/groups/engreitz/Users/ymo/Tools/PerturbNMF/src/Stage3_Interpretation/C_Annotation/AGeneTic/2.Evidence_curation
-RESULT=/oak/stanford/groups/engreitz/Users/ymo/Project/Helen_telohaec/Result/073126_2kG_torch_e8_64
+# Path to your PerturbNMF checkout (export PIPELINE_ROOT=/path/to/PerturbNMF before sbatch)
+: "${PIPELINE_ROOT:?set PIPELINE_ROOT to the PerturbNMF repo root}"
+SCRIPT_DIR="$PIPELINE_ROOT/src/Stage3_Interpretation/C_Annotation/2.Evidence_curation"
+RESULT="/path/to/output_dir/<run_name>"   # edit: run directory with Interpretation/
 INFO_DIR=$RESULT/Interpretation/AGeneTic_test/Gene_info_extended_PerturbNMF_Info
 LIT_DIR=$RESULT/Interpretation/AGeneTic_test/Literature_info_extended_PerturbNMF_Info
 
diff --git a/src/Stage3_Interpretation/C_Annotation/3.Generate/run_html_report.sh b/src/Stage3_Interpretation/C_Annotation/3.Generate/run_html_report.sh
index 68f00c3..8ad0842 100644
--- a/src/Stage3_Interpretation/C_Annotation/3.Generate/run_html_report.sh
+++ b/src/Stage3_Interpretation/C_Annotation/3.Generate/run_html_report.sh
@@ -1,6 +1,6 @@
 #!/bin/bash
 #SBATCH --job-name=agenetic_report
-#SBATCH -p normal
+#SBATCH -p <partition>
 #SBATCH --time=00:10:00
 #SBATCH --cpus-per-task=1
 #SBATCH --mem=2GB
@@ -14,16 +14,18 @@
 # Add --cdn to load Cytoscape.js from the web instead of inlining it (file ~0.4 MB smaller).
 
 set -euo pipefail
-source /oak/stanford/groups/engreitz/Users/ymo/miniforge3/etc/profile.d/conda.sh
+eval "$(conda shell.bash hook)"
 conda activate NMF_Benchmarking
 
-SCRIPT_DIR=/oak/stanford/groups/engreitz/Users/ymo/Tools/PerturbNMF/src/Stage3_Interpretation/C_Annotation/AGeneTic/3.Generate
-RESULT=/oak/stanford/groups/engreitz/Users/ymo/Project/Helen_telohaec/Result/073126_2kG_torch_e8_64
+# Path to your PerturbNMF checkout (export PIPELINE_ROOT=/path/to/PerturbNMF before sbatch)
+: "${PIPELINE_ROOT:?set PIPELINE_ROOT to the PerturbNMF repo root}"
+SCRIPT_DIR="$PIPELINE_ROOT/src/Stage3_Interpretation/C_Annotation/3.Generate"
+RESULT="/path/to/output_dir/<run_name>"   # edit: run directory with Interpretation/
 INFO_DIR=$RESULT/Interpretation/AGeneTic_test/Gene_info_extended_PerturbNMF_Info
 
 python $SCRIPT_DIR/html_report.py \
   --info_dir $INFO_DIR \
   --out $RESULT/Interpretation/AGeneTic_test/Report/AGeneTic_report.html \
-  --dataset_name "Helen teloHAEC 2kG" \
+  --dataset_name "<dataset name>" \
   --top_regulator 5 \
   --top_gene 15
diff --git a/tests/Script/Stage3_Interpretation/A_Plotting/gene_html/test_html_gene_no_perturb.py b/tests/Script/Stage3_Interpretation/A_Plotting/gene_html/test_html_gene_no_perturb.py
index 8c52aed..15426a6 100644
--- a/tests/Script/Stage3_Interpretation/A_Plotting/gene_html/test_html_gene_no_perturb.py
+++ b/tests/Script/Stage3_Interpretation/A_Plotting/gene_html/test_html_gene_no_perturb.py
@@ -8,7 +8,7 @@
 
 Usage:
     eval "$(conda shell.bash hook)" && conda activate NMF_Benchmarking
-    cd /oak/stanford/groups/engreitz/Users/ymo/Tools/PerturbNMF
+    cd /path/to/PerturbNMF
     python -m pytest tests/Script/Stage3_Interpretation/A_Plotting/gene_html/test_html_gene_no_perturb.py -v
 """
 
diff --git a/tests/Script/Stage3_Interpretation/A_Plotting/program_html/test_html_program_no_perturb.py b/tests/Script/Stage3_Interpretation/A_Plotting/program_html/test_html_program_no_perturb.py
index c6e801a..272a3c8 100644
--- a/tests/Script/Stage3_Interpretation/A_Plotting/program_html/test_html_program_no_perturb.py
+++ b/tests/Script/Stage3_Interpretation/A_Plotting/program_html/test_html_program_no_perturb.py
@@ -8,7 +8,7 @@
 
 Usage:
     eval "$(conda shell.bash hook)" && conda activate NMF_Benchmarking
-    cd /oak/stanford/groups/engreitz/Users/ymo/Tools/PerturbNMF
+    cd /path/to/PerturbNMF
     python -m pytest tests/Script/Stage3_Interpretation/A_Plotting/program_html/test_html_program_no_perturb.py -v
 """