diff --git a/DESCRIPTION b/DESCRIPTION
index 8a8f1a6..7aa9607 100644
--- a/DESCRIPTION
+++ b/DESCRIPTION
@@ -1,13 +1,34 @@
 Package: curatedBladderData
 Type: Package
 Title: Bladder Cancer Gene Expression Analysis
-Version: 1.19.2
-Date: 2019-03-28
+Version: 1.49.1
+Date: 2026-10-06
 Author: Markus Riester
 Maintainer: Markus Riester <markus@jimmy.harvard.edu>
-Description: The curatedBladderData package provides relevant functions and data for gene expression analysis in patients with bladder cancer.
-Depends: R (>= 2.10.0), affy
-Suggests: BiocStyle, survival, xtable, sva, genefilter, logging
+Description: Manually curated bladder cancer ExpressionSet objects from 10
+    studies (12 objects, since two studies contributed data on two platforms
+    each), with uniformly processed expression data and curated clinical
+    metadata. Datasets are hosted on Zenodo and downloaded on first use,
+    then cached locally with BiocFileCache. Includes a configurable script
+    for rule-based creation of filtered dataset lists for meta-analysis.
+Depends:
+    R (>= 4.0.0),
+    Biobase
+Imports:
+    BiocFileCache,
+    tools,
+    utils
+Suggests:
+    BiocStyle,
+    genefilter,
+    logging,
+    survival,
+    sva,
+    testthat (>= 3.0.0),
+    xtable
 License: Artistic-2.0
 URL: https://github.com/lima1/curatedBladderData
-biocViews: ExperimentData, CancerData, OvarianCancerData, MicroarrayData, ExpressionData
+BugReports: https://github.com/lima1/curatedBladderData/issues
+biocViews: ExperimentData, CancerData, MicroarrayData, ExpressionData, GEO
+BuildResaveData: no
+Config/testthat/edition: 3
diff --git a/NAMESPACE b/NAMESPACE
index 40ea8e2..0e3864a 100644
--- a/NAMESPACE
+++ b/NAMESPACE
@@ -1 +1,5 @@
-exportPattern("^[[:alpha:]]+")
+export(curatedBladderData)
+import(Biobase)
+importFrom(BiocFileCache, BiocFileCache, bfcrpath, bfcquery, bfcremove)
+importFrom(utils, read.csv)
+importFrom(tools, md5sum, R_user_dir)
diff --git a/NEWS.md b/NEWS.md
new file mode 100644
index 0000000..5752b6e
--- /dev/null
+++ b/NEWS.md
@@ -0,0 +1,39 @@
+# curatedBladderData 1.50.0
+
+## SIGNIFICANT USER-VISIBLE CHANGES
+
+* The 12 datasets are no longer stored inside the package. They are hosted
+  on Zenodo and downloaded individually on first use, then cached locally
+  with `BiocFileCache`. This shrinks the installed package from about
+  75 MB to a few MB.
+* New exported function `curatedBladderData()`: call it with no arguments
+  to list the available datasets, with one dataset name to get an
+  `ExpressionSet`, or with several names to get a named list. `test = TRUE`
+  loads small offline subsets bundled with the package, used by examples,
+  tests and the vignette.
+* Legacy `data(GSE89_eset)` access still works. It downloads through the
+  same cache, deferred until the object is first used, but is deprecated
+  and will be removed in a future release.
+* `inst/extdata/createEsetList.R` now loads datasets through the getter.
+  A new `test.mode` option in the patientselection config selects the
+  offline subsets.
+* The cache is package-specific, at
+  `tools::R_user_dir("curatedBladderData", "cache")`. Downloads are
+  verified against md5 checksums recorded in
+  `inst/extdata/zenodo-manifest.csv`.
+
+## BUG FIXES
+
+* `biocViews` listed `OvarianCancerData`; removed, and `GEO` added.
+* The package-level help page described ovarian cancer, listed the
+  curatedOvarianData author list, and named a different maintainer than
+  DESCRIPTION. Corrected to match DESCRIPTION.
+* Replaced the unused `affy` dependency with `Biobase`, which is where
+  `ExpressionSet`, `exprs()` and `pData()` actually come from.
+* `createEsetList.R` no longer errors when the filters exclude every
+  dataset (`which()` on an empty list).
+
+## INTERNAL
+
+* Added testthat tests and `inst/scripts/` documenting how the Zenodo
+  files, the offline fixtures and the data stubs are generated.
diff --git a/R/getData.R b/R/getData.R
new file mode 100644
index 0000000..4524c41
--- /dev/null
+++ b/R/getData.R
@@ -0,0 +1,118 @@
+## Data retrieval for curatedBladderData: datasets are hosted on Zenodo and
+## downloaded on first use into a package-specific BiocFileCache.
+##
+## NOTE: this file is intentionally duplicated (modulo package name) in
+## curatedCRCData and curatedOvarianData. Keep the copies in sync.
+
+.pkgname <- "curatedBladderData"
+
+## session flags (e.g., show the data() deprecation note only once)
+.pkgenv <- new.env(parent = emptyenv())
+
+## Manifest mapping dataset names to Zenodo file URLs, md5 checksums, and
+## sizes. Regenerated by inst/scripts/make-data.R; the manifest is the ONLY
+## place download URLs live.
+.manifest <- function() {
+    fname <- system.file("extdata", "zenodo-manifest.csv",
+                         package = .pkgname, mustWork = TRUE)
+    utils::read.csv(fname, stringsAsFactors = FALSE)
+}
+
+.cache <- function() {
+    BiocFileCache::BiocFileCache(tools::R_user_dir(.pkgname, "cache"))
+}
+
+## Return the local path to one dataset's .rda, downloading into the cache
+## on first use. The md5 checksum is verified after a fresh download only;
+## on mismatch the corrupt entry is removed and the download retried once.
+.getDataFile <- function(name, retry = TRUE) {
+    meta <- .manifest()
+    i <- match(name, meta$dataset)
+    stopifnot(!is.na(i))
+    bfc <- .cache()
+    known <- BiocFileCache::bfcquery(bfc, meta$url[i], field = "fpath",
+                                     exact = TRUE)
+    fresh <- nrow(known) == 0L
+    if (fresh)
+        message("Downloading ", name, " (",
+                round(meta$size_bytes[i] / 2^20, 1), " MB) from Zenodo...")
+    path <- unname(BiocFileCache::bfcrpath(bfc, meta$url[i]))
+    if (!fresh && !isTRUE(file.size(path) == meta$size_bytes[i])) {
+        # cached file was truncated or modified after download: evict and
+        # re-download (cheap size check; full md5 runs after each download)
+        res <- BiocFileCache::bfcquery(bfc, meta$url[i], field = "fpath",
+                                       exact = TRUE)
+        if (nrow(res))
+            BiocFileCache::bfcremove(bfc, res$rid)
+        if (retry) {
+            message("Cached copy of ", name,
+                    " has the wrong size; re-downloading...")
+            return(.getDataFile(name, retry = FALSE))
+        }
+        stop("Cached file for '", name, "' has the wrong size and could ",
+             "not be refreshed.")
+    }
+    if (fresh) {
+        md5 <- unname(tools::md5sum(path))
+        if (!identical(md5, meta$md5[i])) {
+            res <- BiocFileCache::bfcquery(bfc, meta$url[i], field = "fpath",
+                                           exact = TRUE)
+            if (nrow(res))
+                BiocFileCache::bfcremove(bfc, res$rid)
+            if (retry) {
+                message("Checksum mismatch; retrying the download once...")
+                return(.getDataFile(name, retry = FALSE))
+            }
+            stop("md5 checksum mismatch for '", name, "' downloaded from ",
+                 meta$url[i],
+                 "; the corrupt copy was removed from the cache.")
+        }
+    }
+    path
+}
+
+.loadOne <- function(name, test = FALSE) {
+    path <- if (test) {
+        system.file("extdata", "testdata", paste0(name, ".rda"),
+                    package = .pkgname, mustWork = TRUE)
+    } else {
+        .getDataFile(name)
+    }
+    env <- new.env(parent = emptyenv())
+    objname <- load(path, envir = env)
+    get(objname[1L], envir = env)
+}
+
+## Called by the data/*.R stubs so that legacy data(X) usage keeps working.
+## During this package's own R CMD check, resolve to the bundled test
+## fixture so checking needs no network access; checks of OTHER packages
+## (reverse dependencies) get the real data.
+.stubLoad <- function(name) {
+    if (identical(Sys.getenv("_R_CHECK_PACKAGE_NAME_"), .pkgname))
+        return(.loadOne(name, test = TRUE))
+    if (!isTRUE(.pkgenv$stub_noted)) {
+        .pkgenv$stub_noted <- TRUE
+        message("Note: loading ", .pkgname, " datasets via data() is ",
+                "deprecated and will be removed in a future release; use ",
+                .pkgname, "(\"", name, "\") instead. Datasets are now ",
+                "downloaded on first use and cached locally with ",
+                "BiocFileCache. (This note is shown once per session.)")
+    }
+    .loadOne(name)
+}
+
+#' @export
+curatedBladderData <- function(datasets, test = FALSE) {
+    meta <- .manifest()
+    if (missing(datasets))
+        return(meta$dataset)
+    stopifnot(is.character(datasets), !anyNA(datasets))
+    bad <- setdiff(datasets, meta$dataset)
+    if (length(bad))
+        stop("Unknown dataset(s): ", paste(bad, collapse = ", "),
+             ". Call curatedBladderData() with no arguments to list ",
+             "available datasets.")
+    res <- lapply(datasets, .loadOne, test = test)
+    names(res) <- datasets
+    if (length(res) == 1L) res[[1L]] else res
+}
diff --git a/data/GSE13507_eset.R b/data/GSE13507_eset.R
new file mode 100644
index 0000000..408cca9
--- /dev/null
+++ b/data/GSE13507_eset.R
@@ -0,0 +1,3 @@
+delayedAssign("GSE13507_eset",
+    curatedBladderData:::.stubLoad("GSE13507_eset"),
+    assign.env = environment())
diff --git a/data/GSE13507_eset.rda b/data/GSE13507_eset.rda
deleted file mode 100644
index 5594ced..0000000
Binary files a/data/GSE13507_eset.rda and /dev/null differ
diff --git a/data/GSE1827_eset.R b/data/GSE1827_eset.R
new file mode 100644
index 0000000..c5f3f01
--- /dev/null
+++ b/data/GSE1827_eset.R
@@ -0,0 +1,3 @@
+delayedAssign("GSE1827_eset",
+    curatedBladderData:::.stubLoad("GSE1827_eset"),
+    assign.env = environment())
diff --git a/data/GSE1827_eset.rda b/data/GSE1827_eset.rda
deleted file mode 100644
index 85d4a74..0000000
Binary files a/data/GSE1827_eset.rda and /dev/null differ
diff --git a/data/GSE19915.GPL3883_eset.R b/data/GSE19915.GPL3883_eset.R
new file mode 100644
index 0000000..5baf288
--- /dev/null
+++ b/data/GSE19915.GPL3883_eset.R
@@ -0,0 +1,3 @@
+delayedAssign("GSE19915.GPL3883_eset",
+    curatedBladderData:::.stubLoad("GSE19915.GPL3883_eset"),
+    assign.env = environment())
diff --git a/data/GSE19915.GPL3883_eset.rda b/data/GSE19915.GPL3883_eset.rda
deleted file mode 100644
index bd9b996..0000000
Binary files a/data/GSE19915.GPL3883_eset.rda and /dev/null differ
diff --git a/data/GSE19915.GPL5186_eset.R b/data/GSE19915.GPL5186_eset.R
new file mode 100644
index 0000000..2d61d16
--- /dev/null
+++ b/data/GSE19915.GPL5186_eset.R
@@ -0,0 +1,3 @@
+delayedAssign("GSE19915.GPL5186_eset",
+    curatedBladderData:::.stubLoad("GSE19915.GPL5186_eset"),
+    assign.env = environment())
diff --git a/data/GSE19915.GPL5186_eset.rda b/data/GSE19915.GPL5186_eset.rda
deleted file mode 100644
index 2c0a6ec..0000000
Binary files a/data/GSE19915.GPL5186_eset.rda and /dev/null differ
diff --git a/data/GSE31189_eset.R b/data/GSE31189_eset.R
new file mode 100644
index 0000000..0411de3
--- /dev/null
+++ b/data/GSE31189_eset.R
@@ -0,0 +1,3 @@
+delayedAssign("GSE31189_eset",
+    curatedBladderData:::.stubLoad("GSE31189_eset"),
+    assign.env = environment())
diff --git a/data/GSE31189_eset.rda b/data/GSE31189_eset.rda
deleted file mode 100644
index 11b24c5..0000000
Binary files a/data/GSE31189_eset.rda and /dev/null differ
diff --git a/data/GSE31684_eset.R b/data/GSE31684_eset.R
new file mode 100644
index 0000000..4290ce0
--- /dev/null
+++ b/data/GSE31684_eset.R
@@ -0,0 +1,3 @@
+delayedAssign("GSE31684_eset",
+    curatedBladderData:::.stubLoad("GSE31684_eset"),
+    assign.env = environment())
diff --git a/data/GSE31684_eset.rda b/data/GSE31684_eset.rda
deleted file mode 100644
index 3c817f9..0000000
Binary files a/data/GSE31684_eset.rda and /dev/null differ
diff --git a/data/GSE32894_eset.R b/data/GSE32894_eset.R
new file mode 100644
index 0000000..f7a9042
--- /dev/null
+++ b/data/GSE32894_eset.R
@@ -0,0 +1,3 @@
+delayedAssign("GSE32894_eset",
+    curatedBladderData:::.stubLoad("GSE32894_eset"),
+    assign.env = environment())
diff --git a/data/GSE32894_eset.rda b/data/GSE32894_eset.rda
deleted file mode 100644
index 0453aa3..0000000
Binary files a/data/GSE32894_eset.rda and /dev/null differ
diff --git a/data/GSE37317_eset.R b/data/GSE37317_eset.R
new file mode 100644
index 0000000..c216283
--- /dev/null
+++ b/data/GSE37317_eset.R
@@ -0,0 +1,3 @@
+delayedAssign("GSE37317_eset",
+    curatedBladderData:::.stubLoad("GSE37317_eset"),
+    assign.env = environment())
diff --git a/data/GSE37317_eset.rda b/data/GSE37317_eset.rda
deleted file mode 100644
index 9a9f596..0000000
Binary files a/data/GSE37317_eset.rda and /dev/null differ
diff --git a/data/GSE5287_eset.R b/data/GSE5287_eset.R
new file mode 100644
index 0000000..f393743
--- /dev/null
+++ b/data/GSE5287_eset.R
@@ -0,0 +1,3 @@
+delayedAssign("GSE5287_eset",
+    curatedBladderData:::.stubLoad("GSE5287_eset"),
+    assign.env = environment())
diff --git a/data/GSE5287_eset.rda b/data/GSE5287_eset.rda
deleted file mode 100644
index d426b69..0000000
Binary files a/data/GSE5287_eset.rda and /dev/null differ
diff --git a/data/GSE89_eset.R b/data/GSE89_eset.R
new file mode 100644
index 0000000..35466fa
--- /dev/null
+++ b/data/GSE89_eset.R
@@ -0,0 +1,3 @@
+delayedAssign("GSE89_eset",
+    curatedBladderData:::.stubLoad("GSE89_eset"),
+    assign.env = environment())
diff --git a/data/GSE89_eset.rda b/data/GSE89_eset.rda
deleted file mode 100644
index 981a00e..0000000
Binary files a/data/GSE89_eset.rda and /dev/null differ
diff --git a/data/PMID17099711.GPL8300_eset.R b/data/PMID17099711.GPL8300_eset.R
new file mode 100644
index 0000000..74b9bdd
--- /dev/null
+++ b/data/PMID17099711.GPL8300_eset.R
@@ -0,0 +1,3 @@
+delayedAssign("PMID17099711.GPL8300_eset",
+    curatedBladderData:::.stubLoad("PMID17099711.GPL8300_eset"),
+    assign.env = environment())
diff --git a/data/PMID17099711.GPL8300_eset.rda b/data/PMID17099711.GPL8300_eset.rda
deleted file mode 100644
index aff7e6a..0000000
Binary files a/data/PMID17099711.GPL8300_eset.rda and /dev/null differ
diff --git a/data/PMID17099711.GPL91_eset.R b/data/PMID17099711.GPL91_eset.R
new file mode 100644
index 0000000..ba4c076
--- /dev/null
+++ b/data/PMID17099711.GPL91_eset.R
@@ -0,0 +1,3 @@
+delayedAssign("PMID17099711.GPL91_eset",
+    curatedBladderData:::.stubLoad("PMID17099711.GPL91_eset"),
+    assign.env = environment())
diff --git a/data/PMID17099711.GPL91_eset.rda b/data/PMID17099711.GPL91_eset.rda
deleted file mode 100644
index 6ca90b7..0000000
Binary files a/data/PMID17099711.GPL91_eset.rda and /dev/null differ
diff --git a/external_data_store.txt b/external_data_store.txt
deleted file mode 100644
index 9e67bd4..0000000
--- a/external_data_store.txt
+++ /dev/null
@@ -1,2 +0,0 @@
-data
-
diff --git a/inst/extdata/createEsetList.R b/inst/extdata/createEsetList.R
index 561646e..daf9ee7 100644
--- a/inst/extdata/createEsetList.R
+++ b/inst/extdata/createEsetList.R
@@ -62,7 +62,10 @@ intersectMany <- function(lst){
 }
 
 ##load the esets
-data(list=data(package=package.name)[[3]][,3])
+if (!exists("test.mode")) test.mode <- FALSE
+for (.nm in do.call(package.name, list()))
+    assign(.nm, do.call(package.name, list(datasets = .nm, test = test.mode)))
+rm(.nm)
 
 strEsets <- ls(pattern="^.*_eset$")
 
@@ -142,8 +145,8 @@ if(keep.common.only){
     })
 }
 
-ids.with.missing.data <- which(sapply(esets, function(X)
-sum(!complete.cases(exprs(X))) > 0))
+ids.with.missing.data <- which(vapply(esets, function(X)
+sum(!complete.cases(exprs(X))) > 0, logical(1)))
 loginfo(paste("Ids with missing data:", paste(names(ids.with.missing.data),
 collapse=", ")))
 
diff --git a/inst/extdata/patientselection_all.config b/inst/extdata/patientselection_all.config
index 9f36f90..336f3e3 100755
--- a/inst/extdata/patientselection_all.config
+++ b/inst/extdata/patientselection_all.config
@@ -2,6 +2,11 @@
 ##settings for createEsetList.R
 ##-----------------------------
 
+
+#use the small offline test subsets bundled with the package instead of
+#downloading the full datasets from Zenodo? (for examples/vignettes/tests)
+test.mode <- FALSE
+
 package.name <- "curatedBladderData"
 
 #rescale each gene to z-score?
diff --git a/inst/extdata/testdata/GSE13507_eset.rda b/inst/extdata/testdata/GSE13507_eset.rda
new file mode 100644
index 0000000..96a4bb1
Binary files /dev/null and b/inst/extdata/testdata/GSE13507_eset.rda differ
diff --git a/inst/extdata/testdata/GSE1827_eset.rda b/inst/extdata/testdata/GSE1827_eset.rda
new file mode 100644
index 0000000..9aade87
Binary files /dev/null and b/inst/extdata/testdata/GSE1827_eset.rda differ
diff --git a/inst/extdata/testdata/GSE19915.GPL3883_eset.rda b/inst/extdata/testdata/GSE19915.GPL3883_eset.rda
new file mode 100644
index 0000000..6181a69
Binary files /dev/null and b/inst/extdata/testdata/GSE19915.GPL3883_eset.rda differ
diff --git a/inst/extdata/testdata/GSE19915.GPL5186_eset.rda b/inst/extdata/testdata/GSE19915.GPL5186_eset.rda
new file mode 100644
index 0000000..764fec4
Binary files /dev/null and b/inst/extdata/testdata/GSE19915.GPL5186_eset.rda differ
diff --git a/inst/extdata/testdata/GSE31189_eset.rda b/inst/extdata/testdata/GSE31189_eset.rda
new file mode 100644
index 0000000..6eeceb3
Binary files /dev/null and b/inst/extdata/testdata/GSE31189_eset.rda differ
diff --git a/inst/extdata/testdata/GSE31684_eset.rda b/inst/extdata/testdata/GSE31684_eset.rda
new file mode 100644
index 0000000..ad399f4
Binary files /dev/null and b/inst/extdata/testdata/GSE31684_eset.rda differ
diff --git a/inst/extdata/testdata/GSE32894_eset.rda b/inst/extdata/testdata/GSE32894_eset.rda
new file mode 100644
index 0000000..ad8a1fa
Binary files /dev/null and b/inst/extdata/testdata/GSE32894_eset.rda differ
diff --git a/inst/extdata/testdata/GSE37317_eset.rda b/inst/extdata/testdata/GSE37317_eset.rda
new file mode 100644
index 0000000..a860629
Binary files /dev/null and b/inst/extdata/testdata/GSE37317_eset.rda differ
diff --git a/inst/extdata/testdata/GSE5287_eset.rda b/inst/extdata/testdata/GSE5287_eset.rda
new file mode 100644
index 0000000..0d3cee4
Binary files /dev/null and b/inst/extdata/testdata/GSE5287_eset.rda differ
diff --git a/inst/extdata/testdata/GSE89_eset.rda b/inst/extdata/testdata/GSE89_eset.rda
new file mode 100644
index 0000000..90ed848
Binary files /dev/null and b/inst/extdata/testdata/GSE89_eset.rda differ
diff --git a/inst/extdata/testdata/PMID17099711.GPL8300_eset.rda b/inst/extdata/testdata/PMID17099711.GPL8300_eset.rda
new file mode 100644
index 0000000..bab4d62
Binary files /dev/null and b/inst/extdata/testdata/PMID17099711.GPL8300_eset.rda differ
diff --git a/inst/extdata/testdata/PMID17099711.GPL91_eset.rda b/inst/extdata/testdata/PMID17099711.GPL91_eset.rda
new file mode 100644
index 0000000..34dd886
Binary files /dev/null and b/inst/extdata/testdata/PMID17099711.GPL91_eset.rda differ
diff --git a/inst/extdata/zenodo-manifest.csv b/inst/extdata/zenodo-manifest.csv
new file mode 100644
index 0000000..c28220a
--- /dev/null
+++ b/inst/extdata/zenodo-manifest.csv
@@ -0,0 +1,13 @@
+dataset,filename,url,md5,size_bytes
+GSE13507_eset,GSE13507_eset.rda,https://zenodo.org/records/23192472/files/GSE13507_eset.rda?download=1,9df8c8a2c77600fb3346d7a9d6b34b20,6289556
+GSE1827_eset,GSE1827_eset.rda,https://zenodo.org/records/23192472/files/GSE1827_eset.rda?download=1,3a195c532c1067c1204fd963583d4b2c,3157656
+GSE19915.GPL3883_eset,GSE19915.GPL3883_eset.rda,https://zenodo.org/records/23192472/files/GSE19915.GPL3883_eset.rda?download=1,6d433ad595361f1a893d0f3e07a0f4c3,5705184
+GSE19915.GPL5186_eset,GSE19915.GPL5186_eset.rda,https://zenodo.org/records/23192472/files/GSE19915.GPL5186_eset.rda?download=1,d78aa0d949148cd5a30868453896d058,7141496
+GSE31189_eset,GSE31189_eset.rda,https://zenodo.org/records/23192472/files/GSE31189_eset.rda?download=1,feb209ae4826e1691b7c4d39048b9635,8670296
+GSE31684_eset,GSE31684_eset.rda,https://zenodo.org/records/23192472/files/GSE31684_eset.rda?download=1,3deda3de9e45b169271f472d8c2c2e62,7548196
+GSE32894_eset,GSE32894_eset.rda,https://zenodo.org/records/23192472/files/GSE32894_eset.rda?download=1,43100e65e99bf474f00cb018500df411,14435224
+GSE37317_eset,GSE37317_eset.rda,https://zenodo.org/records/23192472/files/GSE37317_eset.rda?download=1,b0dd231adb053574f487f832e38ad5f5,1791700
+GSE5287_eset,GSE5287_eset.rda,https://zenodo.org/records/23192472/files/GSE5287_eset.rda?download=1,93666b747360690461639477e535ef70,2803624
+GSE89_eset,GSE89_eset.rda,https://zenodo.org/records/23192472/files/GSE89_eset.rda?download=1,93e550b942226b5958f7dad7abefcaa1,1542464
+PMID17099711.GPL8300_eset,PMID17099711.GPL8300_eset.rda,https://zenodo.org/records/23192472/files/PMID17099711.GPL8300_eset.rda?download=1,acd04f2130e55e221919f6491bf60e7b,1923304
+PMID17099711.GPL91_eset,PMID17099711.GPL91_eset.rda,https://zenodo.org/records/23192472/files/PMID17099711.GPL91_eset.rda?download=1,8c88b6bf8f2834efa022e5b3d7fd026a,1994320
diff --git a/inst/scripts/data-integrity-report.Rmd b/inst/scripts/data-integrity-report.Rmd
new file mode 100644
index 0000000..0a2e7fb
--- /dev/null
+++ b/inst/scripts/data-integrity-report.Rmd
@@ -0,0 +1,101 @@
+---
+title: "Data integrity report: Zenodo offloading"
+output: html_document
+params:
+  package: "curatedBladderData"
+  upload_dir: "../../../curatedBladderData-zenodo-upload"
+---
+
+<!-- Verifies the Zenodo data offloading, run manually from inst/scripts/
+     of a package checkout (not built with the package vignettes):
+
+     1. every xz re-serialized object uploaded to Zenodo is identical to
+        the object shipped in the last pre-refactor package version
+        (git tag pre-zenodo-refactor);
+     2. the md5 checksums in inst/extdata/zenodo-manifest.csv match both
+        the local upload files and the checksums Zenodo reports;
+     3. one real download through the getter works and hits the cache on
+        the second call.
+
+     Because the manifest md5s equal the md5s of byte-identical uploads,
+     passing 1 + 2 proves every Zenodo file; only one download needs to
+     be exercised end to end.
+
+     NOTE: intentionally duplicated in curatedOvarianData. -->
+
+```{r setup, include=FALSE}
+knitr::opts_chunk$set(echo = TRUE)
+suppressPackageStartupMessages(library(Biobase))
+pkg <- params$package
+stopifnot(requireNamespace(pkg))
+```
+
+# 1. Uploaded objects vs pre-refactor package objects
+
+```{r compare-objects}
+pre <- tempfile("pre-refactor-")
+system2("git", c("worktree", "add", "--detach", pre, "pre-zenodo-refactor"),
+        stdout = FALSE)
+
+manifest <- read.csv(file.path("..", "extdata", "zenodo-manifest.csv"))
+loadOne <- function(path) {
+    env <- new.env(); nm <- load(path, envir = env); get(nm[1], envir = env)
+}
+cmp <- vapply(manifest$dataset, function(nm) {
+    old <- loadOne(file.path(pre, "data", paste0(nm, ".rda")))
+    new <- loadOne(file.path(params$upload_dir, paste0(nm, ".rda")))
+    isTRUE(all.equal(exprs(old), exprs(new))) &&
+        isTRUE(all.equal(pData(old), pData(new))) &&
+        isTRUE(all.equal(fData(old), fData(new))) &&
+        identical(annotation(old), annotation(new)) &&
+        isTRUE(all.equal(experimentData(old), experimentData(new)))
+}, logical(1))
+system2("git", c("worktree", "remove", "--force", pre))
+data.frame(dataset = manifest$dataset, identical = cmp)
+stopifnot(all(cmp))
+```
+
+# 2. Manifest md5s vs local uploads and Zenodo
+
+```{r compare-md5}
+local_path <- file.path(params$upload_dir, manifest$filename)
+stopifnot(identical(unname(tools::md5sum(local_path)), manifest$md5))
+# size_bytes is not cosmetic: R/getData.R compares it against the cached
+# file on every cache hit, so a wrong size silently evicts and re-downloads
+# a perfectly good file on every call.
+stopifnot(identical(unname(file.size(local_path)), as.numeric(manifest$size_bytes)))
+
+record <- unique(sub(".*records/([0-9A-Za-z_]+)/.*", "\\1", manifest$url))
+stopifnot(length(record) == 1L)
+api <- jsonlite::fromJSON(paste0("https://zenodo.org/api/records/", record))
+zen <- data.frame(filename = api$files$key,
+                  md5 = sub("^md5:", "", api$files$checksum),
+                  size = as.numeric(api$files$size))
+merged <- merge(manifest[, c("filename", "md5", "size_bytes")], zen,
+                by = "filename", suffixes = c(".manifest", ".zenodo"))
+stopifnot(nrow(merged) == nrow(manifest),
+          identical(merged$md5.manifest, merged$md5.zenodo),
+          identical(as.numeric(merged$size_bytes), merged$size))
+merged
+```
+
+# 3. Download smoke test
+
+```{r smoke-test}
+getter <- get(pkg, envir = asNamespace(pkg))
+smallest <- manifest$dataset[which.min(manifest$size_bytes)]
+# use a throwaway cache so the user's real package cache is untouched
+old_cache <- Sys.getenv("R_USER_CACHE_DIR", unset = NA)
+Sys.setenv(R_USER_CACHE_DIR = tempfile("cache-"))
+eset <- getter(smallest)              # downloads
+stopifnot(is(eset, "ExpressionSet"))
+eset2 <- getter(smallest)             # must hit the cache silently
+stopifnot(identical(dim(eset), dim(eset2)))
+if (is.na(old_cache)) Sys.unsetenv("R_USER_CACHE_DIR") else
+    Sys.setenv(R_USER_CACHE_DIR = old_cache)
+dim(eset)
+```
+
+```{r session-info}
+sessionInfo()
+```
diff --git a/inst/scripts/make-data.R b/inst/scripts/make-data.R
new file mode 100644
index 0000000..75485e4
--- /dev/null
+++ b/inst/scripts/make-data.R
@@ -0,0 +1,69 @@
+## Prepare the per-dataset .rda files uploaded to Zenodo, along with the
+## download manifest, the data/ stub files, and data/datalist.
+##
+## The datasets themselves were produced by the curatedBladderData curation
+## pipeline (see the package vignette);
+## this script re-serializes the .rda files that shipped inside the package
+## through version 1.49.x (git tag: pre-zenodo-refactor) with xz compression
+## for hosting on Zenodo.
+##
+## Run from the package root with the full data/*.rda files present:
+##   Rscript inst/scripts/make-data.R            # before upload (placeholder URLs)
+##   Rscript inst/scripts/make-data.R <RECORD>   # after publishing, to finalize URLs
+## Already-compressed files in the output directory are left untouched, so
+## re-running with the record ID only rewrites the manifest. Do NOT
+## regenerate the .rda files after uploading (checksums must match Zenodo).
+##
+## NOTE: this script is intentionally duplicated in curatedCRCData and
+## curatedOvarianData. Keep the copies in sync.
+
+pkg <- unname(read.dcf("DESCRIPTION")[, "Package"])
+record <- commandArgs(trailingOnly = TRUE)
+record <- if (length(record)) record[1L] else "RECORD_ID"
+
+outdir <- file.path("..", paste0(pkg, "-zenodo-upload"))
+stubdir <- file.path(outdir, "stubs")
+dir.create(stubdir, recursive = TRUE, showWarnings = FALSE)
+
+## enumerate from data/ before the refactor, or from the already-compressed
+## upload directory when re-running to finalize URLs
+rdas <- sort(list.files("data", pattern = "\\.rda$"))
+if (length(rdas) == 0L)
+    rdas <- sort(list.files(outdir, pattern = "\\.rda$"))
+stopifnot(length(rdas) > 0L)
+
+rows <- lapply(rdas, function(f) {
+    nm <- sub("\\.rda$", "", f)
+    out <- file.path(outdir, f)
+    if (!file.exists(out)) {
+        env <- new.env(parent = emptyenv())
+        obj <- load(file.path("data", f), envir = env)
+        stopifnot(identical(obj, nm))
+        save(list = nm, envir = env, file = out, compress = "xz")
+        message("compressed ", f)
+    }
+    writeLines(sprintf(
+        'delayedAssign("%s",\n    %s:::.stubLoad("%s"),\n    assign.env = environment())',
+        nm, pkg, nm),
+        file.path(stubdir, paste0(nm, ".R")))
+    data.frame(dataset = nm, filename = f,
+        url = sprintf("https://zenodo.org/records/%s/files/%s?download=1",
+                      record, f),
+        md5 = unname(tools::md5sum(out)), size_bytes = file.size(out))
+})
+manifest <- do.call(rbind, rows)
+
+write.csv(manifest, file.path(outdir, "zenodo-manifest.csv"),
+          row.names = FALSE, quote = FALSE)
+writeLines(manifest$dataset, file.path(stubdir, "datalist"))
+writeLines(sprintf("%s  %12d  %s", manifest$md5, manifest$size_bytes,
+                   manifest$filename),
+           file.path(outdir, "upload-manifest.txt"))
+
+message(nrow(manifest), " datasets; total upload size ",
+        round(sum(manifest$size_bytes) / 2^20), " MB; written to ", outdir)
+if (identical(record, "RECORD_ID"))
+    message("Upload the .rda files to a new Zenodo record, publish it, ",
+            "verify the md5s shown by Zenodo against upload-manifest.txt, ",
+            "then re-run this script with the record ID and copy ",
+            "zenodo-manifest.csv to inst/extdata/.")
diff --git a/inst/scripts/make-test-data.R b/inst/scripts/make-test-data.R
new file mode 100644
index 0000000..1650e0e
--- /dev/null
+++ b/inst/scripts/make-test-data.R
@@ -0,0 +1,40 @@
+## Build the small offline test fixtures in inst/extdata/testdata/: one per
+## dataset, subset to the first <=200 features and <=100 samples with full
+## phenoData retained, so that examples, tests, vignettes, and R CMD check
+## run without network access (see the test= argument of the getter and
+## .stubLoad in R/getData.R).
+##
+## Run from the package root with the full data/*.rda files present
+## (git tag: pre-zenodo-refactor):
+##   Rscript inst/scripts/make-test-data.R
+##
+## NOTE: this script is intentionally duplicated in curatedCRCData and
+## curatedOvarianData. Keep the copies in sync.
+
+suppressPackageStartupMessages(library(Biobase))
+
+dir.create("inst/extdata/testdata", recursive = TRUE, showWarnings = FALSE)
+rdas <- sort(list.files("data", pattern = "\\.rda$"))
+stopifnot(length(rdas) > 0L)
+
+for (f in rdas) {
+    nm <- sub("\\.rda$", "", f)
+    env <- new.env(parent = emptyenv())
+    obj <- load(file.path("data", f), envir = env)
+    stopifnot(identical(obj, nm))
+    eset <- get(nm, envir = env)
+    ## keep the first ~200 features plus genes used in the vignettes
+    ## (survival forest plots, probe-expansion examples)
+    fn <- featureNames(eset)
+    keep <- unique(c(intersect(c("CXCL12", "CXCR4"), fn),
+                     head(grep("///", fn, value = TRUE), 5L),
+                     head(fn, 200L)))
+    fx <- eset[keep, seq_len(min(100L, ncol(eset)))]
+    assign(nm, fx)
+    save(list = nm, file = file.path("inst/extdata/testdata", f),
+         compress = "xz")
+    message("fixture ", nm, ": ", nrow(fx), " x ", ncol(fx))
+}
+message("total fixture size: ",
+        round(sum(file.size(list.files("inst/extdata/testdata",
+                                       full.names = TRUE))) / 2^10), " KB")
diff --git a/man/curatedBladderData-package.Rd b/man/curatedBladderData-package.Rd
index 18f6d53..d6424da 100755
--- a/man/curatedBladderData-package.Rd
+++ b/man/curatedBladderData-package.Rd
@@ -1,33 +1,26 @@
 \name{curatedBladderData-package}
-\alias{curatedBladderData}
+\alias{curatedBladderData-package}
 \docType{package}
 \title{Clinically Annotated Data for the Bladder Cancer Transcriptome}
 \description{The curatedBladderData package provides manually curated
-  clinical data, uniformly processed expression data, and convenience
-  functions for gene expression analysis in patients with ovarian cancer.}
+  clinical data and uniformly processed expression data for gene
+  expression analysis in patients with bladder cancer.}
 \details{
-\tabular{ll}{
-Package: \tab curatedBladderData\cr
-Type: \tab Package\cr
-Version: \tab \Sexpr{packageDescription("curatedBladderData")$Version}\cr
-Date: \tab \Sexpr{packageDescription("curatedBladderData")$Date}\cr
-License: \tab Artistic-2.0\cr
-Depends: \tab R (>= 2.10.0), affy\cr
-}
+  Datasets are hosted on Zenodo and downloaded on first use with
+  \code{\link{curatedBladderData}()}, which caches them locally with
+  \pkg{BiocFileCache}. Legacy \code{data()} access still works but is
+  deprecated and downloads through the same mechanism. See the package
+  vignette for rule-based creation of a list of filtered datasets.
 }
 \author{
-Benjamin F. Ganzfried, Steve Skates, Markus Riester,
-Victoria Wang, Thomas Risch, Benjamin Haibe-Kains, Curtis
-Huttenhower, Svitlana Tyekucheva, Jie Ding, Ina Jazic,
-Michael Birrer, Giovanni Parmigiani, Levi Waldron
-
-Department of Biostatistics and Computational Biology, Dana-Farber Cancer Institute, Harvard School of Public Health
+Markus Riester
 
-Maintainer: Levi Waldron <levi@jimmy.harvard.edu>
+Maintainer: Markus Riester <markus@jimmy.harvard.edu>
 }
-
 \examples{
-##List all datasets:
-data(package="curatedBladderData")
-##
+## List all datasets:
+curatedBladderData()
+## Load one dataset (test=TRUE uses a small offline subset):
+eset <- curatedBladderData("GSE89_eset", test = TRUE)
 }
+\keyword{package}
diff --git a/man/curatedBladderData.Rd b/man/curatedBladderData.Rd
new file mode 100644
index 0000000..bf42804
--- /dev/null
+++ b/man/curatedBladderData.Rd
@@ -0,0 +1,40 @@
+\name{curatedBladderData}
+\alias{curatedBladderData}
+\title{Download curated bladder cancer ExpressionSets}
+\description{
+  Datasets are hosted on Zenodo and downloaded individually on first use,
+  then cached locally with \pkg{BiocFileCache}; later calls load from the
+  cache without network access. Loading datasets with \code{data()} still
+  works but is deprecated in favor of this function.
+}
+\usage{
+curatedBladderData(datasets, test = FALSE)
+}
+\arguments{
+  \item{datasets}{Character vector of dataset names (e.g.
+    \code{"GSE32894_eset"}). If missing, the names of all available
+    datasets are returned and nothing is downloaded.}
+  \item{test}{If \code{TRUE}, load small offline test subsets bundled
+    with the package instead of downloading the full datasets. Intended
+    for examples, tests, and vignettes; the subsets are not suitable
+    for analysis.}
+}
+\value{
+  A single \code{\linkS4class{ExpressionSet}} if one dataset name is
+  given; a named list of \code{ExpressionSet}s if several; a character
+  vector of available dataset names if \code{datasets} is missing.
+}
+\details{
+  Downloaded files are stored in a package-specific cache directory
+  (\code{tools::R_user_dir("curatedBladderData", "cache")}). To remove
+  cached files, delete that directory or manage it with
+  \pkg{BiocFileCache}.
+}
+\examples{
+curatedBladderData()  # list available datasets
+eset <- curatedBladderData("GSE89_eset", test = TRUE)
+dim(eset)
+esets <- curatedBladderData(c("GSE37317_eset", "GSE89_eset"), test = TRUE)
+names(esets)
+}
+\keyword{utilities}
diff --git a/tests/testthat.R b/tests/testthat.R
new file mode 100644
index 0000000..a259495
--- /dev/null
+++ b/tests/testthat.R
@@ -0,0 +1,4 @@
+library(testthat)
+library(curatedBladderData)
+
+test_check("curatedBladderData")
diff --git a/tests/testthat/test-getData.R b/tests/testthat/test-getData.R
new file mode 100644
index 0000000..0427a02
--- /dev/null
+++ b/tests/testthat/test-getData.R
@@ -0,0 +1,85 @@
+# The 12 datasets this package has provided since its initial release;
+# this list guards the Zenodo manifest against accidental row loss.
+expected_datasets <- c(
+    "GSE13507_eset",
+    "GSE1827_eset",
+    "GSE19915.GPL3883_eset",
+    "GSE19915.GPL5186_eset",
+    "GSE31189_eset",
+    "GSE31684_eset",
+    "GSE32894_eset",
+    "GSE37317_eset",
+    "GSE5287_eset",
+    "GSE89_eset",
+    "PMID17099711.GPL8300_eset",
+    "PMID17099711.GPL91_eset"
+)
+
+manifest <- read.csv(system.file("extdata", "zenodo-manifest.csv",
+                                 package = "curatedBladderData"))
+
+test_that("manifest is complete and well-formed", {
+    expect_setequal(manifest$dataset, expected_datasets)
+    expect_false(anyDuplicated(manifest$dataset) > 0)
+    expect_identical(manifest$filename, paste0(manifest$dataset, ".rda"))
+    expect_true(all(grepl("^[0-9a-f]{32}$", manifest$md5)))
+    expect_true(all(grepl(
+        "^https://zenodo\\.org/records/[0-9]+/files/.+\\?download=1$",
+        manifest$url)))
+    expect_true(all(manifest$size_bytes > 0))
+})
+
+test_that("manifest, fixtures, and data/ stubs are in lock-step", {
+    fixtures <- sub("\\.rda$", "", list.files(
+        system.file("extdata", "testdata", package = "curatedBladderData"),
+        pattern = "\\.rda$"))
+    expect_setequal(fixtures, expected_datasets)
+    stubs <- data(package = "curatedBladderData")$results[, "Item"]
+    expect_setequal(stubs, expected_datasets)
+})
+
+test_that("no-argument call lists the datasets", {
+    expect_setequal(curatedBladderData(), expected_datasets)
+})
+
+test_that("getter returns ExpressionSets from offline fixtures", {
+    eset <- curatedBladderData("GSE89_eset", test = TRUE)
+    expect_s4_class(eset, "ExpressionSet")
+    expect_gt(ncol(eset), 0)
+    esets <- curatedBladderData(c("GSE37317_eset", "GSE89_eset"), test = TRUE)
+    expect_type(esets, "list")
+    expect_named(esets, c("GSE37317_eset", "GSE89_eset"))
+    expect_s4_class(esets[[1]], "ExpressionSet")
+})
+
+test_that("unknown dataset names give an informative error", {
+    expect_error(curatedBladderData("NOT_A_DATASET"), "Unknown dataset")
+    expect_error(curatedBladderData("NOT_A_DATASET", test = TRUE),
+                 "Unknown dataset")
+})
+
+test_that("data() stubs create a working binding", {
+    e <- new.env()
+    data("GSE89_eset", package = "curatedBladderData", envir = e)
+    expect_true(exists("GSE89_eset", envir = e))
+    # force the delayed binding under the check guard so .stubLoad() runs
+    # and resolves to the offline fixture (no network)
+    old <- Sys.getenv("_R_CHECK_PACKAGE_NAME_", unset = NA)
+    Sys.setenv("_R_CHECK_PACKAGE_NAME_" = "curatedBladderData")
+    on.exit(if (is.na(old)) Sys.unsetenv("_R_CHECK_PACKAGE_NAME_") else
+        Sys.setenv("_R_CHECK_PACKAGE_NAME_" = old), add = TRUE)
+    expect_s4_class(e$GSE89_eset, "ExpressionSet")
+})
+
+test_that("full download works (opt-in; set RUN_FULL_DOWNLOAD_TESTS=1)", {
+    skip_on_bioc()
+    skip_if(!nzchar(Sys.getenv("RUN_FULL_DOWNLOAD_TESTS")))
+    skip_if_offline("zenodo.org")
+    # smallest dataset in the collection
+    eset <- curatedBladderData("GSE1827_eset")
+    expect_s4_class(eset, "ExpressionSet")
+    # second call must hit the cache (no download message)
+    expect_silent(
+        eset2 <- curatedBladderData("GSE1827_eset"))
+    expect_identical(dim(eset), dim(eset2))
+})
diff --git a/vignettes/curatedBladderData_vignette.rnw b/vignettes/curatedBladderData_vignette.rnw
index 122af2e..351d562 100755
--- a/vignettes/curatedBladderData_vignette.rnw
+++ b/vignettes/curatedBladderData_vignette.rnw
@@ -38,18 +38,38 @@ and clinical data formats.
 In this vignette, we give a short tour of the package and will show how to use
 it efficiently.
 
+\section{Data access and caching}
+
+As of version 1.49.1 the datasets are no longer stored inside the package.
+They are hosted on Zenodo and downloaded individually the first time they
+are requested, then cached locally with \Biocpkg{BiocFileCache}, so later
+uses need no network access.  The cache lives in a package-specific
+directory, \texttt{tools::R\_user\_dir("curatedBladderData", "cache")}; to
+reclaim disk space, delete that directory.
+
+The legacy \texttt{data(GSE89\_eset)} syntax still works, downloading
+through the same cache, but is deprecated in favour of the
+\texttt{curatedBladderData()} function shown below.
+
+Throughout this vignette we pass \texttt{test = TRUE} (and set
+\texttt{test.mode <- TRUE}), which loads small offline subsets bundled with
+the package so that the vignette builds without downloading data.  For real
+analyses, omit these settings.
+
 \section{Load data sets}
 Loading a single dataset is very easy. First we load the package:
 <<example1>>=
 library(curatedBladderData)
 @
-To get a listing of all the datasets, use the \texttt{data} function:
+To get a listing of all the datasets, call \texttt{curatedBladderData()}
+with no arguments:
 <<example1tcgastep2>>=
-data(package="curatedBladderData")
+curatedBladderData()
 @
-Now to load a single dataset, we use the \texttt{data} function again:
-<<example1>>=
-data(GSE89_eset)
+Now to load a single dataset (drop \texttt{test = TRUE} to download the
+full dataset):
+<<example1load>>=
+GSE89_eset <- curatedBladderData("GSE89_eset", test = TRUE)
 GSE89_eset
 @
 The datasets are provided as Bioconductor \texttt{ExpressionSet} objects and
@@ -67,6 +87,7 @@ early stage and normal samples.
 <<example2loadstep1>>=
 source(system.file("extdata", 
 "patientselection_all.config",package="curatedBladderData"))
+test.mode <- TRUE  # use the bundled offline subsets in this vignette
 ls()
 @
 
@@ -208,8 +229,7 @@ a simple way to do it.  For one dataset:
 
 <<simplygetdata, eval=FALSE>>=
 library(curatedBladderData)
-library(affy)
-data(GSE89_eset)
+GSE89_eset <- curatedBladderData("GSE89_eset")
 write.csv(exprs(GSE89_eset), file="GSE89_eset_exprs.csv")
 write.csv(pData(GSE89_eset), file="GSE89_eset_clindata.csv")
 @ 
@@ -220,7 +240,7 @@ Or for several datasets:
 data.to.fetch <- c("GSE89_eset", "GSE37317_eset")
 for (onedata in data.to.fetch){
     print(paste("Fetching", onedata))
-    data(list=onedata)
+    assign(onedata, curatedBladderData(onedata))
     write.csv(exprs(get(onedata)), file=paste(onedata, "_exprs.csv", sep=""))
     write.csv(pData(get(onedata)), file=paste(onedata, "_clindata.csv", sep=""))
 }