diff --git a/.changeset/afraid-pugs-repeat.md b/.changeset/afraid-pugs-repeat.md
new file mode 100644
index 0000000..57c67df
--- /dev/null
+++ b/.changeset/afraid-pugs-repeat.md
@@ -0,0 +1,6 @@
+---
+"@platforma-open/milaboratories.samples-and-data.model": minor
+"@platforma-open/milaboratories.samples-and-data": minor
+---
+
+Export the study setup without its data: template params now always strip datasets and metadata columns down to their configuration, dropping everything keyed by sample or by group (files, sample ids and labels), instead of carrying a fully populated state when every file happened to be a storage reference.
diff --git a/model/src/template_params.ts b/model/src/template_params.ts
index 22cc767..3cd4c9b 100644
--- a/model/src/template_params.ts
+++ b/model/src/template_params.ts
@@ -1,6 +1,4 @@
 import type { BlockParams } from "@platforma-open/milaboratories.samples-and-data.kind";
-import type { ImportFileHandle } from "@platforma-sdk/model";
-import { assertNever, isImportFileHandleIndex } from "@platforma-sdk/model";
 import { isGroupedDataset } from "./args";
 import type { BlockData, DSAny, DSContent, MTColumn } from "./args";
 
@@ -8,27 +6,22 @@ import type { BlockData, DSAny, DSContent, MTColumn } from "./args";
  * Derives the params a project exported as a template hands the block it seeds —
  * the inverse of the data model's `init`.
  *
- * Either the whole study travels or none of its data does. The block's state is
- * one connected thing — datasets key their files by sample or by group, sample
- * groups and multiplexing rules name those samples, metadata values are keyed by
- * them, and `args` holds the pieces to each other — so carrying part of it means
- * choosing which invariants to break. When every file can be resolved from where
- * the block lands, the state goes as it stands; when any file is local to this
- * machine, only the study's setup goes and the new project imports its own data.
+ * The study's setup travels, its data does not: every dataset and every metadata
+ * column goes as configured, emptied of everything keyed by sample or by group.
+ * The block the template seeds is the study as it was set up, waiting for its
+ * files — the user re-adds data into the datasets that were already there, in
+ * place of rebuilding them.
+ *
+ * Data is dropped whether or not it could travel. A file is portable only when
+ * it is a storage reference (`index://` names a `{storageId, path}` any
+ * installation carrying that storage can resolve, while `upload://` carries a
+ * local path signed with the installation's own secret), so carrying files
+ * would make a template mean one thing for one study and another for the next.
+ * Samples go with the files that produced them, so `sampleIds` and
+ * `sampleLabels` stay behind too — a sample without its file is a row with
+ * metadata and no data, for the user to delete.
  */
 export function deriveTemplateParams(data: BlockData): BlockParams {
-  if (data.datasets.every(isDatasetPortable)) {
-    return {
-      datasets: data.datasets,
-      metadata: data.metadata,
-      sampleIds: data.sampleIds,
-      sampleLabelColumnLabel: data.sampleLabelColumnLabel,
-      sampleLabels: data.sampleLabels,
-      h5adFilesToPreprocess: data.h5adFilesToPreprocess,
-      seuratFilesToPreprocess: data.seuratFilesToPreprocess,
-    };
-  }
-
   return {
     datasets: data.datasets.map(stripDatasetData),
     metadata: data.metadata.map(stripColumnValues),
@@ -38,56 +31,6 @@ export function deriveTemplateParams(data: BlockData): BlockParams {
 
 // Internals
 
-/**
- * Tells whether a dataset holds no file that is bound to this machine.
- *
- * A storage reference (`index://`) names a `{storageId, path}` any installation
- * carrying that storage can resolve; an upload handle carries a local path
- * signed with the installation's own secret and resolves nowhere else. A slot
- * with no file yet is neither, so it does not stand in the way.
- */
-function isDatasetPortable(ds: DSAny): boolean {
-  return collectDatasetHandles(ds.content).every((h) => h == null || isImportFileHandleIndex(h));
-}
-
-/**
- * Collects every file slot a dataset holds, empty ones included.
- *
- * Exhaustive over the dataset kinds: a kind added to `DSContent` and not handled
- * here fails to compile on `assertNever`, rather than silently reporting that
- * the new kind holds no files — which would make it look portable and send its
- * local uploads into a template.
- */
-function collectDatasetHandles(content: DSContent): (ImportFileHandle | null | undefined)[] {
-  switch (content.type) {
-    case "Fastq":
-    case "MultiplexedFastq":
-    case "CellRangerMTX":
-      return Object.values(content.data).flatMap((group) => Object.values(group));
-    case "MultilaneFastq":
-      return Object.values(content.data).flatMap((lanes) =>
-        Object.values(lanes).flatMap((group) => Object.values(group)),
-      );
-    case "TaggedFastq":
-      return Object.values(content.data).flatMap((records) =>
-        records.flatMap((record) => Object.values(record.files)),
-      );
-    case "TaggedXsv":
-      return Object.values(content.data).flatMap((records) => records.map((record) => record.file));
-    case "Fasta":
-    case "Xsv":
-    case "H5AD":
-    case "H5":
-    case "Seurat":
-    case "BulkCountMatrix":
-    case "MultiSampleH5AD":
-    case "MultiSampleSeurat":
-      return Object.values(content.data);
-    default:
-      return assertNever(content);
-  }
-}
-
 /**
  * Strips everything a dataset holds per sample or per group, keeping how the
  * dataset is configured: its type, read indices, tag names, barcode tags, xsv