diff --git a/pipeline.ori b/pipeline.ori
new file mode 100644
index 0000000..d433796
--- /dev/null
+++ b/pipeline.ori
@@ -0,0 +1,68 @@
+/*
+  This 'pipeline' is separate from final site.ori,
+  to keep parts of the pipeline private while still allowing better debugging.
+  serve/watch pipeline.ori for debugging.
+  server or copy site.ori to build final output.
+*/
+(all) => {
+  //put the file selection part in a closure, so that the list of all files is kept private.
+    /*
+      list is all the files in current directory.
+      allfiles is all source files: ending with `.md` and not starting with `_`.
+      pubfiles is the subset containing the string `_pub`.
+      files selects either of these depending on the argument passed to this file.
+
+    NOTES:
+
+    - if markdown files are empty, Origami.document() will error.
+    */
+
+    /*
+      this is the external filename method. 
+    */
+  //  allfiles: allfiles.sh() //reads MD files.
+  //  pubfiles: pubfiles.sh(allfiles)
+    (list): ..
+    allfiles: Tree.filter(list, (val, key) => key.endsWith('.md') && !key.startsWith('_'))
+      → (items) => Tree.map(items, {value: (value) => Origami.document(value)}) //if I don't use the verbose syntax I get the filename characters as part of value!
+      → (items) => Tree.map(items, { value: (value, key) => addFilenameData.js(value, key)})
+
+    pubfiles: Tree.filter(allfiles, (val, key) => val.fnd.tags?.includes('pub'))
+
+    /*
+      Now convert to html.
+    */
+    asHtml: Tree.map(all ? allfiles: pubfiles, {
+      value: (value) => Origami.mdHtml(value)
+      key: (value, key) => `${value.fnd.name}.html`
+    })
+
+    /*
+    after converting to html, we can extract a 'summary' from the file.
+    With markdown, the title is taken from the YAML frontmatter,
+    and ends up in a separate field of the document.
+    So we dont have to worry about the title being part of the summary.
+    */
+    withSummary: Tree.map(asHtml, extractSummary.js)
+
+    /*
+    Removing private content coming below the html comment `<!--private-below-->`
+    Only remove this if in public mode. So if this file is called as `site.ori("all")`,
+    then final output is `withSummary`, otherwise it's `privateRemoved`.
+    */
+    privateRemoved: Tree.map(withSummary, removePrivate.js)
+    final: (all ? withSummary : privateRemoved)
+  
+
+  renderedPages: Tree.map(final, (page) => <page.ori>(page, all))
+
+  /*
+  assets are relative to the pubfiles directory
+  */
+  css
+
+  /*
+  Now, I think I have enough to build both the individual pages and the index page!
+  */
+  index.html: indexPage.ori(final, all)
+}
diff --git a/site.ori b/site.ori
index 0380eea..9210863 100644
--- a/site.ori
+++ b/site.ori
@@ -1,71 +1,6 @@
 (all) => {
-  //put the file selection part in a closure, so that the list of all files is kept private.
-    /*
-      list is all the files in current directory.
-      allfiles is all source files: ending with `.md` and not starting with `_`.
-      pubfiles is the subset containing the string `_pub`.
-      files selects either of these depending on the argument passed to this file.
-
-    NOTES:
-
-    - if markdown files are empty, Origami.document() will error.
-    */
-
-    /*
-      this is the external filename method. 
-    */
-  //  allfiles: allfiles.sh() //reads MD files.
-  //  pubfiles: pubfiles.sh(allfiles)
-    (list): ..
-    allfiles: Tree.filter(list, (val, key) => key.endsWith('.md') && !key.startsWith('_'))
-      → (items) => Tree.map(items, {value: (value) => Origami.document(value)}) //if I don't use the verbose syntax I get the filename characters as part of value!
-      → (items) => Tree.map(items, { value: (value, key) => addFilenameData.js(value, key)})
-
-    pubfiles: Tree.filter(allfiles, (val, key) => val.fnd.tags?.includes('pub'))
-
-    /*
-      Now convert to html.
-    */
-    asHtml: Tree.map(all ? allfiles: pubfiles, {
-      value: (value) => Origami.mdHtml(value)
-      key: (value, key) => `${value.fnd.name}.html`
-    })
-
-    /*
-    after converting to html, we can extract a 'summary' from the file.
-    With markdown, the title is taken from the YAML frontmatter,
-    and ends up in a separate field of the document.
-    So we dont have to worry about the title being part of the summary.
-    */
-    withSummary: Tree.map(asHtml, extractSummary.js)
-
-    /*
-    Removing private content coming below the html comment `<!--private-below-->`
-    Only remove this if in public mode. So if this file is called as `site.ori("all")`,
-    then final output is `withSummary`, otherwise it's `privateRemoved`.
-    */
-    privateRemoved: Tree.map(withSummary, removePrivate.js)
-    final: (all ? withSummary : privateRemoved)
-  
-
-  renderedPages: Tree.map(final, (page) => <page.ori>(page, all))
-  /*
-  the rendered pages contain links and images.
-  We want to find those, and insert a tree of their maps into the output.
-  First test is whether that works.
-  Second test is whether the paths will be correct.
-  */
-  hast = Tree.map(renderedPages, parseHtml.js)
-  links:  Tree.flat(hast, 2) → (a) => [...new Set(a)]
-
-  /*
-  assets are relative to the pubfiles directory
-  */
-  css
-
-  /*
-  Now, I think I have enough to build both the individual pages and the index page!
-  */
-  index.html: indexPage.ori(final, all)
-  ...renderedPages/
+   (src): pipeline.ori(all)
+   index.html: src/index.html
+   ...src/renderedPages
+   css: src/css
 }