commit af0dcdc16c0e77e3c7ebe922386c1a9f4b399b72
from: Matthias L. Jugel <leo@thinkberg.com>
date: Sun Jul 13 17:10:48 2025 UTC

make sure to combine duplicate references, display them nicer

commit - 7c634733a6f58eae9d9cbc7242fb529f23c19fe1
commit + af0dcdc16c0e77e3c7ebe922386c1a9f4b399b72
blob - 7eb4af358647391393cd2635a9178d34a6637c94
blob + 83352fe5e971b4612dd21262cc36f42e4effea83
--- configuration.py
+++ configuration.py
@@ -20,14 +20,14 @@ You are an expert assistant that provides well-sourced
 Use Markdown for all formatting (e.g., bold, italics, code blocks, lists, links).
 
 # CITATION REQUIREMENTS:
-* After **EVERY** statement, add `[n]` where n is the source number is was derived from
+* After **EVERY** statement, add `_[n]_` where n is the source number is was derived from
 * Source numbers correspond to the numbered sources below
-* For answers supported by multiple sources, use `[1][2][3]` format
+* For answers supported by multiple sources, use `_[1][2][3]_` format
 * Never cite sources that don't exist in the provided context
 * If information spans multiple sentences, cite after each relevant sentence
 
 ## EXAMPLE OUTPUT FORMAT:
-The renewable energy sector has grown significantly [1]. Solar power installations increased by 20% last year [2]. Wind energy also showed strong growth [1][3].
+The renewable energy sector has grown significantly _[1]_. Solar power installations increased by 20% last year [2]. Wind energy also showed strong growth _[1][3]_.
 """
 
 HUMAN_TEMPLATE: str = r"""
blob - 46ac5804929253742081a57fe74fa51e3e116225
blob + ef43eb0adc92bafd8fe4bdd000effa4dfc3e5f92
--- rag_backend.py
+++ rag_backend.py
@@ -34,12 +34,24 @@ class RagBackend:
             for i, (doc, _score) in enumerate(context_docs)
         ])
 
-        # prepare references for reply
-        references = [
-            f"{':'.join(doc.metadata.get('id').split(':')[:2])}:"
-            for (doc, _score) in context_docs
-        ]
+        # Create a map of document IDs to the indices where they occur
+        ref_map = {}
+        for i, (doc, _score) in enumerate(context_docs):
+            doc_id = ':'.join(doc.metadata.get('id').split(':')[:2])
+            if doc_id in ref_map:
+                ref_map[doc_id].append(i + 1)
+            else:
+                ref_map[doc_id] = [i + 1]
 
+        # Convert the map to a list of formatted strings
+        references = []
+        for doc_id, indices in ref_map.items():
+            # Format as "doc_id" with a mapping of indices
+            references.append({
+                'id': doc_id,
+                'indices': indices
+            })
+
         prompt_template = ChatPromptTemplate.from_messages([
             SystemMessage(content=SYSTEM_PROMPT),
             HumanMessagePromptTemplate.from_template(HUMAN_TEMPLATE)])
blob - 9442611fc77a5381cf224932c347a239d6f3fbf8
blob + aa2e95bb6b0d06817162bdf479a330818c8ec0a4
--- templates/page.html
+++ templates/page.html
@@ -105,33 +105,40 @@
         }
 
         .references-list {
-            list-style-type: none;
             padding-left: 0;
             margin: 0;
-            counter-reset: refs;
+            list-style-type: none;
         }
 
-        .references-list li {
+        .reference-item {
+            display: flex;
+            align-items: center;
             margin-bottom: 8px;
             font-size: 0.85em;
-            counter-increment: refs;
-            margin-left: 5px;
+            white-space: nowrap;
+            overflow: hidden;
         }
 
-        .references-list li::before {
-            content: "[" counter(refs) "] ";
+        .ref-indices {
             font-weight: bold;
+            margin-right: 8px;
+            flex-shrink: 0;
         }
 
+        .ref-link-container {
+            overflow: hidden;
+            text-overflow: ellipsis;
+            flex-grow: 1;
+        }
+
         .references-list a {
             color: #4a90e2;
             text-decoration: none;
-            display: inline-block;
-            max-width: calc(100% - 30px);
+            display: block;
+            width: 100%;
             white-space: nowrap;
             overflow: hidden;
             text-overflow: ellipsis;
-            vertical-align: bottom;
             position: relative;
         }
 
@@ -310,18 +317,35 @@
             refsBox.appendChild(refsTitle);
 
             // Create the reference list
-            const refsList = document.createElement('ol');
+            const refsList = document.createElement('div');
             refsList.className = 'references-list';
 
-            // Add each reference as a list item with link
-            references.forEach((ref, index) => {
-                const refItem = document.createElement('li');
+            // Add each reference as a div with indices and link
+            references.forEach((ref) => {
+                // Create a container for this reference item
+                const refItem = document.createElement('div');
+                refItem.className = 'reference-item';
 
+                // Extract the document ID and indices
+                // Format expected: object with 'id' and 'indices' properties
+                const docId = ref.id;
+                const indices = ref.indices;
+
                 // Parse the reference ID to get filename and page number
-                const idParts = ref.split(':');
+                const idParts = docId.split(':');
                 const filename = idParts[0];
                 const pageNumber = idParts[1] || '1';
 
+                // Create the indices prefix
+                const indicesSpan = document.createElement('span');
+                indicesSpan.className = 'ref-indices';
+                indicesSpan.textContent = indices.map(idx => `[${idx}]`).join(' ');
+                refItem.appendChild(indicesSpan);
+
+                // Create a container for the link to maintain overflow control
+                const linkContainer = document.createElement('div');
+                linkContainer.className = 'ref-link-container';
+
                 // Create the link to the document
                 const refLink = document.createElement('a');
                 refLink.href = `/static/files/${filename}?page=${pageNumber}#page=${pageNumber}`;
@@ -330,7 +354,8 @@
                 refLink.title = `${filename} (page ${pageNumber})`;
                 refLink.target = '_blank';
 
-                refItem.appendChild(refLink);
+                linkContainer.appendChild(refLink);
+                refItem.appendChild(linkContainer);
                 refsList.appendChild(refItem);
             });