opensolr-haystack 0.2.7__tar.gz → 0.2.9__tar.gz

{opensolr_haystack-0.2.7 → opensolr_haystack-0.2.9}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: opensolr-haystack
-Version: 0.2.7
+Version: 0.2.9
 Summary: Haystack integration for Opensolr — managed Apache Solr DocumentStore with server-side embeddings and hybrid BM25+kNN retrieval
 Author-email: Opensolr <support@opensolr.com>
 License: MIT

{opensolr_haystack-0.2.7 → opensolr_haystack-0.2.9}/haystack_integrations/document_stores/opensolr/client.py

@@ -10,6 +10,7 @@ Direct Solr access (select/update) goes to the index's own host, resolved via
 
 from __future__ import annotations
 
+import re
 import json
 from typing import Any, Dict, List, Optional, Tuple
 
@@ -49,16 +50,23 @@ BATCH_EMBED_MAX = 50
 # rules make it lead with the answer and keep the concrete details. Callers
 # can still override everything by passing ``instruction``.
 DEFAULT_RAG_INSTRUCTION = (
-    # Consolidated 2026-08-26. The rule list had grown to eight overlapping lines and the
-    # 3B model drowned in them: it restated the question, opened with a preamble, and
-    # presented items as fit for purposes the context never stated. Tested against the
-    # live model on three cases (answer present, answer absent, yes/no) before shipping.
+    # Length cap and formatting reworked 2026-08-26: the earlier "two to four sentences"
+    # rule produced a dense paragraph that dropped most of what the documents said.
+    # Verified against a real four-document context before shipping — bulleted, far more
+    # detail, zero invented terms. The "do not add from your own knowledge" line is what
+    # keeps it grounded; without it the model pads the list with generic advice.
     "Answer the question using only the context below.\n"
     "Begin with the answer itself, the specific fact, product or detail. "
     "No preamble, no restating the question, no heading.\n"
-    "Then give the concrete supporting details from the context: names, numbers, sizes, dates. "
-    "Write two to four sentences.\n"
-    "If the context holds no answer, say that in the first sentence and name what it does contain instead.\n"
+    "Be thorough: cover every relevant point the context offers, with the concrete details "
+    "\u2014 names, model numbers, measurements, standards, dates.\n"
+    "Format it for reading, in Markdown: short paragraphs, and a bullet list when the answer "
+    "is a set of steps, precautions or options, each bullet opening with a bold lead-in naming "
+    "that item. Never invent generic headings such as \"Overview\", \"Key Points\" or \"Summary\".\n"
+    "Use only what the context states. Do not add advice, products or standards from your own "
+    "knowledge.\n"
+    "If the context holds no answer, say that in the first sentence and name what it does "
+    "contain instead.\n"
     "Never present something as suitable for a purpose the context does not state."
 )
 
@@ -385,11 +393,19 @@ class OpensolrClient:
 
         parts: List[str] = []
         for doc in hits[:docs]:
-            text_words = _flat(doc.get("text")).split()[:words]
+            # Truncate at the Nth word by CUTTING the original string, never by splitting and
+            # re-joining: " ".join(text.split()) throws away every newline and every run of
+            # indentation, so a page arrives as one unbroken line with its paragraphs, lists and
+            # code blocks flattened. Text handed to a model has to reach it as written; the
+            # structure is part of the meaning (2026-08-26).
+            body = _flat(doc.get("text"))
+            matches = list(re.finditer(r"\S+", body))
+            if len(matches) > words:
+                body = body[: matches[words - 1].end()]
             parts.append(
-                _flat(doc.get("title")) + " - "
-                + _flat(doc.get("description")) + " - "
-                + " ".join(text_words) + " - "
+                _flat(doc.get("title")) + "\n"
+                + _flat(doc.get("description")) + "\n"
+                + body + "\n\n"
             )
         return "".join(parts)
 

{opensolr_haystack-0.2.7 → opensolr_haystack-0.2.9}/opensolr_haystack.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: opensolr-haystack
-Version: 0.2.7
+Version: 0.2.9
 Summary: Haystack integration for Opensolr — managed Apache Solr DocumentStore with server-side embeddings and hybrid BM25+kNN retrieval
 Author-email: Opensolr <support@opensolr.com>
 License: MIT

{opensolr_haystack-0.2.7 → opensolr_haystack-0.2.9}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "opensolr-haystack"
-version = "0.2.7"
+version = "0.2.9"
 description = "Haystack integration for Opensolr — managed Apache Solr DocumentStore with server-side embeddings and hybrid BM25+kNN retrieval"
 readme = "README.md"
 license = { text = "MIT" }