sie-haystack 0.7.1__tar.gz → 0.7.3__tar.gz

{sie_haystack-0.7.1 → sie_haystack-0.7.3}/.gitignore

@@ -142,10 +142,8 @@ celerybeat.pid
 
 # Environments
 .env
-# Managed-service cloud provisioning secrets (cloud-provision / deploy/cloud).
-# Vendor credentials live in the env or a gitignored secrets file, never the
-# repo (audit §1/§14). Belt-and-braces so a local copy can never be committed.
-*.sie-cloud-secrets.env
+# Local service credential bundles must never be committed.
+*-secrets.env
 .env.cloud
 .venv
 env/
@@ -244,33 +242,6 @@ override.tf.json
 # tfvars files may contain secrets
 *.tfvars
 *.tfvars.json
-# Reviewed worker-AMI release contracts contain only immutable public package,
-# AMI, and Terraform backend coordinates; the release workflow accepts only
-# committed files from this exact internal path.
-!deploy/terraform/aws/internal-examples/cloud-benchmark-runner/worker-ami/releases/*.tfvars.json
-# Exception: this edge tfvars carries only a Secrets Manager ARN (a resource
-# identifier, not a secret) so the staging-us edge apply needs no manual -var.
-!deploy/cloud/terraform/aws-edge/staging-us.tfvars
-# This dev adoption contract contains only public AWS resource identifiers; it
-# is checked in so a reviewed plan cannot silently target a different orphan
-# VPC/listener through an operator-local tfvars file.
-!deploy/cloud/terraform/dev-controlplane-https-adoption/dev-us.tfvars
-!deploy/cloud/terraform/dev-controlplane-https-adoption/.terraform.lock.hcl
-# Managed release roots commit their exact provider selections. Release
-# preflights and applies use these read-only instead of selecting a newer
-# compatible provider during a long-running deployment.
-!deploy/cloud/terraform/release-infrastructure/.terraform.lock.hcl
-!deploy/cloud/terraform/aws-controlplane/.terraform.lock.hcl
-!deploy/cloud/terraform/aws-edge/.terraform.lock.hcl
-!deploy/cloud/terraform/s3-state/.terraform.lock.hcl
-!deploy/cloud/terraform/vercel/.terraform.lock.hcl
-!deploy/cloud/terraform/console-preview/.terraform.lock.hcl
-!deploy/cloud/terraform/examples/dev/.terraform.lock.hcl
-!deploy/cloud/terraform/examples/dev-eu/.terraform.lock.hcl
-!deploy/cloud/terraform/examples/staging/.terraform.lock.hcl
-!deploy/cloud/terraform/examples/staging-eu/.terraform.lock.hcl
-!deploy/cloud/terraform/examples/prod-us/.terraform.lock.hcl
-# Keep .terraform.lock.hcl for reproducibility (provider versions)
 
 # Node.js
 node_modules/
@@ -294,12 +265,7 @@ tmp/
 .tmp/
 .local/
 
-.requirements-modal.txt
-.mpa-i6pn-*-requirements-modal.txt
 b6_results.json
-# Raw dev-cloud load-test reports are local artifacts, not source-controlled baselines.
-packages/sie_cloud/bench/results/dev-us-*/
-
 # Personal mise overrides (per-developer toolchain tweaks, e.g. rustup
 # component naming differences across rustup versions).
 mise.local.toml

{sie_haystack-0.7.1 → sie_haystack-0.7.3}/PKG-INFO

@@ -1,6 +1,6 @@
-Metadata-Version: 2.4
+Metadata-Version: 2.5
 Name: sie-haystack
-Version: 0.7.1
+Version: 0.7.3
 Summary: SIE integration for Haystack
 Author-email: Superlinked <dev@superlinked.com>
 License: Apache-2.0

{sie_haystack-0.7.1 → sie_haystack-0.7.3}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "sie-haystack"
-version = "0.7.1"
+version = "0.7.3"
 description = "SIE integration for Haystack"
 readme = "README.md"
 requires-python = ">=3.10"

{sie_haystack-0.7.1 → sie_haystack-0.7.3}/src/sie_haystack/rankers.py

@@ -5,11 +5,49 @@ Provides SIERanker for reranking documents by relevance to a query.
 
 from __future__ import annotations
 
+from collections.abc import Mapping
 from typing import Any
 
 from haystack import Document, component
 
 
+def _scores_by_index(results: Mapping[str, Any], count: int) -> list[float]:
+    """Map ScoreResult entries back to input positions by item_id.
+
+    Parses each entry's ``item_id`` defensively: a missing, non-integer,
+    negative, or out-of-range id is skipped (that document keeps its 0.0
+    default) so a malformed entry can neither crash the rerank nor mis-assign a
+    score to the wrong document.
+
+    Args:
+        results: ScoreResult envelope from ``SIEClient.score()``.
+        count: Number of input documents.
+
+    Returns:
+        Scores indexed by input position (0.0 for any unscored/invalid item).
+    """
+    scores = [0.0] * count
+    for entry in results.get("scores", []):
+        item_id = entry.get("item_id", entry.get("index"))
+        # Accept only a genuine integer or an integer string. Reject bool (an int
+        # subclass, int(True) == 1), float (int(1.5) == 1), and non-integer
+        # strings, so a malformed id cannot silently overwrite the wrong position.
+        if isinstance(item_id, bool):
+            continue
+        if isinstance(item_id, int):
+            idx = item_id
+        elif isinstance(item_id, str):
+            try:
+                idx = int(item_id)
+            except ValueError:
+                continue
+        else:
+            continue
+        if 0 <= idx < count:
+            scores[idx] = float(entry.get("score", 0.0))
+    return scores
+
+
 @component
 class SIERanker:
     """Reranks documents by relevance to a query using SIE.
@@ -108,10 +146,19 @@ class SIERanker:
         # Score documents
         results = self.client.score(self._model, query_item, doc_items)
 
+        # ``results`` is a ScoreResult envelope; ranked entries are under
+        # ``results["scores"]`` (each a ScoreEntry keyed by ``item_id`` = input
+        # position, plus ``score``). Map scores back to input order by item_id
+        # rather than zipping positionally — the entries are sorted by relevance,
+        # not by input order. The envelope also carries ``results["request"]``
+        # (request id) and ``results["usage"]`` (token usage); those are
+        # available but intentionally not surfaced through the Haystack contract.
+        score_by_index = _scores_by_index(results, len(documents))
+
         # Build scored documents
         scored_docs = []
-        for doc, result in zip(documents, results, strict=True):
-            score = self._extract_score(result)
+        for idx, doc in enumerate(documents):
+            score = score_by_index[idx]
             # Store score in document metadata
             doc_with_score = Document(
                 id=doc.id,
@@ -131,9 +178,3 @@ class SIERanker:
             ranked_docs = ranked_docs[:effective_top_k]
 
         return {"documents": ranked_docs}
-
-    def _extract_score(self, result: Any) -> float:
-        """Extract score from SDK result."""
-        if isinstance(result, dict):
-            return float(result.get("score", 0.0))
-        return float(getattr(result, "score", 0.0))

{sie_haystack-0.7.1 → sie_haystack-0.7.3}/tests/conftest.py

@@ -155,13 +155,18 @@ def mock_sie_client() -> MagicMock:
             for item in items
         ]
 
-    def mock_score(_model: str, query: Any, items: list[Any], **kwargs: Any) -> list[dict]:
+    def mock_score(_model: str, query: Any, items: list[Any], **kwargs: Any) -> dict[str, Any]:
+        # Mirror the real SDK: a ScoreResult envelope with ranked entries under
+        # "scores", not a bare list.
         query_text = _get_text(query)
         item_dicts = [
             {"id": i.get("id", str(idx)) if isinstance(i, dict) else str(idx), "text": _get_text(i)}
             for idx, i in enumerate(items)
         ]
-        return _create_mock_score_result(query_text, item_dicts, kwargs.get("top_k"))
+        return {
+            "model": _model,
+            "scores": _create_mock_score_result(query_text, item_dicts, kwargs.get("top_k")),
+        }
 
     def mock_extract(_model: str, items: Any, labels: list[str], **_kwargs: Any) -> list[dict]:
         # Extract always returns a list of entities for Haystack

{sie_haystack-0.7.1 → sie_haystack-0.7.3}/tests/test_integration.py

@@ -4,7 +4,7 @@ These tests require a running SIE server and serve as runnable examples.
 Run with: pytest -m integration integrations/sie_haystack/tests/
 
 Prerequisites:
-    mise run serve -d cpu -p 8080
+    mise run serve -- -d cpu -p 8080
 """
 
 from __future__ import annotations

{sie_haystack-0.7.1 → sie_haystack-0.7.3}/tests/test_rankers.py

@@ -2,9 +2,42 @@
 
 from __future__ import annotations
 
+from collections import Counter
+from unittest.mock import MagicMock
+
 from haystack import Document
 from sie_haystack import SIERanker
 
+# Ranked entries reference input positions via item_id, out of input order and
+# sorted by relevance (doc-3 most relevant).
+_RANKED_ENVELOPE = {
+    "model": "test-reranker",
+    "scores": [
+        {"item_id": "3", "score": 0.9, "rank": 0},
+        {"item_id": "1", "score": 0.7, "rank": 1},
+        {"item_id": "4", "score": 0.5, "rank": 2},
+        {"item_id": "0", "score": 0.3, "rank": 3},
+        {"item_id": "2", "score": 0.1, "rank": 4},
+    ],
+}
+
+# Envelope for three inputs where only item_id "1" is usable; the rest are
+# malformed and must be skipped without crashing so their inputs keep the 0.0
+# default. The float 1.5 and bool True come AFTER the valid "1": if int()
+# accepted them (int(1.5) == 1, int(True) == 1) they would overwrite position 1.
+_MALFORMED_ENVELOPE = {
+    "model": "test-reranker",
+    "scores": [
+        {"item_id": "1", "score": 0.8, "rank": 0},
+        {"item_id": "not-an-int", "score": 0.95, "rank": 1},
+        {"item_id": "-1", "score": 0.9, "rank": 2},
+        {"item_id": "99", "score": 0.7, "rank": 3},
+        {"score": 0.5, "rank": 4},
+        {"item_id": 1.5, "score": 0.99, "rank": 5},
+        {"item_id": True, "score": 0.98, "rank": 6},
+    ],
+}
+
 
 class TestSIERanker:
     """Tests for SIERanker component."""
@@ -22,11 +55,44 @@ class TestSIERanker:
         result = ranker.run(query=test_query, documents=haystack_documents)
 
         assert "documents" in result
-        assert len(result["documents"]) == len(haystack_documents)
-        # All results should have scores in metadata
-        for doc in result["documents"]:
-            assert "score" in doc.meta
-            assert isinstance(doc.meta["score"], float)
+        docs = result["documents"]
+        # Every input document comes back exactly as many times as it went in
+        # (Counter checks multiplicity, not just membership). The envelope bug
+        # tried to zip documents against the ScoreResult dict's keys.
+        assert len(docs) == len(haystack_documents)
+        assert Counter(d.content for d in docs) == Counter(d.content for d in haystack_documents)
+        scores = [d.meta["score"] for d in docs]
+        assert all(isinstance(s, float) for s in scores)
+        # Distinct and descending.
+        assert scores == sorted(scores, reverse=True)
+        assert len(set(scores)) == len(scores)
+
+    def test_run_maps_scores_by_item_id(self, mock_sie_client: object) -> None:
+        """Top-ranked document is the relevant one, scores mapped by item_id."""
+        documents = [Document(content=f"doc-{i}") for i in range(5)]
+        mock_sie_client.score = MagicMock(return_value=_RANKED_ENVELOPE)
+        ranker = SIERanker(model="test-reranker")
+        ranker._client = mock_sie_client
+
+        ranked = ranker.run(query="query", documents=documents)["documents"]
+
+        assert [d.content for d in ranked] == ["doc-3", "doc-1", "doc-4", "doc-0", "doc-2"]
+        assert ranked[0].meta["score"] == 0.9
+        assert [d.meta["score"] for d in ranked] == [0.9, 0.7, 0.5, 0.3, 0.1]
+
+    def test_run_skips_malformed_item_id(self, mock_sie_client: object) -> None:
+        """Malformed item_ids are skipped (no crash, no misassignment)."""
+        documents = [Document(content=f"doc-{i}") for i in range(3)]
+        mock_sie_client.score = MagicMock(return_value=_MALFORMED_ENVELOPE)
+        ranker = SIERanker(model="test-reranker")
+        ranker._client = mock_sie_client
+
+        ranked = ranker.run(query="query", documents=documents)["documents"]
+
+        assert len(ranked) == 3
+        by_content = {d.content: d.meta["score"] for d in ranked}
+        assert by_content == {"doc-1": 0.8, "doc-0": 0.0, "doc-2": 0.0}
+        assert ranked[0].content == "doc-1"
 
     def test_run_empty_list(self, mock_sie_client: object) -> None:
         """Test that run handles empty document list."""