embspec 0.1.0__tar.gz

embspec-0.1.0/LICENSE

@@ -0,0 +1,201 @@
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for describing the origin of the Work and
+      reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Support. While redistributing the Work or
+      Derivative Works thereof, You may choose to offer, and charge a
+      fee for, acceptance of support, warranty, indemnity, or other
+      liability obligations and/or rights consistent with this License.
+      However, in accepting such obligations, You may act only on Your
+      own behalf and on Your sole responsibility, not on behalf of any
+      other Contributor, and only if You agree to indemnify, defend,
+      and hold each Contributor harmless for any liability incurred by,
+      or claims asserted against, such Contributor by reason of your
+      accepting any such warranty or support.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright 2026 Mukunda Katta
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+   implied. See the License for the specific language governing
+   permissions and limitations under the License.

embspec-0.1.0/PKG-INFO

@@ -0,0 +1,167 @@
+Metadata-Version: 2.4
+Name: embspec
+Version: 0.1.0
+Summary: Embedding pipeline ops + drift detection for production RAG: index manifests, version assertions, neighbor-stability eval, Drift-Adapter for in-place model migrations.
+Keywords: rag,embeddings,vector,drift,retrieval,production
+Author: Mukunda Katta
+Author-email: Mukunda Katta <mukunda.vjcs6@gmail.com>
+License-Expression: Apache-2.0
+License-File: LICENSE
+Classifier: Development Status :: 4 - Beta
+Classifier: Intended Audience :: Developers
+Classifier: Operating System :: OS Independent
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3 :: Only
+Classifier: Programming Language :: Python :: 3.10
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Programming Language :: Python :: 3.13
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Classifier: Topic :: Software Development :: Libraries :: Python Modules
+Classifier: Typing :: Typed
+Requires-Dist: numpy>=1.24
+Requires-Python: >=3.10
+Project-URL: Homepage, https://github.com/MukundaKatta/embspec
+Project-URL: Source, https://github.com/MukundaKatta/embspec
+Project-URL: Issues, https://github.com/MukundaKatta/embspec/issues
+Project-URL: Changelog, https://github.com/MukundaKatta/embspec/blob/main/CHANGELOG.md
+Description-Content-Type: text/markdown
+
+# embspec
+
+Embedding pipeline ops + drift detection for production RAG.
+
+The single failure mode this library prevents: query encoder upgrade ships before the index is re-encoded; every health check stays 200 OK while retrieval accuracy silently collapses. The [decompressed.io RAG observability post-mortem (2026-03-09)](https://decompressed.io/learn/rag-observability-postmortem) describes this exact bug — \$15K of emergency re-encoding plus 2-5 days of engineer time before someone diagnosed it.
+
+```bash
+pip install embspec
+```
+
+```python
+from embspec import IndexManifest, embed_assert
+
+manifest = IndexManifest.load("s3://my-rag/index-prod/manifest.json")
+
+@embed_assert(manifest, model_id="amazon.titan-embed-text-v2:0", dimension=1024)
+def search(query: str):
+    qv = embed_query(query)
+    return opensearch.knn_search(qv, ...)
+```
+
+If the query encoder ever drifts off the manifest, `search` raises `EmbeddingVersionMismatch` instead of silently returning bad results.
+
+## What's in v0.1
+
+| Primitive | What it does | Anchored in |
+|---|---|---|
+| `IndexManifest` + `EmbeddingSpec` | Single-file source of truth for "what does this index contain" — model id, dimension, version, normalization | [decompressed.io post-mortem](https://decompressed.io/learn/rag-observability-postmortem) |
+| `assert_compatible()` / `embed_assert()` decorator | Fast-fail on encoder/index drift; raise or warn modes | same |
+| `DriftAdapter` | Linear adapter from new-model embeddings to old-model space; lets you swap the query encoder without re-encoding the corpus | [Drift-Adapter, Vejendla 2025 (arxiv:2509.23471)](https://arxiv.org/abs/2509.23471) |
+| `neighbor_stability()` | Compare two retrievers on a frozen probe set; reports overlap, Jaccard, regression list, deploy-safety verdict | [RAGOps survey, Xu et al. 2025 (arxiv:2506.03401)](https://arxiv.org/abs/2506.03401) |
+
+## Why not Evidently / Phoenix / WhyLogs?
+
+- **Evidently** — tabular-ML drift heritage; LLM additions are recent and platform-shaped. Not a drop-in primitive.
+- **Phoenix** — embedding-drift visualization is a sub-feature of a full observability platform. You adopt the platform.
+- **WhyLogs** — generic data-logging primitive; not embedding-aware; last commit 2025-01.
+- **embspec** — three small primitives (`IndexManifest`, `DriftAdapter`, `neighbor_stability`) you compose with whatever vector DB and tracer you already have. No platform, no UI, no agent framework.
+
+## Usage
+
+### Manifest + version assertion
+
+```python
+from datetime import datetime, timezone
+from embspec import IndexManifest, EmbeddingSpec
+
+# When you build the index, write a manifest alongside it
+manifest = IndexManifest(
+    index_name="prod-v3",
+    embedding=EmbeddingSpec(
+        model_id="amazon.titan-embed-text-v2:0",
+        dimension=1024,
+        normalization="l2",
+    ),
+    created_at=datetime.now(timezone.utc),
+    doc_count=8_000_000,
+)
+manifest.save("s3://my-rag/index-prod/manifest.json")
+# (or any local path; manifest.save uses pathlib.Path.write_text)
+
+# At query time, assert the encoder matches before searching
+from embspec import embed_assert
+
+@embed_assert(
+    "s3://my-rag/index-prod/manifest.json",  # path or IndexManifest
+    model_id="amazon.titan-embed-text-v2:0",
+    dimension=1024,
+    mode="raise",  # or "log" for canary rollout
+)
+def search(query: str) -> list[dict]:
+    qv = embed_query(query)
+    return opensearch.knn_search(qv, ...)
+```
+
+### Drift-Adapter for in-place model migration
+
+When you want to upgrade the query encoder without re-encoding 8M docs:
+
+```python
+from embspec import DriftAdapter
+import numpy as np
+
+# Sample, e.g., 50K docs and embed them with both old and new models
+old_emb = embed_with_old_model(sample_texts)  # shape (50000, 1024)
+new_emb = embed_with_new_model(sample_texts)  # shape (50000, 1536)
+
+adapter = DriftAdapter.fit(
+    new_embeddings=new_emb,
+    old_embeddings=old_emb,
+    regularization=0.01,  # ridge; helps when new_emb is rank-deficient
+)
+adapter.save("s3://my-rag/adapters/v3-to-v4.npz")
+
+# At query time, embed with the new model then transform into old space
+qv_new = embed_with_new_model(query)
+qv_compatible = adapter.transform(qv_new)  # shape (1024,)
+results = opensearch.knn_search(qv_compatible, ...)
+```
+
+Per Vejendla 2025, this typically recovers 95-99% retrieval at ~1% of the cost of re-encoding the full corpus.
+
+### Neighbor stability for safe migrations
+
+```python
+from embspec import neighbor_stability
+
+# Run a fixed probe set against both indexes
+old_results = {pid: retrieve_from_v3(q) for pid, q in probes.items()}
+new_results = {pid: retrieve_from_v4(q) for pid, q in probes.items()}
+
+report = neighbor_stability(old_results, new_results, k=10)
+print(f"mean overlap@10: {report.mean_overlap_at_k:.1%}")
+print(f"regressions: {report.regression_count}/{report.n_probes}")
+
+if report.is_safe_to_deploy(min_mean_overlap=0.85, max_regression_fraction=0.05):
+    deploy_v4()
+else:
+    investigate(report.regression_probe_ids)
+```
+
+## What it explicitly does NOT do
+
+- Not a vector database.
+- Not a RAG framework. No retriever, no chunker, no generator.
+- Not a generic ML drift library. Embedding-and-retrieval-shaped only.
+- Not an eval framework. `neighbor_stability` is the one judgment you can make without a labeled gold set; for richer evals use `ragas`, `trulens`, or a tracer.
+- Does not host or serve embeddings.
+
+## Roadmap
+
+- v0.2: `dual_write()` context manager for blue/green index migrations across OpenSearch / pgvector / Pinecone / Qdrant.
+- v0.3: `ChunkingExperiment` A/B harness with optional LLM-judge.
+- v0.4: integration helpers for AWS Bedrock embedding models, OpenAI, Cohere, Voyage.
+
+## License
+
+Apache-2.0. See [LICENSE](./LICENSE).

embspec-0.1.0/README.md

@@ -0,0 +1,138 @@
+# embspec
+
+Embedding pipeline ops + drift detection for production RAG.
+
+The single failure mode this library prevents: query encoder upgrade ships before the index is re-encoded; every health check stays 200 OK while retrieval accuracy silently collapses. The [decompressed.io RAG observability post-mortem (2026-03-09)](https://decompressed.io/learn/rag-observability-postmortem) describes this exact bug — \$15K of emergency re-encoding plus 2-5 days of engineer time before someone diagnosed it.
+
+```bash
+pip install embspec
+```
+
+```python
+from embspec import IndexManifest, embed_assert
+
+manifest = IndexManifest.load("s3://my-rag/index-prod/manifest.json")
+
+@embed_assert(manifest, model_id="amazon.titan-embed-text-v2:0", dimension=1024)
+def search(query: str):
+    qv = embed_query(query)
+    return opensearch.knn_search(qv, ...)
+```
+
+If the query encoder ever drifts off the manifest, `search` raises `EmbeddingVersionMismatch` instead of silently returning bad results.
+
+## What's in v0.1
+
+| Primitive | What it does | Anchored in |
+|---|---|---|
+| `IndexManifest` + `EmbeddingSpec` | Single-file source of truth for "what does this index contain" — model id, dimension, version, normalization | [decompressed.io post-mortem](https://decompressed.io/learn/rag-observability-postmortem) |
+| `assert_compatible()` / `embed_assert()` decorator | Fast-fail on encoder/index drift; raise or warn modes | same |
+| `DriftAdapter` | Linear adapter from new-model embeddings to old-model space; lets you swap the query encoder without re-encoding the corpus | [Drift-Adapter, Vejendla 2025 (arxiv:2509.23471)](https://arxiv.org/abs/2509.23471) |
+| `neighbor_stability()` | Compare two retrievers on a frozen probe set; reports overlap, Jaccard, regression list, deploy-safety verdict | [RAGOps survey, Xu et al. 2025 (arxiv:2506.03401)](https://arxiv.org/abs/2506.03401) |
+
+## Why not Evidently / Phoenix / WhyLogs?
+
+- **Evidently** — tabular-ML drift heritage; LLM additions are recent and platform-shaped. Not a drop-in primitive.
+- **Phoenix** — embedding-drift visualization is a sub-feature of a full observability platform. You adopt the platform.
+- **WhyLogs** — generic data-logging primitive; not embedding-aware; last commit 2025-01.
+- **embspec** — three small primitives (`IndexManifest`, `DriftAdapter`, `neighbor_stability`) you compose with whatever vector DB and tracer you already have. No platform, no UI, no agent framework.
+
+## Usage
+
+### Manifest + version assertion
+
+```python
+from datetime import datetime, timezone
+from embspec import IndexManifest, EmbeddingSpec
+
+# When you build the index, write a manifest alongside it
+manifest = IndexManifest(
+    index_name="prod-v3",
+    embedding=EmbeddingSpec(
+        model_id="amazon.titan-embed-text-v2:0",
+        dimension=1024,
+        normalization="l2",
+    ),
+    created_at=datetime.now(timezone.utc),
+    doc_count=8_000_000,
+)
+manifest.save("s3://my-rag/index-prod/manifest.json")
+# (or any local path; manifest.save uses pathlib.Path.write_text)
+
+# At query time, assert the encoder matches before searching
+from embspec import embed_assert
+
+@embed_assert(
+    "s3://my-rag/index-prod/manifest.json",  # path or IndexManifest
+    model_id="amazon.titan-embed-text-v2:0",
+    dimension=1024,
+    mode="raise",  # or "log" for canary rollout
+)
+def search(query: str) -> list[dict]:
+    qv = embed_query(query)
+    return opensearch.knn_search(qv, ...)
+```
+
+### Drift-Adapter for in-place model migration
+
+When you want to upgrade the query encoder without re-encoding 8M docs:
+
+```python
+from embspec import DriftAdapter
+import numpy as np
+
+# Sample, e.g., 50K docs and embed them with both old and new models
+old_emb = embed_with_old_model(sample_texts)  # shape (50000, 1024)
+new_emb = embed_with_new_model(sample_texts)  # shape (50000, 1536)
+
+adapter = DriftAdapter.fit(
+    new_embeddings=new_emb,
+    old_embeddings=old_emb,
+    regularization=0.01,  # ridge; helps when new_emb is rank-deficient
+)
+adapter.save("s3://my-rag/adapters/v3-to-v4.npz")
+
+# At query time, embed with the new model then transform into old space
+qv_new = embed_with_new_model(query)
+qv_compatible = adapter.transform(qv_new)  # shape (1024,)
+results = opensearch.knn_search(qv_compatible, ...)
+```
+
+Per Vejendla 2025, this typically recovers 95-99% retrieval at ~1% of the cost of re-encoding the full corpus.
+
+### Neighbor stability for safe migrations
+
+```python
+from embspec import neighbor_stability
+
+# Run a fixed probe set against both indexes
+old_results = {pid: retrieve_from_v3(q) for pid, q in probes.items()}
+new_results = {pid: retrieve_from_v4(q) for pid, q in probes.items()}
+
+report = neighbor_stability(old_results, new_results, k=10)
+print(f"mean overlap@10: {report.mean_overlap_at_k:.1%}")
+print(f"regressions: {report.regression_count}/{report.n_probes}")
+
+if report.is_safe_to_deploy(min_mean_overlap=0.85, max_regression_fraction=0.05):
+    deploy_v4()
+else:
+    investigate(report.regression_probe_ids)
+```
+
+## What it explicitly does NOT do
+
+- Not a vector database.
+- Not a RAG framework. No retriever, no chunker, no generator.
+- Not a generic ML drift library. Embedding-and-retrieval-shaped only.
+- Not an eval framework. `neighbor_stability` is the one judgment you can make without a labeled gold set; for richer evals use `ragas`, `trulens`, or a tracer.
+- Does not host or serve embeddings.
+
+## Roadmap
+
+- v0.2: `dual_write()` context manager for blue/green index migrations across OpenSearch / pgvector / Pinecone / Qdrant.
+- v0.3: `ChunkingExperiment` A/B harness with optional LLM-judge.
+- v0.4: integration helpers for AWS Bedrock embedding models, OpenAI, Cohere, Voyage.
+
+## License
+
+Apache-2.0. See [LICENSE](./LICENSE).

embspec-0.1.0/pyproject.toml

@@ -0,0 +1,49 @@
+[project]
+name = "embspec"
+version = "0.1.0"
+description = "Embedding pipeline ops + drift detection for production RAG: index manifests, version assertions, neighbor-stability eval, Drift-Adapter for in-place model migrations."
+readme = "README.md"
+authors = [
+    { name = "Mukunda Katta", email = "mukunda.vjcs6@gmail.com" }
+]
+license = "Apache-2.0"
+license-files = ["LICENSE"]
+requires-python = ">=3.10"
+keywords = ["rag", "embeddings", "vector", "drift", "retrieval", "production"]
+classifiers = [
+    "Development Status :: 4 - Beta",
+    "Intended Audience :: Developers",
+    "Operating System :: OS Independent",
+    "Programming Language :: Python :: 3",
+    "Programming Language :: Python :: 3 :: Only",
+    "Programming Language :: Python :: 3.10",
+    "Programming Language :: Python :: 3.11",
+    "Programming Language :: Python :: 3.12",
+    "Programming Language :: Python :: 3.13",
+    "Topic :: Scientific/Engineering :: Artificial Intelligence",
+    "Topic :: Software Development :: Libraries :: Python Modules",
+    "Typing :: Typed",
+]
+dependencies = [
+    "numpy>=1.24",
+]
+
+[project.urls]
+Homepage = "https://github.com/MukundaKatta/embspec"
+Source = "https://github.com/MukundaKatta/embspec"
+Issues = "https://github.com/MukundaKatta/embspec/issues"
+Changelog = "https://github.com/MukundaKatta/embspec/blob/main/CHANGELOG.md"
+
+[dependency-groups]
+dev = [
+    "pytest>=8.0",
+    "pytest-cov>=5.0",
+]
+
+[build-system]
+requires = ["uv_build>=0.11.7,<0.12.0"]
+build-backend = "uv_build"
+
+[tool.pytest.ini_options]
+testpaths = ["tests"]
+addopts = "-q"

embspec-0.1.0/src/embspec/__init__.py

@@ -0,0 +1,64 @@
+"""embspec — embedding pipeline ops + drift detection for production RAG.
+
+The single primitive this library asserts:
+
+    "The query encoder must use the same embedding model+version as the
+    index it is searching."
+
+The decompressed.io RAG observability post-mortem (2026-03-09) describes
+exactly the failure this library prevents: the query embedding model gets
+upgraded, the index still holds vectors from the old model, every health
+check stays 200 OK while retrieval accuracy silently collapses. Reporter
+spent $15k on emergency re-encoding and 2-5 days of engineer time before
+diagnosis.
+
+Quick start::
+
+    from embspec import IndexManifest, EmbeddingSpec, embed_assert
+
+    # Recorded once when the index was built
+    manifest = IndexManifest.load("s3://my-rag/index-prod/manifest.json")
+
+    @embed_assert(manifest, model_id="amazon.titan-embed-text-v2:0", dimension=1024)
+    def search(query: str):
+        qv = embed_query(query)
+        return opensearch.knn_search(qv, ...)
+
+If the query encoder ever drifts off the manifest, ``search`` raises
+:class:`EmbeddingVersionMismatch` instead of returning silently-degraded
+results.
+"""
+
+from __future__ import annotations
+
+from ._adapter import DriftAdapter
+from ._assert import embed_assert
+from ._errors import (
+    AdapterShapeError,
+    EmbeddingVersionMismatch,
+    EmbspecError,
+    ManifestFormatError,
+)
+from ._manifest import (
+    EmbeddingSpec,
+    IndexManifest,
+    assert_compatible,
+)
+from ._stability import StabilityReport, neighbor_stability
+
+__version__ = "0.1.0"
+
+__all__ = [
+    "AdapterShapeError",
+    "DriftAdapter",
+    "EmbeddingSpec",
+    "EmbeddingVersionMismatch",
+    "EmbspecError",
+    "IndexManifest",
+    "ManifestFormatError",
+    "StabilityReport",
+    "__version__",
+    "assert_compatible",
+    "embed_assert",
+    "neighbor_stability",
+]

embspec-0.1.0/src/embspec/_adapter.py

@@ -0,0 +1,116 @@
+"""Drift-Adapter: linear map from new-model embeddings into old-model embedding space.
+
+Implements the pattern from Vejendla 2025 (arxiv:2509.23471, "Drift-Adapter:
+Closing the Embedding-Drift Gap in Production Vector Stores"). Lets you swap
+the query encoder to a newer embedding model without re-encoding the
+corpus, by fitting a small linear transform on a sample of paired
+(old_model, new_model) embeddings.
+
+Trade-off: the adapter is a least-squares fit so it loses some signal vs.
+true re-encoding. The paper reports 95-99% retrieval recovery with adapter
+sizes ~1% of the corpus. Use it as a cost-saving migration path, then
+re-encode the corpus on a slower schedule (or never if recall holds).
+"""
+
+from __future__ import annotations
+
+from pathlib import Path
+from typing import TYPE_CHECKING
+
+from ._errors import AdapterShapeError
+
+if TYPE_CHECKING:
+    import numpy as np
+    from numpy.typing import NDArray
+
+
+class DriftAdapter:
+    """Linear adapter that maps embeddings from one model into another's space.
+
+    Fitting solves ``W = argmin || X_new @ W - X_old ||_F^2`` where
+    ``X_old`` and ``X_new`` are aligned matrices of paired embeddings.
+    """
+
+    def __init__(self, weight: NDArray) -> None:
+        if weight.ndim != 2:
+            raise AdapterShapeError(f"weight must be 2-D, got shape {weight.shape}")
+        self.weight = weight
+
+    @property
+    def input_dim(self) -> int:
+        return int(self.weight.shape[0])
+
+    @property
+    def output_dim(self) -> int:
+        return int(self.weight.shape[1])
+
+    @classmethod
+    def fit(
+        cls,
+        new_embeddings: NDArray,
+        old_embeddings: NDArray,
+        *,
+        regularization: float = 0.0,
+    ) -> DriftAdapter:
+        """Fit the adapter on aligned (new, old) embedding pairs via least squares.
+
+        ``new_embeddings`` and ``old_embeddings`` must have the same number
+        of rows (``n``) and may have different column counts (the new and
+        old model dimensions). ``regularization`` adds an L2 penalty on the
+        weight matrix (ridge regression); useful for ill-conditioned data.
+        """
+        import numpy as np  # noqa: PLC0415
+
+        if new_embeddings.ndim != 2 or old_embeddings.ndim != 2:
+            raise AdapterShapeError(
+                f"embedding arrays must be 2-D, got {new_embeddings.shape} and {old_embeddings.shape}"
+            )
+        if new_embeddings.shape[0] != old_embeddings.shape[0]:
+            raise AdapterShapeError(
+                f"row counts differ: new={new_embeddings.shape[0]}, old={old_embeddings.shape[0]}; "
+                "embeddings must be paired"
+            )
+        if new_embeddings.shape[0] < new_embeddings.shape[1]:
+            raise AdapterShapeError(
+                f"need at least as many paired samples as new-model dimensions; "
+                f"got {new_embeddings.shape[0]} samples for {new_embeddings.shape[1]} dimensions"
+            )
+
+        if regularization > 0:
+            n_features = new_embeddings.shape[1]
+            xtx = new_embeddings.T @ new_embeddings + regularization * np.eye(n_features)
+            xty = new_embeddings.T @ old_embeddings
+            weight = np.linalg.solve(xtx, xty)
+        else:
+            weight, _, _, _ = np.linalg.lstsq(
+                new_embeddings, old_embeddings, rcond=None
+            )
+        return cls(weight=weight)
+
+    def transform(self, new_embeddings: NDArray) -> NDArray:
+        """Map new-model embeddings into the old-model embedding space."""
+        if new_embeddings.ndim == 1:
+            if new_embeddings.shape[0] != self.input_dim:
+                raise AdapterShapeError(
+                    f"expected vector of dim {self.input_dim}, got {new_embeddings.shape[0]}"
+                )
+            return new_embeddings @ self.weight
+        if new_embeddings.ndim != 2 or new_embeddings.shape[1] != self.input_dim:
+            raise AdapterShapeError(
+                f"expected (n, {self.input_dim}) input, got {new_embeddings.shape}"
+            )
+        return new_embeddings @ self.weight
+
+    def save(self, path: str | Path) -> None:
+        """Save the adapter to a compressed ``.npz`` file."""
+        import numpy as np  # noqa: PLC0415
+
+        np.savez_compressed(str(path), weight=self.weight)
+
+    @classmethod
+    def load(cls, path: str | Path) -> DriftAdapter:
+        """Load an adapter previously saved with :meth:`save`."""
+        import numpy as np  # noqa: PLC0415
+
+        data = np.load(str(path))
+        return cls(weight=data["weight"])

embspec-0.1.0/src/embspec/_assert.py

@@ -0,0 +1,73 @@
+"""Decorator for asserting embedding-spec compatibility on every retrieval call."""
+
+from __future__ import annotations
+
+import functools
+from collections.abc import Callable
+from pathlib import Path
+from typing import Any, Literal, TypeVar
+
+from ._manifest import EmbeddingSpec, IndexManifest
+
+
+F = TypeVar("F", bound=Callable[..., Any])
+
+
+def embed_assert(
+    manifest: IndexManifest | str | Path,
+    *,
+    model_id: str,
+    dimension: int,
+    model_version: str | None = None,
+    normalization: Literal["l2", "none"] = "l2",
+    mode: Literal["raise", "log"] = "raise",
+) -> Callable[[F], F]:
+    """Assert the function uses an embedding spec compatible with the index manifest.
+
+    ``manifest`` may be an :class:`IndexManifest` or a path the manifest is
+    loaded from (the path is resolved on first call so the decorated symbol
+    stays import-cheap).
+
+    ``mode="raise"`` (default) raises :class:`EmbeddingVersionMismatch` on
+    drift. ``mode="log"`` is the safer rollout mode: it never raises but
+    logs via :mod:`warnings` so production stays up while you wire alerts.
+    """
+    spec = EmbeddingSpec(
+        model_id=model_id,
+        dimension=dimension,
+        model_version=model_version,
+        normalization=normalization,
+    )
+
+    state: dict[str, Any] = {"resolved": None}
+
+    def _resolve() -> tuple[IndexManifest, str | None]:
+        if state["resolved"] is None:
+            if isinstance(manifest, IndexManifest):
+                state["resolved"] = (manifest, None)
+            else:
+                path = str(manifest)
+                state["resolved"] = (IndexManifest.load(path), path)
+        return state["resolved"]
+
+    def decorator(fn: F) -> F:
+        @functools.wraps(fn)
+        def wrapper(*args: Any, **kwargs: Any) -> Any:
+            m, path = _resolve()
+            try:
+                m.assert_compatible(spec, manifest_path=path)
+            except Exception:
+                if mode == "log":
+                    import warnings
+
+                    warnings.warn(
+                        f"embspec drift on call to {fn.__qualname__}",
+                        stacklevel=2,
+                    )
+                else:
+                    raise
+            return fn(*args, **kwargs)
+
+        return wrapper  # type: ignore[return-value]
+
+    return decorator

embspec-0.1.0/src/embspec/_errors.py

@@ -0,0 +1,48 @@
+"""Typed errors for embspec."""
+
+from __future__ import annotations
+
+
+class EmbspecError(Exception):
+    """Base class for all embspec errors."""
+
+
+class EmbeddingVersionMismatch(EmbspecError):
+    """The query encoder's embedding spec does not match the index manifest.
+
+    Raised by :func:`embspec.assert_compatible` and the
+    :func:`embspec.embed_assert` decorator. Catching this exception is the
+    fast-fail signal that prevents the silent-accuracy-collapse failure
+    mode described in the decompressed.io RAG observability post-mortem
+    (2026-03-09): the system stays 200 OK with normal latency while
+    retrieval quality silently tanks.
+    """
+
+    def __init__(
+        self,
+        *,
+        index_name: str,
+        manifest_field: str,
+        manifest_value: object,
+        query_value: object,
+        manifest_path: str | None = None,
+    ) -> None:
+        suffix = f" (manifest at {manifest_path})" if manifest_path else ""
+        super().__init__(
+            f"Index {index_name!r} manifest declares {manifest_field}="
+            f"{manifest_value!r} but query encoder uses {query_value!r}{suffix}. "
+            f"Re-encode the corpus or roll the query encoder back."
+        )
+        self.index_name = index_name
+        self.manifest_field = manifest_field
+        self.manifest_value = manifest_value
+        self.query_value = query_value
+        self.manifest_path = manifest_path
+
+
+class ManifestFormatError(EmbspecError):
+    """The manifest file is missing required fields or has an unknown format version."""
+
+
+class AdapterShapeError(EmbspecError):
+    """Embedding tensors have incompatible shapes for fitting or transforming."""

embspec-0.1.0/src/embspec/_manifest.py

@@ -0,0 +1,146 @@
+"""Index manifests track what embedding model + version a vector index was built with.
+
+The manifest is the single source of truth for "what does this index contain";
+asserting against it at every search prevents the silent failure mode where
+a query encoder upgrade ships before the index is re-encoded and accuracy
+collapses while every health check stays green.
+"""
+
+from __future__ import annotations
+
+import json
+from dataclasses import asdict, dataclass, field
+from datetime import datetime, timezone
+from pathlib import Path
+from typing import Any, Literal
+
+from ._errors import EmbeddingVersionMismatch, ManifestFormatError
+
+
+_FORMAT_VERSION: int = 1
+
+
+@dataclass(frozen=True)
+class EmbeddingSpec:
+    """The embedding configuration used to produce a set of vectors."""
+
+    model_id: str
+    dimension: int
+    model_version: str | None = None
+    normalization: Literal["l2", "none"] = "l2"
+
+
+@dataclass(frozen=True)
+class IndexManifest:
+    """Manifest describing the embedding configuration of a vector index."""
+
+    index_name: str
+    embedding: EmbeddingSpec
+    created_at: datetime
+    doc_count: int | None = None
+    extra: dict[str, Any] = field(default_factory=dict)
+
+    def to_dict(self) -> dict[str, Any]:
+        return {
+            "embspec_format_version": _FORMAT_VERSION,
+            "index_name": self.index_name,
+            "embedding": asdict(self.embedding),
+            "created_at": self.created_at.astimezone(timezone.utc).isoformat(),
+            "doc_count": self.doc_count,
+            "extra": self.extra,
+        }
+
+    def save(self, path: str | Path) -> None:
+        Path(path).write_text(json.dumps(self.to_dict(), indent=2, sort_keys=True))
+
+    @classmethod
+    def load(cls, path: str | Path) -> IndexManifest:
+        return cls.from_dict(json.loads(Path(path).read_text()), source=str(path))
+
+    @classmethod
+    def from_dict(cls, data: dict[str, Any], *, source: str | None = None) -> IndexManifest:
+        version = data.get("embspec_format_version")
+        if version != _FORMAT_VERSION:
+            raise ManifestFormatError(
+                f"Unknown embspec_format_version={version!r} in manifest"
+                + (f" at {source}" if source else "")
+            )
+        if "embedding" not in data or "index_name" not in data:
+            raise ManifestFormatError(
+                f"Manifest missing required fields"
+                + (f" at {source}" if source else "")
+            )
+        emb = data["embedding"]
+        embedding = EmbeddingSpec(
+            model_id=emb["model_id"],
+            dimension=int(emb["dimension"]),
+            model_version=emb.get("model_version"),
+            normalization=emb.get("normalization", "l2"),
+        )
+        created_at = (
+            datetime.fromisoformat(data["created_at"])
+            if "created_at" in data
+            else datetime.now(timezone.utc)
+        )
+        return cls(
+            index_name=data["index_name"],
+            embedding=embedding,
+            created_at=created_at,
+            doc_count=data.get("doc_count"),
+            extra=data.get("extra") or {},
+        )
+
+    def assert_compatible(
+        self,
+        spec: EmbeddingSpec,
+        *,
+        manifest_path: str | None = None,
+    ) -> None:
+        """Raise :class:`EmbeddingVersionMismatch` if ``spec`` is not the same as this manifest's embedding.
+
+        Compatibility is exact-match across every field of :class:`EmbeddingSpec`
+        — query and index must use the same model, dimension, version, and
+        normalization. Any drift on any field is a failure mode.
+        """
+        if spec.model_id != self.embedding.model_id:
+            raise EmbeddingVersionMismatch(
+                index_name=self.index_name,
+                manifest_field="embedding.model_id",
+                manifest_value=self.embedding.model_id,
+                query_value=spec.model_id,
+                manifest_path=manifest_path,
+            )
+        if spec.dimension != self.embedding.dimension:
+            raise EmbeddingVersionMismatch(
+                index_name=self.index_name,
+                manifest_field="embedding.dimension",
+                manifest_value=self.embedding.dimension,
+                query_value=spec.dimension,
+                manifest_path=manifest_path,
+            )
+        if spec.model_version != self.embedding.model_version:
+            raise EmbeddingVersionMismatch(
+                index_name=self.index_name,
+                manifest_field="embedding.model_version",
+                manifest_value=self.embedding.model_version,
+                query_value=spec.model_version,
+                manifest_path=manifest_path,
+            )
+        if spec.normalization != self.embedding.normalization:
+            raise EmbeddingVersionMismatch(
+                index_name=self.index_name,
+                manifest_field="embedding.normalization",
+                manifest_value=self.embedding.normalization,
+                query_value=spec.normalization,
+                manifest_path=manifest_path,
+            )
+
+
+def assert_compatible(
+    manifest: IndexManifest,
+    spec: EmbeddingSpec,
+    *,
+    manifest_path: str | None = None,
+) -> None:
+    """Functional alias for :meth:`IndexManifest.assert_compatible`."""
+    manifest.assert_compatible(spec, manifest_path=manifest_path)

embspec-0.1.0/src/embspec/_stability.py

@@ -0,0 +1,103 @@
+"""Neighbor stability: compare two retrievers on a fixed probe set.
+
+This is the missing primitive RAGOps (Xu et al. 2025, arxiv:2506.03401)
+calls out: "existing work provides limited support for observability in
+the retrieval process of RAG applications." A pre/post snapshot of which
+documents come back for a frozen set of probe queries lets you decide
+whether an embedding model change, chunker change, or rerank change is
+safe to deploy.
+
+The function is pure: caller runs both retrievers, passes the result
+dictionaries in. No vector-DB-specific code lives here.
+"""
+
+from __future__ import annotations
+
+from dataclasses import dataclass
+
+
+@dataclass(frozen=True)
+class StabilityReport:
+    """Per-set stability metrics over a probe set."""
+
+    n_probes: int
+    k: int
+    mean_overlap_at_k: float
+    """Mean ``|new ∩ old| / k`` across probes. 1.0 = identical top-k. 0.0 = disjoint."""
+
+    mean_jaccard_at_k: float
+    """Mean Jaccard similarity ``|new ∩ old| / |new ∪ old|`` across probes."""
+
+    regression_probe_ids: tuple[str, ...]
+    """Probe ids whose overlap fell below ``regression_threshold``."""
+
+    @property
+    def regression_count(self) -> int:
+        return len(self.regression_probe_ids)
+
+    def is_safe_to_deploy(
+        self,
+        *,
+        min_mean_overlap: float = 0.85,
+        max_regression_fraction: float = 0.05,
+    ) -> bool:
+        """Heuristic deploy gate.
+
+        Returns True when both: mean overlap is at least ``min_mean_overlap``
+        AND the fraction of regressed probes is at most ``max_regression_fraction``.
+        """
+        if self.n_probes == 0:
+            return False
+        regression_fraction = self.regression_count / self.n_probes
+        return (
+            self.mean_overlap_at_k >= min_mean_overlap
+            and regression_fraction <= max_regression_fraction
+        )
+
+
+def neighbor_stability(
+    old_results: dict[str, list[str]],
+    new_results: dict[str, list[str]],
+    *,
+    k: int = 10,
+    regression_threshold: float = 0.5,
+) -> StabilityReport:
+    """Compute :class:`StabilityReport` from two retrieval result sets.
+
+    Both arguments map ``probe_id -> list of doc_id`` (top-k). Probe ids
+    must agree across the two dicts; any extra ids in either are ignored.
+    The first ``k`` doc ids of each list are compared.
+    """
+    common = set(old_results) & set(new_results)
+    if not common:
+        return StabilityReport(
+            n_probes=0,
+            k=k,
+            mean_overlap_at_k=0.0,
+            mean_jaccard_at_k=0.0,
+            regression_probe_ids=(),
+        )
+
+    overlap_sum = 0.0
+    jaccard_sum = 0.0
+    regressions: list[str] = []
+    for probe_id in sorted(common):
+        old_topk = set(old_results[probe_id][:k])
+        new_topk = set(new_results[probe_id][:k])
+        intersect = len(old_topk & new_topk)
+        union = len(old_topk | new_topk)
+        overlap = intersect / k if k > 0 else 0.0
+        jaccard = intersect / union if union > 0 else 0.0
+        overlap_sum += overlap
+        jaccard_sum += jaccard
+        if overlap < regression_threshold:
+            regressions.append(probe_id)
+
+    n = len(common)
+    return StabilityReport(
+        n_probes=n,
+        k=k,
+        mean_overlap_at_k=overlap_sum / n,
+        mean_jaccard_at_k=jaccard_sum / n,
+        regression_probe_ids=tuple(regressions),
+    )