hev-rerank 0.1.0__tar.gz

hev_rerank-0.1.0/.gitignore

@@ -0,0 +1,5 @@
+.venv/
+__pycache__/
+dist/
+*.egg-info/
+.claude/

hev_rerank-0.1.0/LICENSE

@@ -0,0 +1,202 @@
+
+                                 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 reasonable and customary use in 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 Additional Liability. 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 additional liability.
+
+   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 [yyyy] [name of copyright owner]
+
+   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.

hev_rerank-0.1.0/PKG-INFO

@@ -0,0 +1,122 @@
+Metadata-Version: 2.5
+Name: hev-rerank
+Version: 0.1.0
+Summary: Use Jev (TypeSafe's System One model) as a calibrated reranker: one call, up to 30 documents, a probability per document.
+Project-URL: Repository, https://github.com/hev/jev-rerank
+Project-URL: Results, https://github.com/hev/jev-rerank/blob/main/RESULTS.md
+Author: Adam Hevenor
+License-Expression: Apache-2.0
+License-File: LICENSE
+Keywords: jev,rag,reranker,retrieval,search,typesafe
+Classifier: Development Status :: 3 - Alpha
+Classifier: Intended Audience :: Developers
+Classifier: License :: OSI Approved :: Apache Software License
+Classifier: Programming Language :: Python :: 3
+Classifier: Topic :: Text Processing :: Indexing
+Requires-Python: >=3.10
+Requires-Dist: pyyaml>=6
+Requires-Dist: typesafe-sdk>=0.6
+Provides-Extra: examples
+Requires-Dist: rank-bm25; extra == 'examples'
+Description-Content-Type: text/markdown
+
+# hev-rerank
+
+Use [Jev](https://docs.typesafe.ai), TypeSafe's System One decision model, as a reranker.
+One call carries the query and up to 30 candidate documents; the model answers one
+question per document, in parallel: *is this document relevant to the query?* The answer is
+a calibrated probability, so the score is both a sort key (rerank) and an absolute
+threshold (prune).
+
+This repo is the whole recipe: the [prompt](hev_rerank/prompt.yaml), the request
+[state shape](schema/state.schema.json), a 90-line Python wrapper, and the
+[results](RESULTS.md) of running it against hosted and open-weight rerankers and LLM
+judges on BEIR shortlists.
+
+## Results
+
+nDCG@10, every system permuting the same BM25 top-30 shortlist per query. Jev uses the
+domain-neutral phrasing in `prompt.yaml`, untuned per corpus.
+
+| Reranker | SciFact (300 q) | NFCorpus (323 q) | FiQA (300 q) | p50 / p95 per query | Cost per 300 q |
+|---|---|---|---|---|---|
+| BM25 order, no rerank | 0.667 | 0.310 | 0.234 | | |
+| MiniLM-L6 cross-encoder (local) | 0.682 | 0.336 | 0.323 | | |
+| Cohere rerank-v3.5 | 0.745 | 0.340 | 0.374 | 129–277 ms / 267–698 ms | $0.60 |
+| Mixedbread mxbai-rerank-large-v2 (hosted) | 0.749 | 0.324 | 0.354 | 322–380 ms / 374–460 ms | $1.05 |
+| Voyage rerank-2.5 | 0.755 | 0.357 | 0.395 | 177–187 ms / 255–300 ms | ~$0.15 |
+| Voyage rerank-3 | 0.755 | 0.357 | 0.402 | 173–192 ms / 285–288 ms | ~$0.15 |
+| gpt-5.6-luna (reasoning off), listwise | 0.747 | 0.355 | 0.363 | 3.1–3.2 s / 3.8–5.8 s | ~$0.71 |
+| Claude Haiku 4.5, listwise | 0.723 | | | 2.9 s / 11.6 s | $4.18 |
+| Claude Opus 5 (low effort), listwise | 0.756 | | | 5.0 s / 7.5 s | $28.40 |
+| **Jev, 30 Nouls in one state** | **0.768** | **0.358** | 0.376 | 208–238 ms / 0.8–1.8 s | $0.13–0.19 |
+| Jev, one Noul per (query, doc) pair | 0.772 | 0.358 | 0.376 | 122–144 ms per call | $0.23–0.30 |
+
+Paired-bootstrap per query, Jev's one-state shape against each reranker: at or above Cohere on
+all three corpora (two wins, one tie); above Mixedbread (two wins, one interval touching zero);
+trading with Voyage, the strongest purpose-built reranker here (edge on SciFact, tie on NFCorpus,
+loss on FiQA by ~0.02); above every LLM used as a reranker on cost and latency, and on quality
+except Opus 5, which it ties at ~1/100 the cost.
+
+The honest one-liner: a general decision model with no reranker training lands in the same
+quality and price bracket as the best purpose-built rerankers, a third of Cohere's price, and
+returns a calibrated probability per document that none of them do. Voyage rerank-3 is as
+good and slightly cheaper; what Jev adds over it is the probability and the fact that the same
+call shape covers routing, classification and extraction gating.
+
+Costs are the runs' actual token usage at list price (Voyage measured from its billed
+`usage.total_tokens`). Full tables, confidence intervals, determinism, position-bias and
+calibration checks: [RESULTS.md](RESULTS.md).
+
+## Install
+
+```sh
+pip install hev-rerank        # or: uv add hev-rerank
+export TYPESAFE_API_KEY=...   # https://console.typesafe.ai/settings/keys
+```
+
+## Use
+
+```python
+from hev_rerank import rerank
+
+hits = rerank(
+    "does vitamin D supplementation improve bone density in older adults",
+    [{"title": t, "text": x} for t, x in first_stage_results],   # strings work too
+    top_n=10,          # optional
+    threshold=0.1,     # optional: drop candidates the model is confident are irrelevant
+)
+for h in hits:
+    print(h.score, h.index, h.document["title"])
+```
+
+`rerank` returns `RerankResult(index, score, document)` best-first. Lists longer than
+`max_docs_per_call` (30) are split into concurrent calls under Jev's ~32k-token request
+budget. Scores across calls are comparable because each is an absolute probability.
+
+## The shape, and why
+
+- **State**: `{"query": ..., "documents": {"D00": {...}, "D01": {...}, ...}}`
+  ([schema](schema/state.schema.json)).
+- **Questions**: one `Noul` per document id, from [`prompt.yaml`](hev_rerank/prompt.yaml).
+  All questions in a request are evaluated in parallel, so 30 documents cost about the same
+  wall time as 5.
+- **Score**: the Noul probability. Calibration on the SciFact run: documents scored ≥ 0.9
+  were judged relevant 76% of the time; documents scored < 0.1, 0.5% of the time.
+
+Why not one `Choice` over the documents? It is slightly better when exactly one document is
+relevant (SciFact: +0.01 nDCG@10, +3 pts P@1) and much worse when many are (NFCorpus,
+median 16 relevant per query: 0.316 vs 0.358). A reranker that does not know the corpus
+should ask per-document Nouls.
+
+## Limits
+
+- ~32k tokens per request (state + questions): ~30–50 typical passages per call.
+- Rate limits are undocumented; sustained 429s appeared at ~24 requests in flight.
+- Tail latency on 30-document states (p95 0.8–1.8 s) is higher than specialized hosted
+  rerankers'. Fanning out per-pair calls trades 30× the calls for a p95 near 0.25 s per call.
+- Hosted only; the data leaves your environment.
+
+## License
+
+Apache-2.0.

hev_rerank-0.1.0/README.md

@@ -0,0 +1,100 @@
+# hev-rerank
+
+Use [Jev](https://docs.typesafe.ai), TypeSafe's System One decision model, as a reranker.
+One call carries the query and up to 30 candidate documents; the model answers one
+question per document, in parallel: *is this document relevant to the query?* The answer is
+a calibrated probability, so the score is both a sort key (rerank) and an absolute
+threshold (prune).
+
+This repo is the whole recipe: the [prompt](hev_rerank/prompt.yaml), the request
+[state shape](schema/state.schema.json), a 90-line Python wrapper, and the
+[results](RESULTS.md) of running it against hosted and open-weight rerankers and LLM
+judges on BEIR shortlists.
+
+## Results
+
+nDCG@10, every system permuting the same BM25 top-30 shortlist per query. Jev uses the
+domain-neutral phrasing in `prompt.yaml`, untuned per corpus.
+
+| Reranker | SciFact (300 q) | NFCorpus (323 q) | FiQA (300 q) | p50 / p95 per query | Cost per 300 q |
+|---|---|---|---|---|---|
+| BM25 order, no rerank | 0.667 | 0.310 | 0.234 | | |
+| MiniLM-L6 cross-encoder (local) | 0.682 | 0.336 | 0.323 | | |
+| Cohere rerank-v3.5 | 0.745 | 0.340 | 0.374 | 129–277 ms / 267–698 ms | $0.60 |
+| Mixedbread mxbai-rerank-large-v2 (hosted) | 0.749 | 0.324 | 0.354 | 322–380 ms / 374–460 ms | $1.05 |
+| Voyage rerank-2.5 | 0.755 | 0.357 | 0.395 | 177–187 ms / 255–300 ms | ~$0.15 |
+| Voyage rerank-3 | 0.755 | 0.357 | 0.402 | 173–192 ms / 285–288 ms | ~$0.15 |
+| gpt-5.6-luna (reasoning off), listwise | 0.747 | 0.355 | 0.363 | 3.1–3.2 s / 3.8–5.8 s | ~$0.71 |
+| Claude Haiku 4.5, listwise | 0.723 | | | 2.9 s / 11.6 s | $4.18 |
+| Claude Opus 5 (low effort), listwise | 0.756 | | | 5.0 s / 7.5 s | $28.40 |
+| **Jev, 30 Nouls in one state** | **0.768** | **0.358** | 0.376 | 208–238 ms / 0.8–1.8 s | $0.13–0.19 |
+| Jev, one Noul per (query, doc) pair | 0.772 | 0.358 | 0.376 | 122–144 ms per call | $0.23–0.30 |
+
+Paired-bootstrap per query, Jev's one-state shape against each reranker: at or above Cohere on
+all three corpora (two wins, one tie); above Mixedbread (two wins, one interval touching zero);
+trading with Voyage, the strongest purpose-built reranker here (edge on SciFact, tie on NFCorpus,
+loss on FiQA by ~0.02); above every LLM used as a reranker on cost and latency, and on quality
+except Opus 5, which it ties at ~1/100 the cost.
+
+The honest one-liner: a general decision model with no reranker training lands in the same
+quality and price bracket as the best purpose-built rerankers, a third of Cohere's price, and
+returns a calibrated probability per document that none of them do. Voyage rerank-3 is as
+good and slightly cheaper; what Jev adds over it is the probability and the fact that the same
+call shape covers routing, classification and extraction gating.
+
+Costs are the runs' actual token usage at list price (Voyage measured from its billed
+`usage.total_tokens`). Full tables, confidence intervals, determinism, position-bias and
+calibration checks: [RESULTS.md](RESULTS.md).
+
+## Install
+
+```sh
+pip install hev-rerank        # or: uv add hev-rerank
+export TYPESAFE_API_KEY=...   # https://console.typesafe.ai/settings/keys
+```
+
+## Use
+
+```python
+from hev_rerank import rerank
+
+hits = rerank(
+    "does vitamin D supplementation improve bone density in older adults",
+    [{"title": t, "text": x} for t, x in first_stage_results],   # strings work too
+    top_n=10,          # optional
+    threshold=0.1,     # optional: drop candidates the model is confident are irrelevant
+)
+for h in hits:
+    print(h.score, h.index, h.document["title"])
+```
+
+`rerank` returns `RerankResult(index, score, document)` best-first. Lists longer than
+`max_docs_per_call` (30) are split into concurrent calls under Jev's ~32k-token request
+budget. Scores across calls are comparable because each is an absolute probability.
+
+## The shape, and why
+
+- **State**: `{"query": ..., "documents": {"D00": {...}, "D01": {...}, ...}}`
+  ([schema](schema/state.schema.json)).
+- **Questions**: one `Noul` per document id, from [`prompt.yaml`](hev_rerank/prompt.yaml).
+  All questions in a request are evaluated in parallel, so 30 documents cost about the same
+  wall time as 5.
+- **Score**: the Noul probability. Calibration on the SciFact run: documents scored ≥ 0.9
+  were judged relevant 76% of the time; documents scored < 0.1, 0.5% of the time.
+
+Why not one `Choice` over the documents? It is slightly better when exactly one document is
+relevant (SciFact: +0.01 nDCG@10, +3 pts P@1) and much worse when many are (NFCorpus,
+median 16 relevant per query: 0.316 vs 0.358). A reranker that does not know the corpus
+should ask per-document Nouls.
+
+## Limits
+
+- ~32k tokens per request (state + questions): ~30–50 typical passages per call.
+- Rate limits are undocumented; sustained 429s appeared at ~24 requests in flight.
+- Tail latency on 30-document states (p95 0.8–1.8 s) is higher than specialized hosted
+  rerankers'. Fanning out per-pair calls trades 30× the calls for a p95 near 0.25 s per call.
+- Hosted only; the data leaves your environment.
+
+## License
+
+Apache-2.0.

hev_rerank-0.1.0/RESULTS.md

@@ -0,0 +1,78 @@
+# Jev as a reranker — benchmark results
+
+Generated 2026-09-17. Jev model jev-1.13.0.
+
+BM25 top-K shortlist per query (rank_bm25, stopworded), every reranker permutes the same shortlist. nDCG@10 via ir_measures. Latency is per query (one API call for the single-state shapes; per pair call for the pair shape). Cost is the run's actual token usage at list price ($0.042/M input for Jev; Cohere $2/1k searches; Claude list prices).
+
+## scifact / test — 300 queries
+
+| system | nDCG@10 | P@1 | MRR@10 | p50 | p95 | cost/run |
+|---|---|---|---|---|---|---|
+| jev_pair_generic | 0.772 | 0.710 | 0.756 | 122 ms | 236 ms | $0.28 |
+| jev_batch_generic | 0.768 | 0.697 | 0.749 | 224 ms | 1758 ms | $0.18 |
+| jev_choice | 0.762 | 0.710 | 0.754 |  |  |  |
+| jev_choice_generic | 0.760 | 0.693 | 0.746 | 192 ms | 1621 ms | $0.15 |
+| llm_opus5_low | 0.756 | 0.683 | 0.740 | 4950 ms | 7473 ms | $28.40 |
+| jev_batch | 0.756 | 0.683 | 0.739 |  |  |  |
+| jev_pair | 0.755 | 0.680 | 0.740 |  |  |  |
+| rerank_voyage | 0.755 | 0.657 | 0.734 | 187 ms | 286 ms | $0.21 |
+| rerank_voyage3 | 0.755 | 0.660 | 0.734 | 192 ms | 288 ms | $0.21 |
+| rerank_mixedbread | 0.749 | 0.670 | 0.731 | 379 ms | 460 ms | $1.05 |
+| llm_gpt56_luna_none | 0.747 | 0.653 | 0.729 | 3223 ms | 5824 ms | $0.71 |
+| rerank_cohere | 0.745 | 0.660 | 0.728 | 277 ms | 698 ms | $0.60 |
+| llm_haiku45 | 0.723 | 0.633 | 0.701 | 2899 ms | 11562 ms | $4.18 |
+| local_mxbai-rerank-base-v1 | 0.716 | 0.633 | 0.693 | 8148 ms | 9933 ms |  |
+| local_ms-marco-MiniLM-L-6-v2 | 0.682 | 0.570 | 0.653 | 211 ms | 242 ms |  |
+| bm25 | 0.667 | 0.540 | 0.634 |  |  |  |
+
+## scifact / train — 300 queries
+
+| system | nDCG@10 | P@1 | MRR@10 | p50 | p95 | cost/run |
+|---|---|---|---|---|---|---|
+| jev_choice_generic | 0.776 | 0.713 | 0.766 | 253 ms | 1157 ms | $0.26 |
+| jev_batch_generic | 0.774 | 0.683 | 0.751 | 298 ms | 1721 ms | $0.30 |
+| jev_pair_generic | 0.773 | 0.673 | 0.754 | 117 ms | 221 ms | $0.28 |
+| jev_pair | 0.767 | 0.693 | 0.740 | 126 ms | 239 ms | $0.28 |
+| jev_choice | 0.756 | 0.690 | 0.747 | 202 ms | 1657 ms | $0.16 |
+| jev_batch | 0.752 | 0.667 | 0.732 | 242 ms | 1738 ms | $0.18 |
+| ce_minilm | 0.708 | 0.573 | 0.671 |  |  |  |
+| bm25 | 0.670 | 0.567 | 0.640 |  |  |  |
+
+## nfcorpus / test — 323 queries
+
+| system | nDCG@10 | P@1 | MRR@10 | p50 | p95 | cost/run |
+|---|---|---|---|---|---|---|
+| jev_batch_generic | 0.358 | 0.511 | 0.581 | 238 ms | 1716 ms | $0.19 |
+| jev_pair_generic | 0.358 | 0.523 | 0.590 | 144 ms | 295 ms | $0.30 |
+| rerank_voyage3 | 0.357 | 0.514 | 0.587 | 191 ms | 285 ms | $0.23 |
+| rerank_voyage | 0.357 | 0.517 | 0.584 | 183 ms | 300 ms | $0.23 |
+| llm_gpt56_luna_none | 0.355 | 0.508 | 0.589 | 3062 ms | 4521 ms | $0.76 |
+| rerank_cohere | 0.339 | 0.483 | 0.561 | 170 ms | 404 ms | $0.65 |
+| ce_minilm | 0.336 | 0.492 | 0.567 |  |  |  |
+| rerank_mixedbread | 0.324 | 0.424 | 0.528 | 380 ms | 452 ms | $1.13 |
+| jev_choice_generic | 0.315 | 0.514 | 0.580 | 212 ms | 1724 ms | $0.17 |
+| bm25 | 0.310 | 0.440 | 0.521 |  |  |  |
+
+## fiqa / test — 300 queries
+
+| system | nDCG@10 | P@1 | MRR@10 | p50 | p95 | cost/run |
+|---|---|---|---|---|---|---|
+| rerank_voyage3 | 0.402 | 0.490 | 0.522 | 173 ms | 288 ms | $0.15 |
+| rerank_voyage | 0.395 | 0.483 | 0.517 | 177 ms | 255 ms | $0.15 |
+| jev_pair_generic | 0.376 | 0.440 | 0.486 | 128 ms | 238 ms | $0.23 |
+| jev_batch_generic | 0.376 | 0.443 | 0.483 | 208 ms | 785 ms | $0.13 |
+| rerank_cohere | 0.374 | 0.430 | 0.482 | 129 ms | 267 ms | $0.60 |
+| llm_gpt56_luna_none | 0.363 | 0.393 | 0.462 | 3131 ms | 3774 ms | $0.49 |
+| jev_choice_generic | 0.358 | 0.420 | 0.480 | 191 ms | 772 ms | $0.10 |
+| rerank_mixedbread | 0.354 | 0.380 | 0.447 | 322 ms | 374 ms | $1.05 |
+| ce_minilm | 0.323 | 0.330 | 0.407 |  |  |  |
+| bm25 | 0.234 | 0.200 | 0.279 |  |  |  |
+
+## Validation (scifact/test)
+
+- Determinism: 30 batch requests re-issued, 900 Noul pairs: mean |Δ| 0.004, max 0.14, zero top-1 changes.
+- Position bias: reversed candidate order, per-doc Spearman 0.83 (30-Noul state) / 0.92 (Choice); nDCG@10 within 0.005.
+- Calibration (pair Nouls): p≥0.9 → 76% judged relevant; p<0.1 → 0.5%.
+- Random permutation of the shortlist: nDCG@10 0.12 (floor).
+- Claude Opus 5 refused 12/300 claims (safety classifier); scored as zero in the table, 0.778 with BM25 order substituted on those queries.
+- Jev rate limit: two concurrent jobs (~24 in flight) drew sustained 429s; runs were serialized at 6-wide.

hev_rerank-0.1.0/examples/bm25_then_jev.py

@@ -0,0 +1,15 @@
+"""First-stage BM25 over a handful of passages, then Jev reranks the shortlist."""
+from rank_bm25 import BM25Okapi
+from hev_rerank import rerank
+
+passages = [
+    {"title": "Vitamin D and bone density", "text": "Randomised trial: supplementation raised bone mineral density in adults over 65."},
+    {"title": "Vitamin D sources", "text": "Dietary sources include oily fish, egg yolk and fortified milk."},
+    {"title": "Calcium absorption", "text": "Vitamin D increases intestinal calcium absorption; deficiency impairs it."},
+    {"title": "Sunscreen use", "text": "Sunscreen reduces cutaneous vitamin D synthesis only marginally in practice."},
+]
+query = "does vitamin D supplementation improve bone density in older adults"
+bm25 = BM25Okapi([(p["title"] + " " + p["text"]).lower().split() for p in passages])
+shortlist = [passages[i] for i in sorted(range(len(passages)), key=lambda i: -bm25.get_scores(query.lower().split())[i])]
+for r in rerank(query, shortlist, top_n=3, threshold=0.1):
+    print(f"{r.score:.2f}  {r.document['title']}")

hev_rerank-0.1.0/hev_rerank/__init__.py

@@ -0,0 +1,4 @@
+from .rerank import RerankResult, load_prompt, rerank
+
+__all__ = ["rerank", "RerankResult", "load_prompt"]
+__version__ = "0.1.0"

hev_rerank-0.1.0/hev_rerank/prompt.yaml

@@ -0,0 +1,8 @@
+# The whole "prompt" for using Jev as a reranker. One Noul question per candidate document,
+# all candidates in one state, evaluated in parallel by the model. Domain-neutral phrasing:
+# on SciFact / NFCorpus / FiQA it matched or beat a corpus-tailored phrasing (see RESULTS.md).
+version: generic-1
+question: "Document `documents.{id}` is relevant to `query`: it contains information that answers or directly addresses it."
+criteria:
+  "true": "The document contains information that answers the query or directly addresses what it asks about."
+  "false": "The document is only loosely related, on a similar topic, or does not address what the query asks."

hev_rerank-0.1.0/hev_rerank/rerank.py

@@ -0,0 +1,104 @@
+"""Jev (TypeSafe's System One model) as a reranker.
+
+Shape: the query and up to `max_docs_per_call` candidate documents go into one Jev *state*;
+one Noul ("is this document relevant?") question per document; every question is evaluated
+in parallel by the model, so a 30-document call costs about the same wall time as a 5-document
+call. The Noul is a calibrated probability, so the score is usable as an absolute relevance
+threshold (prune) as well as a sort key (rerank).
+"""
+from __future__ import annotations
+
+import os
+from concurrent.futures import ThreadPoolExecutor
+from dataclasses import dataclass
+from pathlib import Path
+from typing import Any, Iterable, Mapping, Sequence
+
+import yaml
+from typesafe_sdk import Noul, RetryPolicy, TypeSafeClient
+
+PROMPT_PATH = Path(__file__).with_name("prompt.yaml")
+# TypeSafe documents a ~32k-token budget per request (state + questions), ~150k characters of
+# English. Keep headroom for the questions and for non-English text.
+DEFAULT_MAX_STATE_CHARS = 100_000
+
+
+@dataclass(frozen=True)
+class RerankResult:
+    index: int          # position in the input `documents`
+    score: float        # Noul probability the document is relevant, 0..1
+    document: Any       # the input document, unchanged
+
+
+def load_prompt(path: str | os.PathLike[str] | None = None) -> dict:
+    with open(path or PROMPT_PATH) as f:
+        return yaml.safe_load(f)
+
+
+def _doc_state(doc: Any) -> Mapping[str, Any] | str:
+    if isinstance(doc, str):
+        return doc
+    if isinstance(doc, Mapping):
+        return {k: doc[k] for k in ("title", "text", "content", "body") if k in doc} or dict(doc)
+    return str(doc)
+
+
+def _chunk(docs: Sequence[Any], max_docs: int, max_chars: int) -> Iterable[list[int]]:
+    batch: list[int] = []
+    chars = 0
+    for i, d in enumerate(docs):
+        n = len(str(_doc_state(d)))
+        if batch and (len(batch) >= max_docs or chars + n > max_chars):
+            yield batch
+            batch, chars = [], 0
+        batch.append(i)
+        chars += n
+    if batch:
+        yield batch
+
+
+def rerank(
+    query: str,
+    documents: Sequence[Any],
+    *,
+    top_n: int | None = None,
+    threshold: float | None = None,
+    client: TypeSafeClient | None = None,
+    model: str = "jev-latest",
+    prompt: Mapping[str, Any] | None = None,
+    max_docs_per_call: int = 30,
+    max_state_chars: int = DEFAULT_MAX_STATE_CHARS,
+    concurrency: int = 4,
+) -> list[RerankResult]:
+    """Score every document for relevance to `query` and return them best-first.
+
+    documents: strings, or mappings with `title` / `text` (extra keys are passed through
+    unchanged in the result but not sent to the model unless they are title/text/content/body).
+    top_n: keep only the best N. threshold: drop documents scoring below it (the Noul is
+    calibrated: on BEIR shortlists, documents under 0.1 were judged relevant ~0.5% of the time).
+    Long candidate lists are split into several calls, issued concurrently.
+    """
+    if not documents:
+        return []
+    p = prompt or load_prompt()
+    client = client or TypeSafeClient(retry=RetryPolicy(max_retries=6, backoff_max=30.0, timeout=90.0))
+
+    def score_batch(idx: list[int]) -> list[tuple[int, float]]:
+        ids = {f"D{j:02d}": i for j, i in enumerate(idx)}
+        state = {"query": query, "documents": {k: _doc_state(documents[i]) for k, i in ids.items()}}
+        questions = {k: Noul(instructions=p["question"].format(id=k), criteria=p["criteria"]) for k in ids}
+        r = client.system_one(state=state, questions=questions, model=model)
+        return [(ids[k], float(r.answers[k].noul)) for k in ids]
+
+    batches = list(_chunk(documents, max_docs_per_call, max_state_chars))
+    if len(batches) == 1:
+        scored = score_batch(batches[0])
+    else:
+        with ThreadPoolExecutor(concurrency) as ex:
+            scored = [s for part in ex.map(score_batch, batches) for s in part]
+
+    results = [RerankResult(i, s, documents[i]) for i, s in scored]
+    if threshold is not None:
+        results = [r for r in results if r.score >= threshold]
+    results.sort(key=lambda r: -r.score)
+    return results[:top_n] if top_n else results

hev_rerank-0.1.0/pyproject.toml

@@ -0,0 +1,34 @@
+[project]
+name = "hev-rerank"
+version = "0.1.0"
+description = "Use Jev (TypeSafe's System One model) as a calibrated reranker: one call, up to 30 documents, a probability per document."
+readme = "README.md"
+license = "Apache-2.0"
+requires-python = ">=3.10"
+authors = [{ name = "Adam Hevenor" }]
+keywords = ["reranker", "search", "retrieval", "jev", "typesafe", "rag"]
+classifiers = [
+  "Development Status :: 3 - Alpha",
+  "Intended Audience :: Developers",
+  "License :: OSI Approved :: Apache Software License",
+  "Programming Language :: Python :: 3",
+  "Topic :: Text Processing :: Indexing",
+]
+dependencies = ["typesafe-sdk>=0.6", "pyyaml>=6"]
+
+[project.urls]
+Repository = "https://github.com/hev/jev-rerank"
+Results = "https://github.com/hev/jev-rerank/blob/main/RESULTS.md"
+
+[project.optional-dependencies]
+examples = ["rank-bm25"]
+
+[build-system]
+requires = ["hatchling"]
+build-backend = "hatchling.build"
+
+[tool.hatch.build]
+exclude = [".claude/", "dist/"]
+
+[tool.hatch.build.targets.wheel]
+packages = ["hev_rerank"]

hev_rerank-0.1.0/schema/result.schema.json

@@ -0,0 +1,15 @@
+{
+  "$schema": "https://json-schema.org/draft/2020-12/schema",
+  "$id": "https://github.com/hev/jev-rerank/schema/result.schema.json",
+  "title": "Jev rerank result",
+  "type": "array",
+  "description": "Documents best-first. `score` is the Noul probability the document is relevant (calibrated, 0..1).",
+  "items": {
+    "type": "object",
+    "required": ["index", "score"],
+    "properties": {
+      "index": {"type": "integer", "minimum": 0, "description": "Position in the input document list."},
+      "score": {"type": "number", "minimum": 0, "maximum": 1}
+    }
+  }
+}

hev_rerank-0.1.0/schema/state.schema.json

@@ -0,0 +1,23 @@
+{
+  "$schema": "https://json-schema.org/draft/2020-12/schema",
+  "$id": "https://github.com/hev/jev-rerank/schema/state.schema.json",
+  "title": "Jev rerank request state",
+  "description": "The `state` object sent to POST https://api.typesafe.ai/v1/systemone. One Noul question per key of `documents` (see prompt.yaml); question ids match the document keys.",
+  "type": "object",
+  "required": ["query", "documents"],
+  "additionalProperties": false,
+  "properties": {
+    "query": {"type": "string", "description": "The search query or claim."},
+    "documents": {
+      "type": "object",
+      "description": "Candidate documents keyed D00, D01, ... in shortlist order.",
+      "propertyNames": {"pattern": "^D[0-9]{2,}$"},
+      "additionalProperties": {
+        "oneOf": [
+          {"type": "string"},
+          {"type": "object", "properties": {"title": {"type": "string"}, "text": {"type": "string"}}, "additionalProperties": true}
+        ]
+      }
+    }
+  }
+}