szl-engine-bench 0.3.0__tar.gz

szl_engine_bench-0.3.0/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 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 2026 SZL Holdings
+
+   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.

szl_engine_bench-0.3.0/PKG-INFO

@@ -0,0 +1,188 @@
+Metadata-Version: 2.4
+Name: szl-engine-bench
+Version: 0.3.0
+Summary: Honest inference-engine benchmark harness: TTFT, ITL, goodput-at-SLO, fairness gates, and hash-chained receipts.
+License: Apache-2.0
+Requires-Python: >=3.11
+Description-Content-Type: text/markdown
+License-File: LICENSE
+Dynamic: license-file
+
+# szl-engine-bench
+
+An honest, standard-library-only benchmark client for OpenAI-compatible
+streaming endpoints exposed by vLLM, SGLang, llama.cpp, MLX, TGI, and
+Transformers Serve.
+
+Version 0.3.0 reports separate distributions for time to first token/chunk
+(TTFT), inter-chunk latency (ITL), end-to-end latency, and the fraction of
+runs meeting a declared TTFT service-level objective. It does not collapse an
+engine comparison into one winner.
+
+The distinction is deliberate. SGLang's serving guide reports TTFT, TPOT,
+ITL, and throughput as separate online-serving metrics, while vLLM Bench
+separately exposes TTFT/ITL percentiles and SLO goodput. Cache shape, request
+rate, concurrency, model, and hardware can all change the result; a benchmark
+from a different workload is not evidence for this one.
+
+- [SGLang benchmark and profiling guide (source snapshot)](https://github.com/sgl-project/sglang/blob/50ed4c011ffb37e1c16248b99deab7d4dbdde79b/docs/developer_guide/benchmark_and_profiling.md)
+- [vLLM Bench metrics (source snapshot)](https://github.com/vllm-project/vllm-bench/blob/f3131823f167aa1e6f31a48be1ae15a18667a534/README.md#metrics)
+
+## States and measurement semantics
+
+- `MEASURED`: every requested run produced at least two non-empty streaming
+  response chunks and valid monotonic arrival times.
+- `BLOCKED`: the engine endpoint was not configured, or fewer than two engines
+  produced measurements for comparison.
+- `INVALID`: configuration, stream shape, or fairness inputs were unsuitable
+  for the requested metric.
+- `FAILED`: the configured endpoint or streaming protocol actually failed.
+
+ITL is measured between non-empty SSE response chunks. An SSE chunk is not
+assumed to equal one model token. `chunks_per_s` is always labeled as such;
+the request asks the server for streamed usage with
+`stream_options.include_usage=true`, and `tok_per_s` is `UNAVAILABLE` unless
+the endpoint actually supplies an integer `usage.completion_tokens` value.
+This avoids turning packet counts into fake token throughput.
+
+The comparison keeps the v0.1.0 fairness gates: measured engines must use the
+same model string and run count for low-level numerical diagnostics.
+Version 0.3.0 additionally requires the declared identity contract below for
+CLI comparisons; the old model-name check alone is explicitly unqualified. `goodput_at_slo` is the auditable fraction of
+runs whose TTFT is less than or equal to `--slo-ttft-ms`; the boundary is
+inclusive. It is not claimed to be a maximum sustainable request rate.
+
+## Install and verify
+
+```bash
+python -m pip install -e . pytest
+python -m pytest tests -q
+python -m compileall -q engine_bench.py benchmark_manifest.py tests
+python engine_bench.py --engines vllm --runs 1
+```
+
+The last command is a fail-closed smoke test: without a declaration it emits
+a `BLOCKED` result and a valid receipt, even when an endpoint is configured.
+It issues no network request. No model or GPU is needed for the test suite.
+
+The deterministic burst fixture from issue #2 can be reproduced without an
+endpoint:
+
+```bash
+python -c "import engine_bench as e; print(e.run_stats([.05,.06,.07,.4,.45,.5,.55,.6], 8))"
+```
+
+Its TTFT is 50 ms. The inter-chunk gaps are 10, 10, 330, 50, 50, 50, and
+50 ms, so linear interpolation yields ITL p50/p95/p99 of 50, 246, and
+313.2 ms. This is a unit fixture, not a hardware benchmark.
+
+## Exact experiment declarations (v0.3.0)
+
+The default CLI requires `--manifest comparison.json` before contacting any
+configured engine. The complete selected cohort is checked first: a missing or
+changed endpoint, extra engine, mismatched request configuration, invalid digest,
+or different hardware declaration cannot become a successful partial comparison.
+
+The manifest uses the closed `szl.engine-comparison/v1` contract. Its template is
+[`examples/comparison.synthetic.json`](examples/comparison.synthetic.json).
+**Every identity in that file is a synthetic test fixture, not a qualified model,
+hardware profile, or live endpoint.** Populate real operator-reviewed declarations
+before measuring a deployment. The loader never downloads or executes a manifest
+reference and rejects duplicate JSON keys, non-finite values, oversized inputs,
+unknown fields, control characters, mutable revision names, and newline-suffixed
+hashes. SHA checks use full-string matching, not permissive end anchors.
+
+| Declaration | Meaning |
+|---|---|
+| Model/tokenizer revisions | Exact 40-character lowercase Git subjects; not branch names |
+| Weights/tokenizer SHA-256 | Exact single artifact bytes, or SHA-256 of a separately retained canonical multi-file digest manifest |
+| Template/adapter SHA-256 | Deployed template/adapter artifact identity; null adapter means explicitly none |
+| Quantization/precision | Operator-declared configuration; no implicit conversion between engines |
+| Suite/prompt SHA-256 | Retained suite artifact plus SHA-256 of the exact UTF-8 prompt sent by the CLI |
+| Runs/token limit/timeout | Must match the requested invocation; timeout is the client's socket timeout, not a promised total wall-clock deadline |
+| Sampling | Explicit temperature, top-p, and integer seed, included in every actual HTTP request |
+| Engine build/configuration | Separate SHA-256 identities for each engine; these may differ in an engine comparison |
+| Hardware SHA-256 | Same retained hardware-profile digest across the declared cohort |
+| Endpoint SHA-256 | SHA-256 of the exact configured base URL after removing trailing slashes; no credentials allowed |
+
+`benchmark_manifest.endpoint_digest()` computes the endpoint digest. File digests
+refer to exact bytes; they are not a cryptographic demonstration that a remote
+server loaded those bytes. Engine names must match the supported registry. The
+client runs serially, performs no warmup/reset, and requires `concurrency: 1` and
+`cache_state: UNCONTROLLED`; comparing cold and warm caches is not silently claimed.
+It freezes the preflight endpoints and refuses redirects for bound requests.
+
+Every sample commits the exact serialized request bytes. A run binding includes
+the experiment, model/artifact subject, workload, per-engine build/configuration,
+hardware, endpoint, and request hashes. `compare_bound()` rejects altered bindings,
+missing samples, different wire requests, and omitted/failed cohort members.
+No failed engine is discarded to obtain a better-looking comparison.
+
+**Identity evidence remains `OPERATOR_DECLARED_NOT_RUNTIME_ATTESTED`.** A matching
+manifest is not server identity attestation. Comparison reports keep
+`runtime_identity: NOT_ATTESTED`, `quality_evaluation: NOT_PERFORMED`,
+`production_admission: NOT_EVALUATED`, and uncontrolled cache state. They measure
+client-observed latency only and never decide model promotion or an engine winner.
+Different hardware should be evaluated as a separately designed system experiment,
+not misrepresented as an engine-only improvement.
+
+An incomplete or invalid explicit `--manifest` invocation exits 2 after emitting
+its available receipt. The default offline smoke command retains its historical
+zero exit. Library `run_engine()`, `compare()`, and `compare_engines()` remain
+available for numerical diagnostics, labeled `UNBOUND_METRICS_ONLY` in comparisons.
+The CLI requires `--unbound-diagnostics` to explicitly opt into that legacy path;
+it cannot be combined with `--manifest` and is not release admission evidence.
+
+## Measure configured engines
+
+PowerShell:
+
+```powershell
+$env:VLLM_ENDPOINT = "http://127.0.0.1:8000"
+$env:SGLANG_ENDPOINT = "http://127.0.0.1:30000"
+python engine_bench.py --engines vllm sglang --model YOUR_MODEL --runs 10 --max-tokens 64 --slo-ttft-ms 200 --manifest comparison.json
+```
+
+POSIX shells:
+
+```bash
+export VLLM_ENDPOINT=http://127.0.0.1:8000
+export SGLANG_ENDPOINT=http://127.0.0.1:30000
+python engine_bench.py --engines vllm sglang --model YOUR_MODEL --runs 10 --max-tokens 64 --slo-ttft-ms 200 --manifest comparison.json
+```
+
+Supported variables are `VLLM_ENDPOINT`, `SGLANG_ENDPOINT`,
+`LLAMACPP_ENDPOINT`, `MLX_ENDPOINT`, `TGI_ENDPOINT`, and
+`TRANSFORMERS_ENDPOINT`. Each must be an absolute HTTP(S) base URL without
+embedded credentials, query parameters, fragments, or an invalid TCP port.
+Duplicate names in `--engines` are rejected as `INVALID` before any request,
+so receipt maps cannot silently overwrite a result. The client posts to
+`/v1/completions`.
+
+## Receipts
+
+Every CLI invocation emits one canonical-JSON SHA-256 chain receipt using the
+repository's `prev_hash` / `self_hash` schema and the all-zero genesis hash.
+The receipt anchors:
+
+- a canonical SHA-256 of every complete engine result;
+- a canonical SHA-256 of each measured engine's nested ITL gap arrays;
+- the ordered engine selection, benchmark configuration, and a prompt hash; and
+- the comparison verdict and every engine state; and
+- the canonical manifest digest and each exact wire-request/run binding (schema version 3).
+
+Each run sample also includes its raw `itl_gaps_ms` array and matching
+`itl_gaps_sha256`, so the hash can be independently recomputed.
+`UNSIGNED_HONEST` proves integrity and order only; it does not claim signer
+identity or external attestation.
+
+## Verification scope
+
+The test suite covers strict declarations, complete-cohort comparisons, actual
+loopback HTTP request bytes/sampling fields, redirect refusal, preflight zero-
+request behavior, endpoint freezing, bad JSON/digests, failure retention, and
+receipt commitments, in addition to the original streaming and numerical tests.
+Loopback test responses are fixtures: no trained model, remote production system,
+GPU performance, energy measurement, or quality result is represented by CI.
+
+Apache-2.0 · Doctrine v11 · SZL Holdings

szl_engine_bench-0.3.0/README.md

@@ -0,0 +1,178 @@
+# szl-engine-bench
+
+An honest, standard-library-only benchmark client for OpenAI-compatible
+streaming endpoints exposed by vLLM, SGLang, llama.cpp, MLX, TGI, and
+Transformers Serve.
+
+Version 0.3.0 reports separate distributions for time to first token/chunk
+(TTFT), inter-chunk latency (ITL), end-to-end latency, and the fraction of
+runs meeting a declared TTFT service-level objective. It does not collapse an
+engine comparison into one winner.
+
+The distinction is deliberate. SGLang's serving guide reports TTFT, TPOT,
+ITL, and throughput as separate online-serving metrics, while vLLM Bench
+separately exposes TTFT/ITL percentiles and SLO goodput. Cache shape, request
+rate, concurrency, model, and hardware can all change the result; a benchmark
+from a different workload is not evidence for this one.
+
+- [SGLang benchmark and profiling guide (source snapshot)](https://github.com/sgl-project/sglang/blob/50ed4c011ffb37e1c16248b99deab7d4dbdde79b/docs/developer_guide/benchmark_and_profiling.md)
+- [vLLM Bench metrics (source snapshot)](https://github.com/vllm-project/vllm-bench/blob/f3131823f167aa1e6f31a48be1ae15a18667a534/README.md#metrics)
+
+## States and measurement semantics
+
+- `MEASURED`: every requested run produced at least two non-empty streaming
+  response chunks and valid monotonic arrival times.
+- `BLOCKED`: the engine endpoint was not configured, or fewer than two engines
+  produced measurements for comparison.
+- `INVALID`: configuration, stream shape, or fairness inputs were unsuitable
+  for the requested metric.
+- `FAILED`: the configured endpoint or streaming protocol actually failed.
+
+ITL is measured between non-empty SSE response chunks. An SSE chunk is not
+assumed to equal one model token. `chunks_per_s` is always labeled as such;
+the request asks the server for streamed usage with
+`stream_options.include_usage=true`, and `tok_per_s` is `UNAVAILABLE` unless
+the endpoint actually supplies an integer `usage.completion_tokens` value.
+This avoids turning packet counts into fake token throughput.
+
+The comparison keeps the v0.1.0 fairness gates: measured engines must use the
+same model string and run count for low-level numerical diagnostics.
+Version 0.3.0 additionally requires the declared identity contract below for
+CLI comparisons; the old model-name check alone is explicitly unqualified. `goodput_at_slo` is the auditable fraction of
+runs whose TTFT is less than or equal to `--slo-ttft-ms`; the boundary is
+inclusive. It is not claimed to be a maximum sustainable request rate.
+
+## Install and verify
+
+```bash
+python -m pip install -e . pytest
+python -m pytest tests -q
+python -m compileall -q engine_bench.py benchmark_manifest.py tests
+python engine_bench.py --engines vllm --runs 1
+```
+
+The last command is a fail-closed smoke test: without a declaration it emits
+a `BLOCKED` result and a valid receipt, even when an endpoint is configured.
+It issues no network request. No model or GPU is needed for the test suite.
+
+The deterministic burst fixture from issue #2 can be reproduced without an
+endpoint:
+
+```bash
+python -c "import engine_bench as e; print(e.run_stats([.05,.06,.07,.4,.45,.5,.55,.6], 8))"
+```
+
+Its TTFT is 50 ms. The inter-chunk gaps are 10, 10, 330, 50, 50, 50, and
+50 ms, so linear interpolation yields ITL p50/p95/p99 of 50, 246, and
+313.2 ms. This is a unit fixture, not a hardware benchmark.
+
+## Exact experiment declarations (v0.3.0)
+
+The default CLI requires `--manifest comparison.json` before contacting any
+configured engine. The complete selected cohort is checked first: a missing or
+changed endpoint, extra engine, mismatched request configuration, invalid digest,
+or different hardware declaration cannot become a successful partial comparison.
+
+The manifest uses the closed `szl.engine-comparison/v1` contract. Its template is
+[`examples/comparison.synthetic.json`](examples/comparison.synthetic.json).
+**Every identity in that file is a synthetic test fixture, not a qualified model,
+hardware profile, or live endpoint.** Populate real operator-reviewed declarations
+before measuring a deployment. The loader never downloads or executes a manifest
+reference and rejects duplicate JSON keys, non-finite values, oversized inputs,
+unknown fields, control characters, mutable revision names, and newline-suffixed
+hashes. SHA checks use full-string matching, not permissive end anchors.
+
+| Declaration | Meaning |
+|---|---|
+| Model/tokenizer revisions | Exact 40-character lowercase Git subjects; not branch names |
+| Weights/tokenizer SHA-256 | Exact single artifact bytes, or SHA-256 of a separately retained canonical multi-file digest manifest |
+| Template/adapter SHA-256 | Deployed template/adapter artifact identity; null adapter means explicitly none |
+| Quantization/precision | Operator-declared configuration; no implicit conversion between engines |
+| Suite/prompt SHA-256 | Retained suite artifact plus SHA-256 of the exact UTF-8 prompt sent by the CLI |
+| Runs/token limit/timeout | Must match the requested invocation; timeout is the client's socket timeout, not a promised total wall-clock deadline |
+| Sampling | Explicit temperature, top-p, and integer seed, included in every actual HTTP request |
+| Engine build/configuration | Separate SHA-256 identities for each engine; these may differ in an engine comparison |
+| Hardware SHA-256 | Same retained hardware-profile digest across the declared cohort |
+| Endpoint SHA-256 | SHA-256 of the exact configured base URL after removing trailing slashes; no credentials allowed |
+
+`benchmark_manifest.endpoint_digest()` computes the endpoint digest. File digests
+refer to exact bytes; they are not a cryptographic demonstration that a remote
+server loaded those bytes. Engine names must match the supported registry. The
+client runs serially, performs no warmup/reset, and requires `concurrency: 1` and
+`cache_state: UNCONTROLLED`; comparing cold and warm caches is not silently claimed.
+It freezes the preflight endpoints and refuses redirects for bound requests.
+
+Every sample commits the exact serialized request bytes. A run binding includes
+the experiment, model/artifact subject, workload, per-engine build/configuration,
+hardware, endpoint, and request hashes. `compare_bound()` rejects altered bindings,
+missing samples, different wire requests, and omitted/failed cohort members.
+No failed engine is discarded to obtain a better-looking comparison.
+
+**Identity evidence remains `OPERATOR_DECLARED_NOT_RUNTIME_ATTESTED`.** A matching
+manifest is not server identity attestation. Comparison reports keep
+`runtime_identity: NOT_ATTESTED`, `quality_evaluation: NOT_PERFORMED`,
+`production_admission: NOT_EVALUATED`, and uncontrolled cache state. They measure
+client-observed latency only and never decide model promotion or an engine winner.
+Different hardware should be evaluated as a separately designed system experiment,
+not misrepresented as an engine-only improvement.
+
+An incomplete or invalid explicit `--manifest` invocation exits 2 after emitting
+its available receipt. The default offline smoke command retains its historical
+zero exit. Library `run_engine()`, `compare()`, and `compare_engines()` remain
+available for numerical diagnostics, labeled `UNBOUND_METRICS_ONLY` in comparisons.
+The CLI requires `--unbound-diagnostics` to explicitly opt into that legacy path;
+it cannot be combined with `--manifest` and is not release admission evidence.
+
+## Measure configured engines
+
+PowerShell:
+
+```powershell
+$env:VLLM_ENDPOINT = "http://127.0.0.1:8000"
+$env:SGLANG_ENDPOINT = "http://127.0.0.1:30000"
+python engine_bench.py --engines vllm sglang --model YOUR_MODEL --runs 10 --max-tokens 64 --slo-ttft-ms 200 --manifest comparison.json
+```
+
+POSIX shells:
+
+```bash
+export VLLM_ENDPOINT=http://127.0.0.1:8000
+export SGLANG_ENDPOINT=http://127.0.0.1:30000
+python engine_bench.py --engines vllm sglang --model YOUR_MODEL --runs 10 --max-tokens 64 --slo-ttft-ms 200 --manifest comparison.json
+```
+
+Supported variables are `VLLM_ENDPOINT`, `SGLANG_ENDPOINT`,
+`LLAMACPP_ENDPOINT`, `MLX_ENDPOINT`, `TGI_ENDPOINT`, and
+`TRANSFORMERS_ENDPOINT`. Each must be an absolute HTTP(S) base URL without
+embedded credentials, query parameters, fragments, or an invalid TCP port.
+Duplicate names in `--engines` are rejected as `INVALID` before any request,
+so receipt maps cannot silently overwrite a result. The client posts to
+`/v1/completions`.
+
+## Receipts
+
+Every CLI invocation emits one canonical-JSON SHA-256 chain receipt using the
+repository's `prev_hash` / `self_hash` schema and the all-zero genesis hash.
+The receipt anchors:
+
+- a canonical SHA-256 of every complete engine result;
+- a canonical SHA-256 of each measured engine's nested ITL gap arrays;
+- the ordered engine selection, benchmark configuration, and a prompt hash; and
+- the comparison verdict and every engine state; and
+- the canonical manifest digest and each exact wire-request/run binding (schema version 3).
+
+Each run sample also includes its raw `itl_gaps_ms` array and matching
+`itl_gaps_sha256`, so the hash can be independently recomputed.
+`UNSIGNED_HONEST` proves integrity and order only; it does not claim signer
+identity or external attestation.
+
+## Verification scope
+
+The test suite covers strict declarations, complete-cohort comparisons, actual
+loopback HTTP request bytes/sampling fields, redirect refusal, preflight zero-
+request behavior, endpoint freezing, bad JSON/digests, failure retention, and
+receipt commitments, in addition to the original streaming and numerical tests.
+Loopback test responses are fixtures: no trained model, remote production system,
+GPU performance, energy measurement, or quality result is represented by CI.
+
+Apache-2.0 · Doctrine v11 · SZL Holdings