qoptct 0.3.0__tar.gz

qoptct-0.3.0/CHANGELOG.md

@@ -0,0 +1,63 @@
+# Changelog
+
+## 0.3.0 — 2026-11-13 (first PyPI release)
+
+### Changed — package renamed to `qoptct`
+* **The distribution, the import package and the command are now called `qoptct`** (the PyPI name `qopt`
+  belongs to an unrelated project with the import name `qopt`). The project itself is still called qopt.
+  Migration: `pip install qoptct`; `import qopt` → `import qoptct`; `from qopt.qiskit_qopt import QoptPass` →
+  `from qoptct.qiskit_qopt import QoptPass`; command `qopt optimize ...` → `qoptct optimize ...`.
+  Environment variables (`QOPT_CACHE`, `QOPT_PROG_SEC`, ...) and the cache directory `~/.cache/qopt` are unchanged.
+* PyTorch is now optional: `pip install "qoptct[net]"`. Only `--net` (and `QoptPass(net=True)`) needs it and
+  stops with an installation hint if it is missing; the other modes no longer import torch at all.
+* Python ≥ 3.11 (was 3.12); tested with 3.11, 3.12 and 3.14.
+
+### Added
+* Qiskit transpiler stage plugins, both named `qoptct`:
+  `transpile(qc, optimization_method="qoptct")` (Qiskit's optimization stage of the chosen level, then
+  `QoptPass` restricted to the target / basis / coupling map) and `init_method="qoptct"` (`QoptPass` at the
+  Toffoli level, before decomposition).
+* `QoptPass` now works on any circuit: it optimizes each maximal Clifford+T run separately (measurements,
+  barriers, rotations and other gates are kept), replaces a run only if the result is proven equal and strictly
+  cheaper, respects basis gates / coupling map / target, keeps the global phase (runs ≤ 10 qubits), and reports
+  `runs`, `replaced`, CX/T before and after in `pass.last`. Messages and documentation in English.
+* `.github/workflows/tests.yml` (pytest on Ubuntu and macOS against the built wheel) and
+  `.github/workflows/publish.yml` (PyPI Trusted Publishing on a published GitHub release).
+* `MANIFEST.in` (the sdist contains the tests and the redistributable benchmark inputs, never hwb6).
+
+## 0.2.0 — 2026-11-12 (first public release candidate; not yet published)
+
+### Added
+* Measurement-based optimization modes `--mode balanced` and `--mode t` (OpenQASM 3 output with mid-circuit
+  measurements and classically controlled Clifford corrections):
+  * logical-AND compiler with measurement-based uncomputation and exact GF(2) tracking of wire values;
+  * clean-ancilla model `--clean` (wires not listed in `.i` start in |0>);
+  * CX polishing of AND circuits (re-synthesis of unconditioned pieces using |+> and measured-wire freedoms);
+  * hybrid compilation for circuits with genuine H gates (grover_5, qft_4);
+  * recognition of multi-controlled Toffolis (AND ladders, Khattar–Gidney syntheses with AND-based outer pair);
+  * Hadamard gadgets + TODD with deterministic corrections (`--tgad`), partial gadgetization;
+  * optional external VarTODD via `--vartodd` and `QOPT_VARTODD_BIN`.
+* Independent checkers: `verify_qasm` (strict and clean models, no renormalization) and `verify_big`
+  (sum-over-paths, > 22 qubits).
+* `results/`: verified best and Pareto-optimal circuits for 23 benchmarks in three models, `TABLE.md`,
+  `table.csv`, comparison with nine tools (`COMPARISON.md`).
+* `scripts/verify_all.py`, `scripts/reproduce.sh`, `scripts/fetch_benchmarks.sh`.
+* Documentation: `docs/METHOD.md`, `docs/FINDINGS.md`, `paper/preprint.md`.
+
+### Changed
+* Installable package `qopt` with the `qopt` console command; all paths relative to the package; caches in
+  `$QOPT_CACHE` (default `~/.cache/qopt`).
+* Cached data that is keyed by benchmark name (unitary start circuits, hand-made clean-ancilla constructions) is
+  used only if the SHA-256 of the input file matches the benchmark.
+* CLI messages and help in English. The `bench` subcommand of 0.1 was replaced by `scripts/reproduce.sh`.
+
+### Removed / not included
+* `hwb6` (and its optimized circuits): the benchmark file forbids redistribution; `scripts/fetch_benchmarks.sh`
+  downloads it for local use.
+
+### Fixed (found by independent verification, see docs/FINDINGS.md)
+* A 51-T hwb6 circuit of an earlier gadget pipeline was wrong; it was never part of a release.
+
+## 0.1 — 2026-11-10 (internal)
+* Unitary CX/T optimizer: portfolio of evolved rewriting programs, phase folding, neural block resynthesis,
+  proof of equality (unitary / PyZX / numerical) with fallback to the input.

qoptct-0.3.0/CITATION.cff

@@ -0,0 +1,29 @@
+cff-version: 1.2.0
+message: "If you use this software or its results, please cite it as below."
+title: "qopt: a verified Clifford+T circuit optimizer for T and CNOT counts"
+type: software
+version: 0.3.0
+date-released: "2026-11-13"
+license: Apache-2.0
+authors:
+  - given-names: Maksym
+    family-names: Shchehlov
+keywords:
+  - quantum computing
+  - quantum circuit optimization
+  - Clifford+T
+  - T-count
+  - CNOT count
+  - logical AND
+  - measurement-based uncomputation
+  - verification
+abstract: >-
+  qopt reduces the T count and the CNOT count of Clifford+T circuits given as Toffoli networks.
+  It provides a unitary mode (portfolio of evolved rewriting programs, phase folding, optional neural block
+  resynthesis) and measurement-based modes that place Gidney logical-AND gates with measurement-based
+  uncomputation automatically by tracking every wire as a polynomial over GF(2). Every reported circuit is
+  re-checked by an independent state-vector simulator (up to 22 qubits) or a sum-over-paths checker.
+  Python package: pip install qoptct (import qoptct), with a Qiskit transpiler plugin.
+repository-code: "https://github.com/seglovmaksim695-creator/qopt"
+url: "https://github.com/seglovmaksim695-creator/qopt"
+doi: 10.5281/zenodo.24182059

qoptct-0.3.0/LICENSE

@@ -0,0 +1,202 @@
+
+                                 Apache License
+                           Version 2.0, January 2004
+                        https://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
+
+       https://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.

qoptct-0.3.0/MANIFEST.in

@@ -0,0 +1,8 @@
+# Source distribution: package, tests and the redistributable benchmark inputs the tests need.
+include README.md README_RU.md CHANGELOG.md LICENSE NOTICE CITATION.cff
+include bench_qc/*.qc bench_qc/SOURCES.md
+include results/best/strict/tof_3.qasm
+recursive-include tests *.py
+prune _build
+prune results/hardware
+global-exclude *.pyc __pycache__ *hwb*

qoptct-0.3.0/NOTICE

@@ -0,0 +1,11 @@
+qopt — verified Clifford+T circuit optimizer
+Copyright 2026 Maksym Shchehlov
+
+Licensed under the Apache License, Version 2.0 (see LICENSE).
+
+Third-party material:
+* bench_qc/*.qc — benchmark circuits from the Feynman toolkit repository
+  (https://github.com/meamy/feynman, benchmarks/qc; repository license BSD-3-Clause), originally collected by
+  M. Amy, D. Maslov, M. Mosca and others (see bench_qc/SOURCES.md). These files are not covered by the
+  Apache-2.0 license of this project; see bench_qc/SOURCES.md for provenance and checksums.
+* qopt/contrib/c12/mcx.py calls multi-controlled-X syntheses from Qiskit (Apache-2.0) at run time.

qoptct-0.3.0/PKG-INFO

@@ -0,0 +1,282 @@
+Metadata-Version: 2.4
+Name: qoptct
+Version: 0.3.0
+Summary: qopt: verified Clifford+T circuit optimizer (T and CNOT counts) with a Qiskit transpiler plugin
+Author: Maksym Shchehlov
+License-Expression: Apache-2.0
+Project-URL: Homepage, https://github.com/seglovmaksim695-creator/qopt
+Project-URL: Repository, https://github.com/seglovmaksim695-creator/qopt
+Project-URL: Issues, https://github.com/seglovmaksim695-creator/qopt/issues
+Project-URL: Changelog, https://github.com/seglovmaksim695-creator/qopt/blob/main/CHANGELOG.md
+Project-URL: DOI, https://doi.org/10.5281/zenodo.24182059
+Keywords: quantum computing,circuit optimization,Clifford+T,T-count,CNOT count,verification,qiskit,transpiler plugin
+Classifier: Development Status :: 4 - Beta
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Programming Language :: Python :: 3.13
+Classifier: Programming Language :: Python :: 3.14
+Classifier: Operating System :: MacOS
+Classifier: Operating System :: POSIX :: Linux
+Classifier: Topic :: Scientific/Engineering :: Physics
+Classifier: Intended Audience :: Science/Research
+Requires-Python: >=3.11
+Description-Content-Type: text/markdown
+License-File: LICENSE
+License-File: NOTICE
+Requires-Dist: numpy>=2.0
+Requires-Dist: qiskit>=2.0
+Requires-Dist: pyzx>=0.10
+Provides-Extra: net
+Requires-Dist: torch>=2.4; extra == "net"
+Provides-Extra: test
+Requires-Dist: pytest>=8; extra == "test"
+Dynamic: license-file
+
+# qopt — a verified Clifford+T circuit optimizer
+
+[![PyPI](https://img.shields.io/pypi/v/qoptct.svg)](https://pypi.org/project/qoptct/)
+[![tests](https://github.com/seglovmaksim695-creator/qopt/actions/workflows/tests.yml/badge.svg)](https://github.com/seglovmaksim695-creator/qopt/actions/workflows/tests.yml)
+[![DOI](https://zenodo.org/badge/DOI/10.5281/zenodo.24182059.svg)](https://doi.org/10.5281/zenodo.24182059)
+
+> **Package name:** the project is called *qopt*, but on PyPI the name `qopt` belongs to an unrelated project.
+> Since version 0.3.0 the package is therefore installed and imported as **`qoptct`**:
+> `pip install qoptct`, `import qoptct`, command `qoptct`.
+
+`qopt` reduces the **T count** and the **CNOT (CX) count** of Clifford+T circuits given as Toffoli networks
+(Feynman `.qc` format). It has a unitary mode and measurement-based modes, and it **never reports a circuit that
+has not been checked**: unitary outputs are proven equal to the input (or the input is returned), and every
+measurement-based output is re-checked by an independent simulator before it is written.
+
+Repository: <https://github.com/seglovmaksim695-creator/qopt>  · Version 0.3.0 · License Apache-2.0 (benchmark files: see `bench_qc/SOURCES.md`) · Author: Maksym Shchehlov
+
+## Installation
+
+```bash
+pip install qoptct              # Python 3.11+, macOS or Linux; provides the `qoptct` command and the Qiskit plugin
+pip install "qoptct[net]"       # optional: + PyTorch for the neural mode --net
+```
+
+Dependencies: numpy, qiskit ≥ 2.0, pyzx. PyTorch is needed only for `--net`; without it every other mode works
+and `--net` stops with a message that says what to install.
+
+From the repository (to run the tests, the benchmarks and the scripts):
+
+```bash
+git clone https://github.com/seglovmaksim695-creator/qopt.git && cd qopt
+python3 -m venv .venv && . .venv/bin/activate
+pip install -r requirements.txt                      # exact versions used for the results (incl. torch)
+pip install -e ".[test]"
+pytest -q tests                                      # a few minutes
+```
+
+`requirements-lock.txt` pins the full environment used for testing. Caches are written to `$QOPT_CACHE`
+(default `~/.cache/qopt`).
+
+## Usage
+
+```bash
+qoptct optimize bench_qc/tof_3.qc -o tof_3_opt.qc                          # unitary, minimum CX (default --mode cx)
+qoptct optimize bench_qc/tof_3.qc -o tof_3_opt.qc --net --post             # + neural block resynthesis + CX post-pass
+qoptct optimize bench_qc/tof_3.qc -o tof_3.qasm --mode balanced            # fewer T, CX <= 1.5x unitary CX
+qoptct optimize bench_qc/mod_red_21.qc -o m.qasm --mode t --clean          # minimum T, clean-ancilla model
+qoptct optimize X.qc --report report.json                                  # JSON report (also printed to stdout)
+```
+
+Python: `import qoptct; names, ref, best, how = qoptct.optimize("X.qc")`.
+
+| option | meaning |
+|---|---|
+| `--mode cx` (default) | unitary output (`.qc`), minimum CX; proven equal to the input, otherwise the input is returned |
+| `--mode balanced` | measurement-based output (OpenQASM 3) with lower T and CX ≤ `--cx-factor` (1.5) × unitary CX, if one is found and verified |
+| `--mode t` | lowest T found (measurement-based), CX may grow |
+| `--clean` | clean-ancilla model (see below) for the measurement-based modes |
+| `--net` | neural block resynthesis (models in `qoptct/models/`; needs `pip install "qoptct[net]"`) |
+| `--post` | CX post-pass on H-free regions (slow) |
+| `--meas` | older Gidney transformation (clean-ancilla model), kept for compatibility |
+| `--tgad`, `--vartodd` | Hadamard gadgets + TODD; `--vartodd` also tries an external VarTODD binary given by `QOPT_VARTODD_BIN` |
+
+Measurement-based outputs are written only if they pass `verify_qasm` (≤ 22 qubits in total); otherwise the unitary
+result is written. The JSON report says which candidate was used.
+
+## Qiskit plugin
+
+`pip install qoptct` registers two stage plugins for the Qiskit transpiler, both named `qoptct`:
+
+* `optimization_method="qoptct"` — Qiskit's own optimization stage of the chosen level, then `QoptPass` on the
+  Clifford+T parts of the (already translated and routed) circuit;
+* `init_method="qoptct"` (optional, recommended for Toffoli circuits) — `QoptPass` on the circuit as written,
+  while the Toffolis are still Toffolis, before Qiskit's own init stage. Most of the savings come from here.
+
+```python
+from qiskit import QuantumCircuit, transpile
+from qiskit.quantum_info import Operator
+
+qc = QuantumCircuit(5)                                                 # tof_3 from bench_qc/
+qc.ccx(0, 1, 4); qc.ccx(2, 4, 3); qc.ccx(0, 1, 4)
+
+basis = ["h", "x", "z", "s", "sdg", "t", "tdg", "cx"]                 # Clifford+T
+plain = transpile(qc, basis_gates=basis, optimization_level=3)
+ours = transpile(qc, basis_gates=basis, optimization_level=3,
+                 init_method="qoptct", optimization_method="qoptct")
+print(plain.count_ops(), ours.count_ops())
+assert Operator(ours) == Operator(qc)                                  # same operator, incl. global phase
+```
+
+CX/T after `transpile` with a Clifford+T basis (Qiskit 2.6.2, levels 1–3 gave the same numbers):
+
+| circuit | Qiskit level 3 | `optimization_method="qoptct"` | `+ init_method="qoptct"` |
+|---|---|---|---|
+| tof_3 | 18 / 19 | 16 / 15 | 12 / 15 |
+| mod5_4 | 28 / 22 | 25 / 16 | 11 / 8 |
+| barenco_tof_3 | 24 / 24 | 20 / 16 | 16 / 16 |
+
+A part of the circuit is replaced only if the new version is **proven** equal, strictly cheaper (CX, then T),
+and uses only gates and qubit pairs allowed by the basis / coupling map / target; everything else
+(measurements, rotations, other gates) is left as it is. With a coupling map, compare with
+`Operator.from_circuit(out)`. On circuits with up to 10 qubits per part the global phase is kept exactly.
+
+The pass can also be used directly:
+
+```python
+from qiskit.transpiler import PassManager
+from qoptct.qiskit_qopt import QoptPass
+p = QoptPass(); out = PassManager([p]).run(qc); print(p.last)   # runs, replaced, CX/T before and after
+```
+
+`QoptMeasPass(ancillas=[...])` is the measurement-based variant (logical AND, needs qubits known to start in |0>).
+
+## Cost models — read this before comparing numbers
+
+The same benchmark has different "best" numbers in different models. All tables say which model they use.
+
+| model | allowed in the output | input wires | ancillas | checked by |
+|---|---|---|---|---|
+| **unitary** | Clifford+T gates only, same wires | arbitrary state | none | full unitary / PyZX / numerical |
+| **strict** (measurement-based) | + fresh ancillas in \|0>, mid-circuit measurements, classically controlled Clifford gates | **all** `.qc` wires are data in an arbitrary state, including wires the benchmark intends as ancillas | fresh ones must end in a basis state, unentangled | `verify_qasm` (strict), `verify_big` |
+| **clean** (clean ancillas) | as strict | wires listed in the `.i` line are data; **the other `.qc` wires start in \|0> and must end in \|0>** (the intended use of e.g. `tof_n`, `mod_red_21`) | as strict | `verify_qasm(clean=True)` |
+
+* *unitary* is the model of most circuit-optimization papers (Nam et al., PyZX, GUOQ, Quarl, …).
+* *strict* is stronger than *clean*: a strict circuit is valid in the clean model, not vice versa.
+* In both measurement-based models the logical AND costs 4 T and its uncomputation 0 T (Gidney 2018); that is
+  why T counts there are far below unitary ones. They must not be compared with unitary T-count records.
+* Published H-gadget T-count records (TODD/FastTODD, AlphaTensor-Quantum) use yet another model (gadget ancillas,
+  no classical feed-forward on the output). They are listed for orientation only.
+
+## Results (23 benchmarks)
+
+| circuit | qubits | Toffolis | input T/CX | unitary T/CX | strict T/CX (+extra) | clean T/CX (+extra) |
+|---|---|---|---|---|---|---|
+| adder_8 | 24 | 57 | 399/409 | 193/223 | — | — |
+| barenco_tof_10 | 19 | 32 | 224/192 | 72/72 | 59/48 (+3) | 59/48 (+3) |
+| barenco_tof_3 | 5 | 4 | 28/24 | 16/14 | 8/9 (+2) | 8/9 (+2) |
+| barenco_tof_4 | 7 | 8 | 56/48 | 24/24 | 12/13 (+3) | 12/13 (+3) |
+| barenco_tof_5 | 9 | 12 | 84/72 | 32/32 | 16/17 (+4) | 16/17 (+4) |
+| csla_mux_3 | 15 | 10 | 70/80 | 64/64 | 40/70 (+1) | 28/84 (+3) |
+| csum_mux_9 | 30 | 28 | 196/168 | 84/112 | — | — |
+| gf2^4_mult | 12 | 16 | 112/99 | 68/86 | 58/115 (+0) | 36/79 (+1) |
+| gf2^5_mult | 15 | 25 | 175/154 | 115/132 | 90/183 (+0) | 56/124 (+1) |
+| grover_5 | 9 | 48 | 336/288 | 166/160 | 124/155 (+4) | 124/155 (+4) |
+| ham15-low | 17 | 23 | 161/236 | 97/179 | — | 68/177 (+1) |
+| mod5_4 | 5 | 4 | 28/28 | 7/10 | 4/11 (+1) | 4/11 (+0) |
+| mod_mult_55 | 9 | 7 | 49/48 | 35/39 | 20/35 (+2) | 12/24 (+1) |
+| mod_red_21 | 11 | 17 | 119/105 | 73/61 | 44/58 (+5) | 28/48 (+0) |
+| qcla_com_7 | 24 | 29 | 203/186 | 95/111 | 59/565 (+37) | 59/565 (+37) |
+| qcla_mod_7 | 26 | 59 | 413/382 | 237/256 | — | — |
+| qft_4 | 5 | 2 | 69/46 | 67/37 | 53/265 (+42) | 53/265 (+42) |
+| rc_adder_6 | 14 | 11 | 77/93 | 47/65 | 24/54 (+6) | 24/53 (+6) |
+| tof_10 | 19 | 17 | 119/102 | 71/54 | 59/71 (+3) | 36/37 (+1) |
+| tof_3 | 5 | 3 | 21/18 | 15/12 | 8/10 (+2) | 8/9 (+1) |
+| tof_4 | 7 | 5 | 35/30 | 23/18 | 12/15 (+3) | 12/13 (+1) |
+| tof_5 | 9 | 7 | 49/42 | 31/24 | 16/20 (+4) | 16/17 (+1) |
+| vbe_adder_3 | 10 | 10 | 70/70 | 24/28 | 12/33 (+3) | 12/33 (+3) |
+
+Best point per model: unitary = minimum CX; strict and clean = minimum T (then CX, then extra qubits). Other Pareto points (e.g. fewer extra qubits) are in results/TABLE.md.
+
+Totals over the circuits that have a result in the model: unitary: 23 circuits, T 3093→1656, CX 2920→1813; strict: 19 circuits, T 1924→718, CX 1725→1747; clean: 20 circuits, T 2085→671, CX 1961→1781
+
+Full tables with Pareto fronts and provenance: `results/TABLE.md`, machine-readable `results/table.csv`;
+circuits: `results/best/<model>/<name>.qasm` and `results/pareto/<model>/`.
+
+## Hardware check (IBM Quantum)
+
+On `ibm_marrakesh`, the unitary `qopt` circuits returned the correct answer more often than the input circuits on all
+five benchmarks tried (+2 to +10 percentage points), while the measurement-based circuits were worse on current
+hardware (mid-circuit measurement latency). Details: `results/hardware/README.md`.
+
+## Comparison with nine tools (summary; details in `results/COMPARISON.md`)
+
+Tools run on the same files: Qiskit 2.6.1, PyZX 0.10.8, pytket 2.18.5, Cirq 1.7, Q# Resource Estimator 1.31,
+Qualtran 0.7.0, pyLIQTR 1.0.1, Qrisp 0.9.10, PennyLane 0.45.1. All produced circuits were verified with the same
+checker; estimates without circuits are marked as such.
+
+| circuit | best tool circuit by T (as-is or MCX rewrite) | best tool circuit by CX | qopt unitary | qopt strict | qopt clean |
+|---|---|---|---|---|---|
+| barenco_tof_3 | 8/9 +2 (PennyLane 0.45.1, and_ladder; strict) | 8/9 +2 (PennyLane 0.45.1, and_ladder; strict) | 16/14 | 8/9 (+2) | 8/9 (+2) |
+| barenco_tof_4 | 12/13 +3 (PennyLane 0.45.1, and_ladder; strict) | 12/13 +3 (PennyLane 0.45.1, and_ladder; strict) | 24/24 | 12/13 (+3) | 12/13 (+3) |
+| barenco_tof_5 | 16/17 +4 (PennyLane 0.45.1, and_ladder; strict) | 16/17 +4 (PennyLane 0.45.1, and_ladder; strict) | 32/32 | 16/17 (+4) | 16/17 (+4) |
+| csla_mux_3 | 62/161 +0 (PyZX 0.10.8, full_reduce; strict) | 64/72 +0 (Qiskit 2.6.1, as-is; strict) | 64/64 | 40/70 (+1) | 28/84 (+3) |
+| gf2^4_mult | 68/274 +0 (PyZX 0.10.8, full_reduce; strict) | 96/99 +0 (Qiskit 2.6.1, as-is; strict) | 68/86 | 58/115 (+0) | 36/79 (+1) |
+| gf2^5_mult | 115/509 +0 (pytket 2.18.5, zxperm_fpo; strict) | 155/154 +0 (Qiskit 2.6.1, as-is; strict) | 115/132 | 90/183 (+0) | 56/124 (+1) |
+| grover_5 | 166/301 +0 (PyZX 0.10.8, full_reduce; strict) | 296/288 +0 (Qiskit 2.6.1, as-is; strict) | 166/160 | 124/155 (+4) | 124/155 (+4) |
+| ham15-low | 97/322 +0 (PyZX 0.10.8, full_reduce; strict) | 147/236 +0 (Qiskit 2.6.1, as-is; strict) | 97/179 | — | 68/177 (+1) |
+| mod5_4 | 8/20 +0 (PyZX 0.10.8, full_reduce; strict) | 8/20 +0 (PyZX 0.10.8, full_reduce; strict) | 7/10 | 4/11 (+1) | 4/11 (+0) |
+| mod_mult_55 | 35/89 +0 (PyZX 0.10.8, full_reduce; strict) | 45/48 +0 (Qiskit 2.6.1, as-is; strict) | 35/39 | 20/35 (+2) | 12/24 (+1) |
+| mod_red_21 | 73/149 +0 (PyZX 0.10.8, full_reduce; strict) | 107/105 +0 (Qiskit 2.6.1, as-is; strict) | 73/61 | 44/58 (+5) | 28/48 (+0) |
+| qft_4 | 67/55 +0 (PyZX 0.10.8, full_reduce; strict) | 409/45 +0 (Qiskit 2.6.1, as-is; strict) | 67/37 | 53/265 (+42) | 53/265 (+42) |
+| rc_adder_6 | 47/103 +0 (PyZX 0.10.8, full_reduce; strict) | 67/83 +0 (Qiskit 2.6.1, as-is; strict) | 47/65 | 24/54 (+6) | 24/53 (+6) |
+| tof_10 | 36/37 +1 (PennyLane 0.45.1, and_ladder; clean) | 36/37 +1 (PennyLane 0.45.1, and_ladder; clean) | 71/54 | 59/71 (+3) | 36/37 (+1) |
+| tof_3 | 8/9 +1 (PennyLane 0.45.1, and_ladder; clean) | 8/9 +1 (PennyLane 0.45.1, and_ladder; clean) | 15/12 | 8/10 (+2) | 8/9 (+1) |
+| tof_4 | 12/13 +1 (PennyLane 0.45.1, and_ladder; clean) | 12/13 +1 (PennyLane 0.45.1, and_ladder; clean) | 23/18 | 12/15 (+3) | 12/13 (+1) |
+| tof_5 | 16/17 +1 (PennyLane 0.45.1, and_ladder; clean) | 16/17 +1 (PennyLane 0.45.1, and_ladder; clean) | 31/24 | 16/20 (+4) | 16/17 (+1) |
+| vbe_adder_3 | 24/71 +0 (PyZX 0.10.8, full_reduce; strict) | 56/58 +0 (Qiskit 2.6.1, as-is; strict) | 24/28 | 12/33 (+3) | 12/33 (+3) |
+
+Each tool cell: T/CX extra-qubits (tool, variant; model in which the circuit passed verification).
+
+Honest caveats:
+
+* **Given the gate list "as is", no tool replaces Toffolis by logical ANDs with measurement-based uncomputation
+  automatically.** Unitary tools give ~7 T per Toffoli minus phase folding; Q#/Qualtran give only estimates
+  (4 T per Toffoli) without a circuit.
+* **When a human rewrites the circuit as a multi-controlled X** (tof_n, barenco_tof_n), tools reach our numbers or
+  beat them: PennyLane's `TemporaryAND` ladder equals our clean-model tof_n points; on barenco_tof_n Qiskit's
+  Khattar–Gidney syntheses with one ancilla were better than our general compiler until the MCX recognition
+  (`contrib/c12`) was added. Our advantage is automatic discovery in arbitrary networks, not better gadgets.
+* Third-party pitfalls and one wrong circuit of a tool are documented in `docs/FINDINGS.md`.
+
+## How to check the results yourself
+
+```bash
+python scripts/verify_all.py --jobs 2          # re-verifies all 200 rows of results/table.csv (~40 min)
+python scripts/verify_all.py --only tof_3       # one circuit
+sh scripts/reproduce.sh cx out                  # re-run the optimizer on all benchmarks (see script header)
+```
+
+Every qasm in `results/` is self-contained OpenQASM 3 using `h x z s sdg t tdg cx cz`, `measure`, and `if` on
+classical bits; data qubits are `q[0..n-1]` in the order of the `.v` line of the `.qc` file, extra qubits follow
+and start in |0>.
+
+## Limitations
+
+* The checks are randomized: `verify_qasm` uses 30 random input states (10 for 20–22-qubit circuits) and sampled
+  measurement outcomes; `verify_big` samples basis inputs and branches. A wrong circuit could pass only if it acts
+  correctly on all samples. Unitary results ≤ 10 qubits are proven exactly; larger ones rely on PyZX or a
+  numerical check.
+* Measurement-based modes handle at most 22 qubits in total (state-vector check). Larger outputs are discarded.
+* The neural models were trained on random circuits of 2–4 qubits; `--net` gives small gains (≈ 1–2 % CX).
+* Runtime: unitary mode takes seconds for small circuits and up to an hour for the largest benchmarks; the
+  measurement-based modes take minutes. Several rewriting programs have internal time limits, so results can
+  depend on machine speed.
+* Some numbers in `results/` come from longer development searches than a single default run (column `source`
+  in `results/TABLE.md`); a default run may give slightly worse points.
+* The hand-made clean-model constructions (`contrib/c1/best_clean`) apply only to the exact benchmark files.
+* hwb6 is not included (license of the benchmark file).
+
+## Citation
+
+See `CITATION.cff`. A preprint draft is in `paper/preprint.md`.
+
+## Acknowledgements
+
+Benchmarks from the Feynman repository (M. Amy). Logical AND construction: C. Gidney, *Halving the cost of quantum
+addition*, Quantum 2, 74 (2018). Sum-over-paths verification: M. Amy, arXiv:1805.06908.