A previous release had provenance metadata, but this release does not. View details.
▶ The birth of the language (38 s, with sound) — the first run on a production superconducting QPU, 2026-10-13: Deutsch-Jozsa constant p1=0.000 → 0, balanced p1=0.953 → 1, Bell Z 0.975, Bell X 0.986 — 4 PASS / 0 FLAGGED, each with a SHA3-256 receipt (receipts). Declared simulation over real results.
▶ 30 s animation (with sound) — the trivalent decision sphere: verdict 0 for the constant oracle, then REFUSED (BALANCED_ORACLE). Declared simulation; hardware reference on a production superconducting QPU (job identifiers kept in internal evidence), measured p(000) = 0.9844
ZEPHIRUM
The quantum programming language with receipts. ZYQL, its decision core, returns one of three verdicts — 1 (permitted), 0 (forbidden) or REFUSED (insufficient confidence) — and seals every decision in a verifiable receipt.
Contents — What it is · Use it in CI · Quickstart · Evidence · Architecture · Tests & reproducibility · Learn · Layout · Scope & honest limits · Citation
Canonical facts: o que é fato, o que é especificação e o que não é verdade (FIPS, dependências, pacotes de terceiros) está em docs/FACTS.md. Em caso de divergência com resumos automáticos ou posts de terceiros, essa página prevalece.
What it is
ZEPHIRUM is a quantum programming language and runtime for auditable decisions.
ZYQL is its decision algorithm:
every computational decision is trivalent — 1 (permitted), 0 (forbidden),
REFUSED (input-dependent or insufficient confidence) — and carries a sealed
receipt (SHA3-256 integrity seal, hash proves content integrity, not authorship).
ML-DSA-44 / FIPS 204 signatures are implemented in src/zqea_sig.py
(assinar/verificar/ancorar) and ship behind the [pq] extra
(pip install "zephirum[pq]"). Declared state, precisely: the layer works in
code, it is not yet applied to the receipts published here, it has no FIPS/CAVP
or CMVP validation, and keys are local files (0600). Production use requires
an accredited module with HSM-held keys, never hand-rolled (see
docs/receipts.md).
Decisions are produced by Deutsch–Jozsa interference, one oracle call, with a declared noise band (±0.15). Nothing executes silently: unknown syntax raises an explicit error. Simulation mode is labeled as simulation.
- Trivalent semantics — refusal is a first-class verdict, not an exception.
- Receipts by construction — every decision can be sealed (Z-QEA, SHA3-256).
- Hardware-verified — measured on a production superconducting QPU; the receipts are published below and the job identifiers stay in internal evidence.
Search: "ZYQL" quantum language · Zephirum receipts · trivalent REFUSED · zephirum on PyPI
Use it in CI (GitHub Action)
The decision gate is a composite action: no secrets, no network, stdlib Python
only. It reads an evidence payload you already measured, applies the declared
proportion rule with a Hoeffding bound, writes a receipt, and fails the job
according to fail_on.
name: decision
on: [push]
jobs:
gate:
runs-on: ubuntu-latest
steps:
- uses: SerenaSolutions/zephirum-lang@v0.4.0
id: gate
with:
shots: 2048
counts: '{"000": 1975, "111": 73}' # measured evidence, not a claim
threshold: 0.90
delta: 0.05
- run: echo "receipt written to $RUNNER_TEMP/zyql_receipt.json"
| input | required | meaning |
|---|---|---|
shots |
yes | number of samples in the evidence payload |
counts |
yes | observed counts as JSON; must sum to shots, otherwise the gate returns REFUSED with reason CONTRATO_INVALIDO |
threshold |
yes | proportion the evidence must beat, strictly between 0 and 1 |
delta |
no (0.05) |
tolerated error probability in the Hoeffding bound |
fail_on |
no (0,REFUSED) |
verdicts that fail the job, comma separated |
Outputs: verdict (1, 0, REFUSED or INCONCLUSIVE) and receipt (path to
the sealed receipt). Both are exported from the release that ships this section;
on v0.4.0 the receipt is still written to $RUNNER_TEMP/zyql_receipt.json, so
read that file directly.
The rule is declared, not implied: with p_hat = counts[target] / shots and
eps = sqrt(ln(2/delta) / (2*shots)), the verdict is 1 when p_hat - eps > threshold,
0 when p_hat + eps < threshold, and INCONCLUSIVE otherwise. The receipt
records p_hat, eps, the rule version, and an input commitment, so the decision
can be re-derived from the numbers alone.
One semantic note, stated so nobody has to discover it: INCONCLUSIVE does not
fail the job by default, because inconclusive evidence is a legitimate state. If a
pipeline must be decisive to merge, add it explicitly:
fail_on: '0,REFUSED,INCONCLUSIVE'.
Quickstart
Two paths, both tested on 2026-11-09 with Python 3.11 from a clean virtualenv.
If you only want to use it (package published on PyPI, 9 s install, 13 packages):
pip install zephirum
from zyql_quantum_gate import ZYQLQuantumGate
# constant oracle: the decision comes out
print(ZYQLQuantumGate(3, lambda x: 0).run_quantum(shots=2048)["decision"]) # 0
# balanced oracle: the language refuses to promote noise to a conclusion
r = ZYQLQuantumGate(3, lambda x: x ^ 1).run_quantum(shots=2048)
print(r["decision"], r["refusal_reason"]) # REFUSED INSIDE_NOISE_BAND
REFUSED is the point, not a failure: it means the shots did not separate the
hypotheses well enough to justify a claim.
If you want to develop or reproduce the research (needs the repository):
git clone https://github.com/SerenaSolutions/zephirum-lang.git && cd zephirum-lang
pip install -e .
zyql-decide --oracle balanced --shots 512 # CLI runtime, Hoeffding error bound
python -m zyql_lisp examples/hello_world.zph # language front-end + sealed receipt
The baptism of the language — a trivalent greeting, sealed by ZEPHIRUM itself:
;; hello_world.zph — ZEPHIRUM v2 (Lisp)
(hello "Zephirum")
(seal "examples/hello_world.receipt.json")
$ python -m zyql_lisp examples/hello_world.zph
Hello, Zephirum! | decisao = 1
SELO: <sha3-256 of the receipt — timestamp bound, varies per execution>
The hello form prints the greeting and returns the trivalent verdict 1
(a greeting is always permitted). The seal form writes a Z-QEA receipt
(SHA3-256) of everything executed. Source: examples/hello_world.zph.
Evidence
Real-data figure: theoretical braid curves (gold: B_R∘B_S, violet: B_S∘B_R) and the points measured on a production superconducting QPU (vendor-neutral) — 98.1% vs 95.8%. The braid order is distinguishable on silicon. Full detail: docs/PROOFS.md.
Hero art (declared: AI-generated) — braided world-lines of non-Abelian anyons converging into three fusion channels, the geometry the experiments probe.
Proofs of concept — non-Abelian anyons (S3 qutrit miniature)
| Claim | Mode | Result | Receipt |
|---|---|---|---|
| Fusion space is 3-dimensional (qutrit in 2 qubits) | noiseless | exact 1/3+1/3+1/3, forbidden channel vetoed | sealed receipt |
| Braiding is non-Abelian (order matters) | noiseless | commutator L1 = 6 exact; B_R∘B_S → ch2, B_S∘B_R → ch1 | same |
| Noiseless round | noiseless | exact-zero deviation vs theory (canonical rationals) | same |
| ZYQL trivalent decisions 1 / 0 / REFUSED | a production superconducting QPU | 3/3 PASS (job identifiers kept in internal evidence records) | battery receipt |
| Braid order measured on real hardware | a production superconducting QPU | 98.1% vs 95.8% (job identifiers kept in internal evidence records) | master receipt |
| Shots-to-verdict efficiency | a production superconducting QPU | 256 shots to a first-step verdict — 96,88% fewer shots contra a linha de base de 8192 (linha de base declarada: varredura exaustiva com 8192 tiros, não um padrão externo) | receipt 28f406202e9d125a (issue #40) |
Receipt seal (SHA3-256): a44192463ea669564a3da51728c8577e59e5684c67cf2d48f7a95442d548476a — verify it yourself: python3 .github/actions/zqea-verify/zqea_verify.py experiments/s3_qutrit_final_receipt.json
In language: examples/s3_qutrit.zph.
Verified on real hardware (a production superconducting QPU, cited as test laboratory, vendor-neutral, with job identifiers retained in internal evidence; no endorsement claimed):
- Constant purpose: measured p(000)=0.9844 (job identifiers kept in internal evidence records;)
- S3 qutrit miniature: fusion channels (qutrit-in-qubits) + braiding non-commutativity,
noiseless with exact-zero deviation and on hardware: ch2=98.1% (RS) vs ch1=95.8% (SR),
full ZYQL battery 3/3 PASS on the lab device (job identifiers kept in internal evidence records;
Z-QEA receipt sealed, see
experiments/) - Balanced purpose: p(000)=0.0000, orthogonal outcome 93.9%
- Same-task confrontation against an industry-standard quantum SDK (identifiers in internal evidence): Zephirum transpiles to depth 2 / 0 two-qubit gates vs depth 10 / 3 CZ
Benchmark suite: benchmarks/ (public comparative benchmark, issue #28).
Architecture
program (.zph, Lisp) ──► front-end parser ──► canonical decision form
│ │
│ Deutsch–Jozsa interference
│ (one oracle call, noise band ±0.15)
│ │
│ trivalent verdict: 1 / 0 / REFUSED
│ │
└──────────────► Z-QEA receipt ◄─────────────┘
(SHA3-256 seal + machine-readable refusal_reason)
The repository ships the language heart only: kernel (state-vector, [re,im]
float64), OpenQASM 3 transpiler, trivalent gate, quantum types (no-cloning
enforced) and the receipt chain. See Honest scope.
The language:
circuit bell {
qubit[2] q;
H q[0];
CNOT q[0] -> q[1];
}
Tests & reproducibility
- 33 test suites (
tests/): canonical rational battery, certified decisions, frontiers, KEM, Lisp regression, meta, multiple-testing correction, runtime, shot auditor, Z-QEA verifier, hardware rounds. - CI matrix: Python 3.10 and 3.11 on every push and pull request, with
PYTHONHASHSEED=0for reproducible runs; CodeQL static analysis; signed releases with build provenance (SLSA attestation); a dedicated ZQ-GATE self-test workflow. - Statistical discipline: multiple-comparison correction is part of the test battery — verdicts are not claimed on uncorrected p-values.
- Verifiable by construction: every hardware claim above points to a sealed
receipt, with job identifiers retained in internal evidence; receipts are
re-verifiable with
.github/actions/zqea-verify/zqea_verify.py. - Published testing roadmap (fuzzing the front-end, unitary-equivalence assertions, executed notebooks, coverage gate): tracked in issues.
Learn
- docs/LEARN.md — progressive tutorial: first decision to sealed hardware receipts
- docs/COOKBOOK.md — copy-paste recipes, every one executed and recorded
- docs/MIGRATION.md — embedding Zephirum in Python, C and CI pipelines
- docs/PROOFS.md — hardware proofs, figures and receipts
- docs/MATURITY.md — component-by-component maturity
Layout
src/— kernel (state-vector), OpenQASM 3 transpiler, trivalent gate, quantum types (no-cloning enforced), neural RAG layer, runtime (zyql_runtime)tests/— 33 suites (see above)examples/—hello_world.zph,bell.zph,grover.zph,dj_constant.zph,s3_qutrit.zphbenchmarks/— public comparative benchmark suitenotebooks/— scientific presentation layerexperiments/— hardware rounds and sealed receipts (job identifiers in internal evidence)docs/— language reference, receipt specification, proofs, maturity, dossierplayground/— browser playground assets ·site/— docs site ·paper/— preprint sourceszephirum-vscode/— VS Code extension for.zphdocker/— reproducible container
Technical Dossier
A consolidated, receipt-cited dossier of the project — what it is, what has been measured on real quantum hardware, the public m50 commitment, Industry 5.0 applications and the declared honest limits: docs/DOSSIER.md
Self-verified device decision (new, 2026-10-29)
A decisão trivalente agora roda na QPU com a própria QPU certificando a execução:
- Auto-verificação (espelho C†) no mesmo job — taxa de zeros medida vira
selfcheck_p0no recibo; abaixo do limiar ⇒REFUSED. - Evidência sequencial com regra de parada (alpha-spending) declarada antes da execução.
- Recibo com protocolo + contagens por lote +
job_id, selado SHA3-256.
| Oráculo | Veredito | p0 | selfcheck | Shots | job (internal) |
|---|---|---|---|---|---|
| constante | 1 | 0.992 | 0.969 | 128 | internal |
| balanceado | 0 | 0.016 | 0.992 | 128 | internal |
Especificação completa: docs/SELFVERIFIED.md · recibos em experiments/selfcheck_hw_20261029/.
Refusal witness — abstenção com prova (new, 2026-10-29)
REFUSED deixa de ser "não sei" e vira "não sei porque a evidência não fecha":
a TVD entre a distribuição ideal (referência exata) e a medida no dispositivo tem
intervalo de confiança declarado antes da execução.
| Dispositivo | Shots | TVD | Janela | ε | Veredito | job_id |
|---|---|---|---|---|---|---|
| lab device A | 4096 | 0.0317 | [0.000, 0.104] | 0.10 | REFUSED | internal |
| lab device B | 8192 | 0.0198 | [0.000, 0.071] | 0.10 | 1 CERTIFICADO | internal |
O primeiro voo absteve; com o dobro de evidência, certificou. Especificação:
docs/WITNESS.md · recibos em experiments/witness_20261029/.
Hardware protocols (2026-10-29)
Protocolos pré-declarados, executados no plano Open com recibo selado
(docs/PROTOCOLS.md):
| Protocolo | Medida | Resultado | job_id |
|---|---|---|---|
| P-10 fronteira clássica×hardware | TVD vs ideal, n=3..5 | 0.0139 / 0.0188 / 0.0183 | internal |
| P-08 ZNE vs C† | p0 n=4, com/sem ZNE | 0.0197 → 0.0116 (−41%) | internal |
Apps v1 (2026-10-29)
- AuditKit (CPA-01) — certificacao de entropia quantica com recibo: hardware a second lab device, 4096 bits, veredito 1 (identifiers in internal evidence).
- Quantum Regression Seal (QRS-01) — gate de CI/CD que pega miscompilacao silenciosa por comparacao unitaria; action reutilizavel no repo.
Detalhes: docs/APPS.md
App animations (30 s / 38 s, with sound)
CPA-01 AuditKit — entropy certification, 4096 bits on a second lab device (job in internal evidence), verdict 1 ·
QRS-01 Regression Seal — CI/CD gate, identity 1 / phase bug 0 / gate bug 0 ·
Q-ROUTE — live trivalent routing ·
QSR-01 — Mosca inequality: RSA-2048 0, X25519 1, ML-KEM-768 1 ·
Self-verified (C†) — the QPU certifies its own run, selfcheck_p0 0.969 ·
Z-KEM — ML-KEM-768 + AES-256-GCM envelope ·
S3 anyons — braiding measured on a production QPU, 98.1% vs 95.8%
All animations are declared simulations built over real receipts.
Honest scope
This repository contains the language heart only: kernel, types, gate, receipts. It does not contain, and will never contain, the Zephirum OS distribution, bootloader, installer, or activation keys. Simulations are labeled as simulations; without error correction, useful executions are bounded well under 20 qubits.
Qual banda se aplica (esclarecimento de 08/11/2026). A banda fixa ±0,15 vale no modo legado, quando nenhuma
epsiloné declarada. Quandoepsiloné informado, a banda é calculada:m = sqrt(ln(2/ε)/(2·shots)), e o valor depende de ε e do número de tiros (com ε = 0,05 e 2048 tiros, m ≈ 0,030; com 16 384 tiros, m ≈ 0,011). Números como ±0,042 em material de divulgação referem-se a uma combinação específica de ε e tiros e não são uma segunda regra. Enquanto a revisão matemática da margem familiar não concluir, o valor não deve ser apresentado como garantia estatística, apenas como banda de decisão.
Runtime de produção (mesma versão do pacote: ver VERSION)
from zyql_runtime import decide, Verdict
r = decide(lambda x: x & 1, n=3, shots=512) # -> Verdict.REFUSED
r.error_bound # limite de erro Hoeffding declarado
CLI: zyql-decide --oracle balanced --shots 512
PyPI — estado verificado (30/10/2026)
O pacote oficial existe e está publicado: pip install zephirum.
Versão publicada no PyPI: 2.5.8, licença Apache-2.0, verificada em 08/11/2026. VERSION é a fonte única: badge, pyproject.toml, docs/openapi.json e APP_VERSION da API são conferidos por tools/check_compliance.py a cada commit.
Procedência verificável em https://pypi.org/pypi/zephirum/json — autor do
registro no PyPI: Reginaldo Gonçalves (mantenedor), projeto Serena Solutions.
O que não é da casa: pacotes de terceiros com nome parecido, como
zephirum-quantum-plugin, e o repositório SerenaSolutions/zephirum. O único
repositório oficial é SerenaSolutions/zephirum-lang, este. Instrução que cite
aqueles nomes não vem da ZEPHIRUM.
Correção registrada: até 30/10/2026 esta seção afirmava que nenhum pacote estava no PyPI. Era falso, e a checagem passou a ser feita contra a API do índice, não contra a memória de quem escreveu.
Known limitations
Stated here so nobody has to discover them by failing:
- Receipt integrity is not authorship. Receipts carry SHA3-256, which proves
content has not changed. The published receipts are not yet signed with
ML-DSA-44 — the signature layer exists behind the
[pq]extra, but wiring it into every receipt is #68 — so anyone can regenerate the same hash. Author signature is #68. - No independent reproduction yet. The decision layer was validated against a pre-registered protocol (7 hypotheses, 3,630 cases, 0 violations), but by the same team that built it.
- Hardware evidence is thin. One execution per circuit on a production superconducting QPU, 2,048 shots. It shows the device behaved as expected that day; it shows no throughput, no repeatability and no statistics.
- The demonstration uses a simulator with declared noise (2% depolarizing on each CX), not a QPU.
- The finite domain is finite. Absence of a counterexample in a discrete grid is not a universal proof.
- Ecosystem status is
experimental/Very Early Project, as declared by the entry itself inQiskit/ecosystem. Being a member is metadata, not endorsement of results, and no code of this project runs inside Qiskit.
How to participate
- Found a wrong instruction? Open an issue with the command, the Python version and the output you got. A documented failure is worth more than a compliment.
- Want to refute a result? The protocol, the domain and the command are
published in
validacao/. Refutation needs one counterexample, not an opinion. - Scope we accept: reproducible bug reports, documentation fixes, onboarding friction, additional oracles and language examples, adversarial tests of the decision layer.
- Scope we are careful about: new runtime dependencies. The project keeps a
deliberately small dependency set (
lark,numpy,qiskit). - Good first issues are labelled in the issue tracker.
Citation
See CITATION.cff. Software deposit: Zenodo concept DOI 10.5281/zenodo.23632019
(code Apache-2.0, companion thesis CC BY-NC-ND 4.0), (c) Serena Solutions.
License
Apache-2.0 — see LICENSE. A licença inclui concessão explícita de patentes, necessária em computação quântica; a tese companheira permanece CC BY-NC-ND 4.0.
ZEPHIRUM™ e ZYQL™ sao marcas de Serena Solutions. O simbolo (betta quantico) e a identidade visual sao de uso reservado. O codigo e licenciado sob Apache-2.0, que nao concede direito de uso das marcas (Apache-2.0, secao 6). Veja NOTICE.
Metadata
Release files for zephirum 2.5.9
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Loss of provenance
These files do not have provenance metadata, despite files in a prior release including this data.
We recommend further investigation before installing these files. View details.
Source distribution (sdist)
| File | Size | Uploaded | Provenance | Actions |
|---|---|---|---|---|
| zephirum-2.5.9.tar.gz | 156.2 kB | Without provenance | Details for zephirum-2.5.9.tar.gz |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Actions Reset |
|---|---|---|---|---|
| zephirum-2.5.9-py3-none-any.whl | Python 3 | none | any | Details for zephirum-2.5.9-py3-none-any.whl |
Total release size: 263.3 kB
Release files / zephirum-2.5.9.tar.gz
Loss of provenance
This file belongs to a release without provenance metadata, despite a prior release including this data.
We recommend further investigation before installing this file. View details.
| Download URL | zephirum-2.5.9.tar.gz |
|---|---|
| Size | 156.2 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
86b129dbb7f1e48b53af56324a00d6425d0454f055057e92322db46421dd45e3
|
|
BLAKE2b-256 checksum How to use checksums |
580ed6417f9fd09d53e00d43dad90ab4352e08c1632f63b2282b33fad96ebc33
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/7.0.0 CPython/3.11.16
|
Release files / zephirum-2.5.9-py3-none-any.whl
Loss of provenance
This file belongs to a release without provenance metadata, despite a prior release including this data.
We recommend further investigation before installing this file. View details.
| Download URL | zephirum-2.5.9-py3-none-any.whl |
|---|---|
| Size | 107.2 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
77ab350b00cee90d7fed55b784b9a8df763182885fb4ae250fad640004e84229
|
|
BLAKE2b-256 checksum How to use checksums |
27a194ac3b56b38543f47fa9aceb5e388996f3fea72030f21b941c16d65e1fa5
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/7.0.0 CPython/3.11.16
|
Security for zephirum 2.5.9
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
Loss of provenance
A previous release of these files had provenance metadata, but this release does not.
While this could reflect a routine change by the project maintainers, it may also indicate that the release files have been compromised.
Release comparison
- Removed 2.5.8 2 files With provenance
- Added 2.5.9 2 files Without provenance
A note on package security
PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.
To learn more about protecting yourself from malicious actors, visit our package security documentation.