equiframe-benchmark 0.12.4__tar.gz

equiframe_benchmark-0.12.4/CHANGELOG.md

@@ -0,0 +1,317 @@
+## 0.12.4 - portable raw-metadata regression
+
+- Keep the paired release-gate negative fixture distinct from the candidate
+  version. The local 0.12.3 attempt retains its original checks and was not uploaded.
+- Construct explicit LF and CRLF PKG-INFO fixture bytes and verify their distinct
+  raw SHA-256 identities on each platform. Keep production metadata hashing raw;
+  do not normalize line endings or execute a build backend to inventory metadata.
+- Pair exactly with core 0.12.4 and admit its producer in the active schemas,
+  preserving historical schemas, fixtures and scientific outcomes.
+- Require new artifact-bound release checks. Previous 0.12.2 results remain tied
+  to their original artifacts and do not establish this candidate's validation
+  or publication.
+
+## 0.12.2 — paired tagged release
+
+- Pair exactly with core `equiframe 0.12.2`, using Git-derived release versions.
+- Inspect Git-derived versions using the identified wheel or matching `PKG-INFO`
+  metadata, without running a build backend or project code.
+- Preserve previous producer identities while admitting the new producer in
+  current schemas; historical fixtures and scientific outcomes remain unchanged.
+- Use the same reviewed-artifact security and manual publication process as core.
+  The benchmark APIs and scientific protocols retain their Beta status.
+
+## 0.12.0 — exact core pairing and public distribution preparation
+
+- Pair exactly with core `equiframe 0.12.0` in base, engines and ML extras.
+- Reject special files and links during cost-artifact verification instead of
+  blocking on a FIFO; preserve unavailable and mismatched artifact states.
+- Admit the new detector producer identity in current schemas while preserving
+  all historical fixtures and supported producer versions.
+- Include MIT licensing, notices and the same factual manual release checks
+  as the core. The benchmark's scientific protocols and APIs remain experimental;
+  release does not establish accepted research results or external adoption.
+
+## 0.12.0a14 — private Alpha pairing
+
+Pair with core A14's optional EDA. No detector training, thesis metrics or
+historic evidence rewritten. Existing V1 benchmark documents and APIs remain
+unchanged; the V2 additions below are explicit opt-in interfaces.
+
+### Paired principal protocol V2 — source development of 26 October 2026
+
+- Add protocol `paired-principal-2.0` with schema version `2.0`. Each declared
+  method contrast uses its own paired evaluable support with confirmed O;
+  the complete R/M/B/H intersection remains an additional reported support.
+  Preserve the three primary families crossed with all three engine pairs.
+- Add `EvaluationConfigV2`, `EvaluationReportV2`, `ContrastSpecV2`,
+  `FreezeDefinitionV2`, `FreezeManifestV2`, `ProtocolReviewV2`, `StudyResultV2`
+  and `TableExportV2`, with explicit `_v2` evaluation, freeze, study, table,
+  bundle and regeneration entry points. Keep V1 formats and fingerprints intact.
+- Bind final protocol adoption to an explicit review of its exact fingerprint.
+  Keep uncertainty paired by whole groups and preserve undefined metrics,
+  exclusions and unavailable methods without inventing zeroes or labels.
+- Document the new workflow in [PROTOCOL_V2.md](PROTOCOL_V2.md). This change
+  implements the evaluation machinery; it does not approve a protocol, train a
+  model, supply reviewed corpus labels or establish scientific results.
+
+- Support Windows environments created from another virtual environment. Validate
+  up to 16 creator configurations, requiring the current interpreter's base and
+  rejecting cycles, duplicate keys, relative creators, mismatched homes and large
+  or linked configurations. Never execute a configured creator or an unbound command.
+- Preserve rejection of missing scale finalization with safe OS error classes and
+  keep missing-directory I/O recovery separate. Refuse linked operation roots
+  and special inventory members before accepting stored measurements.
+- Extract the synthetic UTC date fixture with Pandas' native microsecond timezone
+  dtype, preserving NULL and date32 for empty outputs without requiring Arrow's
+  separate Windows C++ timezone database. The literal expected dates stay unchanged.
+- Construct the fixture's native UTC Series directly with an explicit
+  microsecond dtype, supported by older and recent pandas stubs under strict
+  Python 3.10 type checking. Preserve the existing precision, timezone, input
+  values and expected dates.
+- Preserve failed native integration evidence when an expected output is supplied.
+  Mark that expected comparison unavailable instead of recording a comparison that
+  never captured its right-hand artifact; independent healthy comparisons remain.
+
+## 0.12.0a13 — package quality and producer identity (2026-10-10)
+
+- Remove a process-test race by capturing OS descendant identities before real
+  cleanup instead of requiring the terminated worker to finish writing a file.
+  Check every captured descendant stops; supervisor behavior remains unchanged.
+
+- Correct detector producer identity: new evidence identifies A13 instead of
+  retaining A11 when executed from A12. Preserve explicit historical producer
+  records and admit A12/A13 without changing `equiframe-if-mad-1.0`, R/M/B/H/O,
+  model parameters, training support or expected outputs.
+- Regenerate current producer-containing schemas and their asset hashes;
+  preserve all historical fixture and receipt bytes. Keep the exact paired
+  core dependency and align current installation guides.
+- Add the source-carried standard-library offline wheel integrity audit and
+  tests for corruption, unsafe paths, Unicode metadata and occupied receipts.
+  Fix strict typing of heterogeneous observations in the option-trace tool.
+- Public distribution, license approval, ecosystem adoption and full platform
+  acceptance remain independent pending gates; no Actions or paid resources.
+
+## 0.12.0a12 — exact core onboarding candidate pairing (2026-10-10)
+
+- Pair the benchmark with exactly core A12 after adding core installation
+  diagnosis and role demos. Dependency ranges and scientific algorithms remain.
+- Previous universality/benchmark receipts retain their A11 identities; this
+  release does not claim new scientific or human-adoption results.
+
+## 0.12.0a11 — exact core candidate pairing (2026-10-09)
+
+- Require the exact new private core A11 candidate for all Arrow/engine/ML extras.
+- Use the same sdist-shipped wheel ZIP creator convention on Windows and Linux.
+- Regenerate schemas containing the new producer identity. Scientific algorithms,
+  independent detector paths and historical expected results remain unchanged.
+- This metadata pairing does not establish a new native universality matrix or
+  independent human acceptance; manual artifact-specific validation is required.
+
+## 0.12.0a10 — remediación del laboratorio de universalidad (2026-10-09)
+
+- Require binary-compatible candidates during minimum/latest benchmark resolution,
+  matching the existing binary-only hash-locked synchronization step.
+- Preserve dependency ranges, native benchmark behavior and scientific protocols.
+- Keep A9 receipts historical; they do not establish A10 support.
+
+## 0.12.0a9 — candidato de universalidad (sin publicación pública)
+
+Nueva identidad distribuida para herramientas/guías revisadas y dependencia directa de typing-extensions en benchmark. A8 permanece histórico. La matriz de veinte celdas requiere recibos nativos; preparación no equivale a soporte observado.
+
+- Target CPython 3.10-3.13 without discarding existing native extras or semantic gates
+- Replace post-3.10 syntax/std-library requirements with typed equivalents; preserve named aliases, historical formats and nanosecond/Decimal semantics
+- Resolve dependency branches by interpreter and propagate the chosen Python through validation tools
+- Preserve the paired native pilot, resource supervisor and benchmark public profiles across the selected interpreters
+- Keep the candidate private Alpha. VM/platform receipts, rights, governance, security coverage and actual independent acceptance remain explicit gates
+
+## 0.12.0a8 - Python compatibility maturity candidate (2026-10-08)
+
+# Private benchmark alpha 0.12.0a7 — 2026-10-07
+
+- Require exactly `equiframe[arrow]==0.12.0a7` and matching engine/ML extras
+- Identify rebuilt artifacts after the minimum-Polars expression composition
+  correction; preserve the benchmark algorithms and historical producer readers
+
+No scientific method, detector independence or evidence format changes.
+
+# Private benchmark alpha 0.12.0a6 — 2026-10-07
+
+- Require exactly `equiframe[arrow]==0.12.0a6` and matching engine/ML extras
+- Identify rebuilt artifacts after the core unsupported-input guard correction;
+  retain benchmark algorithms and the A4 integrity-reader controls unchanged
+
+No scientific method, detector independence or output-model changes.
+
+# Private benchmark alpha 0.12.0a5 — 2026-10-06
+
+- Require exactly core `equiframe[arrow]==0.12.0a5` and matching engine/ML extras
+- Retain the benchmark algorithms and reader controls; identify rebuilt artifacts
+  separately after the core native literal-column-name correction
+- Preserve A4 installation, test and consumer receipts as historical observations
+
+No scientific method or detector independence changes.
+
+# Private benchmark alpha 0.12.0a4 — 2026-10-06
+
+- Enforce the declared boolean type of integrity-verification controls rather
+  than treating numerically equal values such as `verify=0` as an explicit opt-out
+- Preserve explicit supported boolean choices and reject wrong-type controls
+  before they can bypass verification of corrupted retained evidence
+- Require exactly `equiframe[arrow]==0.12.0a4` and matching core engine/ML extras;
+  the coordinated core correction also rejects invalid integrity-reader controls
+- Preserve the prior benchmark artifacts and failing probes, and revalidate the
+  changed distribution separately instead of relabelling earlier benchmark tests
+
+No scientific protocol, detector independence or core artifact format changes.
+The current identified pair is core 0.12.0a4 / benchmark 0.12.0a4.
+
+# Private installed-product alpha 0.12.0a3 — 2026-10-06
+
+- Require exactly `equiframe[arrow]==0.12.0a3`, including matching engine/ML extras
+- Carry the corrected core verification reader boundary and updated current
+  producer schemas while retaining admitted historical producer identities
+- Preserve prior a1/a2 artifacts and receipts; validate the new package pair
+  independently without relabelling earlier installation or scientific evidence
+
+No benchmark detector, label, partition or scientific protocol changes.
+
+# Private installed-product alpha 0.12.0a2 — 2026-10-06
+
+- Require exactly `equiframe[arrow]==0.12.0a2` and retain admitted historical
+  producer identities in detector-bearing schemas
+- Reject wrong numeric types in fixed integer/boolean document metadata,
+  including cost repetitions and scientific-label confirmation; validation
+  does not convert numeric zero into an explicit false confirmation
+- Preserve valid historical bytes and scientific protocols; rerun installed
+  tests against distinct a2 artifacts rather than relabel a1 evidence
+
+# Private installed-product alpha 0.12.0a1 — 2026-10-06
+
+- Pin the exact matching `equiframe[arrow]==0.12.0a1` core and align engine/ML extras
+- Admit the new producer version in embedded detector-bearing current schemas
+  while retaining prior producer values and historical artifact readers
+- Revalidate this separate distribution through normal private pip resolution,
+  isolated wheel/sdist builds, all extras and installed compatibility receipts
+
+No detector, scientific partition, label/reviewer requirement, bootstrap,
+evaluation protocol or historic result is changed. Package installation and
+technical verification do not establish a completed scientific experiment or
+public distribution rights.
+
+# Post-roadmap review: shared integration and bounded repeated work
+
+Private 0.11.1 candidate, identified by commit and artifact hashes.
+
+- Keep snapshot public imports as exact aliases to the generic core implementation;
+  preserve existing manifest schema, strict types and canonical JSON/fingerprints
+- Reuse core fixed-text CLI hints without changing existing machine codes or exit
+  meanings; the standalone pilot retains failure exit 1 with a concise diagnostic
+- Avoid redundant bootstrap recomputation after a report is already validated in
+  the same public operation: table read/write require one evaluation validation
+  instead of two; pairwise construction requires two instead of three
+- Keep revalidation at every public boundary, including altered model instances
+  and JSON; preserve historical report, pairwise and all six CSV bytes
+
+This computation-count reduction is not a performance SLA or scientific result.
+No scientific protocol, detector, split, bootstrap seed, interval definition,
+dependency range or historical schema is changed by this refactoring.
+
+# STEP9 candidate: distribution evidence
+
+The private 0.11.1 candidate now has English entrypoint documentation, an
+artifact-bound dependency inventory/advisory workflow and byte-identical wheel
+and sdist rebuild checks. Runtime APIs and scientific formats are unchanged.
+Public release, project licensing and a completed security assessment remain
+blocked; exact artifact hashes identify the candidate.
+
+# Changelog
+
+## STEP8 candidate (2026-10-06)
+
+- Add experimental bounded engineering scale API/CLI, separate from the frozen
+  scientific paired cost protocol. Deterministic batched workloads, explicit
+  stage boundaries, cold/warm labels, native restricted baselines and raw trials.
+- Supervise worker wall time, sampled RSS/disk and CPU/I/O; retain missing metrics
+  and crashes, verify atomic completion, and clean only owned SQLite temporaries.
+  Reactive resource thresholds are not hard quotas or an untrusted-code sandbox.
+- Record actual isolated-worker producer hashes/versions; differing parent
+  imports fail closed, and missing worker provenance remains explicitly unknown.
+- Installed-distribution verifier also consumes the small scale example.
+  No scientific methodology, dependency, historical format or public release change.
+
+## Sin publicar — compatibilidad STEP6
+
+- Metadatos y lock permiten Python 3.12/3.13; soporte medido solo para las celdas
+  identificadas del candidato, con el mismo núcleo instalado y versiones exactas.
+  No cambia la metodología, los formatos científicos ni la condición experimental.
+
+## Sin publicar — compatibilidad STEP4
+
+- Compruebe presencia de comparaciones con `is not None`; `ComparisonResult`
+  rechaza conversión booleana para impedir aceptación accidental. Consumidores
+  de integración, costes y reproducción conservan sus estados y condiciones.
+  La prueba de expediente rehasheado conserva rechazo temprano de conteos y
+  validación profunda de observaciones contra filas reales. Sin cambios
+  metodológicos, dependencias o formatos.
+
+## 0.11.1 — 2026-10-03
+
+- Reproducción portátil compara conteos/estrategia/clave con las filas capturadas,
+  usando opciones congeladas y referencia peer; lectura local preserva TableArtifact.
+- Selección normal de entrenamiento/calibración comprueba el inventario físico
+  de diferencias; puente Verify exige el tipo de candidato ligado a O.
+- API evaluate_pairwise y PairwiseReport: seis contrastes descriptivos R/M/B/H,
+  soporte confirmado común a cada pareja, denominadores, discordancias y exclusiones.
+  No modifica el freeze, las seis tablas previas ni resultados científicos.
+- Dependencia exacta de equiframe 0.11.1; ejemplo pareado y esquema JSON propios,
+  comprobados también en el consumidor instalado fuera del checkout.
+
+## 0.11.0 — 2026-10-03
+
+- Puente tipado `OracleVerificationBridge`/`oracle_verification_bridge` para
+  conservar identidades, hashes, procedencia y revisión al usar Verify del núcleo.
+  Compatibilidad de `oracle`, `OracleResult` y `evaluate_oracle` mantenida.
+- Dependencia/extras exactos `equiframe==0.11.0`, locks/esquemas compatibles,
+  Twine reproducible y verificador de consumidor limpio ampliado.
+- No altera R/M/B/H, etiquetas, entrenamiento, protocolo científico ni autentica
+  personas por nombres JSON. No publica en PyPI/TestPyPI.
+
+
+## 0.10.1 — 2026-10-03
+
+- Compatibilidad y dependencia fijada en `equiframe 0.10.0`, que añade el
+  migrador nativo. No cambia los protocolos, etiquetas ni algoritmos del estudio.
+- Lock y metadatos sincronizados; evaluación científica, revisión humana,
+  integración real y publicación PyPI conservan sus requisitos pendientes.
+- Los esquemas que incorporan `DetectorVersions` reciben el emisor 0.10.0 y
+  su nuevo default, conservando productores previos y formatos 1.0.
+
+## 0.10.0 — 2026-10-03
+
+- Congelación de protocolos y ejecución de estudios con decisiones, esperados y
+  hashes verificados; las revisiones pendientes conservan su estado.
+- Evaluación R/M/B/H sobre la misma intersección completa, soporte propio,
+  alertas parciales, AP por modelos homogéneos y métricas indefinidas explícitas.
+- Remuestreo de grupos completos con justificación de independencia, percentiles
+  discretos y registro de réplicas indefinidas; no sustituye evidencia científica.
+- Protocolos de costos controlados, tablas regenerables y bundles cerrados con
+  comparación del contenido lógico completo y regenerador explícito.
+- Inventario de fuente/wheel y preparación pública sin conexión a PyPI,
+  publicación, reserva del nombre ni selección de licencia. Los metadatos no
+  acreditan por sí solos instalación, plataforma o reproducción por un compañero.
+- Depende del núcleo 0.9.0. La evaluación científica real, integración externa,
+  matriz de plataformas de esta versión y publicación pública tienen requisitos
+  pendientes y evidencia separada; los fixtures no son aprobaciones reales.
+
+## 0.9.0 — 2026-10-02
+
+- Consumidor de snapshots publicados: recibos estrictos, revisiones, identidad
+  fija, selección de tablas, integridad, tipos/filas y límites cooperativos.
+- Integración de tres motores nativos con comparaciones independientes, esperado
+  fijo, evidencia exclusiva y lector sin reconexión al servicio.
+- Conciliación configurable de grano, importes decimales, monedas y puente;
+  ausencias explícitas, diferencias por adjudicación y totales exactos.
+- API/CLI, cinco esquemas nuevos y ejemplo verificado contra distribución instalada.
+- Depende del núcleo 0.8.1. Continúa pendiente la integración real con las entregas
+  de José/Jhair, la evaluación científica y la publicación pública del paso 19.

equiframe_benchmark-0.12.4/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 Hernán Troya
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.