lacuna-quant 0.13.0__tar.gz

lacuna_quant-0.13.0/CHANGELOG.md

@@ -0,0 +1,410 @@
+# Changelog
+
+All notable changes to Lacuna will be documented here. The project intends to follow [Semantic Versioning](https://semver.org/) once its public release process begins.
+
+## Unreleased
+
+## [0.13.0] - 2026-08-27
+
+### Changed
+
+- Publish future core distributions under the unambiguous PyPI name `lacuna-quant` while
+  preserving the `lacuna` import package and `lacuna` command.
+- Split OIDC publication across protected `pypi` and `pypi-options` GitHub environments so each
+  PyPI project has a unique trusted-publisher identity without storing registry credentials.
+- Add an installation diagnostic that fails clearly when the conflicting `lacuna` distribution is
+  installed beside `lacuna-quant`, preserving diagnostic schema and method version 1.
+- License the repository and all future core and extension distributions under MIT only. Releases
+  already published under the former dual-license terms retain those original grants.
+- Advance `lacuna-options` to `0.2.0` so its dependency metadata targets
+  `lacuna-quant>=0.13,<0.14` instead of the unrelated PyPI `lacuna` project; its import API and
+  analytical semantics are unchanged.
+
+## [0.12.0] - 2026-08-26
+
+### Added
+
+- Add immutable `FactorPanelSchema` and `FactorPanelSemantics` contracts plus
+  `adapt_factor_panel()` for explicit generic Polars, Arrow, and optional named pandas MultiIndex
+  ingestion without filtering, joining, bucketing, label construction, or statistical work.
+- Add materialization/copy, source mapping, unknown-semantics, source-order, and no-methodology
+  evidence for factor-panel adaptation.
+- Add an Alphalens Reloaded migration guide and a small numerical IC fixture pinned and attributed
+  to reviewed commit `f0a07c22`, with named tests for compatible and deliberately divergent cases.
+- Add the cumulative `0.12.x` public API fixture while preserving every earlier compatibility
+  contract and result schema version 1.
+
+### Changed
+
+- Advance the reproducible benchmark artifact to version 6 with grouped/null bucketing,
+  neutralization, multi-lag turnover, diagnostic projection, event-window, and chunked factor-panel
+  cases.
+- Require zero skips in full-extra CI, exercise statistics/report/pandas wheels independently,
+  render Plotly reports in the Chromium revision pinned by Playwright, enforce a same-runner 15%
+  legacy median-regression budget, and run 500-simulation event/decay calibration for nightly and
+  tagged releases.
+
+### Fixed
+
+- Map corrupt ZIP member reads during reproducibility-bundle verification to a concise
+  `ReportError`, so `lacuna bundle verify` rejects damaged archives without exposing a Python
+  traceback.
+
+## [0.11.0] - 2026-08-26
+
+### Added
+
+- Add `fit_decay()` with positive constrained exponential fitting, joint-period stationary block
+  bootstrap, explicit identifiability gates, deterministic RNG provenance, and use-site optional
+  SciPy through `lacuna[statistics]`.
+- Add immutable `PortfolioProjectionResult` and explicit gross/net, leg, weighting, group-neutral,
+  contribution, concentration, coverage, attrition, and target-turnover evidence without portfolio
+  compounding or execution simulation.
+- Add `lacuna.events` with availability-default event anchors, next-observation alignment, window
+  censoring and overlap evidence, and complete-path anchor-cluster stationary-bootstrap pointwise
+  and simultaneous response bands.
+- Add ADR-015, the event-study subsystem contract, the shared v0.11 methodology, and the cumulative
+  `0.11.x` public API fixture.
+
+### Changed
+
+- Retain jointly aligned period/horizon IC and spread tables in descriptive decay evidence so
+  inference resamples each period across all horizons without recomputation.
+- Extend interactive reports with stored decay-fit, portfolio-cohort, and event-response panels.
+
+## [0.10.0] - 2026-08-26
+
+### Added
+
+- Add immutable `BucketSpec` and `SignalTransformResult` APIs for balanced or tie-preserving
+  quantiles, split-aware quantiles, equal-width buckets, explicit edges, thresholds, and visible
+  exclusion policy.
+- Add availability-aware grouped IC/bucket evidence and float64 weighted least-squares signal
+  neutralization with deterministic categorical encoding and rank/fit diagnostics.
+- Add standardized `data_attrition` tables to labels and touched signal methods, plus exact
+  multi-lag rank, autocorrelation, and all-bucket membership turnover while preserving lag-one
+  outputs.
+- Add immutable named evidence to `AuditReport`, study evidence retention and bundle inclusion, and
+  a deterministic self-contained Plotly renderer over stored evidence through `lacuna[report]`.
+- Add the cumulative `0.10.x` public API contract and make historical signature checks accept only
+  additive optional keyword parameters.
+
+### Changed
+
+- Extend `SignalStudy` with explicit bucket, neutralization, grouping, multi-lag, and caller-evidence
+  delegation while preserving the functional APIs as the analytical source of truth.
+
+## [0.9.0] - 2026-08-26
+
+### Added
+
+- Add a published persisted-artifact compatibility matrix and tagged-fixture regression corpus for
+  result schema v1 across core `0.1`–`0.8`, bundle v1 across `0.7`–`0.8`, and standardized audit
+  profile v1 from `0.8`.
+- Add strict non-executing `AuditProfile.from_dict(...)` / `from_json(...)` and standalone
+  `BundleManifest.from_dict(...)` / `from_json(...)` readers with fail-closed version, field,
+  duplicate-key, non-finite, and nested-contract validation.
+- Add a versioned public installation-diagnostics API and `lacuna doctor --strict` gate with stable
+  check codes for package/native identity, Python and wheel support, dependency metadata, packaged
+  schemas, and runtime configuration; diagnostics redact configured paths and never inspect user
+  data or activate plugins.
+- Advance the deterministic benchmark artifact to version 5 with an integrated strategy-audit case
+  spanning labels, signal IC, purged validation, stationary bootstrap, cost stress, point-in-time
+  data, vendor/backtest adapters, and standardized-audit composition; publish the initial profile
+  and retain production behavior because it found no justified semantic-preserving optimization.
+- Add a complete Python API surface reference and versioned coverage manifest; contract tests now
+  require every cumulative root/module export, compatibility fixture, human symbol section, and
+  semantic design route to agree exactly.
+
+### Changed
+
+- Advance `lacuna-options` to `0.1.3` only to widen its compatible core range to
+  `lacuna>=0.5,<0.10`; its independent `0.1.x` Python API and analytical semantics are unchanged.
+- Complete the internal `0.9` roadmap boundary and v1 ledger audit. Every internal v1 definition
+  item now has released or release-gated evidence; `1.0.0` remains blocked by the explicitly
+  external requirement for real users applying Lacuna to independent research stacks.
+
+### Fixed
+
+- Pin repository JSON artifacts to LF checkouts so persisted-fixture SHA-256 and Git blob identities
+  remain byte-stable on Windows as well as Linux and macOS.
+
+## [0.8.0] - 2026-08-26
+
+### Added
+
+- Add versioned standardized audit profiles for signal, strategy, and options research scopes over
+  every released analytical, adapter, plugin, and extension method family.
+- Add explicit required, optional, and not-applicable capability coverage; category, requirement,
+  inventory, and domain-finding evidence tables; and unchanged propagation of source findings.
+- Add a published standardized-audit profile v1 JSON Schema, frozen strategy/profile and additive
+  `0.8.x` public API fixtures, plus real options-extension profile integration.
+- Add strict `AnalysisResult.from_dict(...)` and `from_json(...)` readers with duplicate-key,
+  non-finite, schema, field, enum, timestamp, and version rejection.
+- Add `lacuna audit --scope ... --evidence NAME=PATH` with bounded named evidence, deterministic
+  bundle output, safe overwrite behavior, and existing finding-based exit policies.
+- Add an executable vendor/backtester-to-audit example and installed-wheel standardized-audit,
+  strict-reader, schema-resource, and bundle coverage.
+
+### Changed
+
+- Render standardized reports as categorical evidence assessments with required/optional coverage
+  and no universal cross-phase strategy-quality score; the frozen v0.1 signal score is unchanged.
+- Advance `lacuna-options` to `0.1.2` only to widen its compatible core range to
+  `lacuna>=0.5,<0.9`; its independent `0.1.x` public API remains unchanged.
+- Publish the standardized-audit reference, ADR-014, profile-specific agent review workflow, and an
+  updated v1 ledger that leaves migration/operational hardening and independent real-user evidence
+  as the remaining pre-v1 work.
+
+## [0.7.0] - 2026-08-26
+
+### Added
+
+- Add deterministic `.lacuna` reproducibility bundles containing the canonical audit,
+  Markdown/HTML projections, resolved configuration/environment summaries, and optional structured
+  evidence, provenance, and invocation metadata.
+- Add a published bundle-manifest v1 JSON Schema, frozen additive `0.7.x` API fixture, SHA-256
+  artifact-set/member verification, and Python plus CLI creation/verification workflows.
+- Add bounded hostile-archive validation, supplemental secret/URL/path redaction, fail-closed
+  canonical evidence checks, and explicit integrity-versus-authenticity semantics.
+
+### Changed
+
+- Extend clean-wheel smoke and release archive inspection to require the bundle module, installed
+  manifest schema, and an end-to-end bundle create/verify path.
+- Publish the reproducibility-bundle contract, v1-readiness ledger, architectural trust decision,
+  migration boundary, and `0.8`/`0.9` candidate hardening scopes.
+
+### Fixed
+
+- Escape source-derived HTML metacharacters in Markdown reports so a downstream Markdown renderer
+  cannot interpret hostile finding text as embedded HTML.
+
+## [0.6.0] - 2026-08-26
+
+### Added
+
+- Add DuckDB Arrow-stream ingestion without pandas or SQL construction, and a dependency-free
+  scikit-learn bridge over frozen walk-forward, purged, and combinatorial temporal folds.
+- Add immutable vendor schemas with explicit availability, revision, timezone, adjustment, and
+  identifier declarations, plus generic return/trade/position adapters that require complete
+  backtest semantics.
+- Add domain-specific Python entry-point groups, metadata-only discovery, deterministic conflict
+  selection, explicit trusted-code activation, and protocol-major/capability negotiation with
+  structured evidence.
+- Add the independently versioned `lacuna-options` 0.1 distribution with normalized empirical
+  chains, carry forwards, log-forward moneyness, deterministic delta buckets, and supplied-model IV
+  residual evidence.
+- Freeze additive core `0.6.x` and independent extension `0.1.x` public export/signature fixtures,
+  and add real DuckDB/scikit-learn integration plus options property/adversarial tests.
+
+### Changed
+
+- Extend CI, clean-wheel smoke testing, release-set inspection, checksums, and provenance to cover
+  the separately built options wheel/source distribution without adding core runtime dependencies.
+- Publish the Phase 8 adapter, plugin, options, package-version, security, testing, and post-phase
+  integration contracts throughout the developer and agent handbook.
+
+## [0.5.0] - 2026-08-26
+
+### Added
+
+- Add combinatorial purged K-fold over every held-out group combination, with half-open interval
+  purging, embargo after every test group, explicit combinatorial safety limits, visible group/fold
+  tables, and deterministic complete-path reconstruction.
+- Add unrestricted, within-date, within-group, block, and sign-flip permutation schemes with
+  explicit null contracts, alternatives, deterministic per-replicate streams, and finite-resample
+  p-value correction.
+- Add non-Normal-return Sharpe uncertainty, two-sided confidence bounds, Probabilistic Sharpe Ratio,
+  minimum track-record length, and Deflated Sharpe Ratio over a visible complete trial family and
+  declared effective independent-trial count.
+- Add symmetric CSCV/PBO over synchronous strategy matrices with complete IS/OOS selection evidence,
+  average OOS ranks, logits, deterministic tie policy, partition sensitivity, and bounded
+  combination enumeration.
+- Add joint stationary bootstrap that shares one index path across strategy columns and reports
+  per-strategy means, standard errors, and long-run covariance evidence.
+- Add White Reality Check with joint stationary resampling and Hansen SPA with studentization,
+  the stationary-bootstrap population long-run variance kernel, and separately exposed lower,
+  consistent, and upper null recenterings.
+- Add literal equation-level White/Hansen references, deterministic stream fixtures, CPCV/PBO
+  properties, adapter equivalence, and fixed-seed permutation, Sharpe, PBO, Reality Check, and SPA
+  simulation suites.
+- Publish a comprehensive advanced-inference methodology and agent review contract, and freeze the
+  additive `0.5.x` public API while preserving executable `0.1.x` through `0.4.x` contracts.
+
+### Changed
+
+- Complete roadmap Phase 7 and designate separately optional adapters and extensions as the next
+  `0.6.x` milestone.
+- Advance the reproducible benchmark artifact to version 4 with public CPCV, PBO, Reality Check,
+  and SPA reference cases, stable checksums, workload-specific throughput, and traced memory.
+- Require the advanced-inference implementation, CV surface, and validation surface in source and
+  wheel distributions, and exercise CPCV/PBO/permutation/Sharpe/Reality Check/SPA in clean-wheel
+  smoke tests.
+
+### Fixed
+
+- Use the declared two-sided confidence level for Sharpe confidence bounds while retaining the
+  one-sided confidence quantile for PSR and minimum track-record decisions.
+- Reject permutation-invariant mean statistics for reorder-only permutation schemes, ambiguous PBO
+  partition types, non-numeric DSR trial families, and invalid CPCV integer configuration early with
+  domain-specific errors.
+
+## [0.4.0] - 2026-08-26
+
+### Added
+
+- Add deterministic point-in-time as-of joins with explicit decision/availability semantics,
+  identity matching, exact-boundary and staleness policies, stable left order, explicit unmatched
+  handling, revision-tie resolution, and a verified zero-future-match invariant.
+- Add direct future-data diagnostics with missing-availability preservation, equal-boundary counts,
+  bounded affected-row evidence, and optional absolute materiality.
+- Add structural revision-history diagnostics with unique version identities, monotone publication
+  ordering, per-fact version summaries, and explicit point-in-time/latest-only/unknown source modes.
+- Add three-state survivorship evidence, planted delisting support, half-open membership interval
+  validation, overlap and late-availability detection, and no unknown-to-pass conversion.
+- Add availability-safe historical membership selection that rejects ambiguous interval sources and
+  visibly excludes otherwise-active rows that were not yet observable.
+- Add consecutive universe additions, removals, retention, Jaccard similarity, and drift evidence
+  with explicit source survivorship status.
+- Add reusable `DatasetSpec` contracts and structured validation for required fields, empty inputs,
+  null constraints, key uniqueness, numeric/temporal dtypes, non-finite values, and temporal order.
+- Add exact temporal/revision/survivorship fixtures, generated latest-nonfuture and half-open
+  interval invariants, and eager/lazy Polars, pandas, and Arrow equivalence.
+- Publish a complete point-in-time data-correctness methodology guide and freeze the additive
+  `0.4.x` public API while preserving executable `0.1.x` through `0.3.x` contracts.
+
+### Changed
+
+- Complete roadmap Phase 6 and designate advanced inference as the next guarded milestone.
+- Advance the reproducible benchmark artifact to version 3 with a public point-in-time as-of case,
+  deterministic checksum, left-row throughput, and traced-memory evidence.
+- Require `lacuna.bias` in wheel/source distributions and exercise the availability firewall in
+  clean-wheel smoke tests.
+
+## [0.3.0] - 2026-08-26
+
+### Added
+
+- Add a runtime-checkable cost-model protocol, configurable normalized trade columns, and immutable
+  `CostEstimate` values with named per-trade components, complete/known-only totals, explicit
+  unknown rows, assumptions, findings, units, and stable fingerprints.
+- Add fixed/per-unit/notional commission, observed or assumed half/full spread, fixed/proportional
+  slippage, volatility-scaled slippage, general participation impact, square-root impact, and
+  annualized borrow models.
+- Add component composition with strict row/currency/name reconciliation and safeguards against
+  silently applying costs again to observed execution prices or existing component columns.
+- Add deterministic Cartesian cost grids and explicit correlated `CostScenario` sets with gross/net
+  P&L, return, Sharpe, turnover, component reconciliation, support, and unknown-cost status.
+- Add monotonicity-checked, bracketed all-in-cost break-even solving for net P&L, net return, net
+  Sharpe, and CAGR, including complete solver traces and no silent extrapolation.
+- Add point-in-time or explicitly retrospective liquidity diagnostics with participation coverage,
+  distribution summaries, constraints, and unknown/future-volume evidence.
+- Add multi-scenario square-root-impact capacity curves across strictly increasing capital values,
+  including liquidity coverage, participation, component costs, net performance, and constraint
+  status without inventing one capacity number.
+- Add hand-computed, property, planted, temporal, missing-data, component, adapter, packaging-smoke,
+  and frozen `0.3.x` public API contract tests while preserving the `0.1.x` and `0.2.x` contracts.
+- Add a complete trading-cost/capacity methodology guide and promote the subsystem architecture from
+  target design to implemented behavior.
+
+### Changed
+
+- Advance the implementation roadmap through Phase 5 and designate point-in-time data correctness
+  as the next milestone for the `0.4.x` series.
+- Reuse validated path-independent sufficient statistics and one-time base-model estimates across
+  stress surfaces.
+- Advance the reproducible benchmark artifact to version 2 with a nine-point public cost-stress
+  case measuring scenario throughput, output equivalence checksum, and traced memory.
+- Require `lacuna.costs` in wheels/source distributions and exercise a cost surface in clean-wheel
+  smoke tests.
+
+## [0.2.0] - 2026-08-26
+
+### Added
+
+- Add versioned canonical JSON and SHA-256 research identities with rejection of ambiguous,
+  non-finite, timezone-naive, callable, unordered, and credential-bearing values.
+- Add an append-only SQLite experiment registry for completed, failed, cancelled, retried, and
+  explicitly superseding attempts, plus full eligible-set selection lineage and structured registry
+  snapshots.
+- Add Bonferroni, Holm, Benjamini-Hochberg, and Benjamini-Yekutieli multiple-testing corrections
+  over explicit trial families or current registry attempts.
+- Add deterministic parameter-surface evaluation with failed-point visibility, local neighborhoods,
+  peak isolation, plateau width, grid-boundary evidence, threshold support, and selection/evaluation
+  sample separation.
+- Add seeded continuous parameter perturbation with normal, lognormal, and uniform distributions,
+  bounds, integer rounding, named constraints, rejection accounting, attempt budgets, and registry
+  integration.
+- Add declared half-open subperiod analysis with sample support, confidence interval passthrough,
+  overlap warnings, sign consistency, dispersion, trend, failures, and outcome concentration.
+- Add timestamped universe perturbation with stable membership identities, complete instrument sets,
+  retained-baseline fractions, composition Jaccard distance, sample support, and explicit
+  retrospective-membership findings.
+- Add fixed, strictly trailing expanding/rolling, and explicitly retrospective quantile regime
+  classifiers with source-availability validation and unknown-history states.
+- Add conditional regime evidence with raw/effective sample size, confidence intervals, Sharpe,
+  hit rate, drawdown, net/absolute contribution, leave-one-regime-out totals, overlap semantics, and
+  concentration findings.
+- Publish and regression-test the additive `0.2.x` public API contract while continuing to test
+  preservation of the `0.1.x` contract.
+
+### Changed
+
+- Advance the implementation roadmap through Phase 4 and designate trading realism as the next
+  milestone for the `0.3.x` series.
+- Make release-contract verification select the public API fixture for the release's major/minor
+  series and require new Phase 4 modules in wheels and source distributions.
+
+## [0.1.0] - 2026-08-26
+
+### Changed
+
+- Promote the fully verified `0.1.0-rc.1` implementation and its frozen `0.1.x` public contract to
+  Lacuna's initial stable release.
+- Publish stable version tags as GitHub Releases while retaining prerelease treatment for SemVer
+  candidate tags.
+- Record the maintainer decision to waive independent-user acceptance for `0.1.0` because no testers
+  were available; this release makes no claim of external user validation.
+
+### Removed
+
+- Retire the candidate-specific acceptance protocol and public validation tracker references.
+
+## [0.1.0-rc.1] - 2026-08-26
+
+### Added
+
+- Initial mixed Python/Rust project scaffold.
+- Typed configuration, finding, provenance, and analysis result contracts.
+- Polars-first dataframe normalization boundary.
+- Native bridge diagnostics and checked numerical smoke kernel.
+- Unit, property, integration, Rust, and CI test foundations.
+- Initial project documentation and branding.
+- Comprehensive developer handbook, subsystem architecture contracts, reference material, and coding-agent playbooks derived from the technical specification.
+- Forward-return labels, Pearson/Spearman IC, quantile diagnostics, turnover, decay, and native grouped-rank, bootstrap-reduction, and interval-purge kernels.
+- Expanding and rolling walk-forward splits, interval-aware purged K-fold with embargo, and deterministic IID, moving, circular, and stationary bootstrap inference.
+- Versioned audit rules and scoring with explicit unknown/not-applicable evidence, deterministic JSON/Markdown/self-contained HTML reports, a functional audit API, the `SignalStudy` workflow, and the `lacuna signal` CLI.
+- Published audit-result JSON Schema and golden fixtures, plus pandas/Arrow/lazy adapter equivalence, SciPy reference checks, and randomized native differential/property suites.
+- Reproducible Python end-to-end and Criterion Rust benchmark suites with deterministic generators, equivalence checksums, throughput, memory evidence, and smoke/small/medium dataset tiers.
+- Complete v0.1 methodology documentation, executable signal/purging examples, corrected implementation status, and a packaged copy of the published audit-result schema.
+- Add strict documentation and clean source-distribution/wheel smoke jobs to CI.
+- Redesign CI as a staged, cross-platform gate with frozen installs, immutable action pins, coverage,
+  retained evidence, Dependabot maintenance, and a stable branch-protection result.
+- Add conservative adapter-copy/materialization provenance and execution-operation diagnostics to
+  analytical input evidence.
+- Validate signal, label, and price time/identity schemas, external label intervals, delisting-return
+  dtypes, key compatibility, and runtime policy values before analytical execution.
+- Freeze the `0.1.x` root/module export surface and primary call signatures in an executable public
+  API contract with documented compatibility and migration rules.
+- Build tagged CPython 3.11 stable-ABI wheels for Linux x86_64/aarch64, macOS arm64, and Windows
+  x86_64; smoke-test each target artifact, validate the complete distribution set, publish checksums,
+  and generate GitHub provenance before creating a prerelease.
+- Add an independent-user release-candidate protocol and structured feedback issue form with explicit
+  acceptance criteria and a non-fabricated evidence register.
+- Keep RC1 registry publication disabled because the PyPI `lacuna` distribution is owned by an
+  unrelated project; distribute checksummed and attested wheels through GitHub Releases instead.
+
+### Fixed
+
+- Treat positive and negative floating-point zero as the same tie in native Spearman ranking.