ChebyshevSharp 0.15.2

Documentation patch release. Library code is unchanged from 0.15.0/0.15.1.

Documentation

• Trimmed the American-option case study's embedded ~400-line continuous-state dynamic-programming lecture to a short, problem-specific summary that cross-links the From MDPs to Chebyshev Collocation primer (#224). The empirical results, glossary, boundary picture, Bellman diagnostics, and the GMP19 scope-fence are unchanged; the page drops from 1295 to ~900 lines.

Full notes: https://0xc000005.github.io/ChebyshevSharp/docs/changelog.html

ChebyshevSharp 0.15.1

Documentation patch release. Library code is unchanged from 0.15.0 — the package differs only in version/metadata and the bundled README.

Documentation

• New primer From MDPs to Chebyshev Collocation: a Map of Methods (#222): the general-to-specific theory companion to the American-option case study — MDPs → contraction / value & policy iteration → finite-horizon continuous state → four ways to represent the value function (basis functions, regression, neural networks, discretisation) → comparison — including the compute-once transition-kernel derivation and a finite-difference-vs-collocation comparison.
• Citation/attribution corrections and benchmark-scope disclosures across the docs (#221): fixed the Bermudan recombining-quadratures authors (Lim/Lee/Kim 2014), the Goreinov–Tyrtyshnikov 2001 page range, and the Halperin QLBS year; added Homma & Saltelli (1996), the maxvol-2010 paper, and the Orthogonal Chebyshev Sliding Technique reference; fixed a non-compiling ChebyshevTT snippet.

Full notes: https://0xc000005.github.io/ChebyshevSharp/docs/changelog.html

ChebyshevSharp 0.15.0

[0.15.0] - 2026-06-03 — Continuous-state dynamic programming

Added

• Added ChebyshevSharp.DynamicProgramming, a finite-horizon continuous-state
  Bellman-collocation API for one-dimensional state problems with discrete
  action grids.

Examples

• Added examples/ContinuousStateDynamicProgramming, covering dynamic asset
  allocation and stochastic growth as non-option examples of the generic
  Bellman-collocation API.

Documentation

• Clarified that the American-option case study uses QLNet finite difference
  and binomial engines as the numerical oracle, while QuantEcon
  ContinuousDPs.jl is a Bellman-collocation workflow reference rather than a
  direct American-option parity oracle.
• Added a Continuous-State Dynamic Programming how-to guide that presents the
  generic API before the finance-specific American-option case study and cites
  QuantEcon, ContinuousDPs.jl, CVXPortfolio, and PyPortfolioOpt as related
  documentation patterns.
• Reworked the documentation landing page as a reader-oriented hub with direct
  paths for tutorials, how-to guides, concepts, reference, common workflows, and
  case studies.
• Simplified the top navigation to Docs, API, Release Notes, and GitHub
  while keeping tutorial, how-to, concept, and reference landing pages available
  inside the documentation tree.
• Added the documentation math template used by the American-option tutorial so
  LaTeX formulas render consistently in the DocFX site.

Tests

• Added a Dynamic Chebyshev Bellman expectation regression test for the
  Black-Scholes risk-neutral first-moment identity.
• Added regression tests for the generic finite-horizon Bellman solver, dynamic
  asset-allocation example pattern, stochastic-growth example pattern, input
  validation, and out-of-domain model evaluation.

CI

• Added the continuous-state dynamic-programming console example to the
  validation workflow's runnable example gate.

ChebyshevSharp 0.14.0

Highlights

• Added examples/AmericanOptionDynamicChebyshev, a QLNet-backed American option case-study harness comparing finite-difference and binomial references, Longstaff-Schwartz regression simulation, Stanford-style LSPI, and a reusable Dynamic Chebyshev continuation-value model.
• Added finance regression tests for the American-option reference pricer, Longstaff-Schwartz baseline, LSPI baseline, and Dynamic Chebyshev solver.
• Added the American Option Dynamic Chebyshev case study with optimal-stopping formulas, continuation-value intuition, reproducible commands, accuracy results, speed evidence, and citations.
• Advanced package metadata to 0.14.0 and package validation baseline to 0.13.4.

See docs/docs/changelog.md for the full release entry.

ChebyshevSharp 0.13.4

ChebyshevSharp 0.13.4 — Callable bond risk clone case study

This patch release publishes the callable fixed-rate bond surrogate case study and risk-acceptance harness. It does not intentionally change the public ChebyshevSharp API.

Documentation

• Expanded the callable-bond case study with risk-acceptance validation, full-pillar DV01/PV01 evidence, anchored HDMR and factor-residual trials, state-dependent local-risk, exercise-moneyness residual, and dynamic short-rate-state evidence.
• Added the first risk-acceptable reference-semantics callable tree clone with parallel hybrid effective DV01 acceleration.
• Added references for HDMR and dynamic Chebyshev methods for early-exercise products.

Examples

• Added examples/CallableBondSurrogate, including the QLNet-backed callable baseline, the full-dimensional public wrapper, naive surrogate discovery, structured-alternatives evidence modes, and full 60-pillar DV01 checks.

Release

• Advanced package metadata to 0.13.4 and package validation baseline to 0.13.3.
• Merged via PR #210. Closes issue #209.

ChebyshevSharp 0.13.3

This patch release packages high-dimensional documentation improvements for
ChebyshevSlider and ChebyshevTT. It does not intentionally change public API
or numerical behavior.

Documentation

• Added examples/SliderPartitionValidation, a runnable 8D Slider example that
  compares a correct pairwise partition against an intentionally weak singleton
  partition and shows why held-out validation matters
  (#185, #188).
• Expanded the Slider guide with a pivot/slide mental model, a practical
  partition-selection workflow, and held-out validation code for detecting
  missing cross-group interactions.
• Improved the Tensor Train guide with a core-chain mental model, a complete
  first example that builds, evaluates, validates, and prints ranks/compression,
  plus a TT-Cross acceptance checklist and troubleshooting table.

CI

• Runs the .NET 10 xUnit suite with VSTest parallelization disabled because
  existing verbose-output tests capture Console.Out; this avoids
  nondeterministic writer-disposal failures in CI while preserving full test
  coverage.

Release

• Advanced package metadata to 0.13.3 and updated package validation to
  compare this release against 0.13.2.

ChebyshevSharp 0.13.2

Fixed

• ChebyshevSlider.Build() no longer overflows full-grid diagnostic arithmetic for high-dimensional partitioned sliders whose per-slide grids are tractable (#182, #183).
• ChebyshevSlider.ToString() now reports oversized full-tensor comparisons without throwing for valid slider decompositions (#183).

Documentation

• Improved documentation-site navigation by promoting release notes, adding a project-links page, exposing a top-level GitHub repository link, and adding footer links for GitHub, NuGet, issues, release notes, and contributing (#181).

Release

• Advanced package metadata to 0.13.2 and updated package validation to compare against 0.13.1.

Validation passed in PR #184: format, release build, examples, tests, package validation, DocFX, hosted CI, and Codecov patch coverage.

ChebyshevSharp 0.13.1

Summary

• Packages the completed public documentation audit as a patch release.
• Updates package metadata, getting-started package examples, release-gate documentation, and issue-template placeholders to 0.13.1.
• Sets package validation to compare this patch release against 0.13.0.

Compatibility

This release does not intentionally change public APIs or numerical behavior.

Validation

PR #180 passed format, pack, docs, .NET 8 build, .NET 10 tests, all-tests, and codecov/patch gates before merge.

ChebyshevSharp v0.13.0

[0.13.0] - 2026-05-08 — Audit-driven correctness and validation hardening

This release collects the post-0.12.0 audit fixes. It is primarily a
backward-compatible correctness release, but users may see stricter exceptions
for invalid inputs that were previously accepted, extrapolated, or surfaced as
implementation-level exceptions.

The summary sections below describe user-visible impact. The traceability
inventories at the end of this release section cover the 83 audit pull requests
merged after v0.12.0, this release-preparation pull request, and the 35 issues
closed in the post-0.12.0 release window.

Upgrade Notes

• No public APIs were intentionally removed in this release.
• Invalid domains, dimensions, scalar controls, derivative orders, malformed
  load data, and out-of-domain evaluation points may now throw earlier with
  public argument exceptions instead of producing extrapolated values,
  inconsistent state, or lower-level runtime exceptions.

Changed

• Public evaluation now rejects out-of-domain points across dense, spline,
  slider, and Tensor Train interpolants instead of silently extrapolating
  (#74,
  #76).
• Public scalar algebra now rejects NaN, infinity, zero divisors, and null
  operands consistently before producing invalid interpolant state
  (#129,
  #131).
• Runtime evaluation, derivative registries, and serialized derivative metadata
  now enforce configured maxDerivativeOrder bounds
  (#123,
  #125).
• NuGet package validation now compares this release against 0.12.0.

Fixed

• Adaptive auto-N construction now validates off-grid convergence more robustly,
  rejects non-finite validation samples, and commits rebuilt approximation state
  only after the full adaptive loop succeeds
  (#41,
  #147).
• Dense, spline, slider, and Tensor Train builders now reject non-finite build
  values and invalid numeric controls earlier, with public argument exceptions
  instead of delayed internal failures.
• ChebyshevSpline.Build(), ChebyshevSlider.Build(), and adaptive
  ChebyshevApproximation.Build() now publish replacement state atomically, so
  failed rebuilds preserve the previous built interpolant
  (#143,
  #145,
  #147).
• Public state accessors now return snapshots rather than mutable live arrays,
  including domains, node counts, tensor values, special points, cached
  evaluation points, and result wrapper records
  (#75,
  #77,
  #78,
  #79,
  #80,
  #81,
  #82).
• JSON and .pcb loaders now reject malformed dimensions, tensor lengths,
  oversized binary declarations, invalid spline/slider/TT state, and legacy TT
  dimension metadata before materializing inconsistent objects
  (#72,
  #117,
  #141).
• ChebyshevSpline.SobolIndices() now includes between-piece variance and
  interval-membership interactions instead of only aggregating independent
  per-piece Sobol energies (#84).
• ChebyshevSpline.AutoKnots() now validates public inputs before scanning the
  user function (#86).
• Spline calculus and batch evaluation now reject derivative requests at knot
  boundaries and handle jump root detection more consistently.
• Tensor Train validation now rejects zero-dimensional grids, invalid auto-order
  inputs, invalid round tolerances, non-finite build values, and build calls
  without a source function at the public boundary
  (#121,
  #127,
  #135).
• Tensor Train lifecycle and ownership fixes preserve unbuilt clone state,
  isolate extrusion/slicing result cores from source cores, and repair reduced
  QR handling for wide matrices
  (#137,
  #139).
• Deferred construction, integration results, slider derived results, fixture
  generation, and legacy JSON migration fixtures now preserve their intended
  construction and metadata contracts.

Documentation

• README and docs now distinguish value-only evaluation benchmarks from
  price-plus-Greeks benchmarks and record benchmark provenance
  (#107,
  #109,
  #111).
• Package snippets and current guides now use C# API names and document JSON
  plus portable .pcb serialization (#113).
• DocFX API metadata is generated into ignored build output instead of tracked
  source files, preventing local source-path churn during docs verification
  (#88).

CI and Security

• GitHub Actions are pinned to full-length commit SHAs, DocFX and Stryker tool
  versions are pinned, and the SDK matrix explicitly selects the requested SDK
  (#90,
  #103,
  #105).
• CI now audits transitive NuGet dependencies across target frameworks and
  keeps main branch merges behind required test and Codecov patch gates
  (#115).

Complete PR Inventory

• Release metadata and changelog:
  #148.
• Documentation, community, and benchmarks:
  #30,
  #31,
  #32,
  #44,
  #88,
  #101,
  #107,
  #109,
  #111,
  #113.
• CI, dependency, package, and security gates:
  #33,
  #34,
  #90,
  #103,
  #105,
  #115,
  #117.
• ChebyshevApproximation, adaptive build, and evaluation validation:
  #35,
  #38,
  #41,
  #43,
  #45,
  #46,
  #62,
  #63,
  #64,
  #71,
  #74,
  #76,
  #77,
  #81,
  #147.
• Core numerical tests and fixture provenance:
  #36,
  #37,
  #39,
  #40,
  #93,
  #95,
  #97,
  #99.
• ChebyshevSpline validation, calculus, Sobol, and rebuild lifecycle:
  #47,
  #52,
  #53,
  #54,
  #55,
  #56,
  #57,
  [#59](https:/...

ChebyshevSharp v0.12.0

Release PR: #29

Quality and documentation hardening release.

Highlights:

• Added tensor-shape arithmetic guards across dense, serialization, slicing/extrusion, axis kernels, slider diagnostics, and Tensor Train materialization paths.
• Added deterministic property tests, Stryker.NET mutation-test configuration, runnable examples, contributor documentation, and PR checklist.
• Expanded and reorganized docs around tutorials, how-to guides, concepts, reference material, citations, and validation.
• Clarified Type I Chebyshev node convention and the split from MoCaX/Ruiz-Zeron Lobatto grids.
• Hardened SobolIndices variance handling with a scale-relative noise floor and direct TT nonconstant-energy contraction to preserve low-amplitude sensitivity signals.
• Fixed stale documentation around Slider calculus and TT extrusion/slicing/arithmetic availability.

Validation before release:

• GitHub PR checks passed: Format/Pack/Docs, .NET 8, .NET 10, All Tests Passed, codecov/patch.
• Local Release validation passed: format, build, 1156 tests with coverage, and NuGet pack.

Changelog: https://github.com/0xC000005/ChebyshevSharp/blob/main/docs/docs/changelog.md#0120---2026-05-02--quality-and-documentation-hardening-25-26-27-28