Skip to content

Latest commit

 

History

History
147 lines (137 loc) · 11.3 KB

CHANGELOG.md

File metadata and controls

147 lines (137 loc) · 11.3 KB

Changelog

All notable changes to this project are recorded here, in the format of Keep a Changelog. The project follows Semantic Versioning. Entries are drawn from the work-package index (work-packages/INDEX.md) and the git history; the package that made each change is named in parentheses.

[0.1.0] - 2026-11-17

Added

  • Two release lifecycles. The software (the package, the docs and the root page) is released by the merge of a pull request from dev into main: the release workflow checks admissibility, runs the suite, builds, creates the tag v<version>, publishes to PyPI by trusted publishing, deploys the site and publishes the GitHub release; a failure is named in the run, cleaned up by scripts/release_cleanup.py and reported in an issue; a merge without a new version redeploys the site only. The paper has its own branch (paper), tags (paper-*, published as GitHub releases on a code owner's approval) and change record (paper/CHANGES.md); its tests run in its own workflow and leave the software's suite. /paper/ serves the mode named in paper/mode.toml, which is abstract (the title, the author and the abstract) (WP65, WP66).
  • Release engineering and governance: the release workflow with trusted publishing, a software bill of materials and the version and changelog gates; five enforcement workflows; CONTRIBUTING.md, CODE_OF_CONDUCT.md, SECURITY.md, code owners, and issue and pull-request templates with their admissibility conditions (WP47).
  • Lint and the Linux fast shard run on every push to dev; the full matrix runs on the release pull request and when the release workflow calls it (WP67).
  • Governance as executable: the repository's settings (rulesets, environments, Pages, the code owners' access, the workflow permissions) declared in .github/governance.toml, checked against the live repository by scripts/check_repo_settings.py in a controls job that runs first in the Tests and Release workflows, and attested in each release's notes; GOVERNANCE.md states the release rules, the two reviews and the repository's secondary rules (WP78).
  • The paper's own contribution route: a /paper/ line in .github/CODEOWNERS names the paper's approvers apart from the code owners, a pull request into paper is merged by an approver who is not its author, the pull request template has a paper section, and "Contributing to the paper" in CONTRIBUTING.md and "The paper's route" in GOVERNANCE.md say how the paper is edited, approved, published and reaches the site (WP79).
  • Public-link metadata on every page of the site: a description, a canonical address, Open Graph and Twitter card fields with a preview image, structured data for the software and the site, robots.txt and a sitemap, from one source of values (WP80).
  • Windows as a third platform for the fast and slow test shards, run on dispatch and at gates; the paper's tests run in the paper's workflow (WP34).
  • The package tradespace (alias hts): the posynomial invariant form for a trade space, with the scikit-learn estimator HyperbolicTradeSpace (fit, transform, inverse_transform, score_samples), the frontier chart (Helmert and additive-log-ratio coordinates) and the entropic (log-mean-exp) soft maximum (WP2).
  • The K = 1 (monomial) fit with its solvers, and tests under scikit-learn's estimator checks (WP2).
  • The several-term (K > 1) posynomial fit with sharp term assignment, a mean-slack objective and a sparsity option (WP9), and the fit with soft weights (WP17).
  • Certificates and a stopping declaration printed by a fit (WP13).
  • Conditioning and analysis tools: LogAffinePreconditioner, frontier_in_original_units, select_temperature, TradeSpaceRegressor, holes and frontier targets (find_holes, frontier_targets, knn_gap_scores) (WP3).
  • trade_off_report: the tensions that define the trade-offs, with elasticities (WP3, WP13; named tension_report until WP42).
  • The fit temperature as a path: temperature_path and plot_temperature_path, the user-guide page "The fit temperature", and the small-temperature lemma L22 (WP46).
  • Plotting: the log plane, the 3-D frontier surface, frontier coordinates, tensions, and ternary plots documented in the examples (WP3, WP18).
  • The derivation ledger: plain-SymPy lemma scripts L1 to L22 in derivation/, an axioms ledger, a proof-map checker and its committed record (WP1, WP9, WP13, WP17, WP29, WP42, WP46).
  • The noise result: lemma scripts L18 to L20 (the noise covariance under the conditioning map, the shift bound, noise as a ridge on the exponents) with planted-data tests (WP29).
  • The documentation site (MkDocs): getting started, user guide, theory, API reference generated from docstrings, generated glossary and derivation pages, further reading and a bibliography with resolved references (WP5, WP6, WP8a).
  • A root page and one-command local staging of the site (WP8a, WP14); the root page recast as why, what and how, with the three-metric triangle shaded by the trade-off that binds there and the pairwise slices, both generated at build time from a planted table (WP51).
  • A history and acknowledgements page on the root site, its dated reference list rendered from history.yaml at build time (WP37), with the governance lineage on it (WP52).
  • Journeys and signposting: the docs home leads with the five need links, the user guide has a router page, the README carries the package-only install line and the paper link, and the package docstring names the docs site (WP33).
  • User-guide pages checked by package-user tests: four tasks run from the docs as written, with their fixtures (a plain CSV, a pump SysML v2 model, an export script and its exported table) under tests/fixtures/user-tests/ (WP32).
  • The bridge page, from the chart back to a design (WP41); a user-guide page on the routes back from a tension (WP42).
  • Two sentences on the limitations page: a proof in a theorem prover would fix each lemma's statement over the reals and would not bound the double-precision error, which the certificates check (WP36).
  • Three public case studies, perovskite solar cells, amniote life histories and Broadbalk wheat, each fetched with a hash check, conditioned, fitted and read (WP45), with their conditioning ladders and chains (WP59); their example pages (WP60), redrafted to one template with an arc (WP63); the wheat page's level V against biomass (WP71).
  • Stable reference keys from the docsite into the paper: [text](paper:key) names a label the paper defines, the docs build turns it into a link to the paper page and anchor and fails on an undefined key; the paper's article theme is pinned so a clean checkout builds the paper (WP8b).
  • A user-guide page, "Choosing K": the checks that bear on the number of terms, what each can and cannot say, a temperature recipe with select_temperature, and worked examples on planted tables and the cubesat (WP31).
  • The interface contract between the technical and presentation sides, with its manifest (WP14).
  • The case study (a SysML-modelled illustrative cubesat attitude-control design space): the simulation port, design space, seeded dataset, the SysML v2 model with an experiment ledger, and the experiments module (WP4, WP4b, WP15).
  • The paper as an executed notebook (mystmd), with a supporting-material notebook of computationally verifiable appendices and a container definition (Dockerfile) for the environment of record (WP10).
  • The estimator keyword term_aggregation ("sum" default, "mean"), the count n_distinct_terms on the termination record, and the bound stated for both aggregations (WP21).
  • The case model in two views, configuration and assessment, in one SysML v2 model: requirements, verification cases and a trace from configuration variables to assessed attributes to the traded metrics (WP22); the model-realism loop that added total mass with launch cost as its measure of effectiveness, so every design choice reaches at least one metric (WP23).
  • The case model read in the OpenMBEE view editor, with a harness note for loading and reading it there (WP26), and a stakeholder layer: two stakeholders, one tied to price and one to performance in the operational life, their concerns traced to the requirements and the metrics (WP27).
  • Figures chosen by section, each with the purpose it serves and what it does not show (WP19).
  • Developer tooling for verification in tiers: a fast test suite, a path-to-tests map (scripts/check.py), parallel workers and a sharded Linux run (WP20).
  • LICENSE (MIT, copyright 2026 Dynamical Systems Group) and this changelog.

Changed

  • The package is named tradespace (import name tradespace, distribution name tradespace); it was hyperbolic-trade-space during development and was never released under that name (WP77).
  • The case study moves onto the model of record: the nine-attribute datasets, the settling time by event detection, and the K = 1, 2, 3 record and the paper's numbers regenerated on them inside the container of record; every moved number is listed with its cause in work-packages/swap-evidence.md, and the earlier data stays as the cited comparison (WP16, WP23).
  • The word "facet" is gone from every surface, including the package's public names: the model's objects are monomial terms (term assignment, distinct terms, per-term shares), the paper, the docs, the glossary, the derivations and the interface contract follow (WP24).
  • The tension matrix and the elasticity vocabulary across the package and the docs: tension_report becomes trade_off_report, with tension_matrix and design_tensions beside it (WP42).
  • The case study revised: momentum headroom as the ninth traded metric, the fleet-size variable and the cost model; the datasets regenerated at 3,000 designs per regime and the cost ladders written from them (WP30).
  • The residual pass: patterns that read as machine-written removed from the docs, the glossary and the example pages, with a checker and its baseline (scripts/check_residuals.py) (WP38).
  • Improvement by subtraction: one pass over the paper and the docs removed figures, arguments and paragraphs that did not advance the narrative, each deletion recorded with what it no longer did (WP28); repository tidiness passes removed unused files and closed drift between the docs, the glossary and the code (WP25); release thinning and the repository's organization (WP40); live guidance moved to notes/ and docs/superpowers retired (WP62); the process record left the tree before 0.1, each document indexed with the hash it is read at (WP73).
  • The response to the independent assessment: the paper and docs keep three claims apart (certificate, observed geometry, usefulness outside the observations) (WP21).
  • The fit's formal statements made exact in the derivations and the ledger, following the response to the external review (WP13).

Fixed

  • The case model's views generator accepts a requirement usage under either type name, SatisfyRequirementUsage or VerifyRequirementUsage (the external review's finding 7) (WP74).

Notes

  • Not part of 0.1.0: the trade-off-aware search (WP12, after 0.1); an image on a registry; the PDF export.