axiomize-quantum-skills-2.0 1.2.0

package/CHANGELOG.md

@@ -0,0 +1,518 @@
+# Changelog
+
+All notable changes to Axiomize are documented here. Axiomize follows semantic versioning; release claims are tied to exact-wheel CI/release evidence.
+
+## How to read this file
+
+This changelog covers **two version lineages** that share one history, which used to make
+the version numbers ambiguous. They are now separated:
+
+| Lineage | Range | Where it lives |
+|---|---|---|
+| **Current distribution** `axiomize-quantum-skills-2.0` | 1.0.0 (2026-09-19) → present | [Releases](#releases-axiomize-quantum-skills-20) below, up to and including the current version |
+| **Upstream lineage** of the `Furox-Art/axiomize` repository | 1.0.0 (2026-08-24) → 1.12.2 (2026-09-05), then 2.0.0 and 2.1.0 (2026-09-16) | [Upstream lineage](#upstream-lineage-furox-artaxiomize) further down, marked with an H1 |
+
+Two consequences worth stating plainly:
+
+- The engine code that became this package shipped first as `Furox-Art/axiomize` at version
+  1.12.2. Versions 1.13 through 2.1.0 therefore belong to that repository, **not** to a
+  release of the `axiomize-quantum-skills-2.0` distribution on PyPI.
+- The distribution name `axiomize-quantum-skills-2.0` restarted the numbering at 1.0.0 on
+  2026-09-19. So `1.0.0`, `1.1.0` and `1.2.0` each appear twice below. The earlier of each
+  pair is upstream history; the later one is a real release of this package, identifiable
+  by its date and by the tag on the GitHub Releases page for this repository.
+- The npm package `axiomize-quantum-skills-2.0` is published at `2.0.0`. That number is not
+  part of this lineage and does not correspond to any release here. (Superseded: #25 pinned
+  `package.json` to `1.2.0` and CI now enforces the lockstep, so the next npm release
+  carries the current version. The registry still serves `2.0.0` until that publish
+  happens.)
+
+`python .github/scripts/check_release_contract.py` enforces the version lockstep across
+`pyproject.toml`, `src/axiomize/__init__.py`, `.github/pypi-release-trigger`, `README.md` and
+the first `## [x.y.z]` heading in this file. That is why the first release heading must
+always be the current version.
+
+## Unreleased
+
+### Changed
+
+- **npm status corrected after #25.** #25 fixed `index.js` and pinned `package.json` to
+  `1.2.0`, so the documentation no longer describes the npm CLI as broken in the source.
+  It does describe it accurately for users, because the fix is **not yet published**:
+  `registry.npmjs.org` still serves `2.0.0`, uploaded 2026-09-28, whose `index.js` is the
+  old `proc.on('close', (code) =; }` and whose tarball is 303 files / 2.1 MB. So
+  `npx axiomize-quantum` still fails while `node bin/axiomize-quantum.js` works from a
+  checkout. Verified on the merged tree: `node --check` exits 0 on both JavaScript files,
+  `node index.js --help` exits 0, a bad subcommand exits 2, `require('./index.js')`
+  resolves, and `npm pack --dry-run` produces 6 files / 41.2 kB. PyPI 1.2.0 is the
+  supported install.
+- README, `docs/index.md`, `docs/tutorial.md`, `docs/integrations.md` and
+  `docs/publishing-checklist.md` now say "npm is behind, PyPI is current" instead of either
+  "npm is broken" or "npm works". Benchmark and command provenance now cites the merge
+  commit rather than the pre-merge one, and the 25-case corpus was re-graded on the merged
+  tree: still 25/25 PASS at 10.0/10 with identical lens counts.
+- `SECURITY.md` records the exact API response behind its reporting guidance
+  (`{"enabled":false}`), names the advisories endpoint so a reader can see it does not work
+  yet, and states the single settings change that would fix it. The fallback instructions
+  are unchanged because the setting is genuinely still disabled.
+
+### Added
+
+- `CITATION.cff` so the package can be cited, and `CODE_OF_CONDUCT.md`.
+- Issue templates for bugs, feature requests and questions, plus `.github/ISSUE_TEMPLATE/config.yml`
+  routing security reports and documentation questions to the right pages. Bare issues are
+  now redirected to the templates.
+- `docs/quickstart-model-ir.json`: a complete, copy-pasteable Model IR request used by the
+  README and tutorial quickstarts.
+- A documentation section that shows the engine working: the previously unreferenced
+  `docs/sir-demo.gif`, `docs/sir-example.png` and `docs/report-sample.pdf` are now embedded
+  with captions describing what they show.
+- Repository metadata: topics covering the causal, dimensional, uncertainty and
+  reproducibility surfaces, and a description that states the released version instead of
+  claiming "Axiomize 2.0".
+
+### Fixed
+
+- **Every cross-repository link now points at this repository.** `docs/tutorial.md` told
+  readers to `git clone https://github.com/Furox-Art/axiomize` — a different repository with
+  a different release line — and to copy a skill path from it. It now clones this
+  repository. `docs/example-gallery.md` linked all 13 of its example links plus its
+  "full texts live in" link to the sibling repository, and `docs/benchmark-results.md`
+  linked 11 of its report links there; both now resolve to files in this repository. The
+  gallery's `#the-fifteen-lenses` anchor pointed at a heading that does not exist in the
+  target README; the lens list is now stated locally.
+- `docs/example-gallery.md` covered 11 of the 18 examples in `examples/`. All 18 are now
+  listed, each verified to exist, each linked exactly once.
+- **The tutorial's commands work against the current API.** `docs/tutorial.md` now records
+  `intake`, `policy` and `model` (which did not exist when it was written), and each command
+  in it was executed against 1.2.0 at commit `eb70a47` on CPython 3.12.10 before being
+  documented. It gains a troubleshooting table covering the errors a new user actually hits,
+  including the Model IR wrapper requirement, unknown units, and UTF-8 BOM input.
+- `docs/integrations.md` listed 14 MCP tools; the server exposes 34. It listed 6 REST
+  routes; 27 exist. Both are now accurate, with the route table verified against a live
+  `axiomize serve`. The `{"command":"axiomize","args":["mcp"]}` config is verified by a real
+  `initialize` + `tools/list` handshake.
+- `docs/integrations.md` implied `axiomize solve --json` produces a run id you can pass to
+  `axiomize reproduce`. It does not — the payload has no `run_id`. That is now stated.
+- `docs/integrations.md` listed "Hermes Agent — SUPPORTED (built and tested here)" and
+  checked boxes for Claude Code, Cursor, OpenCode and OpenCode. No field test against any of
+  these is recorded in this repository, so the table now states what was actually exercised:
+  MCP protocol conformance and live REST responses. The unverifiable row and the checkmarks
+  are gone.
+- `docs/integrations.md` and `SECURITY.md` now name the correct token environment variable
+  (`AXIOMIZE_REST_TOKEN`, minimum 16 characters) and recommend the environment over
+  `--auth-token`.
+- `SECURITY.md` referred to GitHub private vulnerability reporting "when available". It is
+  **not** enabled on this repository, so the policy now says so and gives the routes that do
+  work, plus what a maintainer can do to close the gap.
+- `mkdocs.yml` had no `site_url`, so the published `sitemap.xml` was an empty `<urlset/>` —
+  zero URLs, silently. It now emits 20. Added the `search` plugin, so `/search/` resolves
+  instead of 404ing.
+- `mkdocs.yml` listed 6 pages in nav while 14 more were built but unreachable. All are now in
+  the nav, grouped. `mkdocs build --strict` is clean with zero warnings.
+- The documentation site homepage linked only the example gallery and PyPI. It now links
+  the tutorial, integrations, benchmark results, security, contributing, code of conduct,
+  changelog, citation and license.
+- `docs/publishing-checklist.md` was a submission checklist for the sibling repository,
+  including a "README GIF renders" item that could not pass here. It is rewritten for this
+  package, with an explicit "do not claim" table.
+- README no longer shows a PyPI downloads badge. `img.shields.io/pypi/dm/axiomize-quantum-skills-2.0`
+  answers HTTP 200 while rendering `downloads: rate limited by upstream service` (or
+  `downloads: inaccessible`), because shields.io scrapes a third-party download API.
+  The README now links the PyPI and npm project pages, which always answer HTTP 200 and
+  always carry the real numbers. No count is hardcoded. New
+  `tests/test_readme_badges.py` fails if a dynamic download badge or a hand-copied
+  download count reappears.
+- README badge markup: `img.shields.io/badge/python-3.10%2B-informational` rendered the
+  literal text `3.10%2B` instead of `3.10+`. Replaced with `img.shields.io/python-3.10+-blue`.
+- The README described the package as "Axiomize 2.0" and documented `npx axiomize-quantum`
+  as a working quickstart. It is neither: the released line is 1.2.0, and the published npm
+  `index.js` fails `node --check` with a syntax error, so that command cannot run. The
+  README now states the released version and lists the npm breakage under honest limits
+  rather than advertising it as a way in.
+
+### Changed
+
+- README rewritten to document the surface that actually ships: 9 CLI entry points, 14
+  `axiomize` subcommands, 34 MCP tools, the REST routes, and the two installable agent
+  skills. It previously documented none of them.
+- `docs/benchmark-results.md` no longer reports per-wave scores such as a "suite average
+  9.21/10" attributed to blind-test sessions by independent agents. Those figures had no
+  retained transcripts, could not be reproduced, and one was mis-stated in place. The page
+  now carries a provenance block (script, package version, commit, Python, numeric stack)
+  for a run that was actually executed and re-executed while writing it, plus the full
+  25-case result table, the 12 cases that carry a numeric oracle, and an explicit statement
+  of what the automated layer does not check. The stored reports are described as
+  maintainer-authored rather than as independent-agent output.
+- `docs/benchmark-results.md` described "skill version at test time: v1.3.0 content". No 1.3.0
+  exists in this repository's release line; the claim is removed.
+- `docs/example-gallery.md` labels its 18 examples as illustrative worked reports rather than
+  measured results, and states the parameter-provenance conventions (`exo`/`endo`,
+  `lit.`/`data`/`est.`).
+- `docs/worked-examples.md` is labelled as illustrative sketches with no recorded run, and
+  points at the graded corpus for reproducible numbers.
+- This changelog now separates the upstream lineage from this distribution's releases, so a
+  reader can tell which version numbers refer to a real release here.
+
+## Releases (axiomize-quantum-skills-2.0)
+
+## [1.2.0] - 2026-09-30
+
+### Added
+
+- causal estimators now carry external-reference checks against statsmodels: front-door
+  (`x→m`, `m→y` and direct-effect) point estimates and HC1 standard errors match statsmodels
+  OLS, and binary AIPW matches a statsmodels Logit propensity plus two OLS outcome models
+  (PR #19; `tests/test_numerical_reference.py::TestCausalExternalReference`)
+- `exponential` and `student_t` log-likelihoods in the package-native Bayesian engine, both
+  checked against `scipy.stats` (PR #20; `tests/test_bayesian_likelihoods.py`)
+- optional Metropolis-adjusted Langevin sampling via `sampler: langevin` (alias `mala`); a
+  non-finite gradient step falls back to random-walk Metropolis for that iteration. The
+  default sampler is unchanged and remains Metropolis-Hastings (PR #20)
+- benchmark corpus grown from 20 to 25 closed-form cases with formula-checked oracles
+  (PR #21; `benchmarks/ideas.json`, `tests/test_benchmark_ideas.py`):
+  - physics: RC discharge, `10/e` V
+  - biology: three doublings, 8000 cells
+  - chemistry: first-order half-life, `ln(2)/0.1` h
+  - operations: M/M/1 queue wait, 0.8 h
+  - causal: randomized difference in means, 4
+
+## [1.1.1] - 2026-09-29
+
+- Metadata-only patch release: refreshed PyPI keywords, classifiers, and discovery metadata; no functional API changes.
+
+## [1.1.0] - 2026-09-27
+
+### Changed
+
+- calibrated the reasoning branch-controller default `collapse_score` from 0.78 to 0.65 using
+  paired-sweep evidence (`benchmarks/reports/reasoning-threshold-sweep.md`): the 0.78 reference
+  never fired under bounded noisy evidence, while 0.65 fires early in clear regimes with zero
+  premature-wrong collapses; verification/margin/contradiction guards are unchanged
+- `skills/quantum-reasoning/docs/MEASUREMENT.md` documents the calibrated threshold and its evidence
+
+### Added
+
+- offline ablation harness `axiomize.reasoning.ablation` comparing the branch controller against
+  argmax-commit and top-2 beam baselines on deterministic ground-truth episodes, including
+  regime-shift cases; report: `benchmarks/reports/reasoning-ablation.md`
+- paired collapse-threshold sweep + safety-first `recommend_threshold` selector; report:
+  `benchmarks/reports/reasoning-threshold-sweep.md`
+- benchmark corpus broadened to 20 cases (physics/biology/chemistry/operations/causal) with
+  stored blind-test reports for every case
+- causal estimators now carry external-reference checks against statsmodels (IV/2SLS, HC1 OLS)
+- community regression tests for numeric-oracle keyword alternatives (PR #11)
+- README PyPI version/downloads badges
+
+### Housekeeping
+
+- CI action pins bumped (upload-artifact 7.0.1, download-artifact 8.0.1 via Dependabot)
+
+## [1.0.0] - 2026-09-19
+
+### First independent release
+
+- first release under the distribution name `axiomize-quantum-skills-2.0`
+- carries the full merged engine: Model IR, native model families, causal/Bayesian inference,
+  uncertainty and provenance layers, plus `axiomize.reasoning` (quantum-inspired multi-branch
+  hypothesis management) and the self-contained `skills/quantum-reasoning/` skill folder
+- stored blind-test benchmark coverage completed: all 15 cases in `benchmarks/ideas.json`
+  have reports and grade 10/10
+- CI fully green across Python 3.10-3.13 and Linux/macOS/Windows wheel smoke;
+  trusted-publishing release pipeline publishes the exact tested wheel to PyPI
+
+# Upstream lineage: Furox-Art/axiomize
+
+Everything below was released from the `Furox-Art/axiomize` repository, not from the
+`axiomize-quantum-skills-2.0` distribution. No tag below exists on this repository's
+Releases page, and none of these version numbers identifies a build of this package. The
+code described here is what became this package at 1.0.0 on 2026-09-19.
+
+## [2.1.0] - 2026-09-16
+
+### Added
+
+- the quantum-reasoning skill is now a self-contained skill folder, `skills/quantum-reasoning/`, holding `SKILL.md`, `VERSION`, `docs/` and `examples/` (moved from the flat `skills/quantum-reasoning-SKILL.md` file)
+- `axiomize.reasoning.benchmark` is now an importable subpackage; the evaluator and submission validator still run standalone
+- ported benchmark/submission contracts plus CLI and controller edge-branch tests: `tests/test_reasoning_benchmark.py`, `tests/test_reasoning_quantum.py`, `tests/test_reasoning_contracts.py`, `tests/test_reasoning_cli.py` — 89 reasoning tests, 98% coverage across the merged `axiomize.reasoning` and `axiomize.workflow.reasoning_adapter` modules (branch controller and adapter at 100%)
+- `skills/quantum-reasoning/SKILL.md` frontmatter description now carries an explicit trigger phrase so skill loaders route it correctly
+- `dist/` and `build/` are ignored; local wheel builds no longer show up as untracked files
+
+### Changed
+
+- `skills/axiomize/tools/check_skill.py` now validates every skill folder in the repo (frontmatter, folder/name match, trigger phrase and relative links) instead of only `skills/axiomize`
+- both CI and the release workflow run the full quantum-reasoning contract (`tests/test_reasoning_quantum.py` + `tests/test_reasoning_benchmark.py`)
+- the quantum-reasoning docs live only in `skills/quantum-reasoning/docs/`; the duplicated copies under `src/axiomize/reasoning/docs/` were removed
+- relative links in the ported skill docs, examples and benchmark README now resolve inside the merged layout
+- `pyproject.toml` ships the whole skill folder in the wheel as `axiomize/quantum-reasoning/`
+
+## [2.0.0] - 2026-09-16
+
+### Merged
+
+- merged standalone `quantum-reasoning-skill` v0.3.1 into `axiomize.reasoning` (same `axiomize` repo name, now version 2.0)
+  - `src/axiomize/reasoning/branch_controller.py` — deterministic branch score/prune/revive/collapse logic, ported verbatim
+  - `src/axiomize/reasoning/benchmark/` — evaluator, submission validator, seed cases + JSON schemas
+  - `src/axiomize/reasoning/docs/` — MEASUREMENT + COMPATIBILITY contracts
+  - `skills/quantum-reasoning/` — model-facing protocol, docs and examples (moved from the standalone repo layout)
+  - `tests/test_reasoning_quantum.py` — 8 ported tests, all passing
+- previous `quantum-reasoning-skill` repo is superseded; its README now points here and the repo will be archived
+
+## [1.12.2] - 2026-09-05
+
+### Fixed / hardened
+
+- synchronized the public README package line with the actual release version and made release CI enforce README/CHANGELOG/package/trigger version lockstep
+- blocked the Release workflow from publishing when manually dispatched against any ref other than `refs/heads/main`
+- rejected boolean values at shared finite numeric boundaries instead of silently coercing `True`/`False` to `1.0`/`0.0`
+- bounded textual integer parsing before conversion so oversized integer strings fail before interpreter-dependent parsing limits are reached
+- made run-manifest format-version validation exact, rejecting booleans and fractional values that previously could be accepted through `int(...)` coercion
+- closed the superseded unmerged 1.12.0 scientific-maturity draft PR after the shipped 1.12.0/1.12.1 line replaced it
+
+### Release gates
+
+- full Python 3.10/3.11/3.12/3.13 validation
+- security contract and dependency audit
+- source and installed import-graph contracts
+- exact built-wheel CLI/Model IR/export/surrogate/LaTeX/scientific stress checks
+- Ubuntu/Linux, Windows and macOS exact-wheel CLI smoke
+- Trusted Publishing-first PyPI publication and verification
+
+## [1.12.1] - 2026-09-05
+
+### Fixed
+
+- PDE automatic solver planning now uses the same real `FEniCSAdapter.availability()` probe as the executable FEM path.
+- DOLFINx-only installations are correctly routed to the bounded FEM executor instead of being misclassified as SciPy/method-of-lines only.
+- a merely importable but unrunnable legacy FEniCS installation no longer causes the planner to advertise FEM execution incorrectly.
+- public `general_engine.select_solver()` and direct `general_engine_core.select_solver()` now share the same PDE backend decision contract while preserving explicit user solver configuration.
+
+## [1.12.0] - 2026-09-05
+
+### Added
+
+- all-family scientific benchmark/stress matrix with explicit per-case and total runtime budgets
+- permanent exact-installed-wheel scientific stress gate in CI and release preflight
+- Causal Engine 2.0:
+  - DAG cycle validation
+  - explicit/DAG-derived backdoor adjustment
+  - post-treatment-adjustment rejection
+  - AIPW doubly-robust, IPW and outcome-regression estimates for binary treatment
+  - robust continuous-treatment backdoor regression
+  - positivity/overlap, effective-sample-size and covariate-balance diagnostics
+  - bounded intervention/counterfactual predictions
+- package-native Bayesian Engine 2.0:
+  - bounded multi-chain random-walk Metropolis
+  - split R-hat, bulk ESS and MCSE
+  - posterior interval summaries
+  - posterior predictive RMSE, predictive coverage and Bayesian p-values
+- real optional structured FEniCS/DOLFINx FEM executor for bounded scalar Poisson P1 problems on unit intervals/squares
+- numerical-verification contracts for every Model IR family; stochastic between-seed variability remains explicitly separate from numerical error
+- Modelica 3.6 textual export for supported ODE/DAE/algebraic models
+- GraphML network export
+- Graphviz DOT causal-DAG export
+- `axiomize.portable-bundle.v1` export with canonical Model IR SHA-256 integrity metadata
+
+### Changed
+
+- runtime capability discovery now advertises Causal Engine 2.0, Bayesian diagnostics/PPC, all-family numerical verification and extended exports
+- FEniCS availability is based on a real executable backend probe rather than module-name presence
+- README, ROADMAP and CHANGELOG now describe the actual hardened scientific engine rather than the older prompt/skill-only architecture
+
+### Release gates
+
+- Python 3.10/3.11/3.12/3.13 validation
+- security contract and dependency vulnerability audit
+- source and installed import-graph contracts
+- exact built wheel installation and full CLI/Model IR/export/surrogate/LaTeX checks
+- all-family scientific stress matrix
+- Ubuntu/Linux, Windows and macOS exact-wheel CLI smoke
+- Trusted Publishing-first PyPI publication and verification
+
+## [1.11.2] - 2026-09-05
+
+### Security / correctness
+
+- replaced permissive mathematical parsing paths with a bounded AST-whitelist boundary and explicit AST→SymPy construction
+- removed arbitrary `eval()` use from Z3 constraint handling; added solver timeout and denominator-domain guards
+- hardened Model IR namespaces, targets, bounds, finite values, solver settings and schema migrations
+- prevented silent future-schema migration and schema-version relabeling
+- made arbitrary Python and Lean execution explicit-trust operations with reduced environment inheritance and resource/time limits
+- confined REST/MCP run-file access to configured run roots; added request/message/concurrency/read-time limits and safer errors
+- added provider URL/redirect/request/response/timeout hardening
+- added run-state content integrity verification and atomic persistence
+- added hard non-bypassable limits for arrays, draws, networks, optimization, event queues, PDE work, causal matrices and other native executors
+- hardened LaTeX conversion with a mathematical macro allow-list and `-no-shell-escape`
+- corrected finite-horizon SIR validation versus asymptotic final-size theory
+- hardened direct SciPy/CVXPy/CasADi/statsmodels/PyMC/benchmark/playground/CSV/parallel-sweep surfaces
+- preserved first-occurrence order when duplicate data are merged with `sort_time=False`
+
+### CI / supply chain
+
+- immutable commit-SHA pinning for external GitHub Actions
+- permanent security-contract scanner
+- dependency vulnerability audit
+- direct adversarial runtime regressions
+- release verification hardened against PyPI propagation delay without weakening Trusted Publishing or exact-artifact gates
+
+## [1.11.1] - 2026-09-05
+
+### Added
+
+- permanent exact-wheel/CLI matrix on Ubuntu/Linux, Windows and macOS in normal CI and release preflight
+- platform-independent wheel smoke harness
+
+### Fixed
+
+- macOS ARM dependency compatibility by constraining the Darwin Z3 line to compatible 4.x releases
+- release smoke SIR horizon corrected so it tests CLI portability rather than an intentionally truncated asymptotic final-size comparison
+
+## [1.11.0] - 2026-09-05
+
+### Added
+
+- validated polynomial response-surface surrogate/reduced-order models
+- deterministic Latin-hypercube training design through full Model IR execution
+- explicit approval before multiplying full-model simulations
+- untouched holdout validation with RMSE/NRMSE/MAE/max-error/R² thresholds
+- default blocking of out-of-domain surrogate extrapolation
+- exact source-model provenance and dataset hashes
+- CLI/REST/MCP surrogate paths and installed-wheel release smoke
+
+## [1.10.0]
+
+### Added
+
+- numerical/discretization verification separated from parameter/data/structural uncertainty
+- PDE mesh-refinement studies with observed-order/Richardson-style estimates
+- ODE/DAE tolerance refinement
+- approval gating before repeated numerical solves
+- CLI and REST numerical-verification services
+
+## [1.9.0]
+
+### Added
+
+- native advanced-family execution for PDE, index-1 DAE, optimization, control, network, Bayesian, agent-based, discrete-event, hybrid, multiphysics and causal Model IR
+- method-of-lines finite-difference PDE support with Dirichlet/Neumann boundary conditions
+- solver fallback diagnostics and advanced family execution through the stable general-engine facade
+
+## [1.8.0]
+
+### Added
+
+- canonical versioned Model IR / DSL as the single source of truth for deterministic scientific execution
+- migration preview/approval contract and reproducible migration history
+- model-family/domain recommendation and solver selection
+- native algebraic, ODE and stochastic execution
+- generic ODE fitting, scientific constraints/repair, residual diagnostics, AIC/BIC, stability/validity and nondimensionalization planning
+- SINDy-style sparse dynamics discovery, experiment ranking, provenance and portable exports
+- CLI `axiomize model` path and shared REST/MCP application services
+
+## [1.7.0]
+
+### Added
+
+- adaptive weak/medium/strong workflow and user-controlled clarification
+- explicit consumption policy: no silent extra agents, whole-analysis reruns or extra paid calls
+- stronger reproducibility, uncertainty, hypothesis/falsifier and visualization workflow contracts
+
+## [1.6.0]
+
+### Added
+
+- standardized `ScientificTool` interface and live backend availability probes
+- SciPy, SymPy, statsmodels, Z3, CVXPY, CasADi, network/control and Bayesian scientific adapters
+- dimensional validation and explicit validation statuses
+- application services, CLI, REST, MCP, provider abstraction and portable run-state foundations
+
+## [1.5.0] - 2026-08-24
+
+### Added
+
+- first-principles protocol for ideas with no matching archetype
+- novel-domain benchmark reports and novel-territory report appendix
+
+## [1.4.1] - 2026-08-24
+
+### Fixed
+
+- LaTeX converter hardening across worked examples/benchmark reports
+- removed committed TeX build artifacts and expanded `.gitignore`
+
+## [1.4.0] - 2026-08-24
+
+### Added
+
+- LaTeX/PDF export for reports
+- sample rendered report
+
+## [1.3.2] - 2026-08-24
+
+### Added
+
+- animated demo
+- benchmark-report regression wired into CI
+
+## [1.3.1] - 2026-08-24
+
+### Added
+
+- stored blind-test benchmark reports and benchmark results documentation
+- epidemiology and operations domain packs
+
+### Fixed
+
+- stochastic fade-out theory, Erlang-C overflow, CSV check and fit-bound defects
+
+## [1.3.0] - 2026-08-24
+
+### Added
+
+- decision-theory, demographic/actuarial and spatial-statistics lenses
+- report benchmark runner
+- machine-readable fitting output
+- local Gradio playground
+- project-management domain pack
+
+## [1.2.0] - 2026-08-24
+
+### Added
+
+- reliability, statistical-process-control and thermodynamic lenses
+- additional worked examples and domain packs
+- benchmark dataset/scoring rubric
+- CSV quality pre-check
+
+## [1.1.0] - 2026-08-24
+
+### Added
+
+- game-theory, causal-inference and information-theory lenses
+- expanded archetype catalog and worked examples
+- model-comparison diagnostics, report indexing and GitHub Pages
+
+## [1.0.0] - 2026-08-24
+
+First tagged release: multi-perspective modeling workflow, rigor ladder, standardized report, bundled validation/fitting/sweep tools, worked examples and GitHub Actions CI.
+
+[1.12.2]: https://github.com/Furox-Art/axiomize/releases/tag/v1.12.2
+[1.12.1]: https://github.com/Furox-Art/axiomize/releases/tag/v1.12.1
+[1.12.0]: https://github.com/Furox-Art/axiomize/releases/tag/v1.12.0
+[1.11.2]: https://github.com/Furox-Art/axiomize/releases/tag/v1.11.2
+[1.11.1]: https://github.com/Furox-Art/axiomize/releases/tag/v1.11.1
+[1.11.0]: https://github.com/Furox-Art/axiomize/releases/tag/v1.11.0
+[1.10.0]: https://github.com/Furox-Art/axiomize/releases/tag/v1.10.0
+[1.9.0]: https://github.com/Furox-Art/axiomize/releases/tag/v1.9.0
+[1.8.0]: https://github.com/Furox-Art/axiomize/releases/tag/v1.8.0
+[1.7.0]: https://github.com/Furox-Art/axiomize/releases/tag/v1.7.0
+[1.6.0]: https://github.com/Furox-Art/axiomize/releases/tag/v1.6.0
+[1.5.0]: https://github.com/Furox-Art/axiomize/releases/tag/v1.5.0
+[1.4.1]: https://github.com/Furox-Art/axiomize/releases/tag/v1.4.1
+[1.4.0]: https://github.com/Furox-Art/axiomize/releases/tag/v1.4.0
+[1.3.2]: https://github.com/Furox-Art/axiomize/releases/tag/v1.3.2
+[1.3.1]: https://github.com/Furox-Art/axiomize/releases/tag/v1.3.1
+[1.3.0]: https://github.com/Furox-Art/axiomize/releases/tag/v1.3.0
+[1.2.0]: https://github.com/Furox-Art/axiomize/releases/tag/v1.2.0
+[1.1.0]: https://github.com/Furox-Art/axiomize/releases/tag/v1.1.0
+[1.0.0]: https://github.com/Furox-Art/axiomize/releases/tag/v1.0.0

package/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 Furkan
+
+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.

package/README.md

@@ -0,0 +1,276 @@
+# Axiomize Quantum Skills - 2.0
+
+This is not a separate product. It ships [axiomize](https://github.com/Furox-Art/axiomize) and [quantum-reasoning-skill](https://github.com/Furox-Art/quantum-reasoning-skill) together. Install those repositories when you need only one of them. This repository is not a second MCP registration.
+
+Current package line: **1.2.0**
+
+![License](https://img.shields.io/badge/license-MIT-blue)
+![Python](https://img.shields.io/python-3.10%2B-blue)
+![CI](https://github.com/Furox-Art/axiomize-quantum-skills-2.0/actions/workflows/ci.yml/badge.svg)
+[![PyPI](https://img.shields.io/pypi/v/axiomize-quantum-skills-2.0)](https://pypi.org/project/axiomize-quantum-skills-2.0/)
+[![Docs](https://img.shields.io/badge/docs-gh--pages-blue)](https://furox-art.github.io/axiomize-quantum-skills-2.0/)
+[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](LICENSE)
+
+Download numbers live on the package pages themselves, where they are always current:
+[PyPI](https://pypi.org/project/axiomize-quantum-skills-2.0/) ·
+[npm](https://www.npmjs.com/package/axiomize-quantum-skills-2.0)
+
+**PyPI is the supported install. npm is behind:** PyPI serves `1.2.0`; the npm registry
+serves `2.0.0`, published before the npm launcher was fixed, so `npx axiomize-quantum`
+still fails. `package.json` is now pinned to `1.2.0` in lockstep with Python and CI
+enforces it, so the next npm release carries the fix. See
+[Honest limits](#honest-limits).
+
+**Documentation:** <https://furox-art.github.io/axiomize-quantum-skills-2.0/>
+
+A scientific modeling engine with explicit units, dimensional and numerical validation,
+calibration and uncertainty analysis — plus a reasoning protocol that refuses to commit to
+the first plausible answer.
+
+Two independent pieces ship together:
+
+| Piece | What it is | Where |
+|---|---|---|
+| **Axiomize** | Deterministic engine: versioned Model IR, algebraic/ODE/stochastic/PDE/optimization/control/network/Bayesian/causal execution, dimensional checks, fit, sensitivity, uncertainty, portable export | `src/axiomize/`, CLI `axiomize` |
+| **Quantum Reasoning** | A protocol and deterministic branch controller for keeping several hypotheses alive, scoring them against evidence, pruning weak ones, and collapsing only when evidence supports it | `src/axiomize/reasoning/`, CLI `axiomize-reason` |
+
+"Quantum" here is a metaphor for deferring commitment, not quantum computation. It runs
+on classical models and classical hardware. See
+[skills/quantum-reasoning/SKILL.md](skills/quantum-reasoning/SKILL.md).
+
+## See it work
+
+The reference SIR model, animated. beta = 0.3, gamma = 0.1, so R0 = 3.0.
+
+![Animated SIR epidemic curve, R0=3.0](docs/sir-demo.gif)
+
+That curve is produced by the command in step 3 below. Here it is at full resolution, with
+the peak annotated:
+
+![SIR beta=0.3 gamma=0.1 R0=3.00, peak 300,465 at day 61](docs/sir-example.png)
+
+A sample of the LaTeX report the engine converts a finished analysis into:
+[report-sample.pdf](docs/report-sample.pdf) (source:
+[report-sample.tex](docs/report-sample.tex)).
+
+## Quick start (verified against 1.2.0)
+
+### Install
+
+```bash
+pip install axiomize-quantum-skills-2.0
+```
+
+Requires Python 3.10+. NumPy, SciPy, SymPy, NetworkX, statsmodels, Matplotlib, Control,
+CVXPY and CasADi are pulled in automatically; `z3-solver` is included.
+
+### 1. See what is actually installed
+
+Only backends that are genuinely importable are reported as available.
+
+```bash
+axiomize tools
+```
+
+```text
+sympy        1.14.0     available
+scipy        1.18.1     available
+statsmodels  0.15.0     available
+cvxpy        1.9.3      available
+casadi       3.8.1      available
+jax          unknown    UNAVAILABLE (No module named 'jax')
+z3           5.1.0      available
+...
+```
+
+Machine-readable version of the same thing:
+
+```bash
+axiomize capabilities      # -> {"axiomize_version": "1.2.0", ...}
+```
+
+### 2. Run the install-safe benchmark suite
+
+```bash
+axiomize benchmark
+```
+
+```json
+{"status": "PASS", "passed": 12, "total": 12, ...}
+```
+
+### 3. Validate a model with dimensional and cross-checks
+
+```bash
+axiomize validate --N 1000000
+```
+
+```json
+{
+  "status": "PASS",
+  "final_size": 0.9404340723379753,
+  "asymptotic_final_size": 0.9404805152899381,
+  "cross_validation": {"status": "PASS", "difference": 1.1102230246251565e-16},
+  "numeric_theory_check": {"status": "INCONCLUSIVE", ...}
+}
+```
+
+Note the deliberate honesty: cross-validation passes, but the numeric-theory check reports
+`INCONCLUSIVE` and says why — the finite horizon has not yet reached the asymptotic SIR
+state, so it should not be compared against final-size theory yet.
+
+### 4. Validate your own Model IR
+
+```bash
+axiomize model --input-json docs/quickstart-model-ir.json --action validate
+```
+
+```json
+{"status": "PASS", "model_ir": {"name": "member-churn-decay", "domain": "social", ...}}
+```
+
+`docs/quickstart-model-ir.json` is a complete, copy-pasteable request. To design one from
+a plain-language idea instead:
+
+```bash
+axiomize intake "my gym keeps losing members after 3 months; how do I stop the churn?"
+```
+
+```json
+{
+  "status": "NEEDS_INPUT",
+  "questions": [{"field": "system_boundary", "question": "What exactly are we modeling...?"}],
+  "rigor_recommendation": {"level": "medium", "reasons": ["default balanced depth"]}
+}
+```
+
+### 5. Score a reasoning branch
+
+```bash
+axiomize-reason score --evidence 0.9 --verification 0.85
+```
+
+```json
+{"branch": "branch-1", "can_collapse": false, "reason": "leader score below collapse threshold"}
+```
+
+Every command shown above was executed against `axiomize-quantum-skills-2.0` 1.2.0 at
+commit `338bc9e` on CPython 3.12.10 (Windows). The walkthrough continues in
+[docs/tutorial.md](docs/tutorial.md).
+
+## Interfaces
+
+All three call the same core services, so validation behaviour never depends on the caller.
+
+| Interface | Entry point |
+|---|---|
+| CLI | `axiomize` (14 subcommands: `intake`, `policy`, `model`, `clean-data`, `compare-runs`, `solve`, `fit`, `validate`, `tools`, `capabilities`, `reproduce`, `benchmark`, `serve`, `mcp`) |
+| MCP (stdio) | `axiomize mcp` — 34 tools; see [docs/integrations.md](docs/integrations.md) |
+| REST (v1) | `axiomize serve` — loopback-only by default; see [docs/integrations.md](docs/integrations.md) |
+
+MCP host configuration:
+
+```json
+{"command": "axiomize", "args": ["mcp"]}
+```
+
+REST, verified live:
+
+```bash
+axiomize serve --port 8899
+curl http://127.0.0.1:8899/capabilities
+curl -X POST http://127.0.0.1:8899/solve -H 'content-type: application/json' -d '{"beta":0.3,"gamma":0.1,"N":100000}'
+```
+
+Agent skills are shipped as plain Markdown protocols and can be installed independently of
+the Python package:
+
+```bash
+git clone https://github.com/Furox-Art/axiomize-quantum-skills-2.0
+cp -r axiomize-quantum-skills-2.0/skills/axiomize          ~/.config/opencode/skills/
+cp -r axiomize-quantum-skills-2.0/skills/quantum-reasoning ~/.config/opencode/skills/
+```
+
+## What you get
+
+- 15 mathematical perspectives under `skills/axiomize/perspectives/`, plus archetype and
+  first-principles routing.
+- 18 worked examples in [`examples/`](examples/), indexed by
+  [docs/example-gallery.md](docs/example-gallery.md).
+- A 25-case closed-form benchmark corpus with stored reports, graded in CI.
+- Explicit honesty rules: rejected lenses must carry a reason, every quantity carries a
+  unit, and every model states what observation would falsify it.
+
+## Benchmarks
+
+The 25-case corpus is graded in CI by
+`skills/axiomize/tools/benchmark_runner.py`. Result reproduced locally on commit `338bc9e`,
+axiomize 1.2.0, CPython 3.12.10 (Windows), numpy 2.5.3 / scipy 1.18.1:
+
+```bash
+python skills/axiomize/tools/benchmark_runner.py --case physics-rc-discharge --report benchmarks/reports/physics-rc-discharge.md
+```
+
+```text
+automated score: 9/9 -> 10.0/10 (PASS; human rubric layer still required)
+```
+
+All 25 cases pass the automated layer, 10.0/10 each. **This is a contract-compliance
+measurement, not a measure of modeling quality**: the automated layer checks structure
+(literal alternative lists present, lens count, unit column, violation-consequence column,
+falsifiability section, numeric oracle within tolerance). A structurally perfect report
+about a wrong model still scores 10.0/10. The human rubric layer described in
+[benchmarks/rubric.md](benchmarks/rubric.md) is not recorded for these cases. See
+[docs/benchmark-results.md](docs/benchmark-results.md) for the full, provenance-stamped run.
+
+## Honest limits
+
+- The engine does not choose a model for you. `axiomize model --action plan` ranks
+  candidate families and then returns `NEEDS_MODEL_IR`; deterministic execution starts
+  only after you supply an explicit Model IR.
+- Optional backends (`pymc`, `jax`, FEniCS/DOLFINx) are reported `UNAVAILABLE` unless
+  actually installed. Install them with `pip install axiomize-quantum-skills-2.0[full]`.
+- Arbitrary code and theorem elaboration are **not** an OS sandbox. They require explicit
+  trust. See [SECURITY.md](SECURITY.md) and [docs/security.md](docs/security.md).
+- **The npm launcher is fixed in the repository but the fix is not published yet.** The
+  source is correct as of 1.2.0: `index.js` passes `node --check`, resolves the real CLI
+  module `axiomize.cli`, propagates the child's exit code, and the `files` whitelist keeps
+  the tarball at 6 files / 41 kB instead of the previous 303 files / 2.1 MB. The npm
+  registry still serves `2.0.0`, published before the fix, whose `index.js` is the old
+  syntactically invalid one. So `npx axiomize-quantum` **still fails** until a version
+  carrying the fix is published. `package.json` is now pinned to `1.2.0` in lockstep with
+  Python and CI enforces that, so the next npm release carries the fix.
+  Until then, install from PyPI and use the `axiomize` console script directly.
+
+  Working from a checkout, without waiting for a publish:
+
+  ```bash
+  pip install axiomize-quantum-skills-2.0
+  node bin/axiomize-quantum.js --help      # verified working at 1.2.0
+  ```
+
+## Documentation
+
+| Page | Contents |
+|---|---|
+| [Tutorial](docs/tutorial.md) | First session, end to end |
+| [Example gallery](docs/example-gallery.md) | All 18 worked examples with provenance |
+| [Benchmark results](docs/benchmark-results.md) | Reproducible grading run, with commit |
+| [Integrations](docs/integrations.md) | MCP, REST and CLI surfaces |
+| [Worked examples](docs/worked-examples.md) | Modeling + reasoning composed (illustrative) |
+| [Portable export](docs/portable-export.md) | Versioned, integrity-tagged run bundles |
+| [Security](docs/security.md) | Trust boundaries and input surfaces |
+| [Publishing checklist](docs/publishing-checklist.md) | Registry/list submission checklist |
+
+## Contributing
+
+See [CONTRIBUTING.md](CONTRIBUTING.md) for the perspective and example contracts, and
+[CODE_OF_CONDUCT.md](CODE_OF_CONDUCT.md). Security reports: [SECURITY.md](SECURITY.md).
+
+## Citing
+
+See [CITATION.cff](CITATION.cff). Release history: [CHANGELOG.md](CHANGELOG.md).
+
+## License
+
+MIT — see [LICENSE](LICENSE). Copyright (c) 2026 Furkan.

package/bin/axiomize-quantum.js

@@ -0,0 +1,2 @@
+#!/usr/bin/env node  
+require('../index.js').runAxiomizeQuantum(process.argv.slice(2)); 

package/index.js

@@ -0,0 +1,72 @@
+'use strict';
+
+// Thin launcher for the Axiomize Python engine.
+//
+// Two bugs are fixed here relative to the previous revision, both of which made
+// `npx axiomize-quantum` unusable:
+//
+//   1. Syntax error. The old body was `proc.on('close', (code) =; }`, which is
+//      not valid JavaScript, so `node --check` failed and every invocation died
+//      with `SyntaxError: Unexpected token ';'` before running anything.
+//   2. Wrong module target. It spawned `python -m axiomize_quantum.cli`, a
+//      module that does not exist anywhere in this repository. The import
+//      package is `axiomize` (see `[project]` in pyproject.toml) and its CLI
+//      module is `axiomize.cli`, which defines `main()` and a `__main__` guard.
+//
+// The exit code of the child is propagated so that shell callers, CI steps and
+// the npm contract test can all trust `process.exitCode`.
+
+const { spawn } = require('node:child_process');
+
+// `python` on Windows, `python3` elsewhere. Honour an explicit override first so
+// CI and virtualenv users are not at the mercy of PATH ordering.
+function resolvePython() {
+  const override = process.env.AXIOMIZE_PYTHON;
+  if (override) {
+    return override;
+  }
+  return process.platform === 'win32' ? 'python' : 'python3';
+}
+
+// The console-script equivalent is `axiomize`, installed by the wheel. Prefer it
+// when it is on PATH: it is the entry point the packaging metadata actually
+// declares, so using it keeps this launcher honest about what users get.
+function resolveEntryPoint() {
+  if (process.env.AXIOMIZE_CLI) {
+    return { command: process.env.AXIOMIZE_CLI, args: [] };
+  }
+  return { command: resolvePython(), args: ['-m', 'axiomize.cli'] };
+}
+
+function runAxiomizeQuantum(args) {
+  const { command, args: prefix } = resolveEntryPoint();
+  const child = spawn(command, [...prefix, ...args], {
+    stdio: 'inherit',
+    cwd: __dirname,
+  });
+
+  child.on('error', (err) => {
+    process.stderr.write(
+      `axiomize-quantum: failed to launch '${command}': ${err.message}\n` +
+        'Install the Python package first: pip install axiomize-quantum-skills-2.0\n',
+    );
+    process.exitCode = 127;
+  });
+
+  child.on('close', (code, signal) => {
+    if (signal) {
+      process.stderr.write(`axiomize-quantum: terminated by signal ${signal}\n`);
+      process.exitCode = 128;
+      return;
+    }
+    process.exitCode = code === null ? 1 : code;
+  });
+
+  return child;
+}
+
+module.exports = { runAxiomizeQuantum, resolvePython, resolveEntryPoint };
+
+if (require.main === module) {
+  runAxiomizeQuantum(process.argv.slice(2));
+}

package/package.json

@@ -0,0 +1,46 @@
+{
+  "name": "axiomize-quantum-skills-2.0",
+  "version": "1.2.0",
+  "description": "Axiomize 2.0 - scientific modeling engine plus quantum-inspired multi-branch reasoning",
+  "main": "index.js",
+  "bin": {
+    "axiomize-quantum": "./bin/axiomize-quantum.js"
+  },
+  "scripts": {
+    "test": "node --test \".github/tests/**/*.test.js\"",
+    "lint": "node --check index.js && node --check bin/axiomize-quantum.js",
+    "smoke": "node bin/axiomize-quantum.js --help"
+  },
+  "files": [
+    "index.js",
+    "bin/",
+    "README.md",
+    "LICENSE",
+    "CHANGELOG.md"
+  ],
+  "engines": {
+    "node": ">=18.0.0"
+  },
+  "publishConfig": {
+    "tag": "latest",
+    "access": "public",
+    "provenance": true
+  },
+  "keywords": [
+    "modeling",
+    "simulation",
+    "quantum-inspired",
+    "multi-branch",
+    "scientific-computing"
+  ],
+  "author": "Furox-Art",
+  "license": "MIT",
+  "repository": {
+    "type": "git",
+    "url": "git+https://github.com/Furox-Art/axiomize-quantum-skills-2.0.git"
+  },
+  "bugs": {
+    "url": "https://github.com/Furox-Art/axiomize-quantum-skills-2.0/issues"
+  },
+  "homepage": "https://github.com/Furox-Art/axiomize-quantum-skills-2.0#readme"
+}