factormodels 0.1.0__tar.gz

factormodels-0.1.0/AGENTS.md

@@ -0,0 +1,157 @@
+# FactorModels contributor instructions
+
+## Scope and package ownership
+
+FactorModels (FM) owns factor construction, complete fitted factor models,
+generalized least squares (GLS), capital market assumption (CMA) analytics and
+method-specific reports. It uses QuantInvestStrats (`qis`) and FactorLasso for
+shared analytics and statistical estimation. OptimalPortfolios consumes supplied
+alphas and covariance; application-specific recipes, providers and scheduling
+belong to downstream consumers. The library must not import those consumers.
+
+Follow the current task's explicit scope. Historical migration decisions are not
+blanket permission for new public-interface or numerical changes. Keep the
+canonical `factormodels` interface without retired-package forwarding modules or
+an OptimalPortfolios compatibility extra. Numerical changes require explicit
+authorisation and independent reference evidence.
+
+## Environment and outputs
+
+Use the portable source, lint, test, build and documentation commands in
+[CONTRIBUTING.md](CONTRIBUTING.md). Work from an external checkout for builds and
+keep environments, caches, reports and temporary output outside synchronised
+folders. Select the declared dependency groups and use the reviewed lockfile.
+
+On the registered maintainer Windows/OneDrive workspace, use
+`C:\Python\FactorModels312\Scripts\python.exe`. Never create or use a virtual
+environment, cache, build or temporary output tree inside OneDrive. Before manual
+Python or check commands, dot-source the workspace
+`ArturSepp/scripts/repo_governance/Enter-AgentRepo.ps1` for this repository. The
+registered launcher provides `check`, `test`, `verify` and `docs` tasks. Generated
+state belongs under the external `AGENT_LOCAL_ROOT`; project `uv` operations must
+set `UV_PROJECT_ENVIRONMENT` to the external environment explicitly. Preserve any
+frozen historical environment and use a separate environment for candidate work.
+
+## Public API and tests
+
+Use Google-style docstrings, explicit module exports and root `__all__`. The root
+public export list is the canonical package application programming interface
+(API). Test canonical objects and their contracts. Keep deterministic synthetic
+fixtures free of private data and provider dependencies. Source-adjacent tests
+are shipped for installed-wheel checks; top-level tests hold package acceptance
+contracts. Examples use an explicit `run_local_test(case)` entry point and run
+offline unless they explicitly describe an optional acquisition action.
+
+Every fitted result states cutoff, ordering, return transformations, annualisation,
+currency when supplied, selected residual policy and relevant provenance. The
+`qis.RiskModel` is an interchange view, not the complete fitted result. Preserve
+full selected covariance; label diagonal-reference decompositions explicitly.
+
+Read the [shared documentation standard](https://github.com/ArturSepp/ArturSepp/blob/main/docs/documentation_standard.md)
+before substantive reader-facing documentation edits. Maintain `LICENSE`,
+supplemental notices, `CITATION.cff`, the README citation and `CHANGELOG.md`.
+Version metadata must agree. Keep generated analyses and private inputs out of
+source distributions and the public repository.
+
+## Generated shared instructions
+
+The block below is generated from the maintainer's shared instruction source.
+Amend that source and run its scoped producer; do not hand-edit the block or
+change unrelated repository emissions. The current task's authorisation defines
+which files and public-interface changes are in scope.
+
+<!-- ===== SHARED AGENT CORE (consumer variant) — begin =====
+     Generated from SHARED_AGENT_CORE.md in the maintainer's project knowledge. Do not hand-edit
+     between these markers — propose the change to the maintainer instead. Variants: builder
+     (qis) / consumer / standalone. Last synced 2026-10-20, agent core v1.6; scoped FactorModels amendment -->
+
+## Domain invariants
+
+Not inferable from any single file, and the source of numerically wrong code that runs clean:
+
+- **No look-ahead, anywhere in a backtest path.** A weight decided at *t* is applied over
+  *[t, t+1]*. Estimation is point-in-time: `MeanAdjType.INSAMPLE` subtracts a full-sample mean
+  and is therefore forward-looking — correct for a descriptive exhibit, wrong inside a backtest.
+- **Return convention is stated, never implied** — `qis.to_returns(..., is_log_returns=...)`.
+  Annualisation follows from the frequency; never silently switch convention, frequency, or
+  annualisation factor.
+- **Sharpe has three explicitly labelled conventions** in `qis`; excess variants need
+  `PerfParams.rates_data`. State which one a number uses.
+- **`qis.BootstrapType.STATIONARY` wraps circularly from qis 5.1.0.** Any result resampled under
+  an earlier version does not reproduce.
+- One convention per concept across the stack. If two packages disagree, that is a bug to
+  report, not a difference to accommodate.
+
+## Use the stack before you write it
+
+This package consumes `qis` (analytics, backtesting, risk, stress testing and reporting) and
+`factorlasso` (statistical estimation and fitted-data containers). FactorModels owns the
+complete factor construction and estimation pipeline that connects them. Its finite outputs
+are factor definitions, net asset value (NAV) histories and fitted loadings/covariance for an asset universe; document
+which construction stages are available and each object's downstream usage. OptimalPortfolios consumes
+explicit alphas and covariance for optimisation. Reimplementing an upstream capability is
+a defect, not a convenience.
+Triggers — stop and check the export list before writing: backtest, rebalance, turnover,
+drawdown, Sharpe, volatility target, bootstrap, resample, unsmooth, covariance, correlation,
+regime, hedge ratio, factsheet, tracking error, realised tracking error, information ratio,
+benchmark beta, marginal contribution, risk contribution.
+
+- **The hard stop:** a `for` loop over dates accumulating a position, a weight or a P&L is
+  `qis.backtest_model_portfolio`. The hand-rolled version gets drift adjustment wrong — `qis`
+  holds *units* between rebalancings, not weights.
+- **Never hand-roll `d' Σ d`, a beta ratio, or a tracking-error (TE) decomposition.** Ex-ante tracking error,
+  factor exposures, benchmark beta and marginal TE come from `qis.RiskModel`, built from
+  fitted FactorModels output via `factormodels.build_risk_model`; supplied covariance uses
+  `qis.RiskModel(covar={date: covariance})`. Realised ex-post TE is
+  `qis.compute_ewma_realised_tracking_error` (EWMA series), and whole-sample TE/IR scalars are
+  `qis.compute_te_ir_errors`.
+- **Never invent a symbol.** If a function, class, or keyword argument is not in the export
+  list, it does not exist. Check in one line —
+  `python -c "import qis; print([n for n in dir(qis) if 'unsmooth' in n.lower()])"`;
+  `qis.api.CORE_API` is the documented core and `help(qis.<symbol>)` gives the arguments. Say a
+  symbol is missing rather than producing code that calls it.
+- **If you genuinely must reimplement**, name the rejected stack symbol and why, in a comment on
+  the line above the definition — that turns a silent divergence into a reviewable decision.
+- Never introduce `quantstats`, `pyfolio`, `empyrical`, `ffn`, `bt`, or an ad-hoc statistics
+  layer.
+
+## Verification loop
+
+- Plan → patch → verify. Name the verification command and its result when proposing a patch.
+- A second pass is mandatory where a plausible patch can be numerically wrong and still run
+  clean: estimation windows, weight normalisation, annualisation, constraint construction,
+  anything resampled. Verify against a reference computed a different way, and say which.
+- Prove a new test fails before trusting that it passes: reintroduce the defect, watch it fail,
+  restore.
+
+## Escalation and scope
+
+- Stop and propose before proceeding when a change would exceed roughly five files, alter a
+  public signature, or touch a numerical path.
+- Never change numerical results, random seeds, or computed values unless the change is the
+  request.
+- A public-signature change carries a `CHANGELOG.md` entry and a version bump in the same
+  change. Removing a keyword argument from a function taking `**kwargs` is a silent break — the
+  caller's keyword is swallowed and nothing raises. Treat it as breaking.
+- Do not refactor beyond the requested scope. Propose the wider change; do not perform it.
+
+## Concurrent sessions
+
+More than one agent or session may work on this checkout at the same time, so a file can change
+between your read of it and your write.
+
+- Re-read a file from disk immediately before editing it. Never write a file from an earlier
+  read: a whole-file write from a stale copy silently reverts another session's work.
+- Prefer minimal anchored edits over whole-file replacement. If the on-disk content is not what
+  you expected, stop and reconcile your change onto the current content rather than overwrite.
+
+## Agent-generated artifacts
+
+All agent-generated roadmaps, execution plans, audits, reports, handoffs, and other working
+outputs live under the repository-root `agents/` directory, which is local and ignored by Git.
+Never create `ROADMAP_*.md`, `Claude outputs/`, `Codex outputs/`, or similar agent-output
+artifacts at the repository root. Name feature roadmaps `agents/ROADMAP_<feature>.md`. An
+execution request names the file and stage. A stage is complete when its stated verification
+command passes; its out-of-scope list is binding.
+
+<!-- ===== SHARED AGENT CORE — end ===== -->

factormodels-0.1.0/CHANGELOG.md

@@ -0,0 +1,120 @@
+# Changelog
+
+FactorModels constructs multi-asset tradable factor models and provides
+generalized least squares (GLS) and capital market assumption (CMA) diagnostics.
+The first public release is 0.1.0; earlier entries record unpublished development.
+
+## 0.1.0 — first public release (2026-10-21)
+
+- Publish the canonical `factormodels` distribution and public FactorModels source.
+- Provide factor construction, fitted asset loadings and covariance, rolling models,
+  GLS/CMA analytics and explicit QIS/OptimalPortfolios handoffs.
+- Retain the verified candidate algorithms, estimator outputs and numerical defaults.
+- Use public OSS consumer verification and PyPI installation instructions.
+
+- Require Python 3.11 or later and Pillow 13 or later to exclude the known
+  Pillow vulnerabilities whose fixed release no longer supports Python 3.10.
+- Check the new minimum interpreter in CI and enforce both floors in installed
+  package metadata. Estimator source, defaults and numerical conventions are unchanged.
+
+## 0.1.0rc4 — unpublished FactorModels rename candidate (2026-10-20)
+
+- Rename the distribution and import namespace from `tradablefactors` to
+  `factormodels`, displayed as FactorModels (FM). No forwarding package is installed.
+- Retain the public class/member names, fitted estimator outputs, construction
+  settings, residual policies, numerical conventions and dependency resolutions.
+- Describe factors as supplied price/level histories with explicit return
+  transformations. Bundle schema versions and saved-fit input formats are unchanged.
+- Adapt integrations to the new namespace while retaining the existing repository
+  URL.
+- Add portable contributor setup, locked source checks, installed-artifact
+  verification and strict documentation build instructions.
+
+## 0.1.0rc3 — unpublished release candidate (2026-10-19)
+
+- Connect the fitted-model quickstart to GLS/CMA through an explicit second-order
+  mean bridge, retaining annual log covariance as the declared risk approximation.
+- Export `annual_arithmetic_from_log` for labelled mean and variance inputs.
+  Reconcile arithmetic alpha using both asset and factor variance adjustments.
+- Export fitted-model GLS/CMA tables and a PDF when the teaching assumptions and
+  positive diagonal residual geometry apply; retain full model tables otherwise.
+- Clarify citation wording, synthetic shrinkage and downstream use. Preserve
+  fitted estimator values, seeds, selected covariance policies and dependencies.
+
+## 0.1.0rc2 — unpublished release candidate (2026-10-19)
+
+This candidate provides factor definitions and net asset value (NAV) construction, fitted asset
+loadings and annual covariance at a stated cutoff, rolling model estimates,
+GLS factor-premium inference, and CMA diagnostics with method-specific reports.
+Completed model objects retain labels, conventions and provenance for downstream
+risk analysis and portfolio optimisation.
+
+Changes since `0.1.0rc1`:
+
+- Add labelled, copy-returning `FactorModelSnapshot.loadings` and
+  `factor_covariance` properties without changing estimation or covariance values.
+- Install and export `synthetic_bundle()` for an offline quickstart that works
+  without a source checkout. Preserve its existing seed, inputs and settings.
+- Keep log-return bundle results separate from arithmetic-excess GLS/CMA reports.
+  Bundle examples export fitted tables and explain why no arithmetic audit is
+  produced. A separate arithmetic example demonstrates premia, diagnostics and
+  explicit report export with consistent cash, alpha and covariance conventions.
+- Clarify the package's completed outputs and downstream use in the README,
+  and add the multi-asset tradable factor (MATF) working-paper citation and package discovery metadata.
+- Retain estimator outputs, numerical library implementations and dependency
+  versions.
+
+## 0.1.0rc1 — unpublished release candidate (2026-10-19)
+
+- Align package, citation and README version metadata without numerical or API
+  changes.
+- Retain the MIT source licence, supplemental notices and verified MIT FactorLasso
+  dependency floor. Dependency versions and resolution remain unchanged.
+
+## 0.1.0.dev3 — development (2026-10-18)
+
+- Require the verified MIT FactorLasso release `>=1.0.4.post1` in base installs
+  and the lock, excluding earlier GPL releases without changing estimation or
+  covariance behaviour. Keep optimiser installation a separate consumer step.
+- Ship supplemental OpenBLAS/LAPACK notices and document the reviewed Windows
+  compiler-runtime exceptions and vendor redistributables separately from MIT source.
+- Document direct TradableFactors replacements for former OptimalPortfolios
+  estimator calls, FactorLasso aliases and result-context helpers; compatibility
+  forwarding APIs are not provided.
+- Add the installed, versioned six-file bundle API with explicit loading,
+  in-memory fitting and new-directory persistence. Unsupported or unversioned
+  settings are rejected; historical bundles remain in their recorded environments.
+- Provide offline synthetic, supplied-bundle and public-proxy reference workflows
+  with inspectable results and separate downstream QuantInvestStrats (`qis`) and
+  OptimalPortfolios use.
+- Document completed model fields, units, cutoffs and mutability; move factor
+  estimation and GLS/CMA method documentation to its owning package.
+
+## 0.1.0.dev2 — development (2026-10-18)
+
+- Add opportunity GLS diagnostics, alpha-uncertainty inference and universe
+  diagnostics formerly in OptimalPortfolios. Return prepared arrays for downstream
+  solving.
+- Add `prepare_portfolio_context` for QIS risk and optional diagnostic-table
+  handoff, preserving full residual covariance and native factor coordinates.
+- Add rolling cluster extraction and stability-pooled score preparation for
+  downstream consumers.
+- Use FactorLasso for statistical estimation and QIS for general risk/reporting
+  interfaces, without introducing an optimiser dependency.
+
+## 0.1.0.dev1 — development
+
+- Add factor families, basket/model specifications and assembly from an explicitly supplied catalogue.
+- Add basket construction, volatility targeting, composite NAVs, supported
+  prehistory splicing and loading-sign constraints with explicit numerical policies.
+- Accept caller-supplied label aliases and catalogues without a data-provider
+  dependency.
+- Add offline construction examples and independent held-unit, warm-up,
+  shared-gate and splice verification.
+
+## 0.1.0.dev0 — development
+
+- Extract GLS, Sharpe accounting, CMA identification and conditional inference from OptimalPortfolios.
+- Introduce explicit current/rolling fitted results and cutoff-aware factor covariance estimation.
+- Export the selected total covariance to existing QIS risk interfaces with stated decomposition limits.
+- Preserve first-party MIT attribution.

factormodels-0.1.0/CITATION.cff

@@ -0,0 +1,38 @@
+cff-version: 1.2.0
+message: "Please cite the working paper for the MATF methodology and this software metadata for the implementation."
+type: software
+repository-code: "https://github.com/ArturSepp/FactorModels"
+url: "https://github.com/ArturSepp/FactorModels"
+title: "FactorModels: Financial factor models and GLS analytics"
+authors:
+  - family-names: Sepp
+    given-names: Artur
+    orcid: "https://orcid.org/0000-0002-7038-1748"
+license: MIT
+version: 0.1.0
+date-released: "2026-10-21"
+keywords:
+  - MATF
+  - multi-asset
+  - tradable-factors
+  - factor-models
+  - generalized-least-squares
+  - covariance-estimation
+  - capital-market-assumptions
+preferred-citation:
+  type: report
+  title: "Capital Market Assumptions and Strategic Asset Allocation Using Multi-Asset Tradable Factors"
+  authors:
+    - family-names: Sepp
+      given-names: Artur
+      orcid: "https://orcid.org/0000-0002-7038-1748"
+    - family-names: Hansen
+      given-names: Emilie H.
+    - family-names: Kastenholz
+      given-names: Mika
+  year: 2026
+  publisher:
+    name: SSRN
+  doi: "10.2139/ssrn.6785958"
+  url: "https://papers.ssrn.com/sol3/papers.cfm?abstract_id=6785958"
+  notes: "Working paper. Date written: 2026-05-17."

factormodels-0.1.0/CLAUDE.md

@@ -0,0 +1,8 @@
+# FactorModels agent entry point
+
+Read [AGENTS.md](AGENTS.md) for package ownership, numerical invariants and the
+generated shared instructions. Use [CONTRIBUTING.md](CONTRIBUTING.md) for portable
+source, installed-artifact and documentation checks. Its registered maintainer
+section applies only to that workspace; Windows/OneDrive environments and output
+must remain external. Follow the current task's scope and preserve concurrent
+work. Do not infer permission to publish from a successful local check.

factormodels-0.1.0/CONTRIBUTING.md

@@ -0,0 +1,141 @@
+# Contributing
+
+*Author: [Artur Sepp](https://github.com/ArturSepp) / First recorded: [2026-10-18](https://github.com/ArturSepp/tradablefactors/commit/2e23f5a9334fe2c44a5acfa028e0ac3e368f93a0)*
+
+Contribute to [FactorModels](https://github.com/ArturSepp/FactorModels) using a
+reviewed source checkout and the dependency versions in `uv.lock`. See
+[AGENTS.md](AGENTS.md) for numerical and package boundaries and
+[CITATION.cff](CITATION.cff) for software attribution.
+
+## Prepare an external checkout
+
+Use Python 3.12 and `uv` for the reference contributor commands below. The package
+supports Python 3.11 and later; continuous integration (CI) checks its declared
+platform/interpreter matrix separately. Obtain an authorised checkout of the
+revision you intend to test, and run these commands from its root. A source
+archive also works for the direct checks below; Git-based shared checkers require
+a checkout.
+
+Keep this working checkout outside OneDrive or other synchronised folders.
+Builders can write metadata beside source files, so redirecting only the final
+artifact directory is insufficient. Choose one setup block for your shell; it
+places environments, caches and generated results outside the checkout.
+
+On Linux or macOS:
+
+```sh
+export FM_SOURCE="$PWD"
+export FM_WORK="$(mktemp -d)"
+export UV_PROJECT_ENVIRONMENT="$FM_WORK/venv"
+export UV_CACHE_DIR="$FM_WORK/cache/uv"
+export PYTHONPYCACHEPREFIX="$FM_WORK/cache/python"
+export NUMBA_CACHE_DIR="$FM_WORK/cache/numba"
+export MPLCONFIGDIR="$FM_WORK/cache/matplotlib"
+export MPLBACKEND=Agg
+export PYTEST_ADDOPTS="-o cache_dir='$FM_WORK/cache/pytest' --basetemp='$FM_WORK/tests'"
+```
+
+On Windows, use PowerShell and an external environment under `C:\Python`:
+
+```powershell
+$env:FM_SOURCE = (Get-Location).Path
+$env:FM_WORK = Join-Path $env:LOCALAPPDATA ('FactorModels\contributor-' + [guid]::NewGuid().ToString('N'))
+$env:UV_PROJECT_ENVIRONMENT = 'C:\Python\FactorModelsContributor312'
+$env:UV_CACHE_DIR = Join-Path $env:FM_WORK 'cache\uv'
+$env:PYTHONPYCACHEPREFIX = Join-Path $env:FM_WORK 'cache\python'
+$env:NUMBA_CACHE_DIR = Join-Path $env:FM_WORK 'cache\numba'
+$env:MPLCONFIGDIR = Join-Path $env:FM_WORK 'cache\matplotlib'
+$env:MPLBACKEND = 'Agg'
+$env:PYTEST_ADDOPTS = "-o cache_dir='$($env:FM_WORK)/cache/pytest' --basetemp='$($env:FM_WORK)/tests'"
+```
+
+Use a different `C:\Python` environment name if that contributor environment is
+already in use. Do not sync an unrelated or frozen historical environment.
+
+## Source checks and documentation
+
+After the setup block, these commands work in either shell:
+
+```console
+uv sync --locked --python 3.12 --group test --group lint --extra docs
+uv run --no-sync ruff check src tests tools
+uv run --no-sync interrogate
+uv run --no-sync deptry src
+uv run --no-sync pytest
+```
+
+The `test` and `lint` dependency groups supply the reviewed tools; the `docs`
+extra supplies Sphinx and its extensions. `--locked` rejects a stale lockfile
+instead of changing the dependency cohort. Keep intentional dependency updates
+separate from unrelated source changes.
+
+Build the source distribution and then its wheel with the default `build`
+sequence, and build HTML with warnings treated as errors. On Linux or macOS:
+
+```sh
+uv run --no-sync python -m build --outdir "$FM_WORK/dist"
+uv run --no-sync python -m sphinx -E -W --keep-going -b html docs "$FM_WORK/html"
+```
+
+The equivalent PowerShell commands are:
+
+```powershell
+uv run --no-sync python -m build --outdir "$env:FM_WORK\dist"
+uv run --no-sync python -m sphinx -E -W --keep-going -b html docs "$env:FM_WORK\html"
+```
+
+## Installed-artifact checks
+
+An editable source install does not verify wheel contents. Install the newly
+built wheel into a separate environment and run from an empty directory, using
+its absolute Python path. For a reproducible acceptance run, export the locked
+runtime and test requirements with `uv export --locked --group test --no-dev
+--no-emit-project`, install that export, then install the wheel with `--no-deps`.
+Do not use `uv run` from the source project for the installed check: it selects
+the project's editable environment.
+
+Run `python -m pytest --pyargs factormodels` with that installed interpreter.
+Also pass the absolute source paths for `tests/test_bundle_load.py`,
+`tests/test_examples.py`, `tests/test_readme.py`, `tests/test_documented_contracts.py` and
+`tests/test_dependency_floor.py`. Those top-level acceptance contracts complement
+the shipped source-adjacent tests; their collection counts need not match.
+Confirm `factormodels.__file__` resolves inside the installed environment.
+The wheel job in [.github/workflows/ci.yml](.github/workflows/ci.yml) records the
+installed execution contract.
+
+## Numerical and contribution rules
+
+Keep tests deterministic and offline. Every numerical change needs an independent
+reference calculation, and a new regression check must detect its controlled
+counterexample. Preserve input units, cutoffs, dates, labels, seeds, return
+transformations, annualisation and the selected residual policy. Compare changed
+interfaces through their numerical outputs. The library must not import its
+optimisation or application consumers.
+
+Offline examples require neither acquisition services nor an optimiser package.
+Public data acquisition and the OptimalPortfolios composed example are explicit,
+separate actions. Use the runnable sources linked from the
+[reference workflows](docs/reference_workflows.md).
+
+Contribute source, tests and documentation. Keep local `agents/` records,
+production bundles, market-data snapshots, generated reports and manuscripts out
+of Git and distributions. Preserve the source licence and supplemental notices;
+see the [licensing guide](docs/licensing.md).
+
+## Registered maintainer workspace
+
+On the maintainer's Windows/OneDrive checkout, use
+`C:\Python\FactorModels312\Scripts\python.exe` and the registered
+`ArturSepp/scripts/repo_governance/Enter-AgentRepo.ps1` helper before manual Python
+or check commands. The helper sets the external `AGENT_LOCAL_ROOT` and routes
+caches and temporary state to local storage. Never create or use a virtual
+environment inside OneDrive, and never run a project `uv` command there without
+an explicit external `UV_PROJECT_ENVIRONMENT`.
+
+Use the registered `Invoke-Repo.ps1 -Repo FactorModels -Task verify` or `-Task docs`
+launcher when that maintainer workspace is available. The shared documentation
+checker validates the staged selection by default; for unstaged work use
+`.github/oss_checks.py docs --working-tree --python <external-python>` with an
+external `--output-dir`, after the helper. Builds use the existing external
+source-export path. These registered helpers are local conveniences, not
+prerequisites for the direct contributor commands above.

factormodels-0.1.0/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2024 Artur Sepp
+
+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.

factormodels-0.1.0/MANIFEST.in

@@ -0,0 +1,10 @@
+include AGENTS.md CLAUDE.md CONTRIBUTING.md CITATION.cff CHANGELOG.md uv.lock
+recursive-include docs *.md *.py
+recursive-include examples *.py
+recursive-include tests *.py *.json
+recursive-include tools *.py
+include tools/navigation_refs.json
+recursive-include src *.md
+prune agents
+global-exclude *.pyc
+global-exclude *.pyo