twindex 0.2.0__tar.gz

twindex-0.2.0/CHANGELOG.md

@@ -0,0 +1,54 @@
+# Changelog
+
+All notable changes to this project are documented here. The format follows
+[Keep a Changelog](https://keepachangelog.com/en/1.1.0/), and the project uses
+[Semantic Versioning](https://semver.org/).
+
+## [0.2.0] - 2026-11-07
+
+### Added
+- **General models.** `Model(degree, source, nonlinearity)`: Bernstein
+  coefficients of any degree up to 20, a polynomial source term, and
+  nonlinearities `Σ a_m u^m` with odd `m` and `a_m ≥ 0`. The original problem is
+  `V1_MODEL`. The theory is generalised accordingly (THEORY §1, §3–§5).
+- **Certificate format v2** with an explicit `problem.model` block. Version 1
+  certificates are still read and checked unchanged.
+- **Affine candidate families** (THEORY Theorem 3′): a box certificate can carry
+  sensitivity polynomials, so that one certificate covers a whole parameter box
+  with residual bound `Σ_α h^α ‖R_α‖`. Around `θ̄ = 500` this certifies boxes with
+  5× the half-width (625× the volume) of the plain bound.
+- **`twindex audit`**: certify how well a parameter can be recovered from your
+  own measurements (CSV), noise level, prior and model.
+- Model-aware solver, twin search and SIVIA.
+- Affine SIVIA: pavings use sensitivity families by default. On the full
+  `[0,1000]⁴` prior they now prove most leaves incompatible, but the compatible
+  set is too thin for axis-aligned boxes, so certified upper bounds remain the
+  prior width (see the study). A local example proves `L = U = 50` with one affine leaf.
+- `twindex.fit`: certify any callable (a PINN, a solver) as an exact candidate.
+- Inhomogeneous boundary values `u(0) = α`, `u(1) = β`.
+- A redesigned landing page with a hero figure, typeset equations and a results gallery.
+- Sensor/noise studies for targets θ₀, θ₁ and θ₂ with certified results, and an
+  executable tutorial.
+- `mypy --strict` on the trusted core, a 90% coverage gate, and CI on Linux
+  (Python 3.10–3.13), macOS and Windows.
+
+## [0.1.0] - 2026-11-07
+
+First public release.
+
+### Added
+- Exact trusted core (standard library only, no floating point): rational
+  parsing, exact polynomials, certified Bernstein sup-norm bounds, the model
+  problem, a strict certificate format (v1), and the checker with VALID /
+  INVALID / INCONCLUSIVE verdicts.
+- Certificate kinds: `witness`, `pair`, `box` and `paving` (SIVIA-style
+  two-sided bounds on the target diameter).
+- Untrusted search side: a Chebyshev–Newton solver, exact candidate conversion,
+  a local twin search and an adaptive SIVIA driver.
+- `twindex` command line: `check`, `example`, `search`, `sivia`.
+- Ten example certificates covering every kind and every verdict, replayed in
+  CI on Python 3.10–3.13.
+- Full proofs in `docs/THEORY.md`; interface contract in `docs/DESIGN.md`.
+
+[0.2.0]: https://github.com/Arnav-M10/Twindex/releases/tag/v0.2.0
+[0.1.0]: https://github.com/Arnav-M10/Twindex/releases/tag/v0.1.0

twindex-0.2.0/CITATION.cff

@@ -0,0 +1,19 @@
+cff-version: 1.2.0
+message: "If you use Twindex in your research, please cite it as below."
+title: "Twindex: exact, replayable certificates of data-compatible ambiguity for a PDE inverse problem"
+type: software
+authors:
+  - family-names: Mittal
+    given-names: Arnav
+version: 0.2.0
+date-released: 2026-11-07
+license: MIT
+repository-code: "https://github.com/Arnav-M10/Twindex"
+url: "https://arnav-m10.github.io/Twindex/"
+keywords:
+  - inverse problems
+  - identifiability
+  - certified computation
+  - exact arithmetic
+  - partial differential equations
+  - set inversion

twindex-0.2.0/CODE_OF_CONDUCT.md

@@ -0,0 +1,12 @@
+# Code of conduct
+
+This project follows the
+[Contributor Covenant, version 2.1](https://www.contributor-covenant.org/version/2/1/code_of_conduct/).
+
+In short: be respectful, assume good faith, and keep discussion about the work.
+Harassment of any kind is not tolerated.
+
+Report unacceptable behaviour privately to the maintainer through
+[GitHub private reporting](https://github.com/Arnav-M10/Twindex/security/advisories/new)
+or by contacting the repository owner on GitHub. Reports are handled
+confidentially.

twindex-0.2.0/CONTRIBUTING.md

@@ -0,0 +1,25 @@
+# Contributing
+
+* Read [docs/DESIGN.md](docs/DESIGN.md) first; it is the interface contract and says who owns which module.
+* The trusted core (`rational`, `poly`, `model`, `certificate`, `checker`) uses the standard library only and never touches floats.
+* Work on a branch, then open a pull request. Every pull request is reviewed by the owner of each module it touches.
+* Before merging: `ruff check .` and `pytest` must pass.
+
+Setup:
+
+```bash
+python3 -m venv .venv
+.venv/bin/pip install -e ".[dev]"
+.venv/bin/pytest
+```
+
+Docs (MkDocs Material, deployed to GitHub Pages from `main`):
+
+```bash
+.venv/bin/pip install -e ".[docs]"
+.venv/bin/mkdocs serve
+```
+
+Releasing: bump `version` in `pyproject.toml` and `twindex/__init__.py`, update
+`CHANGELOG.md` and `CITATION.cff`, then publish a GitHub release tagged `vX.Y.Z`.
+The `release` workflow builds, verifies and uploads to PyPI through trusted publishing.

twindex-0.2.0/LICENSE

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

twindex-0.2.0/MANIFEST.in

@@ -0,0 +1,8 @@
+include LICENSE README.md CONTRIBUTING.md
+include mkdocs.yml CHANGELOG.md CITATION.cff SECURITY.md CODE_OF_CONDUCT.md
+recursive-include docs *.md
+recursive-include docs/assets *.svg
+recursive-include scripts *.py
+recursive-include examples *.json *.md *.py
+recursive-include tests *.py
+recursive-include twindex/data *.json

twindex-0.2.0/PKG-INFO

@@ -0,0 +1,84 @@
+Metadata-Version: 2.4
+Name: twindex
+Version: 0.2.0
+Summary: Exact, replayable certificates that noisy PDE data cannot tell two parameter sets apart.
+Author: Arnav Mittal
+License-Expression: MIT
+Project-URL: Homepage, https://github.com/Arnav-M10/Twindex
+Project-URL: Documentation, https://arnav-m10.github.io/Twindex/
+Project-URL: Source, https://github.com/Arnav-M10/Twindex
+Project-URL: Issues, https://github.com/Arnav-M10/Twindex/issues
+Keywords: inverse problems,identifiability,certified computation,computer-assisted proof,exact arithmetic,PDE,PINN,set inversion,optimal recovery
+Classifier: Development Status :: 4 - Beta
+Classifier: Intended Audience :: Science/Research
+Classifier: Operating System :: OS Independent
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3 :: Only
+Classifier: Programming Language :: Python :: 3.10
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Programming Language :: Python :: 3.13
+Classifier: Topic :: Scientific/Engineering :: Mathematics
+Classifier: Topic :: Scientific/Engineering :: Physics
+Requires-Python: >=3.10
+Description-Content-Type: text/markdown
+License-File: LICENSE
+Provides-Extra: search
+Requires-Dist: numpy>=1.24; extra == "search"
+Provides-Extra: studies
+Requires-Dist: numpy>=1.24; extra == "studies"
+Requires-Dist: matplotlib>=3.7; extra == "studies"
+Provides-Extra: docs
+Requires-Dist: mkdocs-material>=9.5; extra == "docs"
+Requires-Dist: mkdocstrings[python]>=0.26; extra == "docs"
+Requires-Dist: numpy>=1.24; extra == "docs"
+Provides-Extra: dev
+Requires-Dist: numpy>=1.24; extra == "dev"
+Requires-Dist: pytest>=7; extra == "dev"
+Requires-Dist: hypothesis>=6; extra == "dev"
+Requires-Dist: ruff>=0.4; extra == "dev"
+Requires-Dist: build>=1; extra == "dev"
+Requires-Dist: twine>=5; extra == "dev"
+Requires-Dist: setuptools>=77; extra == "dev"
+Requires-Dist: mypy>=1.10; extra == "dev"
+Requires-Dist: pytest-cov>=5; extra == "dev"
+Dynamic: license-file
+
+# Twindex
+
+**Your inverse solver returned a parameter. Twindex proves whether the data could have determined it.**
+
+Least squares, Bayesian inference and physics-informed neural networks always
+*return* a parameter. None of them can tell you whether a completely different
+parameter fits your measurements just as well. Twindex finds such *twins* for
+the boundary value problem
+
+    -u''(x) + q(x; theta) f(u(x)) = g(x)  on (0, 1),   u(0) = alpha,  u(1) = beta,
+    q(x; theta) = sum_j theta_j B_j(x)   (Bernstein basis, theta >= 0),
+    f(u) = sum of a_m u^m over odd m, a_m >= 0,
+
+and proves it: it writes a JSON certificate that a small exact checker
+(standard-library Python, no floating point, type-checked with mypy --strict)
+replays and verifies. A VALID pair certificate proves that **no method** can
+recover the parameter from those data with worst-case error below half the
+certified gap.
+
+```console
+$ pip install twindex              # exact checker + CLI, zero dependencies
+$ pip install "twindex[search]"    # + numerical search, audits and fitting (numpy)
+
+$ twindex example theta500_pair > pair.json
+$ twindex check pair.json
+pair.json: VALID: ... every estimator of theta_0 has worst-case error >= 500 on these data
+
+$ twindex audit --data measurements.csv --epsilon 1/1000 --prior 0,1000 --target 1 --output audit/
+```
+
+* Certificates: witness, pair (twins), box, and paving (two-sided bounds).
+* Your model: any coefficient degree, polynomial sources, odd-power
+  nonlinearities, inhomogeneous boundary values.
+* `twindex.fit`: check any callable (a PINN, scipy, FEM) as an exact candidate.
+* Every step is proved: https://arnav-m10.github.io/Twindex/THEORY/
+
+Documentation: https://arnav-m10.github.io/Twindex/ ·
+Source: https://github.com/Arnav-M10/Twindex

twindex-0.2.0/README.md

@@ -0,0 +1,277 @@
+<p align="center">
+  <picture>
+    <source media="(prefers-color-scheme: dark)" srcset="https://raw.githubusercontent.com/Arnav-M10/Twindex/main/docs/assets/logo-dark.svg">
+    <img alt="Twindex" src="https://raw.githubusercontent.com/Arnav-M10/Twindex/main/docs/assets/logo-light.svg" width="560">
+  </picture>
+</p>
+
+<p align="center">
+  <a href="https://github.com/Arnav-M10/Twindex/actions/workflows/ci.yml"><img alt="CI" src="https://github.com/Arnav-M10/Twindex/actions/workflows/ci.yml/badge.svg"></a>
+  <a href="https://pypi.org/project/twindex/"><img alt="PyPI" src="https://img.shields.io/pypi/v/twindex.svg"></a>
+  <a href="https://pypi.org/project/twindex/"><img alt="Python" src="https://img.shields.io/badge/python-3.10%20%7C%203.11%20%7C%203.12%20%7C%203.13-blue.svg"></a>
+  <a href="https://arnav-m10.github.io/Twindex/"><img alt="Docs" src="https://img.shields.io/badge/docs-online-informational.svg"></a>
+  <img alt="Typed core" src="https://img.shields.io/badge/core-mypy%20--strict-2f81f7.svg">
+  <a href="LICENSE"><img alt="License: MIT" src="https://img.shields.io/badge/license-MIT-green.svg"></a>
+</p>
+
+<p align="center">
+  <b>Your inverse solver returned a parameter. Twindex proves whether the data could have determined it.</b><br>
+  <a href="https://arnav-m10.github.io/Twindex/">Documentation</a> ·
+  <a href="https://arnav-m10.github.io/Twindex/tutorial/">Tutorial</a> ·
+  <a href="https://arnav-m10.github.io/Twindex/audit/">Audit your data</a> ·
+  <a href="docs/THEORY.md">Theory</a> ·
+  <a href="examples/">Examples</a>
+</p>
+
+---
+
+<p align="center">
+  <img src="docs/assets/hero.png" alt="Two very different coefficients q(x) give solutions that agree at every sensor to within the noise" width="100%">
+</p>
+
+Least squares, Bayesian inference and physics-informed neural networks will
+always *return* a parameter. None of them can tell you whether a **completely
+different** parameter fits your measurements just as well. Twindex finds such
+*twins* and proves it: it writes an exact certificate that a small checker,
+standard-library Python with no floating point, replays and verifies.
+
+> [!IMPORTANT]
+> The figure above is a real certificate (`examples/theta500_pair.json`). Two
+> coefficient fields that differ by up to 1000 produce solutions that agree at all
+> five sensors to within the noise level $`\varepsilon = 10^{-3}`$. Therefore **no
+> method** can recover $`\theta_0`$ from those data with worst-case error below 500.
+> Anyone can re-check the claim with `twindex check`.
+
+**Contents:**
+[Why Twindex](#why-twindex) ·
+[What it certifies](#what-twindex-certifies) ·
+[Why you can trust a VALID](#why-you-can-trust-a-valid) ·
+[Installation](#installation) ·
+[Quick start](#quick-start) ·
+[Features](#features) ·
+[Results](#results) ·
+[How it works](#how-it-works) ·
+[Documentation](#documentation) ·
+[Citing](#citing-twindex)
+
+## Why Twindex
+
+PINN libraries such as DeepXDE and PhysicsNeMo, Bayesian samplers and classical
+least squares are built to **estimate** parameters. Twindex answers a different
+question, *could any estimate be trusted?*, and it answers with a proof.
+
+| | returns | uncertainty | proves that the data cannot pin the parameter down |
+|---|---|---|---|
+| least squares, PINNs | one estimate $`\hat\theta`$ | none | no |
+| Bayesian inference | a posterior | probabilistic, depends on the prior and likelihood | no |
+| **Twindex** | **certified twins** $`\theta, \theta'`$ | **worst case, under bounded noise** | **yes, by an exact certificate** |
+
+Twindex complements estimators rather than replacing them: run your solver, then
+ask Twindex whether a very different answer fits the data just as well.
+
+## What Twindex certifies
+
+Twindex works with the boundary value problem
+
+$$
+-u''(x) + q(x;\theta)\, f\big(u(x)\big) = g(x), \qquad x \in (0,1), \qquad u(0)=\alpha,\quad u(1)=\beta,
+$$
+
+$$
+q(x;\theta) = \sum_{j=0}^{d} \theta_j \binom{d}{j} x^j (1-x)^{d-j}, \qquad
+f(u) = \sum_{m \text{ odd}} a_m u^m \quad (a_m \ge 0), \qquad \theta \in \Theta \subseteq [0,\infty)^{d+1}.
+$$
+
+Here $`g`$ is any polynomial, and the coefficient degree $`d`$, the nonlinearity
+and the boundary values are yours to choose. Given sensors $`x_i`$, measurements
+$`y_i`$ and a noise bound $`\varepsilon`$, the **data-compatible set** is
+
+$$
+\mathcal{C}_y = \Big\lbrace\, \theta \in \Theta \;:\; \max_i \big\vert u_\theta(x_i) - y_i \big\vert \le \varepsilon \,\Big\rbrace.
+$$
+
+| certificate | proves |
+|---|---|
+| `witness` | one parameter $`\theta \in \mathcal{C}_y`$ |
+| `pair` | twins $`\theta, \theta' \in \mathcal{C}_y`$, so every estimator $`T`$ satisfies $`\sup_{\vartheta \in \mathcal{C}_y} \lvert T(y) - \vartheta_k \rvert \ge \tfrac12 \lvert \theta_k - \theta'_k \rvert`$ |
+| `box` | a whole box of parameters lies inside $`\mathcal{C}_y`$, or misses it entirely |
+| `paving` | a covering of $`\Theta`$ with certified two-sided bounds $`L \le \mathrm{diam}_k \mathcal{C}_y \le U`$ |
+
+## Why you can trust a VALID
+
+Every verdict follows from one theorem, proved in [docs/THEORY.md](docs/THEORY.md).
+If a polynomial $`p`$ matches the boundary values exactly and its residual is
+bounded,
+
+$$
+\rho \;\ge\; \max_{x \in [0,1]} \Big\vert -p''(x) + q(x;\theta)\, f\big(p(x)\big) - g(x) \Big\vert,
+$$
+
+then the **true** solution satisfies, for every $`x \in [0,1]`$,
+
+$$
+\big\vert u_\theta(x) - p(x) \big\vert \;\le\; \rho \, \frac{x(1-x)}{2}.
+$$
+
+<details>
+<summary><b>Proof sketch</b> (complete proofs in <a href="docs/THEORY.md">docs/THEORY.md</a>)</summary>
+
+<br>
+
+Let $`e = u_\theta - p`$ and $`r = -p'' + q f(p) - g`$. Because $`f`$ is a sum of odd
+powers with nonnegative coefficients, $`f(u) - f(p) = (u - p)\,H(u, p)`$ with
+$`H \ge 0`$. Hence
+
+$$
+-e'' + c\,e = -r, \qquad c = q\,H(u_\theta, p) \ge 0, \qquad e(0) = e(1) = 0.
+$$
+
+The barrier $`b(x) = \rho\,x(1-x)/2`$ satisfies $`-b'' + c\,b \ge \rho \ge \lvert r \rvert`$, so
+$`w_\pm = b \mp e`$ satisfy $`-w_\pm'' + c\,w_\pm \ge 0`$ with zero boundary values.
+A one-dimensional maximum principle gives $`w_\pm \ge 0`$, that is, $`\lvert e \rvert \le b`$.
+
+</details>
+
+There is no convergence assumption and no floating point. $`\rho`$ is computed
+from Bernstein coefficients with exact rational subdivision, and every sensor
+inequality is decided in exact arithmetic. The checker, about 2,000 lines of
+standard-library Python and type-checked with `mypy --strict`, gives one of
+three verdicts:
+
+* **VALID**: the claim is proved;
+* **INVALID**: the certificate is malformed or its claim is false;
+* **INCONCLUSIVE**: nothing was proved either way.
+
+## Installation
+
+```console
+$ pip install twindex              # exact checker + CLI, zero dependencies
+$ pip install "twindex[search]"    # + numerical search, audits and fitting (numpy)
+```
+
+## Quick start
+
+**Replay a certificate.**
+
+```console
+$ twindex example theta500_pair > pair.json
+$ twindex check pair.json
+pair.json: VALID: both witnesses are data-compatible; D_Q(y) >= 1000, so every estimator of theta_0 has worst-case error >= 500 on these data
+```
+
+**Audit your own measurements** (a CSV file with columns `x,y`):
+
+```console
+$ twindex audit --data measurements.csv --epsilon 1/1000 --prior 0,1000 --target 1 --output audit/
+theta_1 cannot have a uniform absolute-error guarantee better than +/-500 from these data (certified).
+```
+
+**Check a PINN** (or any solver) as a black box:
+
+```python
+from twindex.fit import certify_callable
+
+report = certify_callable(lambda x: my_pinn(x), problem, theta_hat)
+print(report.verdict)   # VALID: theta_hat really is consistent with the data
+```
+
+## Features
+
+* **Exact, replayable certificates.** Plain JSON that anyone can re-check with
+  `twindex check`. Floats are rejected everywhere, even inside JSON.
+* **Your model.** Bernstein coefficients of any degree, polynomial sources,
+  odd-power nonlinearities $`f(u) = \sum_m a_m u^m`$, and inhomogeneous boundary
+  values.
+* **Audits.** `twindex audit` turns measurements, a noise level and a prior into
+  a certified statement about how well a parameter can be recovered.
+* **Bring your own solver.** `twindex.fit` turns any callable (a PINN, scipy,
+  FEM) into an exact candidate.
+* **Two-sided bounds.** SIVIA-style pavings, with affine candidate families that
+  certify whole parameter boxes at once.
+* **Honest by construction.** The search is a heuristic; the checker is the
+  judge. INCONCLUSIVE is never reported as a disproof.
+
+## Results
+
+<p align="center">
+  <img src="docs/assets/results.png" alt="Certified ambiguity of theta_0 against sensor count and noise level" width="80%">
+</p>
+
+How much do more sensors and less noise buy? Certified lower bounds on the
+ambiguity of $`\theta_0`$, for data generated at $`\bar\theta = (500, 500, 500, 500)`$
+with prior $`[0, 1000]^4`$, rounded down ([full study](https://arnav-m10.github.io/Twindex/study/)):
+
+| sensors | $`\varepsilon = 10^{-2}`$ | $`\varepsilon = 10^{-3}`$ | $`\varepsilon = 10^{-4}`$ | $`\varepsilon = 10^{-5}`$ |
+|---:|---:|---:|---:|---:|
+| 3  | 1000 | 1000 | 749 | 718 |
+| 5  | 1000 | 1000 | 763 | 85 |
+| 9  | 1000 | 1000 | 766 | 97 |
+| 17 | 1000 | 1000 | 761 | 85 |
+
+Even 17 sensors at noise $`10^{-3}`$ cannot narrow $`\theta_0`$ below its entire
+prior range. These are certified lower bounds from a local search; they are
+not optimal.
+
+## How it works
+
+1. **Search** (untrusted, floating point). A Chebyshev–Newton solver and a
+   local search propose twins. Each solution is rounded to an exact rational
+   polynomial that matches the boundary values exactly.
+2. **Bound** (trusted, exact). The residual $`\rho`$ is bounded from Bernstein
+   coefficients with exact de Casteljau subdivision.
+3. **Transfer.** The barrier theorem moves the bound from the candidate to the
+   true PDE solution.
+4. **Decide.** At each sensor,
+   $`\lvert p(x_i) - y_i \rvert + \rho\, x_i(1-x_i)/2 \le \varepsilon`$ proves
+   compatibility, and $`\lvert p(x_i) - y_i \rvert - \rho\, x_i(1-x_i)/2 > \varepsilon`$
+   proves incompatibility.
+
+## Documentation
+
+| | |
+|---|---|
+| [Getting started](https://arnav-m10.github.io/Twindex/getting-started/) | install, check, search |
+| [Tutorial](https://arnav-m10.github.io/Twindex/tutorial/) | certify your own twins in five minutes |
+| [Audit](https://arnav-m10.github.io/Twindex/audit/) | your data, your model, a certified answer |
+| [Theory](docs/THEORY.md) | complete proofs of every step |
+| [Certificate format](docs/DESIGN.md) | JSON schema and verdict rules |
+| [API reference](https://arnav-m10.github.io/Twindex/api/) | every public function |
+
+## Scope
+
+* One space dimension and steady problems with a monotone nonlinearity. The
+  barrier argument needs $`q \ge 0`$ and a nondecreasing $`f`$.
+* Bounded (worst-case) noise, not a statistical model. A certificate speaks
+  about the data $`y`$ it declares.
+* The search is a heuristic. A VALID result is a proof; failing to find twins
+  proves nothing.
+
+## Related work
+
+Twindex combines known ideas: optimal recovery (Micchelli–Rivlin; Foucart–Liao),
+state and parameter estimation for parametric PDEs (Cohen, Dahmen, Mula,
+Nichols), set inversion by interval analysis (Jaulin–Walter), certified
+reduced-basis possibility regions (Patera and coauthors), and maximum-principle
+a posteriori bounds (Nochetto et al.). What it adds is an exact, independently
+replayable checker. See [THEORY §9](docs/THEORY.md#9-related-work).
+
+## Contributing
+
+Contributions are welcome; start with [CONTRIBUTING.md](CONTRIBUTING.md) and the
+[design contract](docs/DESIGN.md). A soundness bug (VALID for a false claim) is
+treated as a security issue; please report it privately ([SECURITY.md](SECURITY.md)).
+
+## Citing Twindex
+
+```bibtex
+@software{twindex,
+  author  = {Mittal, Arnav},
+  title   = {Twindex: exact, replayable certificates of data-compatible ambiguity for a PDE inverse problem},
+  year    = {2026},
+  url     = {https://github.com/Arnav-M10/Twindex}
+}
+```
+
+## License
+
+[MIT](LICENSE). Built by Arnav Mittal with Claude Code and Codex.

twindex-0.2.0/SECURITY.md

@@ -0,0 +1,20 @@
+# Security policy
+
+Twindex is a verifier, so a **soundness bug** (any input that makes the checker
+report VALID for a claim that is false) is treated as a security issue.
+
+## Reporting
+
+Please report soundness bugs and other vulnerabilities privately through
+[GitHub private vulnerability reporting](https://github.com/Arnav-M10/Twindex/security/advisories/new).
+Include a certificate or a Python snippet that reproduces the problem. Do not
+open a public issue for a soundness bug before a fix is released.
+
+Expect an acknowledgement within a week. Fixes ship with a regression test and
+a changelog entry.
+
+## Supported versions
+
+| version | supported |
+|---------|-----------|
+| 0.1.x   | yes       |