index.html

---
layout: landing
title: "Burning Cost - ML and data science research in UK personal lines insurance"
description: "10 flagship Python libraries for UK personal lines insurance pricing. Proxy discrimination auditing (FCA Consumer Duty), model validation and governance, DML causal inference, conformal prediction, Whittaker-Henderson smoothing, telematics HMM, GAMLSS, and Buhlmann-Straub credibility. sklearn-compatible, actuarially tested."
---

<!-- Hero -->
<section class="hero">
  <div class="hero-bg">
    <div class="hero-orb hero-orb--1"></div>
    <div class="hero-orb hero-orb--2"></div>
    <div class="hero-orb hero-orb--3"></div>
  </div>
  <div class="hero-inner">
    <div class="hero-label">
      <span class="hero-label-dot"></span>
      ML research &middot; UK personal lines insurance
    </div>
    <h1>Your <em>GBM</em> outperforms.<br>Your <em>GLM</em> is still live.</h1>
    <p class="hero-sub">Ten libraries for the hard problems in UK pricing. Free, MIT-licensed, Databricks-native. All libraries also work standalone via pip — Databricks is optional.</p>
    <div class="hero-cta">
      <a href="https://github.com/burning-cost" target="_blank" class="btn btn-primary">Browse Libraries on GitHub</a>
      <a href="/blog/" class="btn btn-outline">Read the Blog</a>
    </div>

    <!-- Animated terminal -->
    <div class="hero-terminal">
      <div class="hero-terminal-bar">
        <div class="term-dots">
          <span></span><span></span><span></span>
        </div>
        <span class="term-title">shap_relativities_demo.py</span>
      </div>
      <div class="hero-terminal-body" id="hero-term-body"><span class="term-cursor"></span></div>
    </div>
  </div>
</section>

<!-- Stats strip -->
<div class="stats-strip">
  <div class="stats-inner">
    <div class="stat">
      <a href="/top-10/" style="text-decoration:none;color:inherit"><span class="stat-number">10</span>
      <span class="stat-label">Flagship Libraries</span></a>
    </div>
    <div class="stat">
      <a href="https://github.com/burning-cost" style="text-decoration:none;color:inherit"><span class="stat-number">34</span>
      <span class="stat-label">Total Libraries</span></a>
    </div>
    <div class="stat">
      <a href="/benchmarks/" style="text-decoration:none;color:inherit"><span class="stat-number">14,600+</span>
      <span class="stat-label">Tests</span></a>
    </div>
    <div class="stat">
      <span class="stat-number">524</span>
      <span class="stat-label">Articles</span>
    </div>
    <div class="stat">
      <span class="stat-number">25,000+</span>
      <span class="stat-label">PyPI Downloads / month (March 2026)</span>
    </div>
  </div>
</div>

<!-- Social proof / PyPI downloads -->
<section class="pypi-downloads">
  <div class="pypi-downloads-inner">
    <div class="pypi-downloads-header fade-up">
      <div class="section-label section-label--light">Built for practitioners, used by practitioners</div>
      <h2>25,000 downloads in March 2026 across 34 libraries</h2>
      <p>Real adoption from pricing teams &mdash; each download is a pip install on someone&apos;s work machine or Databricks cluster. Without-mirrors counts only, stripping CI bots and PyPI mirrors.</p>
    </div>

    <div class="downloads-grid fade-up">
      <div class="downloads-table">
        <div class="downloads-table-header">
          <span>Library</span>
          <span>Downloads / month</span>
        </div>
        <div class="downloads-row downloads-row--top">
          <span class="dl-name">insurance-causal</span>
          <div class="dl-bar-wrap"><div class="dl-bar" style="width:100%"></div><span class="dl-count">2,003</span></div>
        </div>
        <div class="downloads-row">
          <span class="dl-name">insurance-fairness</span>
          <div class="dl-bar-wrap"><div class="dl-bar" style="width:99%"></div><span class="dl-count">1,992</span></div>
        </div>
        <div class="downloads-row">
          <span class="dl-name">insurance-monitoring</span>
          <div class="dl-bar-wrap"><div class="dl-bar" style="width:81%"></div><span class="dl-count">1,622</span></div>
        </div>
        <div class="downloads-row">
          <span class="dl-name">insurance-causal-policy</span>
          <div class="dl-bar-wrap"><div class="dl-bar" style="width:75%"></div><span class="dl-count">1,495</span></div>
        </div>
        <div class="downloads-row">
          <span class="dl-name">insurance-gam</span>
          <div class="dl-bar-wrap"><div class="dl-bar" style="width:56%"></div><span class="dl-count">1,120</span></div>
        </div>
        <div class="downloads-row">
          <span class="dl-name">insurance-optimise</span>
          <div class="dl-bar-wrap"><div class="dl-bar" style="width:55%"></div><span class="dl-count">1,099</span></div>
        </div>
        <div class="downloads-row">
          <span class="dl-name">insurance-conformal</span>
          <div class="dl-bar-wrap"><div class="dl-bar" style="width:50%"></div><span class="dl-count">1,010</span></div>
        </div>
        <div class="downloads-row">
          <span class="dl-name">insurance-quantile</span>
          <div class="dl-bar-wrap"><div class="dl-bar" style="width:48%"></div><span class="dl-count">968</span></div>
        </div>
        <div class="downloads-row">
          <span class="dl-name">insurance-credibility</span>
          <div class="dl-bar-wrap"><div class="dl-bar" style="width:45%"></div><span class="dl-count">909</span></div>
        </div>
        <div class="downloads-row">
          <span class="dl-name">insurance-severity</span>
          <div class="dl-bar-wrap"><div class="dl-bar" style="width:42%"></div><span class="dl-count">846</span></div>
        </div>
        <div class="downloads-total">
          <span>Top 10 total</span>
          <span>13,064 / month</span>
        </div>
      </div>

      <div class="social-proof-aside">
        <div class="proof-card">
          <div class="proof-number">34</div>
          <div class="proof-label">libraries on PyPI</div>
          <div class="proof-sub">covering every stage of the UK pricing workflow</div>
        </div>
        <div class="proof-card">
          <div class="proof-number">14,600+</div>
          <div class="proof-label">automated tests</div>
          <div class="proof-sub">actuarial correctness checks, not just unit tests</div>
        </div>
        <div class="proof-card">
          <div class="proof-number">515+</div>
          <div class="proof-label">practitioner articles</div>
          <div class="proof-sub">methods explained by people who have used them in sign-off meetings</div>
        </div>
        <div class="proof-cta">
          <a href="https://github.com/burning-cost" target="_blank" class="btn btn-primary">Browse Libraries on GitHub</a>
          <a href="/blog/" class="btn btn-outline">Read the Blog</a>
        </div>
      </div>
    </div>
  </div>
</section>

<!-- Demo notebook -->
<section class="demo-notebook">
  <div class="demo-notebook-inner fade-up">
    <div class="section-label">Try it now</div>
    <h2>Five libraries, one dataset, one notebook</h2>
    <p>A runnable example that fits a CatBoost frequency model, extracts SHAP factor tables, runs a proxy discrimination audit, monitors for drift, and attaches conformal prediction intervals &mdash; all on the same synthetic UK motor dataset. Opens in Google Colab with no local setup required.</p>
    <div class="demo-notebook-cta">
      <a href="https://colab.research.google.com/github/burning-cost/insurance-pricing-demo/blob/main/insurance-pricing-pipeline.ipynb" target="_blank">
        <img src="https://colab.research.google.com/assets/colab-badge.svg" alt="Open In Colab" style="height:32px;">
      </a>
      <a href="https://github.com/burning-cost/insurance-pricing-demo" target="_blank" class="btn btn-outline" style="margin-left:1rem;">View on GitHub</a>
    </div>
    <p class="demo-notebook-libs">
      Uses:
      <a href="https://github.com/burning-cost/shap-relativities">shap-relativities</a> &middot;
      <a href="https://github.com/burning-cost/insurance-fairness">insurance-fairness</a> &middot;
      <a href="https://github.com/burning-cost/insurance-monitoring">insurance-monitoring</a> &middot;
      <a href="https://github.com/burning-cost/insurance-conformal">insurance-conformal</a>
    </p>
  </div>
</section>

<!-- Problem statement -->
<section class="problem">
  <div class="problem-inner fade-up">
    <div class="section-label">The problem we solve</div>
    <h2>The missing piece is not technical skill. It is tooling that bridges the two.</h2>
    <p>Most UK pricing teams have adopted GBMs but are still taking GLM outputs to production. The GBM sits on a server outperforming the production model, but the outputs are not in a form that a rating engine, regulator, or pricing committee can work with. The model never makes it to rates.</p>

    <div class="problem-highlight">
      <p>Each library here solves one specific problem in the pricing workflow. Actuarial tests are included. Outputs use the formats pricing teams already recognise: factor tables, Lorenz curves, A/E ratios, movement-capped rate changes.</p>
    </div>

    <p>sklearn-compatible where it matters. Documented by people who have sat in the same sign-off meetings you have.</p>
  </div>
</section>

<!-- Code previews -->
<section class="code-preview">
  <div class="code-preview-bg"></div>
  <div class="code-preview-inner">
    <div class="code-preview-header fade-up">
      <div class="section-label">See it in practice</div>
      <h2>Three lines to a factor table. Five to validated splits.</h2>
      <p>Real API calls from the libraries. Not wrappers around wrappers. Each one does the specific thing a pricing team needs.</p>
    </div>

    <div class="fade-up">
      <div class="code-tabs">
        <button class="code-tab active" data-tab="shap">shap-relativities</button>
        <button class="code-tab" data-tab="cv">insurance-cv</button>
        <button class="code-tab" data-tab="rate">insurance-optimise</button>
      </div>

      <div id="panel-shap" class="code-panel active">
        <div class="code-block-single">
          <div class="code-block-single-body"><span class="tok-kw">from</span> <span class="tok-cls">shap_relativities</span> <span class="tok-kw">import</span> <span class="tok-cls">SHAPRelativities</span>

<span class="tok-var">sr</span> <span class="tok-op">=</span> <span class="tok-cls">SHAPRelativities</span><span class="tok-pn">(</span><span class="tok-var">model</span><span class="tok-pn">,</span> <span class="tok-var">X_train</span><span class="tok-pn">)</span>
<span class="tok-var">factors</span> <span class="tok-op">=</span> <span class="tok-var">sr</span><span class="tok-pn">.</span><span class="tok-fn">fit_transform</span><span class="tok-pn">(</span><span class="tok-var">X_test</span><span class="tok-pn">)</span>

<span class="tok-cm"># Returns multiplicative factor tables in GLM format</span>
<span class="tok-cm"># Same structure as exp(beta) from your Emblem model</span>
<span class="tok-var">factors</span><span class="tok-pn">.</span><span class="tok-fn">head</span><span class="tok-pn">()</span>
<span class="tok-cm">#  vehicle_age  relativity  ci_lower  ci_upper</span>
<span class="tok-cm">#  0            1.000       0.982     1.018</span>
<span class="tok-cm">#  1            0.912       0.901     0.923</span>

<span class="tok-cm">#  3            0.793       0.780     0.807</span>
<span class="tok-cm">#  4+           0.754       0.739     0.769</span></div>
          <div class="code-block-caption-single">Factor tables, confidence intervals, exposure weighting, reconstruction validation. Output goes straight into a pricing committee pack.</div>
        </div>
      </div>

      <div id="panel-cv" class="code-panel">
        <div class="code-block-single">
          <div class="code-block-single-body"><span class="tok-kw">from</span> <span class="tok-cls">insurance_cv</span> <span class="tok-kw">import</span> <span class="tok-cls">InsuranceTemporalCV</span>
<span class="tok-kw">from</span> <span class="tok-cls">sklearn.model_selection</span> <span class="tok-kw">import</span> <span class="tok-fn">cross_val_score</span>

<span class="tok-var">cv</span> <span class="tok-op">=</span> <span class="tok-cls">InsuranceTemporalCV</span><span class="tok-pn">(</span>
    <span class="tok-var">n_splits</span><span class="tok-op">=</span><span class="tok-num">5</span><span class="tok-pn">,</span>
    <span class="tok-var">ibnr_buffer_months</span><span class="tok-op">=</span><span class="tok-num">6</span>
<span class="tok-pn">)</span>
<span class="tok-var">scores</span> <span class="tok-op">=</span> <span class="tok-fn">cross_val_score</span><span class="tok-pn">(</span>
    <span class="tok-var">model</span><span class="tok-pn">,</span> <span class="tok-var">X</span><span class="tok-pn">,</span> <span class="tok-var">y</span><span class="tok-pn">,</span>
    <span class="tok-var">cv</span><span class="tok-op">=</span><span class="tok-var">cv</span><span class="tok-pn">,</span>
    <span class="tok-var">scoring</span><span class="tok-op">=</span><span class="tok-str">"poisson_deviance"</span>
<span class="tok-pn">)</span>

<span class="tok-cm"># Walk-forward splits - no future data leaks into training folds</span>
<span class="tok-cm"># IBNR buffer prevents immature periods contaminating validation</span>
<span class="tok-fn">print</span><span class="tok-pn">(</span><span class="tok-str">f"CV deviance: {scores.mean():.4f} ± {scores.std():.4f}"</span><span class="tok-pn">)</span></div>
          <div class="code-block-caption-single">Walk-forward splits with configurable IBNR buffers. Temporally correct: no future data leaks into training folds. sklearn-compatible API.</div>
        </div>
      </div>

      <div id="panel-rate" class="code-panel">
        <div class="code-block-single">
          <div class="code-block-single-body"><span class="tok-kw">from</span> <span class="tok-cls">insurance_optimise</span> <span class="tok-kw">import</span> <span class="tok-cls">RateOptimiser</span>

<span class="tok-var">opt</span> <span class="tok-op">=</span> <span class="tok-cls">RateOptimiser</span><span class="tok-pn">(</span>
    <span class="tok-var">current_rates</span><span class="tok-pn">,</span>
    <span class="tok-var">technical_rates</span><span class="tok-pn">,</span>
    <span class="tok-var">exposure</span>
<span class="tok-pn">)</span>
<span class="tok-var">result</span> <span class="tok-op">=</span> <span class="tok-var">opt</span><span class="tok-pn">.</span><span class="tok-fn">optimise</span><span class="tok-pn">(</span>
    <span class="tok-var">max_movement</span><span class="tok-op">=</span><span class="tok-num">0.10</span><span class="tok-pn">,</span>
    <span class="tok-var">target_lr_improvement</span><span class="tok-op">=</span><span class="tok-num">0.03</span>
<span class="tok-pn">)</span>

<span class="tok-cm"># Efficient frontier as a linear programme</span>
<span class="tok-cm"># Respects ±10% movement cap per segment</span>
<span class="tok-fn">print</span><span class="tok-pn">(</span><span class="tok-str">f"LR improvement: {result.lr_delta:.1%}"</span><span class="tok-pn">)</span>
<span class="tok-cm"># LR improvement: 2.8%  (within movement constraints)</span></div>
          <div class="code-block-caption-single">Formulates the efficient frontier as a linear programme. Respects movement caps per segment, targets aggregate loss ratio improvement.</div>
        </div>
      </div>
    </div>
  </div>
</section>

<!-- Who this is for -->
<section class="personas">
  <div class="personas-inner">
    <div class="personas-header fade-up">
      <div class="section-label">Who this is for</div>
      <h2>Built for people who know the problem from the inside</h2>
      <p>These libraries assume you understand insurance pricing. They do not explain what a GLM is.</p>
    </div>
    <div class="persona-grid fade-up-children">

      <div class="persona-card">
        <div class="persona-icon persona-icon--blue">PA</div>
        <div class="persona-title">Pricing actuaries moving from Emblem or Radar to Python</div>
        <p class="persona-desc">You know the techniques. These libraries give you Python equivalents that produce outputs in the same formats you already use: factor tables, A/E ratios, Lorenz curves.</p>
      </div>

      <div class="persona-card">
        <div class="persona-icon persona-icon--purple">DS</div>
        <div class="persona-title">Data scientists joining an insurance pricing team</div>
        <p class="persona-desc">You have the ML skills but lack the actuarial context. These libraries encode that context: correct cross-validation for IBNR, credibility-weighted factors, fairness tests that map to FCA requirements.</p>
      </div>

      <div class="persona-card">
        <div class="persona-icon persona-icon--teal">PM</div>
        <div class="persona-title">Pricing managers evaluating modern tooling</div>
        <p class="persona-desc">You need to know what is production-ready and what is a research prototype. Each library here has actuarial tests, a clear scope, and outputs a pricing team lead can explain to a committee.</p>
      </div>

      <div class="persona-card">
        <div class="persona-icon persona-icon--amber">AR</div>
        <div class="persona-title">Academic researchers working on insurance pricing methods</div>
        <p class="persona-desc">We implement recent literature: Manna et al. (2025) on conformal prediction, BYM2 spatial models, variance-weighted non-conformity scores. Reproducible, documented, testable.</p>
      </div>

    </div>
  </div>
</section>

<!-- Libraries -->
<section class="libraries">

  <div class="section-header fade-up">
    <div class="section-label">Flagship libraries</div>
    <h2>Ten tools for the problems that matter most</h2>
    <p>These are the libraries we think are genuinely differentiated. Each addresses a specific hard problem in UK pricing — regulatory compliance, causal inference, uncertainty quantification, smoothing — where no adequate open-source Python tooling existed before. The full portfolio of 34 libraries is below.</p>
  </div>

  <!-- Flagship grid -->
  <div class="lib-grid lib-grid--flagship fade-up-children">

    <a href="https://github.com/burning-cost/insurance-fairness" target="_blank" class="lib-card lib-card--flagship">
      <span class="lib-card-badge">FCA Consumer Duty</span>
      <span class="lib-card-name">insurance-fairness</span>
      <span class="lib-card-desc">Proxy discrimination auditing for insurance pricing models. Quantifies indirect discrimination from rating variables correlated with protected characteristics. FCA Consumer Duty documentation support built in.</span>
      <span class="lib-card-tags">
        <span class="lib-tag lib-tag--accent">FCA Consumer Duty</span>
        <span class="lib-tag">proxy discrimination</span>
        <span class="lib-tag">bias metrics</span>
      </span>
    </a>

    <a href="https://github.com/burning-cost/insurance-governance" target="_blank" class="lib-card lib-card--flagship">
      <span class="lib-card-badge">PRA SS1/23</span>
      <span class="lib-card-name">insurance-governance</span>
      <span class="lib-card-desc">PRA SS1/23-compliant model validation reports. Bootstrap Gini CI, Poisson A/E CI, double-lift charts, renewal cohort test. 9-section HTML/JSON output ready for model risk committees and the PRA.</span>
      <span class="lib-card-tags">
        <span class="lib-tag lib-tag--accent">PRA SS1/23</span>
        <span class="lib-tag">model governance</span>
        <span class="lib-tag">MRM</span>
      </span>
    </a>

    <a href="https://github.com/burning-cost/insurance-causal" target="_blank" class="lib-card lib-card--flagship">
      <span class="lib-card-name">insurance-causal</span>
      <span class="lib-card-desc">Double machine learning for deconfounding rating factors. Standard GLM coefficients are biased wherever rating variables correlate with distribution channel or policyholder behaviour. DML fixes that bias without a structural model.</span>
      <span class="lib-card-tags">
        <span class="lib-tag lib-tag--accent">causal inference</span>
        <span class="lib-tag">DML</span>
        <span class="lib-tag">deconfounding</span>
      </span>
    </a>

    <a href="https://github.com/burning-cost/insurance-whittaker" target="_blank" class="lib-card lib-card--flagship">
      <span class="lib-card-name">insurance-whittaker</span>
      <span class="lib-card-desc">Whittaker-Henderson smoothing for experience rating tables. 1D, 2D, and Poisson variants with REML lambda selection and Bayesian credible intervals. Smooths age curves, NCD scales, and vehicle group relativities. Pure NumPy/SciPy.</span>
      <span class="lib-card-tags">
        <span class="lib-tag lib-tag--accent">REML</span>
        <span class="lib-tag">smoothing</span>
        <span class="lib-tag">experience rating</span>
      </span>
    </a>

    <a href="https://github.com/burning-cost/insurance-telematics" target="_blank" class="lib-card lib-card--flagship">
      <span class="lib-card-name">insurance-telematics</span>
      <span class="lib-card-desc">End-to-end pipeline from raw 1Hz GPS/accelerometer data to GLM-compatible risk scores. HMM driving state classification (cautious/normal/aggressive), Bühlmann-Straub credibility aggregation to driver level, Poisson GLM integration.</span>
      <span class="lib-card-tags">
        <span class="lib-tag lib-tag--accent">HMM</span>
        <span class="lib-tag">telematics</span>
        <span class="lib-tag">UBI</span>
      </span>
    </a>

    <a href="https://github.com/burning-cost/insurance-conformal" target="_blank" class="lib-card lib-card--flagship">
      <span class="lib-card-name">insurance-conformal</span>
      <span class="lib-card-desc">Distribution-free prediction intervals with finite-sample coverage guarantees. Variance-weighted non-conformity scores produce tighter intervals than the naive approach with identical coverage. Solvency II SCR bounds included.</span>
      <span class="lib-card-tags">
        <span class="lib-tag lib-tag--accent">conformal prediction</span>
        <span class="lib-tag">uncertainty</span>
        <span class="lib-tag">Solvency II</span>
      </span>
    </a>

    <a href="https://github.com/burning-cost/insurance-credibility" target="_blank" class="lib-card lib-card--flagship">
      <span class="lib-card-name">insurance-credibility</span>
      <span class="lib-card-desc">Bühlmann-Straub credibility in Python with mixed-model equivalence checks. Practical for capping thin segments, stabilising NCD factors, and blending a new model with an incumbent rate. The actuarial solution to the thin data problem.</span>
      <span class="lib-card-tags">
        <span class="lib-tag lib-tag--accent">Bühlmann-Straub</span>
        <span class="lib-tag">thin data</span>
        <span class="lib-tag">NCD</span>
      </span>
    </a>

    <a href="https://github.com/burning-cost/insurance-frequency-severity" target="_blank" class="lib-card lib-card--flagship">
      <span class="lib-card-name">insurance-frequency-severity</span>
      <span class="lib-card-desc">Sarmanov copula joint frequency-severity modelling with GLM marginals. IFM estimation, analytical premium correction, Garrido conditional severity, dependence tests. Tests the independence assumption every pricing model makes. 80+ tests.</span>
      <span class="lib-card-tags">
        <span class="lib-tag lib-tag--accent">Sarmanov copula</span>
        <span class="lib-tag">joint modelling</span>
        <span class="lib-tag">dependence</span>
      </span>
    </a>

    <a href="https://github.com/burning-cost/insurance-gam" target="_blank" class="lib-card lib-card--flagship">
      <span class="lib-card-name">insurance-gam</span>
      <span class="lib-card-desc">EBM and Neural Additive Model for interpretable deep learning in insurance pricing. Shape functions per rating factor give the transparency of a GLM with the predictive power of a neural network. Outputs factor tables for pricing committee review.</span>
      <span class="lib-card-tags">
        <span class="lib-tag lib-tag--accent">EBM</span>
        <span class="lib-tag">NAM</span>
        <span class="lib-tag">interpretable ML</span>
      </span>
    </a>

    <a href="https://github.com/burning-cost/insurance-monitoring" target="_blank" class="lib-card lib-card--flagship">
      <span class="lib-card-name">insurance-monitoring</span>
      <span class="lib-card-desc">Model drift detection for insurance pricing - exposure-weighted PSI/CSI, actual-vs-expected ratios, Gini drift z-test. Sequential testing for anytime-valid champion/challenger A/B tests. PIT calibration monitoring and feature-interaction drift attribution.</span>
      <span class="lib-card-tags">
        <span class="lib-tag lib-tag--accent">drift detection</span>
        <span class="lib-tag">sequential testing</span>
        <span class="lib-tag">A/E ratios</span>
      </span>
    </a>

  </div>

  <div class="gs-cta-row fade-up">
    <a href="/tools/" class="btn btn-primary">View all 34 libraries</a>
    <a href="/guide/" class="btn btn-outline">Which library do I need?</a>
  </div>

  <!-- Full portfolio (collapsed) -->
  <div class="section-header fade-up" style="margin-top: 3rem;">
    <div class="section-label">Full library portfolio</div>
    <h2>The complete pricing workflow, covered</h2>
    <p>All 34 libraries. Each solves one well-defined problem. Actuarial tests included. sklearn-compatible where it matters.</p>
  </div>

  <!-- Workflow diagram -->
  <div class="workflow fade-up">
    <div class="workflow-track">
      <div class="workflow-stage">
        <div class="workflow-node">
          <div class="workflow-node-icon">&#128201;</div>
          <div class="workflow-node-name">Data</div>
        </div>
        <div class="workflow-lib-labels">
          <span class="workflow-lib">insurance-cv</span>
          <span class="workflow-lib">ins-synthetic</span>
          <span class="workflow-lib">ins-datasets</span>
        </div>
      </div>
      <div class="workflow-arrow">&#8594;</div>
      <div class="workflow-stage">
        <div class="workflow-node">
          <div class="workflow-node-icon">&#129504;</div>
          <div class="workflow-node-name">Model</div>
        </div>
        <div class="workflow-lib-labels">
          <span class="workflow-lib">ins-credibility</span>
          <span class="workflow-lib">bayesian-pricing</span>
          <span class="workflow-lib">ins-multilevel</span>
          <span class="workflow-lib">ins-spatial</span>
          <span class="workflow-lib">ins-trend</span>
          <span class="workflow-lib">ins-anam</span>
          <span class="workflow-lib">ins-dispersion</span>
          <span class="workflow-lib">ins-experience</span>
        </div>
      </div>
      <div class="workflow-arrow">&#8594;</div>
      <div class="workflow-stage">
        <div class="workflow-node">
          <div class="workflow-node-icon">&#128269;</div>
          <div class="workflow-node-name">Interpret</div>
        </div>
        <div class="workflow-lib-labels">
          <span class="workflow-lib">shap-relativities</span>
          <span class="workflow-lib">ins-interactions</span>
          <span class="workflow-lib">ins-causal</span>
        </div>
      </div>
      <div class="workflow-arrow">&#8594;</div>
      <div class="workflow-stage">
        <div class="workflow-node">
          <div class="workflow-node-icon">&#10003;</div>
          <div class="workflow-node-name">Validate</div>
        </div>
        <div class="workflow-lib-labels">
          <span class="workflow-lib">ins-governance</span>
          <span class="workflow-lib">ins-conformal</span>
          <span class="workflow-lib">ins-monitoring</span>
        </div>
      </div>
      <div class="workflow-arrow">&#8594;</div>
      <div class="workflow-stage">
        <div class="workflow-node">
          <div class="workflow-node-icon">&#9878;</div>
          <div class="workflow-node-name">Compliance</div>
        </div>
        <div class="workflow-lib-labels">
          <span class="workflow-lib">ins-fairness</span>
          <span class="workflow-lib">ins-causal-policy</span>
          <span class="workflow-lib">ins-governance</span>
        </div>
      </div>
      <div class="workflow-arrow">&#8594;</div>
      <div class="workflow-stage">
        <div class="workflow-node">
          <div class="workflow-node-icon">&#128200;</div>
          <div class="workflow-node-name">Deploy</div>
        </div>
        <div class="workflow-lib-labels">
          <span class="workflow-lib">ins-optimise</span>
          <span class="workflow-lib">ins-causal</span>
          <span class="workflow-lib">ins-deploy</span>
        </div>
      </div>
    </div>
  </div>

  <details class="lib-portfolio-details">
    <summary class="lib-portfolio-summary">Show full portfolio (34 libraries)</summary>

    <div class="lib-group">
      <div class="lib-group-header">
        <span class="lib-group-icon">&#128269;</span>
        <span class="lib-group-label">Model interpretation</span>
      </div>
      <div class="lib-grid fade-up-children">
        <a href="https://github.com/burning-cost/shap-relativities" target="_blank" class="lib-card">
          <span class="lib-card-name">shap-relativities</span>
          <span class="lib-card-desc">Extract multiplicative rating factor tables from CatBoost models using SHAP values. Same output format as exp(&beta;) from a GLM: factor tables, confidence intervals, exposure weighting, reconstruction validation.</span>
          <span class="lib-card-tags">
            <span class="lib-tag">SHAP</span>
            <span class="lib-tag">CatBoost</span>
            <span class="lib-tag">factor tables</span>
          </span>
        </a>
      </div>
    </div>

    <div class="lib-group">
      <div class="lib-group-header">
        <span class="lib-group-icon">&#10003;</span>
        <span class="lib-group-label">Validation &amp; deployment</span>
      </div>
      <div class="lib-grid fade-up-children">
        <a href="https://github.com/burning-cost/insurance-cv" target="_blank" class="lib-card">
          <span class="lib-card-name">insurance-cv</span>
          <span class="lib-card-desc">Temporally-correct cross-validation for insurance pricing models. Walk-forward splits with configurable IBNR buffers, Poisson and Gamma deviance scorers, sklearn-compatible API.</span>
          <span class="lib-card-tags">
            <span class="lib-tag">walk-forward</span>
            <span class="lib-tag">sklearn</span>
            <span class="lib-tag">IBNR</span>
          </span>
        </a>
        <a href="https://github.com/burning-cost/insurance-conformal" target="_blank" class="lib-card">
          <span class="lib-card-name">insurance-conformal</span>
          <span class="lib-card-desc">Distribution-free prediction intervals for insurance GBMs. Variance-weighted non-conformity scores produce narrower intervals than the naive approach with identical coverage guarantees.</span>
          <span class="lib-card-tags">
            <span class="lib-tag">conformal prediction</span>
            <span class="lib-tag">GBM</span>
          </span>
        </a>
        <a href="https://github.com/burning-cost/insurance-conformal-ts" target="_blank" class="lib-card">
          <span class="lib-card-name">insurance-conformal-ts</span>
          <span class="lib-card-desc">Conformal prediction for non-exchangeable claims time series. ACI, EnbPI, SPCI, and MSCP methods with Poisson and negative binomial non-conformity scores. Coverage guarantees without exchangeability.</span>
          <span class="lib-card-tags">
            <span class="lib-tag">conformal prediction</span>
            <span class="lib-tag">time series</span>
            <span class="lib-tag">ACI</span>
          </span>
        </a>
        <a href="https://github.com/burning-cost/insurance-monitoring" target="_blank" class="lib-card">
          <span class="lib-card-name">insurance-monitoring</span>
          <span class="lib-card-desc">Three-layer model monitoring for deployed pricing models. Exposure-weighted PSI/CSI, segmented A/E ratios with IBNR adjustment, and a formal Gini z-test to distinguish recalibration from refit signals.</span>
          <span class="lib-card-tags">
            <span class="lib-tag">PSI/CSI</span>
            <span class="lib-tag">A/E ratios</span>
            <span class="lib-tag">Gini z-test</span>
          </span>
        </a>
        <a href="https://github.com/burning-cost/insurance-governance" target="_blank" class="lib-card">
          <span class="lib-card-name">insurance-governance</span>
          <span class="lib-card-desc">PRA SS1/23 compliant model validation reports. Bootstrap Gini CI, Poisson A/E CI, double-lift charts, renewal cohort test, 9-section HTML/JSON output ready for model governance.</span>
          <span class="lib-card-tags">
            <span class="lib-tag">PRA SS1/23</span>
            <span class="lib-tag">model governance</span>
            <span class="lib-tag">HTML report</span>
          </span>
        </a>
        <a href="https://github.com/burning-cost/insurance-deploy" target="_blank" class="lib-card">
          <span class="lib-card-name">insurance-deploy</span>
          <span class="lib-card-desc">Champion/challenger pricing framework. Shadow mode deployment, SHA-256 deterministic routing, SQLite quote log, bootstrap likelihood ratio comparison, ENBP audit trail for ICOBS 6B.2.</span>
          <span class="lib-card-tags">
            <span class="lib-tag">champion/challenger</span>
            <span class="lib-tag">shadow mode</span>
            <span class="lib-tag">audit trail</span>
          </span>
        </a>
      </div>
    </div>

    <div class="lib-group">
      <div class="lib-group-header">
        <span class="lib-group-icon">&#9881;</span>
        <span class="lib-group-label">Techniques</span>
      </div>
      <div class="lib-grid fade-up-children">
        <a href="https://github.com/burning-cost/insurance-credibility" target="_blank" class="lib-card">
          <span class="lib-card-name">insurance-credibility</span>
          <span class="lib-card-desc">Buhlmann-Straub credibility in Python, with mixed-model equivalence checks. Practical for capping thin segments, stabilising NCD factors, and blending a new model with an incumbent rate.</span>
          <span class="lib-card-tags">
            <span class="lib-tag">Buhlmann-Straub</span>
            <span class="lib-tag">Polars</span>
          </span>
        </a>

        <a href="https://github.com/burning-cost/bayesian-pricing" target="_blank" class="lib-card">
          <span class="lib-card-name">bayesian-pricing</span>
          <span class="lib-card-desc">Hierarchical Bayesian models for thin-data pricing segments. Partial pooling across risk groups, with credibility factor output in a format that maps back to traditional actuarial review.</span>
          <span class="lib-card-tags">
            <span class="lib-tag">hierarchical Bayes</span>
            <span class="lib-tag">partial pooling</span>
          </span>
        </a>
        <a href="https://github.com/burning-cost/insurance-interactions" target="_blank" class="lib-card">
          <span class="lib-card-name">insurance-interactions</span>
          <span class="lib-card-desc">Tools for detecting, quantifying, and presenting interaction effects in insurance pricing models: the effects a main-effects-only GLM cannot see.</span>
          <span class="lib-card-tags">
            <span class="lib-tag">interaction effects</span>
            <span class="lib-tag">GLM diagnostics</span>
          </span>
        </a>
        <a href="https://github.com/burning-cost/insurance-causal" target="_blank" class="lib-card">
          <span class="lib-card-name">insurance-causal</span>
          <span class="lib-card-desc">Causal inference methods for insurance pricing. Separating genuine risk signal from confounded association, relevant wherever rating factors are correlated with distribution channel or policyholder behaviour.</span>
          <span class="lib-card-tags">
            <span class="lib-tag">causal inference</span>
            <span class="lib-tag">deconfounding</span>
          </span>
        </a>
        <a href="https://github.com/burning-cost/insurance-spatial" target="_blank" class="lib-card">
          <span class="lib-card-name">insurance-spatial</span>
          <span class="lib-card-desc">Spatial territory ratemaking using BYM2 models. Geographically smoothed relativities that borrow strength across adjacent areas, particularly useful for postcode-level home and motor models with thin data.</span>
          <span class="lib-card-tags">
            <span class="lib-tag">BYM2</span>
            <span class="lib-tag">spatial smoothing</span>
            <span class="lib-tag">postcode</span>
          </span>
        </a>
        <a href="https://github.com/burning-cost/insurance-multilevel" target="_blank" class="lib-card">
          <span class="lib-card-name">insurance-multilevel</span>
          <span class="lib-card-desc">Two-stage CatBoost + REML random effects for high-cardinality group factors like broker, scheme, or fleet. Credibility-weighted adjustments with ICC diagnostics.</span>
          <span class="lib-card-tags">
            <span class="lib-tag">mixed effects</span>
            <span class="lib-tag">CatBoost</span>
            <span class="lib-tag">credibility</span>
          </span>
        </a>
        <a href="https://github.com/burning-cost/insurance-distill" target="_blank" class="lib-card">
          <span class="lib-card-name">insurance-distill</span>
          <span class="lib-card-desc">GBM-to-GLM distillation. Fits a surrogate Poisson or Gamma GLM to CatBoost predictions and exports multiplicative factor tables that Radar or Emblem can consume directly.</span>
          <span class="lib-card-tags">
            <span class="lib-tag">distillation</span>
            <span class="lib-tag">CatBoost</span>
            <span class="lib-tag">factor tables</span>
          </span>
        </a>
        <a href="https://github.com/burning-cost/insurance-survival" target="_blank" class="lib-card">
          <span class="lib-card-name">insurance-survival</span>
          <span class="lib-card-desc">Survival models for insurance retention. Cure rate models, CLV, lapse tables, insurance exposure handling, MLflow wrapper. Wraps lifelines with insurance-specific gaps filled.</span>
          <span class="lib-card-tags">
            <span class="lib-tag">lapse</span>
            <span class="lib-tag">retention</span>
            <span class="lib-tag">CLV</span>
          </span>
        </a>
        <a href="https://github.com/burning-cost/insurance-causal-policy" target="_blank" class="lib-card">
          <span class="lib-card-name">insurance-causal-policy</span>
          <span class="lib-card-desc">Synthetic difference-in-differences for causal rate change evaluation. Event study, HonestDiD sensitivity, FCA evidence pack. Did your rate change actually cause the movement you see?</span>
          <span class="lib-card-tags">
            <span class="lib-tag">SDID</span>
            <span class="lib-tag">causal inference</span>
            <span class="lib-tag">rate change</span>
          </span>
        </a>
        <a href="https://github.com/burning-cost/insurance-trend" target="_blank" class="lib-card">
          <span class="lib-card-name">insurance-trend</span>
          <span class="lib-card-desc">Loss cost trend analysis with structural break detection. Identifies regime changes in claims frequency and severity, producing annualised trend factors with confidence intervals for reserving and pricing review.</span>
          <span class="lib-card-tags">
            <span class="lib-tag">trend analysis</span>
            <span class="lib-tag">structural breaks</span>
            <span class="lib-tag">loss cost</span>
          </span>
        </a>
        <a href="https://github.com/burning-cost/insurance-gam" target="_blank" class="lib-card">
          <span class="lib-card-name">insurance-gam</span>
          <span class="lib-card-desc">Actuarial Neural Additive Model for interpretable deep learning in insurance pricing. Shape functions per rating factor give the transparency of a GLM with the predictive power of a neural network.</span>
          <span class="lib-card-tags">
            <span class="lib-tag">neural additive model</span>
            <span class="lib-tag">interpretable ML</span>
            <span class="lib-tag">deep learning</span>
          </span>
        </a>
        <a href="https://github.com/burning-cost/insurance-glm-tools" target="_blank" class="lib-card">
          <span class="lib-card-name">insurance-glm-tools</span>
          <span class="lib-card-desc">Automated GLM factor level clustering for insurance pricing. R2VF algorithm collapses 500 vehicle makes or 350 occupation codes into pricing bands. Ridge ranking for nominals, fused lasso for fusion, unpenalised GLM refit. BIC lambda selection. Monotonicity enforcement.</span>
          <span class="lib-card-tags">
            <span class="lib-tag">fused lasso</span>
            <span class="lib-tag">categorical banding</span>
            <span class="lib-tag">GLM</span>
          </span>
        </a>
        <a href="https://github.com/burning-cost/insurance-thin-data" target="_blank" class="lib-card">
          <span class="lib-card-name">insurance-thin-data</span>
          <span class="lib-card-desc">Foundation model pricing for thin segments. TabPFN v2 and TabICLv2 backend abstraction with a unified sklearn-compatible API, automated GLM benchmark, PDP relativities in standard factor table format, CommitteeReport output. When a segment has 200 policies, prior-fitted networks outperform GLMs without hyperparameter search. 72 tests.</span>
          <span class="lib-card-tags">
            <span class="lib-tag">TabPFN</span>
            <span class="lib-tag">thin data</span>
            <span class="lib-tag">foundation model</span>
          </span>
        </a>
      </div>
    </div>

    <div class="lib-group">
      <div class="lib-group-header">
        <span class="lib-group-icon">&#128200;</span>
        <span class="lib-group-label">Commercial</span>
      </div>
      <div class="lib-grid fade-up-children">
        <a href="https://github.com/burning-cost/insurance-optimise" target="_blank" class="lib-card">
          <span class="lib-card-name">insurance-optimise</span>
          <span class="lib-card-desc">Constrained rate change optimisation for UK personal lines. Formulates the efficient frontier between loss ratio improvement and movement cap constraints as a linear programme.</span>
          <span class="lib-card-tags">
            <span class="lib-tag">linear programming</span>
            <span class="lib-tag">movement caps</span>
          </span>
        </a>


      </div>
    </div>

    <div class="lib-group">
      <div class="lib-group-header">
        <span class="lib-group-icon">&#128202;</span>
        <span class="lib-group-label">Tail risk &amp; distributions</span>
      </div>
      <div class="lib-grid fade-up-children">
        <a href="https://github.com/burning-cost/insurance-quantile" target="_blank" class="lib-card">
          <span class="lib-card-name">insurance-quantile</span>
          <span class="lib-card-desc">Actuarial tail risk quantile/expectile regression. QuantileGBM wrapping CatBoost MultiQuantile, per-risk TVaR, large loss loading, ILF tables, exceedance curves, isotonic crossing fix.</span>
          <span class="lib-card-tags">
            <span class="lib-tag">quantile GBM</span>
            <span class="lib-tag">TVaR</span>
            <span class="lib-tag">tail risk</span>
          </span>
        </a>
        <a href="https://github.com/burning-cost/insurance-distributional" target="_blank" class="lib-card">
          <span class="lib-card-name">insurance-distributional</span>
          <span class="lib-card-desc">Distributional GBMs for insurance. TweedieGBM, GammaGBM, ZIPGBM, NegBinomialGBM. Smyth-J&oslash;rgensen double GLM for dispersion. Per-risk volatility scoring.</span>
          <span class="lib-card-tags">
            <span class="lib-tag">Tweedie</span>
            <span class="lib-tag">distributional</span>
            <span class="lib-tag">volatility</span>
          </span>
        </a>
        <a href="https://github.com/burning-cost/insurance-synthetic" target="_blank" class="lib-card">
          <span class="lib-card-name">insurance-synthetic</span>
          <span class="lib-card-desc">Vine copula synthetic portfolio generation with insurance-specific marginals. Exposure-aware Poisson, constraint enforcement, fidelity reporting with TSTR Gini gap.</span>
          <span class="lib-card-tags">
            <span class="lib-tag">vine copula</span>
            <span class="lib-tag">synthetic data</span>
            <span class="lib-tag">fidelity</span>
          </span>
        </a>
        <a href="https://github.com/burning-cost/insurance-datasets" target="_blank" class="lib-card">
          <span class="lib-card-name">insurance-datasets</span>
          <span class="lib-card-desc">Synthetic UK motor insurance datasets with known data generating process. Useful for model validation benchmarks, library demos, and training exercises.</span>
          <span class="lib-card-tags">
            <span class="lib-tag">synthetic data</span>
            <span class="lib-tag">UK motor</span>
            <span class="lib-tag">DGP</span>
          </span>
        </a>
        <a href="https://github.com/burning-cost/insurance-distributional-glm" target="_blank" class="lib-card">
          <span class="lib-card-name">insurance-distributional-glm</span>
          <span class="lib-card-desc">GAMLSS for Python: model ALL distribution parameters as functions of covariates — mean, dispersion, shape, zero-inflation. Seven families, RS algorithm with backtracking. Pure NumPy/SciPy.</span>
          <span class="lib-card-tags">
            <span class="lib-tag">GAMLSS</span>
            <span class="lib-tag">distributional GLM</span>
            <span class="lib-tag">NumPy/SciPy</span>
          </span>
        </a>
        <a href="https://github.com/burning-cost/insurance-dispersion" target="_blank" class="lib-card">
          <span class="lib-card-name">insurance-dispersion</span>
          <span class="lib-card-desc">Double GLM for joint mean-dispersion modelling. Standard GLMs assign a single scalar phi to all observations — fleet brokers and personal lines get identical volatility assumptions. DGLM fits a second regression for phi so dispersion varies by risk segment. Alternating IRLS, REML correction, overdispersion LRT, actuarial factor tables. 79 tests.</span>
          <span class="lib-card-tags">
            <span class="lib-tag">double GLM</span>
            <span class="lib-tag">dispersion</span>
            <span class="lib-tag">REML</span>
          </span>
        </a>
        <a href="https://github.com/burning-cost/insurance-severity" target="_blank" class="lib-card">
          <span class="lib-card-name">insurance-severity</span>
          <span class="lib-card-desc">Composite severity regression with spliced body/tail distributions, covariate-dependent thresholds, ILF estimation, and TVaR. The body and tail obey different physics; this fits them separately while letting the threshold vary by covariate. 106 tests.</span>
          <span class="lib-card-tags">
            <span class="lib-tag">spliced distributions</span>
            <span class="lib-tag">ILF</span>
            <span class="lib-tag">TVaR</span>
          </span>
        </a>
      </div>
    </div>

    <div class="lib-group">
      <div class="lib-group-header">
        <span class="lib-group-icon">&#9878;</span>
        <span class="lib-group-label">Compliance</span>
      </div>
      <div class="lib-grid fade-up-children">
        <a href="https://github.com/burning-cost/insurance-fairness" target="_blank" class="lib-card">
          <span class="lib-card-name">insurance-fairness</span>
          <span class="lib-card-desc">Proxy discrimination detection for insurance pricing models. Measures of disparate impact, fairness-accuracy trade-off analysis, FCA Consumer Duty documentation support.</span>
          <span class="lib-card-tags">
            <span class="lib-tag">FCA Consumer Duty</span>
            <span class="lib-tag">proxy discrimination</span>
          </span>
        </a>
      </div>
    </div>

    <div class="lib-group">
      <div class="lib-group-header">
        <span class="lib-group-icon">&#128663;</span>
        <span class="lib-group-label">Telematics</span>
      </div>
      <div class="lib-grid fade-up-children">
        <a href="https://github.com/burning-cost/insurance-telematics" target="_blank" class="lib-card">
          <span class="lib-card-name">insurance-telematics</span>
          <span class="lib-card-desc">End-to-end pipeline from raw 1Hz GPS/accelerometer data to GLM-compatible risk scores. HMM driving state classification (cautious/normal/aggressive), Buhlmann-Straub credibility aggregation to driver level, Poisson GLM integration. TripSimulator for synthetic fleet generation.</span>
          <span class="lib-card-tags">
            <span class="lib-tag">HMM</span>
            <span class="lib-tag">telematics</span>
            <span class="lib-tag">UBI</span>
          </span>
        </a>
      </div>
    </div>

    <div class="lib-group">
      <div class="lib-group-header">
        <span class="lib-group-icon">&#128202;</span>
        <span class="lib-group-label">Trend &amp; time series</span>
      </div>
      <div class="lib-grid fade-up-children">
        <a href="https://github.com/burning-cost/insurance-dynamics" target="_blank" class="lib-card">
          <span class="lib-card-name">insurance-dynamics</span>
          <span class="lib-card-desc">GAS (Generalised Autoregressive Score) models for dynamic insurance pricing. Exposure-weighted Poisson, Gamma, Beta, and NegBin variants. GAS(1,1) MLE, trend_index output in development factor format, panel data across rating cells. First maintained Python GAS library.</span>
          <span class="lib-card-tags">
            <span class="lib-tag">GAS models</span>
            <span class="lib-tag">dynamic pricing</span>
            <span class="lib-tag">trend</span>
          </span>
        </a>
      </div>
    </div>

    <div class="lib-group">
      <div class="lib-group-header">
        <span class="lib-group-icon">&#128269;</span>
        <span class="lib-group-label">Advanced modelling</span>
      </div>
      <div class="lib-grid fade-up-children">
        <a href="https://github.com/burning-cost/insurance-whittaker" target="_blank" class="lib-card">
          <span class="lib-card-name">insurance-whittaker</span>
          <span class="lib-card-desc">Whittaker-Henderson smoothing for experience rating tables. 1D, 2D, and Poisson variants with REML/GCV/AIC lambda selection and Bayesian CIs. Smooths noisy age curves, NCD scales, and vehicle group relativities. Pure NumPy/SciPy.</span>
          <span class="lib-card-tags">
            <span class="lib-tag">smoothing</span>
            <span class="lib-tag">experience rating</span>
            <span class="lib-tag">REML</span>
          </span>
        </a>
        <a href="https://github.com/burning-cost/insurance-frequency-severity" target="_blank" class="lib-card">
          <span class="lib-card-name">insurance-frequency-severity</span>
          <span class="lib-card-desc">Sarmanov copula joint frequency-severity modelling with GLM marginals. IFM estimation, analytical premium correction, Garrido conditional severity, dependence tests. Tests the independence assumption every pricing model makes. 80+ tests.</span>
          <span class="lib-card-tags">
            <span class="lib-tag">Sarmanov copula</span>
            <span class="lib-tag">joint modelling</span>
            <span class="lib-tag">dependence</span>
          </span>
        </a>
      </div>
    </div>

    <div class="lib-group">
      <div class="lib-group-header">
        <span class="lib-group-icon">&#128203;</span>
        <span class="lib-group-label">Covariate shift</span>
      </div>
      <div class="lib-grid fade-up-children">
        <a href="https://github.com/burning-cost/insurance-covariate-shift" target="_blank" class="lib-card">
          <span class="lib-card-name">insurance-covariate-shift</span>
          <span class="lib-card-desc">Density ratio correction for insurance book shifts. CatBoost, RuLSIF, and KLIEP importance weighting, LR-QR conformal prediction under shift, FCA SUP 15.3 diagnostics. When your book mix changes, your model weights need to change too. 138 tests.</span>
          <span class="lib-card-tags">
            <span class="lib-tag">covariate shift</span>
            <span class="lib-tag">density ratio</span>
            <span class="lib-tag">importance weighting</span>
          </span>
        </a>
      </div>
    </div>

  </details>

</section>

<!-- Start here -->
<section class="start-here">
  <div class="start-here-inner">
    <div class="start-here-header fade-up">
      <div class="section-label">New here?</div>
      <h2>Start here</h2>
      <p>If you are a pricing actuary visiting for the first time, these are the posts worth reading first. Picked for coverage, not for clicks.</p>
    </div>
    <div class="start-here-grid fade-up-children">
      <a class="start-here-card" href="/2026/03/23/getting-started-three-libraries-one-workflow/">
        <div class="sh-tag">Getting started</div>
        <div class="sh-title">Getting Started: Three Libraries, One Workflow</div>
        <div class="sh-desc">Use insurance-causal, insurance-conformal, and insurance-monitoring together on one model. The practical entry point.</div>
      </a>


      <a class="start-here-card" href="/2026/03/22/fca-proxy-discrimination-python-testing-guide/">
        <div class="sh-tag">Compliance</div>
        <div class="sh-title">FCA Proxy Discrimination Testing in Python</div>
        <div class="sh-desc">EP25/2 requires non-life insurers to test for proxy discrimination. Here is exactly how to run that test with audit-ready output.</div>
      </a>
      <a class="start-here-card" href="/2026/03/22/three-methods-proxy-discrimination-detection-compared/">
        <div class="sh-tag">Methodology</div>
        <div class="sh-title">Three Ways to Detect Proxy Discrimination: When They Agree and When They Don&rsquo;t</div>
        <div class="sh-desc">Mutual information, proxy R-squared, and SHAP proxy scores all flag proxies but catch different things. How to interpret conflicting signals.</div>
      </a>
      <a class="start-here-card" href="/2026/03/22/insurance-model-monitoring-gini-ae-double-lift-python/">
        <div class="sh-tag">Monitoring</div>
        <div class="sh-title">Insurance Model Monitoring: Gini Drift, A/E Ratios and Double-Lift Curves</div>
        <div class="sh-desc">What pricing actuaries actually monitor and how to do it in Python. Gini stability, A/E by segment, PSI, and double-lift.</div>
      </a>
      <a class="start-here-card" href="/2026/03/25/how-to-build-double-lift-chart-python/">
        <div class="sh-tag">How-to</div>
        <div class="sh-title">How to Build a Double-Lift Chart in Python</div>
        <div class="sh-desc">The standard UK actuarial tool for comparing two models. Bin by prediction ratio, compute A/E per decile, plot with matplotlib.</div>
      </a>
      <a class="start-here-card" href="/2026/03/24/how-to-export-factor-table-to-excel-python/">
        <div class="sh-tag">How-to</div>
        <div class="sh-title">How to Export a Factor Table to Excel in Python</div>
        <div class="sh-desc">Extract CatBoost relativities with shap-relativities and write formatted Excel output ready for pricing committee review or Radar import.</div>
      </a>

      <a class="start-here-card" href="/2026/03/22/python-insurance-pricing-cookbook/">
        <div class="sh-tag">Reference</div>
        <div class="sh-title">Python Insurance Pricing Cookbook: 20 Recipes</div>
        <div class="sh-desc">20 short code recipes for common pricing tasks, each using a real API from one of our libraries. Copy and adapt.</div>
      </a>
    </div>
  </div>
</section>

<!-- Training Course CTA -->
<section class="course-cta">
  <div class="course-cta-inner">
    <div class="section-label">Structured learning</div>
    <h2>Insurance Pricing in Python &mdash; 12-module training course</h2>
    <p>A guided path through the full pricing workflow &mdash; GLMs to GAMs, causal inference, fairness auditing, and conformal prediction &mdash; in 12 hands-on modules with real pricing notebooks. Free and open source on GitHub.</p>
    <a href="/course/" class="btn btn-primary">View the Course &rarr;</a>
  </div>
</section>

<!-- Blog posts -->
<section class="posts">
  <div class="posts-inner">
    <div class="posts-header fade-up">
      <div class="section-label">From the blog</div>
      <h2>Practitioner articles on insurance pricing</h2>
    </div>

    <div class="post-grid fade-up-children">
      {% assign featured_posts = site.posts | where: "featured", true %}
      {% for post in featured_posts limit:9 %}
      <div class="post-card">
        <div class="post-date">{{ post.date | date: "%d %b %Y" }}</div>
        <div class="post-title"><a href="{{ post.url }}">{{ post.title }}</a></div>
        {% if post.description %}
        <div class="post-excerpt">{{ post.description }}</div>
        {% elsif post.excerpt %}
        <div class="post-excerpt">{{ post.excerpt | strip_html | truncate: 140 }}</div>
        {% endif %}
        <a href="{{ post.url }}" class="post-read-more">Read article &rarr;</a>
      </div>
      {% endfor %}
    </div>

    <div class="post-more">
      <a href="/blog/" class="btn btn-outline-light">All articles &rarr;</a>
    </div>
  </div>
</section>

<!-- Recent articles list: plain crawlable links so Googlebot can discover the full archive -->
<section class="recent-articles" style="background:#f7f8fc;padding:2.5rem 1.5rem;">
  <div style="max-width:1100px;margin:0 auto;">
    <div style="font-size:0.65rem;font-weight:700;letter-spacing:0.12em;text-transform:uppercase;color:#5a6278;margin-bottom:1.25rem;">Recent articles</div>
    <ul style="list-style:none;display:grid;grid-template-columns:repeat(auto-fill,minmax(300px,1fr));gap:0.5rem 2rem;">
      {% for post in site.posts limit:30 %}
      <li style="font-size:0.875rem;padding:0.35rem 0;border-bottom:1px solid #e4e7ef;">
        <a href="{{ post.url | relative_url }}" style="color:#1a1d2e;text-decoration:none;" onmouseover="this.style.color='#4f7ef8'" onmouseout="this.style.color='#1a1d2e'">{{ post.title }}</a>
        <span style="font-size:0.7rem;color:#8b92a8;margin-left:0.5rem;">{{ post.date | date: "%d %b %Y" }}</span>
      </li>
      {% endfor %}
    </ul>
    <div style="margin-top:1.25rem;">
      <a href="/blog/" style="font-size:0.875rem;color:#4f7ef8;">View all {{ site.posts | size }} articles &rarr;</a>
    </div>
  </div>
</section>