index.html

---
layout: landing
title: "Burning Cost - Open-source pricing tools for UK insurance teams"
description: "14 Python libraries covering the full pricing workflow. From walk-forward cross-validation to constrained rate optimisation. Built by practitioners, for UK personal lines pricing teams."
---

<!-- 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>
      Open source &middot; UK insurance pricing
    </div>
    <h1>Your <em>GBM</em> outperforms.<br>Your <em>GLM</em> is still live.</h1>
    <p class="hero-sub">14 Python libraries that bridge the gap: from walk-forward cross-validation to constrained rate optimisation. Written for teams that already know what a GLM is.</p>
    <div class="hero-cta">
      <a href="/course/" class="btn btn-primary">Training course &rarr;</a>
      <a href="https://github.com/burningcost" target="_blank" class="btn btn-outline">Browse on GitHub</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">
      <span class="stat-number">14</span>
      <span class="stat-label">Libraries</span>
    </div>
    <div class="stat">
      <span class="stat-number">600+</span>
      <span class="stat-label">Tests</span>
    </div>
    <div class="stat">
      <span class="stat-number">17</span>
      <span class="stat-label">Articles</span>
    </div>
    <div class="stat">
      <span class="stat-number">8</span>
      <span class="stat-label">Course modules</span>
    </div>
  </div>
</div>

<!-- 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">rate-optimiser</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-fn">print</span><span class="tok-pn">(</span><span class="tok-str">f"Reconstruction R² = {sr.reconstruction_r2:.4f}"</span><span class="tok-pn">)</span>
<span class="tok-cm"># Reconstruction R² = 0.9973</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">rate_optimiser</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">Open-source libraries</div>
    <h2>The full pricing workflow, covered</h2>
    <p>Each library 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 &amp; Features</div>
        </div>
        <div class="workflow-lib-labels">
          <span class="workflow-lib">insurance-cv</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 Fitting</div>
        </div>
        <div class="workflow-lib-labels">
          <span class="workflow-lib">credibility</span>
          <span class="workflow-lib">bayesian-pricing</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">Interpretation</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">insurance-distill</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">Validation</div>
        </div>
        <div class="workflow-lib-labels">
          <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>
        </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">Rates &amp; Commercial</div>
        </div>
        <div class="workflow-lib-labels">
          <span class="workflow-lib">rate-optimiser</span>
          <span class="workflow-lib">ins-demand</span>
          <span class="workflow-lib">experience-rating</span>
        </div>
      </div>
    </div>
  </div>

  <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/burningcost/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>
      <a href="https://github.com/burningcost/insurance-distill" target="_blank" class="lib-card">
        <span class="lib-card-name">insurance-distill</span>
        <span class="lib-card-desc">GBM-to-GLM distillation for insurance pricing. Fit a GLM that approximates a GBM's predictions, giving you a transparent, auditable model that regulators and rating engines can work with.</span>
        <span class="lib-card-tags">
          <span class="lib-tag">model distillation</span>
          <span class="lib-tag">GLM</span>
          <span class="lib-tag">GBM</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</span>
    </div>
    <div class="lib-grid fade-up-children">
      <a href="https://github.com/burningcost/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/burningcost/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. Implements the variance-weighted non-conformity score from Manna et al. (2025), producing 30% 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/burningcost/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>
    </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/burningcost/credibility" target="_blank" class="lib-card">
        <span class="lib-card-name">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/burningcost/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/burningcost/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/burningcost/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/burningcost/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>
    </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/burningcost/rate-optimiser" target="_blank" class="lib-card">
        <span class="lib-card-name">rate-optimiser</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>
      <a href="https://github.com/burningcost/insurance-demand" target="_blank" class="lib-card">
        <span class="lib-card-name">insurance-demand</span>
        <span class="lib-card-desc">Demand and conversion modelling for insurance pricing. Price elasticity curves, own-price and cross-price effects, integration with rate optimisation.</span>
        <span class="lib-card-tags">
          <span class="lib-tag">price elasticity</span>
          <span class="lib-tag">conversion</span>
        </span>
      </a>
      <a href="https://github.com/burningcost/experience-rating" target="_blank" class="lib-card">
        <span class="lib-card-name">experience-rating</span>
        <span class="lib-card-desc">NCD/bonus-malus systems, experience modification factors, and schedule rating for UK insurance pricing. Principled adjustment of technical rates based on individual risk history.</span>
        <span class="lib-card-tags">
          <span class="lib-tag">NCD</span>
          <span class="lib-tag">bonus-malus</span>
          <span class="lib-tag">schedule rating</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/burningcost/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>

</section>

<!-- Course -->
<section class="course">
  <div class="course-bg"></div>
  <div class="course-inner">
    <div class="course-left fade-up">
      <div class="course-badge">Training course</div>
      <h2>Modern Insurance Pricing with Python and Databricks</h2>
      <p>Eight modules written for pricing actuaries and analysts at UK personal lines insurers. Every module covers a real pricing problem, not a generic data science tutorial adapted to insurance. You work through real Databricks notebooks, on synthetic data that behaves like the real thing.</p>

      <div class="price-cards price-cards--single">
        <div class="price-card price-card--featured price-card--full">
          <div class="price-card-label">Full course + all tools</div>
          <div class="price-card-price">&pound;295</div>
          <div class="price-card-desc">All 8 modules &middot; all future updates &middot; all Burning Cost tools</div>
          <div class="price-card-meta">One-time payment &mdash; no subscription</div>
        </div>
      </div>

      <a href="/course/" class="btn btn-primary">See the full course &rarr;</a>    </div>

    <div class="course-right fade-up">
      <ul class="course-modules-list">
        <li>
          <span class="mod-num">01</span>
          <span class="mod-title">Databricks for pricing teams</span>
        </li>
        <li>
          <span class="mod-num">02</span>
          <span class="mod-title">GLMs in Python: the bridge from Emblem</span>
        </li>
        <li>
          <span class="mod-num">03</span>
          <span class="mod-title">GBMs for insurance pricing</span>
        </li>
        <li>
          <span class="mod-num">04</span>
          <span class="mod-title">SHAP relativities</span>
        </li>
        <li>
          <span class="mod-num">05</span>
          <span class="mod-title">Conformal prediction intervals</span>
        </li>
        <li>
          <span class="mod-num">06</span>
          <span class="mod-title">Credibility and Bayesian pricing</span>
        </li>
        <li>
          <span class="mod-num">07</span>
          <span class="mod-title">Constrained rate optimisation</span>
        </li>
        <li>
          <span class="mod-num">08</span>
          <span class="mod-title">End-to-end pipeline capstone</span>
        </li>
      </ul>
    </div>
  </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">
      {% for post in site.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>