Model-repo review round: correct what the site could not support

connect/index.html

@@ -285,7 +285,9 @@ <h1 class="reveal page-title" style="--d:1;">
       <p class="lede reveal" style="--d:2;">
         Three routes to the same models. Start with MCP below — it needs no
         install, no PolicyEngine account and no API key, though some clients
-        gate custom connectors by plan. Use the CLI or Python when you want the
+        gate custom connectors by plan. That is true of the hosted route; a
+        <em>local</em> UK population or reform-scoring run additionally needs a
+        HuggingFace token, as below. Use the CLI or Python when you want the
         models in a script, or need one of the local-only members.
       </p>
     </div>
@@ -368,10 +370,19 @@ <h2>Connect in three steps.</h2>
           takes around 20 seconds while the tax-benefit models load; a UK
           forecast runs a couple of minutes on first call. Other tools
           respond faster.</span></p>
+        <p class="vmet-notes-line"><span class="vmet-notes-key">uk microdata</span>
+          <span>Running a UK <em>population</em> or reform score locally needs
+          <code>HUGGING_FACE_TOKEN</code> from an account with access to the
+          private <code>policyengine/policyengine-uk-data-private</code>
+          dataset. UK household calls, every US call, and every hosted call
+          need nothing.</span></p>
         <p class="vmet-notes-line"><span class="vmet-notes-key">local only</span>
           <span>OG-UK long-run scoring (<code>score_reform</code> with
           <code>model='og'</code>) and <code>dynamic_reform_impact</code> take
-          tens of minutes — the hosted server points you to the local CLI.
+          tens of minutes and are excluded from the hosted image;
+          <code>dynamic_reform_impact</code> returns the two-step local recipe,
+          while <code>model='og'</code> returns an import error — use the local
+          CLI for both.
           The three <code>define_*</code> tools likewise return run instructions rather than results when called on the hosted server &mdash; the unlicensed upstream is never hosted. Everything else here runs hosted.</span></p>
       </div>
     </div>
@@ -443,7 +454,7 @@ <h2>Use Python directly.</h2>
       </div>
       <div class="model-pane compact-pane" id="model-panel-ogmicro" role="tabpanel" aria-labelledby="model-tab-ogmicro" data-model="ogmicro" hidden>
         <span class="tag tag-local">local only</span><h3>Dynamic reform scoring</h3>
-        <p>Experimental, UK only: an OG-UK steady-state solve feeds long-run wage changes back into a second microsimulation run. Local only, and until <a href="https://github.com/PSLmodels/OG-UK/issues/68">OG-UK#68</a> the solve needs its own environment — run <code>pe-macro og-score --json &gt; og.json</code> there, then <code>pe-macro dynamic-score --og-payload og.json</code>.</p>
+        <p>Experimental, UK only: an OG-UK steady-state solve feeds long-run wage changes back into a second microsimulation run. Local only, and until <a href="https://github.com/PSLmodels/OG-UK/issues/68">OG-UK#68</a> the solve needs its own environment — run <code>pe-macro og-score --reform '{"gov.hmrc.income_tax.rates.uk[0].rate":0.21}' --json &gt; og.json</code> there, then <code>pe-macro dynamic-score --reform '{"gov.hmrc.income_tax.rates.uk[0].rate":0.21}' --og-payload og.json</code> &mdash; both steps need the same <code>--reform</code>.</p>
         <a class="btn btn-primary" href="/olg/code#run">Open dynamic scoring setup</a>
       </div>
       <div class="model-pane compact-pane" id="model-panel-define" role="tabpanel" aria-labelledby="model-tab-define" data-model="define" hidden>
@@ -529,15 +540,6 @@ <h2>Use Python directly.</h2>
     });
   }
 
-  document.addEventListener("click", function (e) {
-    var btn = e.target.closest("[data-copy]"); if (!btn) return;
-    var container = btn.closest(".code-window") || btn.parentElement;
-    var code = container.querySelector("code");
-    navigator.clipboard.writeText(code.textContent).then(function () {
-      btn.textContent = "copied";
-      setTimeout(function () { btn.textContent = "copy"; }, 1400);
-    });
-  });
 })();
 </script>
 <script>

define/code/index.html

@@ -129,7 +129,7 @@ <h4>Clone to rebuild the validation record</h4>
         <h4>The path to hosting</h4>
         <p class="opt-note">a clean-room Python reimplementation of the published Model Manual v1.1 is the route to a hostable model — protocol and milestones in <a href="https://github.com/PolicyEngine/define-uk-model/blob/main/REIMPLEMENTATION.md">REIMPLEMENTATION.md</a>, upstream code used only as a numerical oracle.</p>
         <p>Four slices have landed. The §2.2 accounting core &mdash; the transactions and balance-sheet matrices &mdash; is satisfied identically at the manual's own initial values. §3.2 adds the high-level macroeconomic block, §3.3.1 domestic production, and §3.3.2 the power generation sector: <strong>118 equations, Eqs. (21)&ndash;(138), contiguous</strong>. §3.3.3 input&ndash;output and the comparison against the upstream oracle are outstanding, so <strong>nothing here is hostable yet</strong> and the hosted tools continue to return run instructions.</p>
-        <p>Reimplementing from the manual is also what surfaces its gaps, and roughly <strong>thirty-five</strong> are now pinned by tests rather than absorbed into a tolerance &mdash; a net-lending residual the manual itself says &ldquo;should equal 0&rdquo;, identities that fail at the manual's own initial values, and symbols the equations use that the parameter tables never list. The largest is a blocker: <strong>§3.3.2 as published cannot be simulated forward.</strong> Six parameters it uses are absent from Table 5, and two variables are never defined anywhere in the manual, so credit rationing collapses to a constant and desired power-sector investment turns negative. That is the section the climate-policy scenarios run through, and it is <a href="https://github.com/PolicyEngine/define-uk-model/blob/main/REIMPLEMENTATION.md">recorded in full</a> for the authors.</p>
+        <p>Reimplementing from the manual is also what surfaces its gaps, and <strong>thirteen</strong> are now pinned by tests as machine-readable gap records rather than absorbed into a tolerance, with more written up in the reimplementation notes &mdash; a net-lending residual the manual itself says &ldquo;should equal 0&rdquo;, identities that fail at the manual's own initial values, and symbols the equations use that the parameter tables never list. The largest is a blocker: <strong>§3.3.2 as published cannot be simulated forward.</strong> Six parameters it uses are absent from Table 5, and two variables are never defined anywhere in the manual, so credit rationing collapses to a constant and desired power-sector investment turns negative. That is the section the climate-policy scenarios run through, and it is <a href="https://github.com/PolicyEngine/define-uk-model/blob/main/REIMPLEMENTATION.md">recorded in full</a> for the authors.</p>
         <div class="meth-inline">
         <div class="meth-sec-title">development</div>
         <div class="codeblock"><button class="cb-copy" data-copy>copy</button><pre><code>git clone https://github.com/PolicyEngine/define-uk-model
@@ -174,16 +174,6 @@ <h4>The path to hosting</h4>
 
 <script src="/reveal.js?v=2" defer></script>
 <script>
-document.addEventListener("click", function (e) {
-  var btn = e.target.closest("[data-copy]"); if (!btn) return;
-  var code = btn.parentElement.querySelector("code");
-  navigator.clipboard.writeText(code.textContent).then(function () {
-    btn.textContent = "copied";
-    setTimeout(function () { btn.textContent = "copy"; }, 1400);
-  });
-});
-</script>
-<script>
 // Code scrollytelling: highlight the step at the reading line and
 // show its code panel in the sticky aside.
 (function () {

define/validation/index.html

@@ -55,7 +55,7 @@
     <div class="hero-inner">
       <p class="eyebrow reveal" style="--d:0">model 07 — ecological stock-flow consistent · define-uk · UK · experimental</p>
       <h1 class="reveal page-title" style="--d:1;">
-        Validated to the ceiling the publications allow.
+        Validated where the publications allow it, and no further.
       </h1>
       <p class="lede reveal" style="--d:2">
         The manual's macro block replicates; the emissions path diverges and
@@ -94,7 +94,7 @@ <h2>The replication gates, check by check.</h2>
           <tr>
             <th scope="row">Manual Table&nbsp;4, macro block — real GDP growth, unemployment, population 16+, labour force at 2025/2030/2040, S1 baseline</th>
             <td>population and labour force &plusmn;0.01m; growth &plusmn;0.31pp; unemployment &plusmn;0.2pp</td>
-            <td><strong>PASS</strong> (2026-08-04) — growth exact at 2030/2040, &minus;0.30pp at 2025</td>
+            <td><strong>PASS</strong> (2026-08-04) — growth exact at 2030/2040, &minus;0.30pp at 2025 against a &plusmn;0.31pp gate: passing by 0.01pp is not the same as matching</td>
           </tr>
           <tr>
             <th scope="row">Manual Table&nbsp;4, total emissions, S1 baseline</th>
@@ -109,7 +109,7 @@ <h2>The replication gates, check by check.</h2>
           <tr>
             <th scope="row">FMM-2023 anchors — green public investment GDP peak; baseline 2030 emissions</th>
             <td>wide (pre-1.0 vintage): GDP peak 0.7&ndash;1.3%; emissions 330&ndash;352 MtCO2e</td>
-            <td><strong>PASS</strong> — peak +0.92% vs published &asymp;+1%; 342.8 vs &ldquo;just under 350&rdquo;</td>
+            <td><strong>PASS</strong> — peak +0.92% vs published &asymp;+1%; 342.8 vs &ldquo;just under 350&rdquo;. The tests assert membership of the wide bands, not these two figures</td>
           </tr>
           <tr>
             <th scope="row">Published numeric scenario results for v1.1</th>
@@ -235,11 +235,22 @@ <h2>The fiscal multiplier, stated honestly.</h2>
         Our own computation of the model's cumulative
         green-public-investment multiplier — cumulative ΔGDP over cumulative
         ΔSPEND_GVT, nominal, over the full simulation — is
-        <strong>1.78</strong>. IMF estimates of green-spending multipliers
-        run 1.1&ndash;1.5 (Batini et al. 2021); the OBR's capital-spending
-        impact multiplier is &asymp;1.0. Above but near the IMF range, as the
-        demand-led closure predicts, and labelled accordingly wherever a
-        delta is served.
+        <strong>1.78</strong>. It is <em>not</em> comparable to the published
+        green-spending multipliers it sat beside here — an IMF range of
+        1.1&ndash;1.5 and the OBR's &asymp;1.0 capital-spending impact
+        multiplier. Three things all push the same way. Ours is a
+        <em>cumulative ratio to 2040</em> set against short-horizon
+        multipliers. Its denominator <code>SPEND_GVT</code> is
+        <em>endogenous</em>: government investment is only 12.8% of it, and
+        social benefits fall when unemployment does, so a scenario that cuts
+        unemployment shrinks the denominator and biases the ratio up. And the
+        closure is demand-led with Kaldor&ndash;Verdoorn productivity and real
+        GDP not held to a supply ceiling, so a ratio above 1 over fifteen
+        years is what the model implies rather than a finding. The clean
+        comparison — cumulative ΔGDP over cumulative ΔGCF_GVTG — has not been
+        computed; until it is, treat 1.78 as upper-leaning. The caveats travel
+        with every served delta (<code>GPI_MULTIPLIER_CAVEATS</code>, gated by
+        test).
       </p>
       <p>
         The upstream repository's own <code>Multiplier_Summary.csv</code> is

economy/topics/growth/index.html

@@ -41,7 +41,6 @@
 <nav class="crumbs mono" aria-label="You are here">
     <a href="/">Home</a>
     <a href="/economy">Economy</a>
-    <a href="/economy/topics">Topics</a>
     <span aria-current="page">Growth</span>
 </nav>
 
@@ -148,7 +147,7 @@ <h2>The model view, and the same breath its limits.</h2>
       </div>
       <p class="callout">
         <strong>Predictive validation: weak.</strong>
-        Against a no-change random walk the model looked strong on CPI (0.63 at h=1), but a driftless walk on a trending log level is too weak a benchmark: against a random walk WITH DRIFT the CPI ratio becomes 0.83 at h=1 and 1.03 at h=8, i.e. no better than naive. Bank Rate is the one series that improves under the harder benchmark (0.79 at h=1, p=0.018) and is the defensible forecasting claim. UK GDP is not distinguishable from either benchmark (p=0.38-0.67), and excluding six Covid-target origins its ratio falls to 0.77. The frozen-edge run gives 0.32pp RMSE from a single origin.
+        No demonstrated skill against a random walk WITH DRIFT on any variable at any horizon once the 64 variable-by-horizon tests are adjusted together: minimum Benjamini-Hochberg q = 0.36, and 0.40 under the published specification. A driftless walk on a trending log level is too weak a benchmark, so the strong-looking CPI result (0.63 at h=1) becomes 0.83 at h=1 and 1.03 at h=8 against drift. Bank Rate comes closest (0.79 at h=1, unadjusted p=0.018 -- the smallest of the 64) but does not survive the adjustment. UK GDP is not distinguishable from drift at any horizon (p=0.33-0.43). Two caveats on figures quoted elsewhere: the ex-Covid ratio of 0.77 and the p=0.38-0.67 range are against the weaker no-change benchmark, not drift. Separately, the rolling evaluation had estimated without the six Covid dummies that every published forecast carries; under the published specification UK GDP goes 1.06 to 0.99 at h=1 and 1.12 to 0.95 at h=8 -- level with naive rather than worse, still not better. The frozen-edge run gives 0.32pp RMSE from a single origin.
         <a href="/svar/validation">Read the validation page →</a>
       </p>
       <p class="chooser-note">

economy/topics/inflation/index.html

@@ -41,7 +41,6 @@
 <nav class="crumbs mono" aria-label="You are here">
     <a href="/">Home</a>
     <a href="/economy">Economy</a>
-    <a href="/economy/topics">Topics</a>
     <span aria-current="page">Inflation</span>
 </nav>
 
@@ -137,7 +136,7 @@ <h2>The model view, and the same breath its limits.</h2>
       </div>
       <p class="callout">
         <strong>Predictive validation: weak.</strong>
-        Against a no-change random walk the model looked strong on CPI (0.63 at h=1), but a driftless walk on a trending log level is too weak a benchmark: against a random walk WITH DRIFT the CPI ratio becomes 0.83 at h=1 and 1.03 at h=8, i.e. no better than naive. Bank Rate is the one series that improves under the harder benchmark (0.79 at h=1, p=0.018) and is the defensible forecasting claim. UK GDP is not distinguishable from either benchmark (p=0.38-0.67), and excluding six Covid-target origins its ratio falls to 0.77. The frozen-edge run gives 0.32pp RMSE from a single origin.
+        No demonstrated skill against a random walk WITH DRIFT on any variable at any horizon once the 64 variable-by-horizon tests are adjusted together: minimum Benjamini-Hochberg q = 0.36, and 0.40 under the published specification. A driftless walk on a trending log level is too weak a benchmark, so the strong-looking CPI result (0.63 at h=1) becomes 0.83 at h=1 and 1.03 at h=8 against drift. Bank Rate comes closest (0.79 at h=1, unadjusted p=0.018 -- the smallest of the 64) but does not survive the adjustment. UK GDP is not distinguishable from drift at any horizon (p=0.33-0.43). Two caveats on figures quoted elsewhere: the ex-Covid ratio of 0.77 and the p=0.38-0.67 range are against the weaker no-change benchmark, not drift. Separately, the rolling evaluation had estimated without the six Covid dummies that every published forecast carries; under the published specification UK GDP goes 1.06 to 0.99 at h=1 and 1.12 to 0.95 at h=8 -- level with naive rather than worse, still not better. The frozen-edge run gives 0.32pp RMSE from a single origin.
         <a href="/svar/validation">Read the validation page →</a>
       </p>
     </div>

economy/topics/jobs/index.html

@@ -41,7 +41,6 @@
 <nav class="crumbs mono" aria-label="You are here">
     <a href="/">Home</a>
     <a href="/economy">Economy</a>
-    <a href="/economy/topics">Topics</a>
     <span aria-current="page">Jobs</span>
 </nav>
 

economy/topics/public-finances/index.html

@@ -41,7 +41,6 @@
 <nav class="crumbs mono" aria-label="You are here">
     <a href="/">Home</a>
     <a href="/economy">Economy</a>
-    <a href="/economy/topics">Topics</a>
     <span aria-current="page">Public finances</span>
 </nav>
 

economy/topics/rates/index.html

@@ -41,7 +41,6 @@
 <nav class="crumbs mono" aria-label="You are here">
     <a href="/">Home</a>
     <a href="/economy">Economy</a>
-    <a href="/economy/topics">Topics</a>
     <span aria-current="page">Rates and gilts</span>
 </nav>
 

economy/topics/reform/index.html

@@ -41,7 +41,6 @@
 <nav class="crumbs mono" aria-label="You are here">
     <a href="/">Home</a>
     <a href="/economy">Economy</a>
-    <a href="/economy/topics">Topics</a>
     <span aria-current="page">Tax and benefit reform</span>
 </nav>