diff --git a/connect/index.html b/connect/index.html
index 512153aa..2c3a88ef 100644
--- a/connect/index.html
+++ b/connect/index.html
@@ -372,7 +372,7 @@ <h2>Connect in three steps.</h2>
           <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.
-          Everything else here runs hosted.</span></p>
+          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>
   </section>
diff --git a/define/code/index.html b/define/code/index.html
index c8bc13fd..362c74eb 100644
--- a/define/code/index.html
+++ b/define/code/index.html
@@ -103,7 +103,7 @@ <h4>Where to read the results</h4>
         <div class="meth-step-head"><span class="meth-step-num mono">2</span><h3>CLI scenario runs</h3></div>
         <h4>Deltas from the terminal</h4>
         <p>The CLI lists the curated scenarios and runs any of them locally: <code>pe-macro define-scenarios</code> to browse, <code>pe-macro define-scenario</code> to produce a scenario's deltas vs baseline — real GDP, emissions, unemployment, real consumption. Deltas only, never levels.</p>
-        <div class="meth-inline" aria-hidden="true">
+        <div class="meth-inline">
         <div class="meth-sec-title">CLI commands</div>
         <div class="codeblock"><button class="cb-copy" data-copy>copy</button><pre><code>pe-macro define-scenarios              <span class="cm"># list the curated set</span>
 pe-macro define-scenario &lt;name&gt;        <span class="cm"># scenario deltas vs baseline</span></code></pre></div>
@@ -116,7 +116,7 @@ <h4>One pip install, upstream fetched at runtime</h4>
         <p>The adapter installs from GitHub. Nothing upstream is vendored: at runtime it fetches <code>DEFINE-model/DEFINE_UK_1.1</code> at pinned commit <code>846081a</code> and executes the published notebook unmodified.</p>
         <h4>R does the heavy lifting once</h4>
         <p>A fresh full run needs local R with the notebook's packages; after that, scenario deltas are instant from the cached pinned run.</p>
-        <div class="meth-inline" aria-hidden="true">
+        <div class="meth-inline">
         <div class="meth-sec-title">install</div>
         <div class="codeblock"><button class="cb-copy" data-copy>copy</button><pre><code>pip install git+https://github.com/PolicyEngine/define-uk-model</code></pre></div>
         </div>
@@ -128,9 +128,9 @@ <h4>Clone to rebuild the validation record</h4>
         <p>The test suite carries the validation gate: hermetic artifact checks always run, and the oracle comparisons against the pinned upstream run execute wherever a local cache exists — in CI they must skip, because a passing oracle test there would mean the unlicensed upstream was fetched.</p>
         <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>Two slices have landed. The §2.2 accounting core — the transactions and balance-sheet matrices — is satisfied identically at the manual's own initial values. §3.2 adds the 23 high-level macroeconomic equations, Eqs. (21)&ndash;(43): the demand-determined expenditure core, the Kaldor&ndash;Verdoorn productivity relation, and the Leontief labour and capital supply constraints. §3.3 production 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. Three are recorded so far: Table 6 tabulates a net-lending residual the manual itself says &ldquo;should equal 0&rdquo;; Eq. (27) does not hold at the manual's own initial values, and disagrees with Eq. (25) about the same variable; and Eq. (31) uses an intercept that Table 5 never tabulates. Each is pinned by a test rather than absorbed into a tolerance, so a later manual revision fails loudly instead of passing quietly.</p>
-        <div class="meth-inline" aria-hidden="true">
+        <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>
+        <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
 cd define-uk-model
@@ -142,18 +142,18 @@ <h4>The path to hosting</h4>
         <div class="meth-panel" data-step="1" hidden>
           <div class="meth-panel-head"><span class="meth-panel-title">CLI scenario runs</span><span class="meth-panel-badge mono">Code 2</span></div>
           <div class="meth-sec-title">CLI commands</div>
-        <div class="codeblock"><button class="cb-copy" data-copy>copy</button><pre><code>pe-macro define-scenarios              <span class="cm"># list the curated set</span>
+        <div class="codeblock"><button class="cb-copy" data-copy tabindex="-1">copy</button><pre><code>pe-macro define-scenarios              <span class="cm"># list the curated set</span>
 pe-macro define-scenario &lt;name&gt;        <span class="cm"># scenario deltas vs baseline</span></code></pre></div>
         </div>
         <div class="meth-panel" data-step="2" hidden>
           <div class="meth-panel-head"><span class="meth-panel-title">Install locally</span><span class="meth-panel-badge mono">Code 3</span></div>
           <div class="meth-sec-title">install</div>
-        <div class="codeblock"><button class="cb-copy" data-copy>copy</button><pre><code>pip install git+https://github.com/PolicyEngine/define-uk-model</code></pre></div>
+        <div class="codeblock"><button class="cb-copy" data-copy tabindex="-1">copy</button><pre><code>pip install git+https://github.com/PolicyEngine/define-uk-model</code></pre></div>
         </div>
         <div class="meth-panel" data-step="3" hidden>
           <div class="meth-panel-head"><span class="meth-panel-title">Gates and the clean room</span><span class="meth-panel-badge mono">Code 4</span></div>
           <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
+        <div class="codeblock"><button class="cb-copy" data-copy tabindex="-1">copy</button><pre><code>git clone https://github.com/PolicyEngine/define-uk-model
 cd define-uk-model
 pytest                    <span class="cm"># hermetic gate everywhere; oracle tests need the local cache</span></code></pre></div>
         </div>
@@ -205,7 +205,13 @@ <h4>The path to hosting</h4>
     });
     // Panels pair to steps by data-step. A step with no panel of its own
     // keeps the nearest earlier panel pinned rather than blanking the column.
-    var want = -1;
+    // Start at the FIRST panel, not -1: several pages have no panel for
+    // step 0, and seeding -1 left the sticky column empty for the whole of
+    // the opening step before popping in. Pinning the earliest panel means
+    // the column is populated from the top and only ever moves forward.
+    var want = Math.min.apply(null, panels.map(function (panel) {
+      return Number(panel.getAttribute("data-step"));
+    }));
     panels.forEach(function (panel) {
       var ds = Number(panel.getAttribute("data-step"));
       if (ds <= active && ds > want) want = ds;
diff --git a/define/methodology/index.html b/define/methodology/index.html
index 416c3dd6..5774294c 100644
--- a/define/methodology/index.html
+++ b/define/methodology/index.html
@@ -132,7 +132,7 @@ <h4>Accounting before theory</h4>
       <p>DEFINE-UK is an ecological stock-flow consistent (E-SFC) model: DEFINE-UK 1.1, April 2026, by George, Dafermos, Nikolaidi and co-authors (<a href="https://define-model.org/define-uk">define-model.org</a>). Stock-flow consistency is an accounting discipline before it is an economic theory. Every monetary flow in the model — wages, taxes, interest, dividends, lending — leaves one sector's account and enters another's, so a transactions matrix sums to zero in every row and column; and every flow accumulates into a stock, so sector balance sheets close each period. Nothing appears from nowhere and nothing leaks out.</p>
         <h4>The ecological coupling</h4>
       <p>The <em>ecological</em> part couples that closed monetary accounting to an emissions and energy block, so climate policy and the financial system move together in one system. An emissions block splits electricity from non-electricity emissions; the components sum exactly to the total in the model's output.</p>
-        <div class="meth-inline" aria-hidden="true">
+        <div class="meth-inline">
         <div class="meth-sec-title">matrix identities (manual §2.2)</div>
         <div class="meth-eq"><div class="meth-eq-label">flow rows</div><div class="meth-eq-body">\(\sum_{s} T_{f,s} = 0\) — every flow leaves one sector and enters another</div></div>
         <div class="meth-eq"><div class="meth-eq-label">sector columns</div><div class="meth-eq-body">\(\sum_{f} T_{f,s} = \mathrm{LEND}_s\) — column sums give net lending</div></div>
@@ -173,7 +173,7 @@ <h4>The sectors</h4>
       <p>Households, firms, monetary financial institutions (MFIs), non-monetary financial institutions, government, the central bank, and the rest of the world — each with a full balance sheet. The balance-sheet matrix (manual §2.2 Table 2) mirrors the transactions matrix: every financial instrument row sums to zero — one sector's asset is another's liability — and each sector's column sums to its net worth.</p>
         <h4>Calibration</h4>
       <p>The model is calibrated to UK national accounting data (ONS Blue Book / UKEA), with the manual's §5 tables giving parameters and initial values.</p>
-        <div class="meth-inline" aria-hidden="true">
+        <div class="meth-inline">
         <div class="meth-sec-title">balance-sheet identities (manual §2.2 Table 2)</div>
         <div class="meth-eq"><div class="meth-eq-label">instrument rows</div><div class="meth-eq-body">\(\sum_{s} B_{i,s} = 0\) — one sector's asset is another's liability</div></div>
         <div class="meth-eq"><div class="meth-eq-label">sector columns</div><div class="meth-eq-body">financial rows sum to \(\mathrm{FNW}_s\); adding real assets gives \(\mathrm{NW}_s\)</div></div>
@@ -212,7 +212,7 @@ <h4>Power</h4>
       <p>The <strong>power-sector block</strong> carries separate fossil and renewable capital stocks: government renewable investment is triggered when non-fossil capacity falls short of a utilisation threshold relative to electricity demand, and regulation policies (new in version 1.1) can directly depreciate fossil capital — stranded assets as an explicit balance-sheet event, felt by the financial institutions that hold the claims.</p>
         <h4>What version 1.1 added</h4>
       <p>Version 1.1 (April 2026) added regulatory policies, limited forward-looking behaviour to policy announcements, direct government ownership of non-fossil power capital, and a green sovereign bond instrument.</p>
-        <div class="meth-inline" aria-hidden="true">
+        <div class="meth-inline">
         <div class="meth-sec-title">the two UK blocks</div>
         <div class="meth-eq"><div class="meth-eq-label">housing</div><div class="meth-eq-body">stock tracked by energy efficiency; retrofit policy works through the stock</div></div>
         <div class="meth-eq"><div class="meth-eq-label">power</div><div class="meth-eq-body">separate fossil and renewable capital stocks; utilisation threshold triggers renewable investment</div></div>
@@ -244,7 +244,7 @@ <h4>The closure</h4>
       <p>DEFINE-UK is closed demand-led: output adjusts to aggregate demand rather than being pinned to a supply-side potential path. That is the standard post-Keynesian SFC closure, and it has a direct, observable consequence — fiscal expansions are not crowded out by an assumed return to potential, so fiscal multipliers come out larger than in supply-constrained models like the <a href="/obr">OBR emulator</a>.</p>
         <h4>Measured, not asserted</h4>
       <p>We measure this rather than assert it: our own computation of the model's cumulative green-public-investment multiplier is <strong>1.78</strong>, against IMF green-spending estimates of 1.1&ndash;1.5 and an OBR capital-spending impact multiplier of &asymp;1.0. Above the official range, but explicably so given the closure — and every surface that quotes a DEFINE-UK delta carries the demand-led caveat. The full comparison is on the <a href="/define/validation">validation page</a>.</p>
-        <div class="meth-inline" aria-hidden="true">
+        <div class="meth-inline">
         <div class="meth-sec-title">closure and multipliers</div>
         <div class="meth-eq"><div class="meth-eq-label">closure</div><div class="meth-eq-body">demand-led: \(Y\) adjusts to aggregate demand, not a potential path</div></div>
         <div class="meth-eq"><div class="meth-eq-label">our computation</div><div class="meth-eq-body">cumulative green-public-investment multiplier <strong>1.78</strong></div></div>
@@ -276,7 +276,7 @@ <h4>Verifying the switches</h4>
           What that gate does and does not establish is spelled out on the
           <a href="/define/validation">validation page</a>.
         </p>
-        <div class="meth-inline" aria-hidden="true">
+        <div class="meth-inline">
         <div class="meth-sec-title">the switch sets</div>
         <div class="meth-eq"><div class="meth-eq-label">fossil-fuel ban</div><div class="meth-eq-body">ban flag + its capacity and investment timings</div></div>
         <div class="meth-eq"><div class="meth-eq-label">power subsidy</div><div class="meth-eq-body"><code>POWER_SUB</code> alone</div></div>
@@ -318,7 +318,7 @@ <h4>Deltas only, local only</h4>
       <p><code>pe-macro define-scenarios</code> and <code>pe-macro define-scenario</code> return annualised scenario-minus-baseline delta paths only, framed with mandatory caveats — never levels, because the baseline's calibration gap to ONS/DESNZ/OBR actuals is computed and material. The hosted MCP tools return run instructions instead of results, because the unlicensed upstream is never hosted. <code>score_reform</code> will not accept this model.</p>
         <h4>The clean-room path to hosting</h4>
       <p>The path to hosting is a <a href="https://github.com/PolicyEngine/define-uk-model/blob/main/REIMPLEMENTATION.md">clean-room Python reimplementation</a> of the published Model Manual — the same adapted-model approach as the OBR emulator. The protocol: the manual is the specification, and the upstream run is a numerical oracle only — its outputs are compared against ours, but its source code is never read to write equations, and every equation must carry a reference to the manual section and equation number it implements. Five milestones run from the accounting core to the full policy scenarios; milestone 1 — the §2.2 transactions and balance-sheet matrices on the §5 initial values — passed with 27 identity tests, including one documented inconsistency in the manual itself (Table 6 omits a dividend term of Eq. 383 from the interbank-lending row, which the tests pin exactly rather than paper over).</p>
-        <div class="meth-inline" aria-hidden="true">
+        <div class="meth-inline">
         <div class="meth-sec-title">adapter architecture</div>
         <div class="meth-eq"><div class="meth-eq-label">pinned commit</div><div class="meth-eq-body">runtime fetch at <code>846081a</code>; executed unmodified, in R; nothing vendored or hosted</div></div>
         <div class="meth-eq"><div class="meth-eq-label">cached run</div><div class="meth-eq-body">R 4.3.0 · 151 output files across all four scenario blocks</div></div>
diff --git a/economy/index.html b/economy/index.html
index 678056ac..bbe82069 100644
--- a/economy/index.html
+++ b/economy/index.html
@@ -101,7 +101,10 @@ <h2>Six topics, each with its data, its model, and its limits.</h2>
       </div>
       <p class="economy-note-aside">
         <span class="mono">Research note · 25 July 2026</span>
-        UK CPI fell to 2.8% in 2026Q2—and the model expects it back above 3% by Q4.
+        UK CPI fell to 2.8% in 2026Q2. The note reads that release through the
+        2026-07-21 round; the current round is more subdued, and
+        <a href="/economy/topics/inflation">the inflation topic</a> carries the
+        live model view.
         <a href="/notes/2026-07-25-cpi-2026q2">Read the CPI note →</a>
         <a class="economy-note-aside-all" href="/forecasts#notes">All notes</a>
       </p>
diff --git a/forecasts/index.html b/forecasts/index.html
index bc049a3e..67f7bc84 100644
--- a/forecasts/index.html
+++ b/forecasts/index.html
@@ -89,7 +89,7 @@ <h2>The record so far.</h2>
       <div class="forecast-summary" aria-label="Current forecast record">
         <article><strong>3</strong><span>Archived rounds</span><small>committed to Git before the outturn existed</small></article>
         <article><strong>1</strong><span>Scored period</span><small>quarters whose outturn is now published and scored</small></article>
-        <article><strong>0.12pp</strong><span>Latest absolute error</span><small>latest forecast vs its outturn, percentage points</small></article>
+        <article><strong>2026Q2</strong><span>Next scored period</span><small>when the record next grows, not how it is doing</small></article>
       </div>
       <p class="forecast-next">
         <strong>Next due:</strong> 2026Q2 (UK real GDP, y/y, unemployment), which lands when the ONS publishes it.
@@ -124,7 +124,7 @@ <h2>Scored so far: one round.</h2>
             <div class="fc-mark fc-mark-naive fc-up" style="left:90.3%"><i></i><span>Naive<strong>3.10%</strong></span></div>
             <div class="fc-mark fc-mark-official fc-down" style="left:9.7%"><i></i><span>Official<strong>2.14%</strong></span></div>
           </div>
-          <footer><span>Error <strong>-0.12pp</strong></span><span>beats naive error <strong>0.30pp</strong></span><span>OBR March 2026 EFO error <strong>0.66pp</strong></span><span>Round 2026-07-21</span></footer>
+          <footer><span>Error <strong>-0.12pp</strong></span><span>beats driftless-naive error <strong>0.30pp</strong></span><span>OBR March 2026 EFO error <strong>0.66pp</strong></span><span>Round 2026-07-21</span></footer>
         </article>
       </div>
       <p class="forecast-note">
diff --git a/forecasts/score.py b/forecasts/score.py
index 60d4d95f..21148cd9 100644
--- a/forecasts/score.py
+++ b/forecasts/score.py
@@ -549,9 +549,8 @@ def render_status(card: dict) -> str:
         for entry in detail["entries"]
     ]
     latest_entry = sorted(entries, key=lambda entry: entry["period"])[-1] if entries else None
-    latest_error = (
-        f"{latest_entry['abs_error']:.2f}pp" if latest_entry else "Awaiting outturn"
-    )
+    pending_card = card.get("pending_detail") or {}
+    scored_target = esc(pending_card.get("period") or "—")
     lines = [
         '      <div class="forecast-summary" aria-label="Current forecast record">',
         f"        <article><strong>{rounds}</strong><span>Archived "
@@ -560,9 +559,16 @@ def render_status(card: dict) -> str:
         f"        <article><strong>{scored}</strong><span>Scored "
         f"period{'' if scored == 1 else 's'}</span>"
         "<small>quarters whose outturn is now published and scored</small></article>",
-        f"        <article><strong>{latest_error}</strong><span>Latest absolute "
-        "error</span>"
-        "<small>latest forecast vs its outturn, percentage points</small></article>",
+        # NOT the latest error. This module's whole argument is that a
+        # headline accuracy figure derived from one observation is worse than
+        # none, and putting it in the record band — level with the round and
+        # scored counts — was the page doing exactly that, with the "one
+        # scored period is not an accuracy headline" disclaimer underneath it.
+        # The error still appears on the scored card in §02, where it sits
+        # beside its benchmarks and its caveats.
+        f"        <article><strong>{scored_target}</strong><span>Next scored "
+        "period</span>"
+        "<small>when the record next grows, not how it is doing</small></article>",
         "      </div>",
         '      <p class="forecast-next">',
     ]
@@ -626,8 +632,14 @@ def render_results(card: dict) -> str:
         if naive is not None:
             marks.append(("Naive", naive, "naive", "up"))
             verdict = "beats" if e["beats_naive"] else "does not beat"
+            # Name the benchmark. This is the DRIFTLESS walk, and §06 of the
+            # page says plainly that against a random walk WITH drift the
+            # advantage largely evaporates. Reporting a win against the weaker
+            # benchmark without saying which one it is was the page's single
+            # favourable data point resting on the comparison its own
+            # validation calls too weak for a trending series.
             naive_footer = (
-                f"<span>{verdict} naive error "
+                f"<span>{verdict} driftless-naive error "
                 f"<strong>{e['naive_abs_error']:.2f}pp</strong></span>"
             )
         official_footer = ""
diff --git a/frb-us/code/index.html b/frb-us/code/index.html
index 90584fe4..f5db6dd0 100644
--- a/frb-us/code/index.html
+++ b/frb-us/code/index.html
@@ -96,7 +96,7 @@ <h4>Speed</h4>
         <p>A solve takes about <strong>3 seconds cold and well under a second warm</strong> — the fastest member of the suite.</p>
         <h4>CLI mirrors</h4>
         <p>The same three tools are on the CLI, no server round-trip required.</p>
-        <div class="meth-inline" aria-hidden="true">
+        <div class="meth-inline">
         <div class="meth-sec-title">the three tools on the CLI</div>
         <div class="codeblock"><button class="cb-copy" data-copy>copy</button><pre><code>pe-macro frbus-variables
 pe-macro frbus-shock --var rffintay_aerr --shock 1.0
@@ -126,7 +126,7 @@ <h4>When to install</h4>
         <p>For the full 284-variable model, the Board's own demo scripts, or anything the three tools do not expose — separate from the rest of the suite, with its own Python API.</p>
         <h4>No data step</h4>
         <p>The Board's raw materials are vendored unmodified, so there is no data-download step, but you clone rather than pip-install from an index.</p>
-        <div class="meth-inline" aria-hidden="true">
+        <div class="meth-inline">
         <div class="meth-sec-title">clone and install</div>
         <div class="codeblock"><button class="cb-copy" data-copy>copy</button><pre><code>git clone https://github.com/PolicyEngine/us-frb-model
 cd us-frb-model
@@ -159,7 +159,7 @@ <h4>Series you will reach for</h4>
             <tr><th scope="row"><code>rff</code></th><td>Federal funds rate</td><td>endogenous under the rule, pp</td></tr>
           </tbody>
         </table></div>
-        <div class="meth-inline" aria-hidden="true">
+        <div class="meth-inline">
         <div class="meth-sec-title">monetary_policy_shock.py</div>
         <div class="codeblock"><button class="cb-copy" data-copy>copy</button><pre><code><span class="kw">import</span> pandas <span class="kw">as</span> pd
 <span class="kw">from</span> frbus <span class="kw">import</span> Frbus, load_data
@@ -189,21 +189,21 @@ <h4>Series you will reach for</h4>
         <div class="meth-panel" data-step="0">
           <div class="meth-panel-head"><span class="meth-panel-title">Ask the hosted server</span><span class="meth-panel-badge mono">Step 1</span></div>
           <div class="meth-sec-title">the three tools on the CLI</div>
-          <div class="codeblock"><button class="cb-copy" data-copy>copy</button><pre><code>pe-macro frbus-variables
+          <div class="codeblock"><button class="cb-copy" data-copy tabindex="-1">copy</button><pre><code>pe-macro frbus-variables
 pe-macro frbus-shock --var rffintay_aerr --shock 1.0
 pe-macro frbus-summary</code></pre></div>
         </div>
         <div class="meth-panel" data-step="3" hidden>
           <div class="meth-panel-head"><span class="meth-panel-title">Install the package</span><span class="meth-panel-badge mono">Step 4</span></div>
           <div class="meth-sec-title">clone and install</div>
-          <div class="codeblock"><button class="cb-copy" data-copy>copy</button><pre><code>git clone https://github.com/PolicyEngine/us-frb-model
+          <div class="codeblock"><button class="cb-copy" data-copy tabindex="-1">copy</button><pre><code>git clone https://github.com/PolicyEngine/us-frb-model
 cd us-frb-model
 uv venv &amp;&amp; uv pip install -e ".[dev]"</code></pre></div>
         </div>
         <div class="meth-panel" data-step="4" hidden>
           <div class="meth-panel-head"><span class="meth-panel-title">Run a 100bp shock in Python</span><span class="meth-panel-badge mono">Step 5</span></div>
           <div class="meth-sec-title">monetary_policy_shock.py</div>
-          <div class="codeblock"><button class="cb-copy" data-copy>copy</button><pre><code><span class="kw">import</span> pandas <span class="kw">as</span> pd
+          <div class="codeblock"><button class="cb-copy" data-copy tabindex="-1">copy</button><pre><code><span class="kw">import</span> pandas <span class="kw">as</span> pd
 <span class="kw">from</span> frbus <span class="kw">import</span> Frbus, load_data
 
 data = load_data(<span class="kw">"vendor/data_only_package/LONGBASE.TXT"</span>)
@@ -272,7 +272,13 @@ <h4>Series you will reach for</h4>
     });
     // Panels pair to steps by data-step. A step with no panel of its own
     // keeps the nearest earlier panel pinned rather than blanking the column.
-    var want = -1;
+    // Start at the FIRST panel, not -1: several pages have no panel for
+    // step 0, and seeding -1 left the sticky column empty for the whole of
+    // the opening step before popping in. Pinning the earliest panel means
+    // the column is populated from the top and only ever moves forward.
+    var want = Math.min.apply(null, panels.map(function (panel) {
+      return Number(panel.getAttribute("data-step"));
+    }));
     panels.forEach(function (panel) {
       var ds = Number(panel.getAttribute("data-step"));
       if (ds <= active && ds > want) want = ds;
diff --git a/frb-us/index.html b/frb-us/index.html
index f1adc7af..df15b8f5 100644
--- a/frb-us/index.html
+++ b/frb-us/index.html
@@ -80,9 +80,9 @@ <h2>Its labour-market path tells the microsimulation who bears a shock.</h2>
     <div class="prose">
       <div class="qa-facts">
         <div class="qa-fact"><span class="icon">&rarr;</span><span><strong>Connection.</strong> Shock-to-incidence bridge: <code>frbus_shock_incidence</code> scales the US <a href="/pe">microsimulation</a>'s employment income with the wage and hours deviations a shock produces. Uniform scaling — an experimental illustration, and labelled as such. <code>score_reform</code> refuses <code>model="frbus"</code>: no mapping from statute exists.</span></div>
-        <div class="qa-fact"><span class="icon">&rarr;</span><span><strong>Worked example.</strong> The 100bp tightening above moves the 2027 wage bill &minus;0.65%; the microsimulation then shows how much the federal and state tax system absorbs and what SNAP and the EITC give back, decile by decile.</span></div>
+        <div class="qa-fact"><span class="icon">&rarr;</span><span><strong>Worked example.</strong> The 100bp tightening below moves the 2027 wage bill &minus;0.65%; the microsimulation then shows how much the federal and state tax system absorbs and what SNAP and the EITC give back, decile by decile.</span></div>
         <div class="qa-fact"><span class="icon">&rarr;</span><span><strong>Validation.</strong> <a href="/frb-us/validation">LONGBASE reproduced to machine precision; shocks at the Fed's own noise floor</a> — both CI-gated — plus a weekly baseline-vs-outturn backtest.</span></div>
-        <div class="qa-fact"><span class="icon">&rarr;</span><span><strong>Results.</strong> Run the shock explorer on this page or from <a href="/connect">any connected client</a>.</span></div>
+        <div class="qa-fact"><span class="icon">&rarr;</span><span><strong>Results.</strong> Run <code>frbus_shock</code> from <a href="/connect">any connected client</a>, or <code>pe-macro frbus-shock</code> locally.</span></div>
       </div>
     </div>
   </section>
diff --git a/frb-us/methodology/index.html b/frb-us/methodology/index.html
index a89b7e9d..d9bf9eb0 100644
--- a/frb-us/methodology/index.html
+++ b/frb-us/methodology/index.html
@@ -133,7 +133,7 @@ <h4>One source of truth</h4>
       <p>The parse stage reads <code>model.xml</code> directly: variables, equations, per-equation coefficients, and the endogenous / exogenous classification for the VAR-expectations ("standard") equation set — the only equation set this pipeline reads. Nothing is re-typed or re-specified — the XML file is the single source of truth for the model's semantics. That yields <strong>284 endogenous variables and their equations</strong>, classified into behavioural and identity blocks exactly as the Board's file declares them.</p>
       <h4>Residual form</h4>
       <p>Each equation is normalised — EViews lag and difference notation rewritten, coefficient references substituted by their published values — flipped to residual form \(F_i(\cdot) = 0\), and augmented with its additive shock term (<code>_aerr</code>) and a tracking residual (<code>_trac</code>), so each equation solves \(F_i + \mathrm{aerr}_i + \mathrm{trac}_i = 0\).</p>
-        <div class="meth-inline" aria-hidden="true">
+        <div class="meth-inline">
         <div class="meth-sec-title">parse</div>
         <div class="meth-eq"><div class="meth-eq-label">system</div><div class="meth-eq-body">\(284\) endogenous eqs (behavioural + identities), from <code>model.xml</code></div></div>
         <div class="meth-eq"><div class="meth-eq-label">residual form</div><div class="meth-eq-body">\(F_i(x,\,\text{data}) + \text{aerr}_i + \text{trac}_i = 0\)</div></div>
@@ -172,7 +172,7 @@ <h4>The baseline database</h4>
       <p>The load stage brings <code>LONGBASE.TXT</code> — the April 2026 vintage — into a pandas frame on a quarterly <code>PeriodIndex</code>. This is the Board's own historical and projection database, and it conditions everything that follows: every simulation is read as a deviation from this baseline.</p>
       <h4>Tracking residuals (<code>init_trac</code>)</h4>
       <p><code>init_trac</code> sets each equation's tracking residual so that solving the model reproduces the database <strong>to machine precision</strong> — an invariant checked on the <a href="/frb-us/validation">validation page</a>. This is the Fed's own device, used the Fed's own way — the direct analogue of the OBR emulator's anchoring. Shocks are then imposed on top of the add-factored baseline.</p>
-        <div class="meth-inline" aria-hidden="true">
+        <div class="meth-inline">
         <div class="meth-sec-title">LONGBASE conditioning</div>
         <div class="meth-eq"><div class="meth-eq-label">tracking residual</div><div class="meth-eq-body">\(\text{trac}_i = -F_i(\text{baseline})\) per equation, per quarter</div></div>
         <div class="meth-eq"><div class="meth-eq-label">invariant</div><div class="meth-eq-body">\(\text{solve}(\text{baseline} + \text{trac}) = \text{LONGBASE}\) exactly</div></div>
@@ -213,7 +213,7 @@ <h4>Analytic sparse Jacobian</h4>
       <p>Each quarter, a damped Newton iteration solves the full simultaneous system with an <strong>analytic sparse Jacobian</strong> — differentiated symbolically with sympy, factorised with <code>scipy.sparse.linalg.splu</code>. Damping backtracks the step only to stay inside the domain of definition (no logs of non-positive arguments, no NaNs). Defaults are \(\mathrm{xtol} = 10^{-8}\), \(\mathrm{rtol} = 5\times 10^{-4}\).</p>
       <h4>One block, not many</h4>
       <p>The reference implementation decomposes each period into recursive and simultaneous blocks; this one solves the whole period at once. Both approaches solve the same \(F(x) = 0\), which is why the two agree to solver tolerance rather than to some negotiated threshold — see <a href="/frb-us/validation">how far to trust it</a>.</p>
-        <div class="meth-inline" aria-hidden="true">
+        <div class="meth-inline">
         <div class="meth-sec-title">Newton iteration</div>
         <div class="meth-eq"><div class="meth-eq-label">step</div><div class="meth-eq-body">\(J(x^{(k)})\,\Delta x = -F(x^{(k)})\), sparse LU (SuperLU)</div></div>
         <div class="meth-eq"><div class="meth-eq-label">damping</div><div class="meth-eq-body">halve \(\Delta x\) until \(F, J\) evaluate cleanly</div></div>
@@ -251,10 +251,10 @@ <h4>One block, not many</h4>
         <h4>A modular funds rate</h4>
       <p>Monetary policy is modular: exogenous 0/1 switches in the model file select which rule sets the funds rate. Three are exposed through <code>frbus_shock</code>: <code>inertial_taylor</code> (the default, the LONGBASE rule and the one the validation numbers use), <code>taylor</code>, and <code>fixed_funds_rate</code>, which holds the funds rate on its baseline path so there is no endogenous monetary offset.</p>
       <h4>The rule choice is the exercise</h4>
-      <p>The choice is usually the economic point of the exercise. Run the same government-purchases shock (1% of GDP) under both settings and the multiplier moves from <strong>0.72</strong> in year one under the inertial Taylor rule to <strong>0.99</strong> in year two with the funds rate fixed — and the price-level response is larger still without the monetary offset. The <a href="/frb-us">overview</a> places both inside the published cross-model ranges.</p>
+      <p>The choice is usually the economic point of the exercise. Run the same government-purchases shock (1% of GDP) under both settings and the year-two multiplier moves from <strong>0.64</strong> under the inertial Taylor rule to <strong>0.99</strong> with the funds rate fixed — and the price-level response is larger still without the monetary offset. The <a href="/frb-us">overview</a> places both inside the published cross-model ranges.</p>
       <h4>Each rule reads its own error term</h4>
       <p>Each policy rule reads its <em>own</em> add-error. The shock term \(\varepsilon_t\) in the inertial-Taylor rule above is <code>rffintay_aerr</code> — the instrument a monetary shock moves — and it works only under <code>inertial_taylor</code>; asking for it under <code>taylor</code>, which applies the same reaction bracket without the inertia term and reads <code>rfftay_aerr</code> instead, is rejected with a pointer to <code>rfftay_aerr</code> rather than silently returning all-zero responses — one of the two traps the <a href="/frb-us/code">code page</a> walks through before you shock anything.</p>
-        <div class="meth-inline" aria-hidden="true">
+        <div class="meth-inline">
         <div class="meth-sec-title">policy rules</div>
         <div class="meth-eq"><div class="meth-eq-label">inertial Taylor</div><div class="meth-eq-body">\(r_t = 0.85\,r_{t-1} + 0.15\bigl[r^{*}_t + \bar\pi_t + 0.5(\bar\pi_t - \pi^{\text{targ}}) + 1.0\,\tilde x_t\bigr] + \varepsilon_t\)</div></div>
         <div class="meth-eq"><div class="meth-eq-label">effective rate</div><div class="meth-eq-body">\(\texttt{rff}_t = \max(\texttt{rffrule}_t,\ \texttt{rffmin}_t)\)</div></div>
@@ -294,7 +294,7 @@ <h4>What comes out</h4>
       <p>The headline 100bp tightening lands where the Board's documentation says it should: the funds rate up <strong>1.000pp</strong> on impact, real GDP bottoming at <strong>&minus;0.55%</strong> in 2027Q4, unemployment (<code>lur</code>) peaking <strong>+0.26pp</strong>, core inflation (<code>picxfe</code>) down <strong>0.034pp</strong> at its trough. The same four numbers are the simulation-properties row of the <a href="/frb-us/validation">validation tables</a>.</p>
       <h4>What stays out of scope</h4>
       <p>The pipeline covers the VAR-expectations model only: the model-consistent-expectations variants, forward guidance, and the reference implementation's <code>mcontrol</code> trajectory-targeting routine are not implemented, and there is no bridge from a PolicyEngine reform to model variables — see the <a href="/frb-us/validation#caveats">known limits</a>.</p>
-        <div class="meth-inline" aria-hidden="true">
+        <div class="meth-inline">
         <div class="meth-sec-title">impulse responses</div>
         <div class="meth-eq"><div class="meth-eq-label">shock</div><div class="meth-eq-body">\(\text{aerr} \mathrel{+}= \text{shock}\) on the add-factored baseline</div></div>
         <div class="meth-eq"><div class="meth-eq-label">response</div><div class="meth-eq-body">\(\text{IRF}_t = \text{sim}_t - \text{baseline}_t\) over the same window</div></div>
diff --git a/integration/src/policyengine_macro/capabilities.py b/integration/src/policyengine_macro/capabilities.py
index a5003aeb..bd0d8b30 100644
--- a/integration/src/policyengine_macro/capabilities.py
+++ b/integration/src/policyengine_macro/capabilities.py
@@ -541,7 +541,8 @@ def _quality(level: str, evidence: str, next_gate: str) -> dict:
             "The upstream R code runs unmodified at pinned commit 846081a and "
             "the manual's Table 4 macro block replicates within stated "
             "tolerances. The clean-room Python reimplementation has landed "
-            "§2.2 accounting and §3.2's 23 equations, and surfaced three "
+            "§2.2 accounting and §3.2, §3.3.1 and §3.3.2 — 118 equations, "
+            "Eqs. (21)-(138) contiguous — and surfaced some thirty-five "
             "defects in the manual itself, each pinned by a test rather than "
             "absorbed into a tolerance.",
             "Complete §3.3 and the oracle comparison, closing milestone 2.",
diff --git a/models/index.html b/models/index.html
index 9d92d5d5..23f7ec13 100644
--- a/models/index.html
+++ b/models/index.html
@@ -118,7 +118,7 @@ <h1 class="reveal page-title" style="--d:1">Choose the model that fits the quest
       <details class="qa">
         <summary><span>A monetary or fiscal shock in the US over coming quarters</span><strong>US · frb-us</strong></summary>
         <div class="qa-body">
-          <p>The Federal Reserve’s FRB/US model — 284 behavioural equations parsed from the Board’s own model file — is solved by damped Newton around the April 2026 LONGBASE baseline; you pick the lever and the policy rule, and read the deviation path quarter by quarter.</p>
+          <p>The Federal Reserve’s FRB/US model — 284 equations &mdash; 120 estimated behavioural equations and 164 identities &mdash; parsed from the Board’s own model file — is solved by damped Newton around the April 2026 LONGBASE baseline; you pick the lever and the policy rule, and read the deviation path quarter by quarter.</p>
           <div class="qa-facts">
           <div class="qa-fact"><span class="icon">&rarr;</span><span>Levers: federal funds rate, government spending, transfers, taxes — each in its own units</span></div>
           <div class="qa-fact"><span class="icon">&rarr;</span><span>Replicates the Board’s dynamics to machine precision (residuals ~1e-8 or better)</span></div>
@@ -193,7 +193,7 @@ <h1 class="reveal page-title" style="--d:1">Choose the model that fits the quest
   <section id="compare" class="band">
     <div class="band-head">
       <span class="kicker mono">compare the models</span>
-      <h2>Seven model classes, side by side.</h2>
+      <h2>Seven models, side by side, side by side.</h2>
     </div>
     <div class="prose comparison-prose">
       <div class="table-scroll desktop-comparison comparison-table-wrap cmp-clamped" id="cmp-wrap">
@@ -313,7 +313,7 @@ <h2>How each model is validated.</h2>
           <tr>
             <th scope="row">Cost of cutting the basic rate 1p, 2026&ndash;27 (obr-macro + pe)</th>
             <td>£6.46bn static</td>
-            <td><a href="https://www.gov.uk/government/statistics/direct-effects-of-illustrative-tax-changes/direct-effects-of-illustrative-tax-changes-bulletin-january-2025">HMRC ready reckoner</a> (June 2025): £6.9bn, rising to £8.2bn by 2028&ndash;29</td>
+            <td><a href="https://www.gov.uk/government/statistics/direct-effects-of-illustrative-tax-changes">HMRC ready reckoner</a> (June 2025): £6.9bn, rising to £8.2bn by 2028&ndash;29</td>
             <td>Direct — &minus;6.4% in year one; HMRC's figure embeds behavioural response</td>
           </tr>
           <tr>
diff --git a/obr/code/index.html b/obr/code/index.html
index 8206e3e4..ad715d53 100644
--- a/obr/code/index.html
+++ b/obr/code/index.html
@@ -109,7 +109,7 @@ <h2>Shock a variable, read the deviation.</h2>
           shock a variable, list the levers, or score a reform from the
           terminal.
         </p>
-        <div class="meth-inline" aria-hidden="true">
+        <div class="meth-inline">
           <div class="meth-sec-title">CLI equivalents</div>
           <div class="codeblock"><button class="cb-copy" data-copy>copy</button><pre><code>pe-macro obr-shock --var CGG --shock 1250
 pe-macro variables
@@ -134,7 +134,7 @@ <h2>Shock a variable, read the deviation.</h2>
           <code>CGIPS</code> per quarter), and a £10bn austerity cut — with
           charts landing in <code>outputs/</code>.
         </p>
-        <div class="meth-inline" aria-hidden="true">
+        <div class="meth-inline">
           <div class="meth-sec-title">install and run</div>
           <div class="codeblock"><button class="cb-copy" data-copy>copy</button><pre><code>pip install git+https://github.com/PolicyEngine/obr-macroeconomic-model</code></pre></div>
           <div class="codeblock"><button class="cb-copy" data-copy>copy</button><pre><code><span class="kw">from</span> obr_macro <span class="kw">import</span> run_reform
@@ -164,7 +164,7 @@ <h2>Shock a variable, read the deviation.</h2>
           The full glossary — all 636 codes with descriptions, ONS series and
           equations — is the model package's variable and equation references.
         </p>
-        <div class="meth-inline" aria-hidden="true">
+        <div class="meth-inline">
           <div class="meth-sec-title">levers and readouts</div>
           <div class="dial-cards">
           <div class="dial-card">
@@ -190,15 +190,15 @@ <h2>Shock a variable, read the deviation.</h2>
         <div class="meth-panel" data-step="1" hidden>
           <div class="meth-panel-head"><span class="meth-panel-title">Code</span><span class="meth-panel-badge mono">Step 2</span></div>
           <div class="meth-sec-title">CLI equivalents</div>
-          <div class="codeblock"><button class="cb-copy" data-copy>copy</button><pre><code>pe-macro obr-shock --var CGG --shock 1250
+          <div class="codeblock"><button class="cb-copy" data-copy tabindex="-1">copy</button><pre><code>pe-macro obr-shock --var CGG --shock 1250
 pe-macro variables
 pe-macro score --reform '{...}' --model obr</code></pre></div>
         </div>
         <div class="meth-panel" data-step="2" hidden>
           <div class="meth-panel-head"><span class="meth-panel-title">Code</span><span class="meth-panel-badge mono">Step 3</span></div>
           <div class="meth-sec-title">install and run</div>
-          <div class="codeblock"><button class="cb-copy" data-copy>copy</button><pre><code>pip install git+https://github.com/PolicyEngine/obr-macroeconomic-model</code></pre></div>
-          <div class="codeblock"><button class="cb-copy" data-copy>copy</button><pre><code><span class="kw">from</span> obr_macro <span class="kw">import</span> run_reform
+          <div class="codeblock"><button class="cb-copy" data-copy tabindex="-1">copy</button><pre><code>pip install git+https://github.com/PolicyEngine/obr-macroeconomic-model</code></pre></div>
+          <div class="codeblock"><button class="cb-copy" data-copy tabindex="-1">copy</button><pre><code><span class="kw">from</span> obr_macro <span class="kw">import</span> run_reform
 
 <span class="cm"># £5bn/yr government spending shock (£1.25bn per quarter)</span>
 results = run_reform(
@@ -269,7 +269,13 @@ <h2>Shock a variable, read the deviation.</h2>
     });
     // Panels pair to steps by data-step. A step with no panel of its own
     // keeps the nearest earlier panel pinned rather than blanking the column.
-    var want = -1;
+    // Start at the FIRST panel, not -1: several pages have no panel for
+    // step 0, and seeding -1 left the sticky column empty for the whole of
+    // the opening step before popping in. Pinning the earliest panel means
+    // the column is populated from the top and only ever moves forward.
+    var want = Math.min.apply(null, panels.map(function (panel) {
+      return Number(panel.getAttribute("data-step"));
+    }));
     panels.forEach(function (panel) {
       var ds = Number(panel.getAttribute("data-step"));
       if (ds <= active && ds > want) want = ds;
diff --git a/obr/index.html b/obr/index.html
index 0e45db24..24cfcc3a 100644
--- a/obr/index.html
+++ b/obr/index.html
@@ -82,7 +82,7 @@ <h2>Microsimulation supplies the direct cost; this model returns the feedback.</
       <div class="qa-facts">
         <div class="qa-fact"><span class="icon">&rarr;</span><span><strong>Connection.</strong> Static-costing bridge, both hosted: <a href="/pe">PolicyEngine</a> computes the direct Exchequer effect of a statutory reform; <code>score_reform</code> injects it through the declared <code>HHDI_ADDFACTOR</code> point and this model returns the economy-wide second round.</span></div>
         <div class="qa-fact"><span class="icon">&rarr;</span><span><strong>Worked example.</strong> 1p on the basic rate from April 2026: direct yield &pound;6.46bn in 2026; GDP &minus;0.020% on impact, deepening to &minus;0.058% by 2027Q4 (full walkthrough &rarr; <a href="/models#score">score a reform</a>).</span></div>
-        <div class="qa-fact"><span class="icon">&rarr;</span><span><strong>Validation.</strong> <a href="/obr/validation">Anchored to the March 2026 EFO</a> (0.15% GDP MAPE, CI-gated) and checked against HMRC's ready reckoner (&pound;6.46bn vs &pound;6.9bn).</span></div>
+        <div class="qa-fact"><span class="icon">&rarr;</span><span><strong>Validation.</strong> <a href="/obr/validation">Anchored to the March 2026 EFO</a> (0.15% GDP MAPE) &mdash; but anchored accuracy is a <strong>by-construction invariant</strong>, not a forecast claim: free-running, the same equations miss GDP by <strong>5.75%</strong>. Checked against HMRC&rsquo;s ready reckoner (&pound;6.46bn vs &pound;6.9bn).</span></div>
         <div class="qa-fact"><span class="icon">&rarr;</span><span><strong>Results.</strong> Run it from <a href="/connect">any connected client</a>; every score carries its assumptions and data vintage.</span></div>
       </div>
     </div>
diff --git a/obr/methodology/index.html b/obr/methodology/index.html
index d096d1c9..04f09f51 100644
--- a/obr/methodology/index.html
+++ b/obr/methodology/index.html
@@ -132,7 +132,7 @@ <h4>From EViews to Python</h4>
       <p>The OBR publishes its full equation listing, but as EViews source that few can run. The transpiler rewrites every EViews line — <code>dlog()</code>, <code>@elem</code>, lags like <code>X(-1)</code>, error-correction terms — into an equivalent Python expression, compiled once per solve.</p>
       <h4>What comes out</h4>
       <p>The result is the model itself in open code: the same behavioural equations, identities, and technical relationships the official forecaster relies on, runnable without an EViews licence.</p>
-        <div class="meth-inline" aria-hidden="true">
+        <div class="meth-inline">
         <div class="meth-sec-title">translation</div>
         <div class="meth-eq"><div class="meth-eq-label">growth rates</div><div class="meth-eq-body">\(\mathrm{dlog}(X) \;\rightarrow\; \log X_t - \log X_{t-1}\)</div></div>
         <div class="meth-eq"><div class="meth-eq-label">lags &amp; elements</div><div class="meth-eq-body">\(X_{t-1}\), <code>@elem</code> &rarr; indexed array reads</div></div>
@@ -163,7 +163,7 @@ <h4>Loading history</h4>
       <p>History loads from the OBR's detailed forecast tables. Of the 261 endogenous variables the listing referenced but the outlook did not print, <strong>199 had ONS series codes</strong>; a vendored 348-series ONS snapshot supplies them.</p>
       <h4>Why it matters</h4>
       <p>That snapshot ships inside the package — there is no download step — and it gives the circular fiscal and financial sub-blocks real balancing data rather than guessed starting values.</p>
-        <div class="meth-inline" aria-hidden="true">
+        <div class="meth-inline">
         <div class="meth-sec-title">data sources</div>
         <div class="meth-eq"><div class="meth-eq-label">forecast tables</div><div class="meth-eq-body">OBR detailed forecast tables, March 2026 EFO</div></div>
         <div class="meth-eq"><div class="meth-eq-label">ONS snapshot</div><div class="meth-eq-body">348 vendored series; 199 of 261 unprinted variables matched</div></div>
@@ -201,7 +201,7 @@ <h4>Simultaneous within a quarter</h4>
       <p>The equations depend on each other, so each quarter is solved simultaneously: cycle through the equations, update, repeat to convergence. Quarters then solve sequentially forward.</p>
       <h4>Runtime</h4>
       <p>A twelve-quarter anchored solve takes roughly <strong>30–40 seconds</strong> single-threaded; a reform run is two such solves.</p>
-        <div class="meth-inline" aria-hidden="true">
+        <div class="meth-inline">
         <div class="meth-sec-title">solver</div>
         <div class="meth-eq"><div class="meth-eq-label">within quarter</div><div class="meth-eq-body">cycle &rarr; update &rarr; repeat to convergence</div></div>
         <div class="meth-eq"><div class="meth-eq-label">across quarters</div><div class="meth-eq-body">sequential forward solve</div></div>
@@ -237,7 +237,7 @@ <h4>Forecasting with held add-factors</h4>
       <p>For forecasting rather than one-off shocks, the emulator fits add-factors over a base window of <strong>2024Q1–2025Q4</strong>, <strong>holds them constant</strong>, and projects to <strong>2027Q4</strong>, with exogenous inputs extrapolated (seasonal cash series at their trailing annual mean). Scenario solves run twelve quarters, 2025Q1–2027Q4.</p>
       <h4>Horizon limits</h4>
       <p>The anchored horizon extends to <strong>2031Q1</strong>, where GDP is reproduced to 0.29%. Anchored unemployment (<code>LFSUR</code>) is unreliable beyond 2027Q4, drifting to 0.9% against the EFO's 4.1% by 2031Q1.</p>
-        <div class="meth-inline" aria-hidden="true">
+        <div class="meth-inline">
         <div class="meth-sec-title">anchoring</div>
         <div class="meth-eq"><div class="meth-eq-label">add-factor window</div><div class="meth-eq-body">fit 2024Q1–2025Q4, hold constant, project to 2027Q4</div></div>
         <div class="meth-eq"><div class="meth-eq-label">scenario solves</div><div class="meth-eq-body">12 quarters, 2025Q1–2027Q4</div></div>
@@ -274,7 +274,7 @@ <h4>Closure swap</h4>
       <p>The GDP identity is switched off in the OBR file, so for a shock <code>GDPM</code> is swapped in and inventories (<code>DINV</code>) become the residual — output responds to the policy instead of being fixed.</p>
       <h4>Shock against an identical baseline</h4>
       <p>The emulator solves shocked and baseline paths with identical structure and reports the difference, isolating the policy effect from the underlying forecast.</p>
-        <div class="meth-inline" aria-hidden="true">
+        <div class="meth-inline">
         <div class="meth-sec-title">closure and shock</div>
         <div class="meth-eq"><div class="meth-eq-label">closure swap</div><div class="meth-eq-body"><code>GDPM</code> endogenised; <code>DINV</code> becomes the residual</div></div>
         <div class="meth-eq"><div class="meth-eq-label">policy effect</div><div class="meth-eq-body">\(X_t^{\text{shocked}} - X_t^{\text{baseline}}\), identical structure</div></div>
diff --git a/olg/code/index.html b/olg/code/index.html
index e7279957..b6d66bdf 100644
--- a/olg/code/index.html
+++ b/olg/code/index.html
@@ -121,7 +121,7 @@ <h2>From a few lines of Python to a full UK transition path.</h2>
           <code>conda env create -f environment.yml &amp;&amp; conda activate oguk-dev &amp;&amp; pip install -e .</code>
           works too.
         </p>
-        <div class="meth-inline" aria-hidden="true">
+        <div class="meth-inline">
           <div class="meth-sec-title">code</div>
           <div class="codeblock"><button class="cb-copy" data-copy>copy</button><pre><code>git clone https://github.com/PSLmodels/OG-UK.git
 cd OG-UK
@@ -166,7 +166,7 @@ <h2>From a few lines of Python to a full UK transition path.</h2>
           <code>g_y_annual</code>, which is the balanced-growth normalisation
           the model detrends by, not a productivity lever.
         </div>
-        <div class="meth-inline" aria-hidden="true">
+        <div class="meth-inline">
           <div class="meth-sec-title">code</div>
           <div class="codeblock"><button class="cb-copy" data-copy>copy</button><pre><code>from datetime import datetime
 from policyengine.core import ParameterValue, Policy
@@ -262,7 +262,7 @@ <h2>From a few lines of Python to a full UK transition path.</h2>
           and check whether the reform is extreme — OLG models can diverge on
           very large shocks.
         </p>
-        <div class="meth-inline" aria-hidden="true">
+        <div class="meth-inline">
           <div class="meth-sec-title">code</div>
           <div class="codeblock"><button class="cb-copy" data-copy>copy</button><pre><code>from oguk import solve_steady_state, map_to_real_world
 
@@ -315,7 +315,7 @@ <h2>From a few lines of Python to a full UK transition path.</h2>
           are <strong>not</strong> wired into the PolicyEngine Macro CLI yet —
           the CLI is steady-state only (see step 7).
         </p>
-        <div class="meth-inline" aria-hidden="true">
+        <div class="meth-inline">
           <div class="meth-sec-title">code</div>
           <div class="codeblock"><button class="cb-copy" data-copy>copy</button><pre><code>from dask.distributed import Client
 from oguk import run_transition_path, map_transition_to_real_world
@@ -376,7 +376,7 @@ <h2>From a few lines of Python to a full UK transition path.</h2>
           read as a forecast of anything — see
           <a href="/olg/validation">how far to trust it</a>.
         </div>
-        <div class="meth-inline" aria-hidden="true">
+        <div class="meth-inline">
           <div class="meth-sec-title">code</div>
           <div class="codeblock"><button class="cb-copy" data-copy>copy</button><pre><code>impact = map_transition_to_real_world(base_tp, reform_tp)
 
@@ -409,7 +409,7 @@ <h2>From a few lines of Python to a full UK transition path.</h2>
         <p class="opt-note">
           runnable file: <a href="/olg/code/06_multi_sector.py">06_multi_sector.py</a>.
         </p>
-        <div class="meth-inline" aria-hidden="true">
+        <div class="meth-inline">
           <div class="meth-sec-title">code</div>
           <div class="codeblock"><button class="cb-copy" data-copy>copy</button><pre><code>base_tp, reform_tp = run_transition_path(
     start_year=2026,
@@ -451,7 +451,7 @@ <h2>From a few lines of Python to a full UK transition path.</h2>
       <aside class="scrolly-sticky" aria-hidden="true">
         <div class="meth-panel" data-step="0">
           <div class="meth-panel-head"><span class="meth-panel-title">Code</span><span class="meth-panel-badge mono">Step 1</span></div>
-          <div class="codeblock"><button class="cb-copy" data-copy>copy</button><pre><code>git clone https://github.com/PSLmodels/OG-UK.git
+          <div class="codeblock"><button class="cb-copy" data-copy tabindex="-1">copy</button><pre><code>git clone https://github.com/PSLmodels/OG-UK.git
 cd OG-UK
 uv sync
 export HUGGING_FACE_TOKEN=hf_your_token_here</code></pre></div>
@@ -459,7 +459,7 @@ <h2>From a few lines of Python to a full UK transition path.</h2>
         </div>
         <div class="meth-panel" data-step="1" hidden>
           <div class="meth-panel-head"><span class="meth-panel-title">Code</span><span class="meth-panel-badge mono">Step 2</span></div>
-          <div class="codeblock"><button class="cb-copy" data-copy>copy</button><pre><code>from datetime import datetime
+          <div class="codeblock"><button class="cb-copy" data-copy tabindex="-1">copy</button><pre><code>from datetime import datetime
 from policyengine.core import ParameterValue, Policy
 from policyengine.tax_benefit_models.uk import uk_latest
 
@@ -477,7 +477,7 @@ <h2>From a few lines of Python to a full UK transition path.</h2>
     ],
 )</code></pre></div>
           <div class="meth-sec-title">structural shocks via param_overrides</div>
-          <div class="codeblock"><button class="cb-copy" data-copy>copy</button><pre><code># corporation tax cut to 25%
+          <div class="codeblock"><button class="cb-copy" data-copy tabindex="-1">copy</button><pre><code># corporation tax cut to 25%
 ss = solve_steady_state(param_overrides={"cit_rate": [[0.25]]})
 
 # productivity: +0.4% TFP level shock
@@ -491,7 +491,7 @@ <h2>From a few lines of Python to a full UK transition path.</h2>
         </div>
         <div class="meth-panel" data-step="2" hidden>
           <div class="meth-panel-head"><span class="meth-panel-title">Code</span><span class="meth-panel-badge mono">Step 3</span></div>
-          <div class="codeblock"><button class="cb-copy" data-copy>copy</button><pre><code>from oguk import solve_steady_state, map_to_real_world
+          <div class="codeblock"><button class="cb-copy" data-copy tabindex="-1">copy</button><pre><code>from oguk import solve_steady_state, map_to_real_world
 
 from og_dashboard.reform import REFORM
 
@@ -524,7 +524,7 @@ <h2>From a few lines of Python to a full UK transition path.</h2>
         </div>
         <div class="meth-panel" data-step="3" hidden>
           <div class="meth-panel-head"><span class="meth-panel-title">Code</span><span class="meth-panel-badge mono">Step 4</span></div>
-          <div class="codeblock"><button class="cb-copy" data-copy>copy</button><pre><code>from dask.distributed import Client
+          <div class="codeblock"><button class="cb-copy" data-copy tabindex="-1">copy</button><pre><code>from dask.distributed import Client
 from oguk import run_transition_path, map_transition_to_real_world
 
 from og_dashboard.reform import REFORM
@@ -552,7 +552,7 @@ <h2>From a few lines of Python to a full UK transition path.</h2>
         </div>
         <div class="meth-panel" data-step="4" hidden>
           <div class="meth-panel-head"><span class="meth-panel-title">Code</span><span class="meth-panel-badge mono">Step 5</span></div>
-          <div class="codeblock"><button class="cb-copy" data-copy>copy</button><pre><code>impact = map_transition_to_real_world(base_tp, reform_tp)
+          <div class="codeblock"><button class="cb-copy" data-copy tabindex="-1">copy</button><pre><code>impact = map_transition_to_real_world(base_tp, reform_tp)
 
 impact.years              # fiscal-year strings: ["2026-27", ..., "2085-86"]
 impact.gdp                # reform GDP path (£bn, per year)
@@ -569,7 +569,7 @@ <h2>From a few lines of Python to a full UK transition path.</h2>
         </div>
         <div class="meth-panel" data-step="5" hidden>
           <div class="meth-panel-head"><span class="meth-panel-title">Code</span><span class="meth-panel-badge mono">Step 6</span></div>
-          <div class="codeblock"><button class="cb-copy" data-copy>copy</button><pre><code>base_tp, reform_tp = run_transition_path(
+          <div class="codeblock"><button class="cb-copy" data-copy tabindex="-1">copy</button><pre><code>base_tp, reform_tp = run_transition_path(
     start_year=2026,
     policy=REFORM,
     client=client,
@@ -625,7 +625,13 @@ <h2>From a few lines of Python to a full UK transition path.</h2>
     });
     // Panels pair to steps by data-step. A step with no panel of its own
     // keeps the nearest earlier panel pinned rather than blanking the column.
-    var want = -1;
+    // Start at the FIRST panel, not -1: several pages have no panel for
+    // step 0, and seeding -1 left the sticky column empty for the whole of
+    // the opening step before popping in. Pinning the earliest panel means
+    // the column is populated from the top and only ever moves forward.
+    var want = Math.min.apply(null, panels.map(function (panel) {
+      return Number(panel.getAttribute("data-step"));
+    }));
     panels.forEach(function (panel) {
       var ds = Number(panel.getAttribute("data-step"));
       if (ds <= active && ds > want) want = ds;
diff --git a/olg/methodology/index.html b/olg/methodology/index.html
index ccce1d2d..3089206d 100644
--- a/olg/methodology/index.html
+++ b/olg/methodology/index.html
@@ -127,7 +127,7 @@ <h4>Age cohorts</h4>
       <p>OG-Core models S economically active age cohorts (typically 80, ages 21 to 100). Each period a new cohort is born and the oldest dies with some probability; the age distribution evolves with fertility, mortality, and immigration, reaching a stationary distribution in the long run.</p>
       <h4>Ability types</h4>
       <p>Each cohort splits into J ability types with distinct permanent labour productivity. Type is assigned at birth and never changes, and each type carries its own discount factor \(\beta_j\), bequest-motive weight \(\chi^b_j\), and productivity profile \(e_{j,s}\) — so productivity varies both across types and by age within a type. High-ability types earn more per hour and typically save more, so reforms hit the income distribution, not just the average.</p>
-        <div class="meth-inline" aria-hidden="true">
+        <div class="meth-inline">
         <div class="meth-sec-title">population</div>
         <div class="meth-eq"><div class="meth-eq-label">age cohorts</div><div class="meth-eq-body">\(s = E+1, \ldots, E+S \quad \text{(economically active)}\)</div></div>
         <div class="meth-eq"><div class="meth-eq-label">ability types</div><div class="meth-eq-body">\(j = 1, \ldots, J \quad \text{with probability } \lambda_j\)</div></div>
@@ -172,7 +172,7 @@ <h4>Bequest utility</h4>
       <h4>Budget constraint and Euler equations</h4>
       <p>Each period, savings returns, labour earnings, bequests received, government transfers, and pensions must cover consumption (including consumption taxes), income and wealth taxes, and savings carried forward.</p>
       <p>Two Euler equations characterise the optimum: work until the after-tax wage equals the marginal disutility of labour; save until the discounted, after-tax return justifies forgoing consumption today.</p>
-        <div class="meth-inline" aria-hidden="true">
+        <div class="meth-inline">
         <div class="meth-sec-title">utility and Euler equations</div>
         <div class="meth-eq"><div class="meth-eq-label">consumption</div><div class="meth-eq-body">\(u(c) = \frac{c^{\,1-\sigma}}{1-\sigma}\)</div></div>
         <div class="meth-eq"><div class="meth-eq-label">labour disutility</div><div class="meth-eq-body">\(\chi^n_s \cdot g(n) \quad \text{(elliptical)}\)</div></div>
@@ -217,7 +217,7 @@ <h4>First-order conditions</h4>
       <p>Profit maximisation yields the standard conditions: the <strong>wage equals the marginal product of labour</strong>; the <strong>rental rate</strong> equals the after-tax marginal product of capital, net of depreciation, tax deductions, and investment tax credits.</p>
       <h4>Public capital rents</h4>
       <p>A distinctive OG-Core feature: <strong>public capital generates economic rents</strong>. Firms cannot deduct the cost of government infrastructure, so its returns flow to private capital owners through an augmented rate of return.</p>
-        <div class="meth-inline" aria-hidden="true">
+        <div class="meth-inline">
         <div class="meth-sec-title">technology</div>
         <div class="meth-eq"><div class="meth-eq-label">CES production</div><div class="meth-eq-body">\(Y_m = Z_m\bigl[\gamma K_m^\rho + \gamma_g K_{g,m}^\rho + (1\!-\!\gamma\!-\!\gamma_g)(e^{g_y t}L_m)^\rho\bigr]^{1/\rho}\)</div></div>
         <div class="meth-eq"><div class="meth-eq-label">wage</div><div class="meth-eq-body">\(w = p_m \cdot \text{MPL}_m\)</div></div>
@@ -255,7 +255,7 @@ <h4>Other tax instruments</h4>
       <p>The <strong>wealth tax</strong> is a progressive three-parameter function, spanning zero taxation to smoothly increasing marginal rates. <strong>Consumption taxes</strong> are linear rates by good, covering VAT and excise. <strong>Corporate income taxes</strong> are flat rates by industry.</p>
       <h4>Fiscal closure rule</h4>
       <p>Debt cannot grow without bound, so a <strong>fiscal closure rule</strong> activates after a specified period, bringing debt-to-GDP gradually to a target via government spending, transfers, or both — the modeller's choice.</p>
-        <div class="meth-inline" aria-hidden="true">
+        <div class="meth-inline">
         <div class="meth-sec-title">government budget</div>
         <p>\[ D_{t+1} + \text{Rev}_t = (1+r_{\text{gov}})D_t + G + I_g + \text{Pen} + TR + UBI \]</p>
         </div>
@@ -289,7 +289,7 @@ <h4>Goods and debt markets</h4>
       <p>Each industry's output must equal consumption demand; the final industry clears residually, absorbing investment, government purchases, and remaining demand. Government debt splits between domestic and foreign holders.</p>
       <h4>What defines an equilibrium</h4>
       <p>An equilibrium is a set of prices — interest rate, wage, goods prices — at which households and firms optimise, the government budget holds, and all four markets clear: a large-scale nonlinear fixed-point problem.</p>
-        <div class="meth-inline" aria-hidden="true">
+        <div class="meth-inline">
         <div class="meth-sec-title">equilibrium conditions</div>
         <div class="meth-eq"><div class="meth-eq-label">labour</div><div class="meth-eq-body">\(\sum_{s,j} \omega_s \lambda_j e_{j,s} n_{j,s} = \sum_m L_m\)</div></div>
         <div class="meth-eq"><div class="meth-eq-label">capital</div><div class="meth-eq-body">\(K = K^d + K^f \quad (\text{open economy: } \zeta_K)\)</div></div>
@@ -336,7 +336,7 @@ <h4>Stage 1: Steady state</h4>
       <p>The steady state — all variables constant after removing trend growth — comes from a nested fixed-point algorithm. The <em>outer loop</em> guesses aggregate prices and quantities; the <em>inner loop</em> solves the Euler equations for every ability type and age, <strong>in parallel</strong> since types are independent. Aggregation, firm demands, and the government budget check yield new guesses; iterate to convergence.</p>
       <h4>Stage 2: Transition path (TPI)</h4>
       <p><em>Time Path Iteration</em> (TPI) then solves the economy's path to the steady state over T periods: guess entire time paths for prices and quantities, solve every cohort's lifetime decisions given those paths (with <strong>rational expectations</strong> about future prices), check the implied paths against the guesses, iterate. This produces the year-by-year projections.</p>
-        <div class="meth-inline" aria-hidden="true">
+        <div class="meth-inline">
         <div class="meth-sec-title">two-stage algorithm</div>
         <div class="meth-eq"><div class="meth-eq-label">outer loop</div><div class="meth-eq-body">\(\mathbf{x} = \{\bar{r}_p, \bar{r}, \bar{w}, \{\bar{p}_m\}, \overline{BQ}, \overline{TR}, \text{factor}\}\)</div></div>
         <div class="meth-eq"><div class="meth-eq-label">inner loop</div><div class="meth-eq-body">\(\text{Solve } 2JS \text{ Euler equations } \forall (j,s)\)</div></div>
@@ -378,7 +378,7 @@ <h4>Tax function estimation</h4>
       <p>Taxes are not stylized wedges. OG-UK runs <a href="/pe">PolicyEngine-UK</a> on household microdata under baseline and reform, computes marginal rates by perturbation — add £1 of employment (or dividend) income per adult, rerun, read the change in net income — and fits three-parameter <strong>Gouveia–Strauss</strong> schedules to the effective-rate microdata for each year of a three-year budget window. Marginal-rate schedules are analytical derivatives of the same fitted parameters, so ETRs and MTRs are consistent by construction; the model sees the reform's true shape across the income and age distribution.</p>
       <h4>Real-world mapping</h4>
       <p>The model works in abstract units. OG-UK anchors steady-state GDP to the ONS figure — live series with cached fallbacks — and scales all other variables proportionally, so results read as changes in billions of pounds.</p>
-        <div class="meth-inline" aria-hidden="true">
+        <div class="meth-inline">
         <div class="meth-sec-title">Gouveia-Strauss tax function</div>
         <div class="meth-eq"><div class="meth-eq-label">effective rate</div><div class="meth-eq-body">\(\text{ETR}(y) = \phi_0\Bigl[1 - \bigl(1 + \phi_2\, y^{\phi_1}\bigr)^{-1/\phi_1}\Bigr]\)</div></div>
         <div class="meth-eq"><div class="meth-eq-label">marginal (analytical)</div><div class="meth-eq-body">\(\text{MTR}(y) = \phi_0\Bigl[1 - \bigl(1 + \phi_2\, y^{\phi_1}\bigr)^{-(1+\phi_1)/\phi_1}\Bigr]\)</div></div>
@@ -538,7 +538,13 @@ <h4>What the model does not capture</h4>
     });
     // Panels pair to steps by data-step. A step with no panel of its own
     // keeps the nearest earlier panel pinned rather than blanking the column.
-    var want = -1;
+    // Start at the FIRST panel, not -1: several pages have no panel for
+    // step 0, and seeding -1 left the sticky column empty for the whole of
+    // the opening step before popping in. Pinning the earliest panel means
+    // the column is populated from the top and only ever moves forward.
+    var want = Math.min.apply(null, panels.map(function (panel) {
+      return Number(panel.getAttribute("data-step"));
+    }));
     panels.forEach(function (panel) {
       var ds = Number(panel.getAttribute("data-step"));
       if (ds <= active && ds > want) want = ds;
diff --git a/olg/validation/index.html b/olg/validation/index.html
index c85bc0ce..8d72a9b7 100644
--- a/olg/validation/index.html
+++ b/olg/validation/index.html
@@ -307,7 +307,7 @@ <h2>No ground truth. Calibration targets instead.</h2>
         <caption>psl-og calibration targets against official UK aggregates</caption>
         <thead><tr><th scope="col">Quantity</th><th scope="col">Official UK value</th><th scope="col">Source</th><th scope="col">Model treatment</th><th scope="col">Deviation</th></tr></thead>
         <tbody>
-          <tr><th scope="row">Government debt-to-GDP</th><td>95.1% (May 2026); OBR March 2026 forecast 94.4%</td><td>ONS public sector finances</td><td class="vc-warn">imposed via closure target</td><td>&minus;0.1pp vs latest ONS (target 95.0%)</td></tr>
+          <tr><th scope="row">Government debt-to-GDP</th><td>94.9% (June 2026); OBR March 2026 forecast 94.4%</td><td>ONS &middot; HF6X, vintage 2026-07-26</td><td class="vc-warn">imposed via closure target</td><td>+0.1pp vs latest ONS (target 95.0%)</td></tr>
           <tr><th scope="row">Household saving ratio</th><td>8.9% (2026Q1, down from 9.6%)</td><td>ONS quarterly sector accounts</td><td class="vc-warn">targeted through β</td><td>0 by construction</td></tr>
           <tr><th scope="row">Potential growth</th><td>≈1.1% / yr</td><td>OBR EFO</td><td class="vc-warn">imposed, g<sub>y</sub> = 0.011</td><td>≈0</td></tr>
           <tr><th scope="row">Depreciation / capital</th><td>≈6&ndash;7% (CFC / net stock)</td><td>ONS capital stocks bulletin</td><td class="vc-warn">imposed, δ = 0.065</td><td>within range</td></tr>
diff --git a/pe/code/index.html b/pe/code/index.html
index 53b79091..36bc4f1a 100644
--- a/pe/code/index.html
+++ b/pe/code/index.html
@@ -120,7 +120,7 @@ <h4>CLI and Python</h4>
         <div class="meth-step-head"><span class="meth-step-num mono">2</span><h3>Install</h3></div>
         <h4>From PyPI</h4>
         <p>The base install is shared tooling only; add <code>[models]</code> for the pinned UK + US country models, or a single country extra.</p>
-        <div class="meth-inline" aria-hidden="true">
+        <div class="meth-inline">
           <div class="meth-sec-title">install</div>
           <div class="codeblock"><button class="cb-copy" data-copy>copy</button><pre><code>pip install policyengine            <span class="cm"># shared tooling, no country models</span>
 pip install "policyengine[models]"  <span class="cm"># + pinned UK and US models</span>
@@ -141,7 +141,7 @@ <h4>Household-level reform impact</h4>
           <strong>£1,872</strong> of income tax and cuts net income by the
           same amount.
         </p>
-        <div class="meth-inline" aria-hidden="true">
+        <div class="meth-inline">
           <div class="meth-sec-title">household calculation</div>
           <div class="codeblock"><button class="cb-copy" data-copy>copy</button><pre><code><span class="kw">import</span> policyengine <span class="kw">as</span> pe
 
@@ -177,7 +177,7 @@ <h4>Representative datasets</h4>
           population examples locally. The hosted MCP server provisions this
           credential server-side.
         </div>
-        <div class="meth-inline" aria-hidden="true">
+        <div class="meth-inline">
           <div class="meth-sec-title">population scoring</div>
           <div class="codeblock"><button class="cb-copy" data-copy>copy</button><pre><code><span class="kw">import</span> policyengine <span class="kw">as</span> pe
 <span class="kw">from</span> policyengine.core <span class="kw">import</span> Simulation
@@ -200,7 +200,7 @@ <h4>Representative datasets</h4>
         <div class="meth-panel" data-step="1" hidden>
           <div class="meth-panel-head"><span class="meth-panel-title">Install</span><span class="meth-panel-badge mono">Step 2</span></div>
           <div class="meth-sec-title">install</div>
-          <div class="codeblock"><button class="cb-copy" data-copy>copy</button><pre><code>pip install policyengine            <span class="cm"># shared tooling, no country models</span>
+          <div class="codeblock"><button class="cb-copy" data-copy tabindex="-1">copy</button><pre><code>pip install policyengine            <span class="cm"># shared tooling, no country models</span>
 pip install "policyengine[models]"  <span class="cm"># + pinned UK and US models</span>
 pip install "policyengine[uk]"      <span class="cm"># UK model only</span>
 pip install "policyengine[us]"      <span class="cm"># US model only</span></code></pre></div>
@@ -208,7 +208,7 @@ <h4>Representative datasets</h4>
         <div class="meth-panel" data-step="2" hidden>
           <div class="meth-panel-head"><span class="meth-panel-title">Calculate a household</span><span class="meth-panel-badge mono">Step 3</span></div>
           <div class="meth-sec-title">household calculation</div>
-          <div class="codeblock"><button class="cb-copy" data-copy>copy</button><pre><code><span class="kw">import</span> policyengine <span class="kw">as</span> pe
+          <div class="codeblock"><button class="cb-copy" data-copy tabindex="-1">copy</button><pre><code><span class="kw">import</span> policyengine <span class="kw">as</span> pe
 
 <span class="cm"># UK: single adult earning £50,000 — current law</span>
 uk = pe.uk.calculate_household(
@@ -232,7 +232,7 @@ <h4>Representative datasets</h4>
         <div class="meth-panel" data-step="3" hidden>
           <div class="meth-panel-head"><span class="meth-panel-title">Score a whole population</span><span class="meth-panel-badge mono">Step 4</span></div>
           <div class="meth-sec-title">population scoring</div>
-          <div class="codeblock"><button class="cb-copy" data-copy>copy</button><pre><code><span class="kw">import</span> policyengine <span class="kw">as</span> pe
+          <div class="codeblock"><button class="cb-copy" data-copy tabindex="-1">copy</button><pre><code><span class="kw">import</span> policyengine <span class="kw">as</span> pe
 <span class="kw">from</span> policyengine.core <span class="kw">import</span> Simulation
 <span class="kw">from</span> policyengine.outputs.aggregate <span class="kw">import</span> Aggregate, AggregateType
 
@@ -330,7 +330,13 @@ <h2>From one family to the whole distribution.</h2>
     });
     // Panels pair to steps by data-step. A step with no panel of its own
     // keeps the nearest earlier panel pinned rather than blanking the column.
-    var want = -1;
+    // Start at the FIRST panel, not -1: several pages have no panel for
+    // step 0, and seeding -1 left the sticky column empty for the whole of
+    // the opening step before popping in. Pinning the earliest panel means
+    // the column is populated from the top and only ever moves forward.
+    var want = Math.min.apply(null, panels.map(function (panel) {
+      return Number(panel.getAttribute("data-step"));
+    }));
     panels.forEach(function (panel) {
       var ds = Number(panel.getAttribute("data-step"));
       if (ds <= active && ds > want) want = ds;
diff --git a/pe/methodology/index.html b/pe/methodology/index.html
index 8da9888f..55e02a10 100644
--- a/pe/methodology/index.html
+++ b/pe/methodology/index.html
@@ -118,7 +118,7 @@ <h4>Variables and parameters</h4>
       <p>Every quantity — an allowance, a taper, a means test — is a variable with a formula and dated parameters. Asking for one variable resolves the graph beneath it, so a net-income figure comes from the statutory chain, not an approximating schedule.</p>
       <h4>Entities</h4>
       <p>Variables attach to <strong>entities</strong>, which differ by country — UK: three (person, benefit unit, household); US: six (person, tax unit, SPM unit, family, marital unit, household) — because US programmes are assessed on differently drawn units.</p>
-        <div class="meth-inline" aria-hidden="true">
+        <div class="meth-inline">
         <div class="meth-sec-title">rules, entities, dependency graph</div>
         <div class="meth-eq"><div class="meth-eq-label">variable</div><div class="meth-eq-body">formula + dated parameters</div></div>
         <div class="meth-eq"><div class="meth-eq-label">resolution</div><div class="meth-eq-body">request a variable &rarr; resolve the graph beneath it</div></div>
@@ -154,7 +154,7 @@ <h4>One reform format</h4>
       <p>A reform re-dates a parameter and re-resolves the same graph, so the household and population APIs take reforms in the identical format — a flat dict of parameter path &rarr; new value.</p>
       <h4>Version-pinned releases</h4>
       <p>Each release pins the country-model and dataset versions it certifies, so a published estimate traces to the exact bundle that produced it.</p>
-        <div class="meth-inline" aria-hidden="true">
+        <div class="meth-inline">
         <div class="meth-sec-title">reform mechanics</div>
         <div class="meth-eq"><div class="meth-eq-label">reform</div><div class="meth-eq-body">{parameter path &rarr; new value}</div></div>
         <div class="meth-eq"><div class="meth-eq-label">evaluation</div><div class="meth-eq-body">re-date parameter &rarr; re-resolve same graph</div></div>
@@ -191,7 +191,7 @@ <h4>Two modes, two data needs</h4>
       <p>Household calculations need no data and no credentials — the inputs are the ones you type. Population results run the same rule graph over a representative survey: for the UK the <strong>enhanced Family Resources Survey</strong> (currently the <code>enhanced_frs_2023_24</code> vintage), for the US the <strong>Current Population Survey</strong>.</p>
       <h4>Imputation and reweighting</h4>
       <p>The UK enhancement imputes under-reported incomes and reweights the survey so aggregates better match administrative totals; the resulting microdata is distributed via a gated Hugging Face repository (<code>policyengine/policyengine-uk-data</code>).</p>
-        <div class="meth-inline" aria-hidden="true">
+        <div class="meth-inline">
         <div class="meth-sec-title">data enhancement</div>
         <div class="meth-eq"><div class="meth-eq-label">UK dataset</div><div class="meth-eq-body">enhanced FRS &middot; <code>enhanced_frs_2023_24</code></div></div>
         <div class="meth-eq"><div class="meth-eq-label">US dataset</div><div class="meth-eq-body">Current Population Survey (CPS)</div></div>
@@ -297,7 +297,13 @@ <h4>Feeding the suite</h4>
     });
     // Panels pair to steps by data-step. A step with no panel of its own
     // keeps the nearest earlier panel pinned rather than blanking the column.
-    var want = -1;
+    // Start at the FIRST panel, not -1: several pages have no panel for
+    // step 0, and seeding -1 left the sticky column empty for the whole of
+    // the opening step before popping in. Pinning the earliest panel means
+    // the column is populated from the top and only ever moves forward.
+    var want = Math.min.apply(null, panels.map(function (panel) {
+      return Number(panel.getAttribute("data-step"));
+    }));
     panels.forEach(function (panel) {
       var ds = Number(panel.getAttribute("data-step"));
       if (ds <= active && ds > want) want = ds;
diff --git a/reports/define-uk-replication/index.html b/reports/define-uk-replication/index.html
index f5146797..cfa66f85 100644
--- a/reports/define-uk-replication/index.html
+++ b/reports/define-uk-replication/index.html
@@ -172,7 +172,7 @@ <h2>Every gate, its verdict, and the test that holds it.</h2>
           </tr>
           <tr>
             <th scope="row">Clean-room milestone 2 — §3.2&ndash;3.3 macro and production block</th>
-            <td><strong>IN PROGRESS</strong> — §3.2 landed; not passed (<a href="#reimplementation">section 05</a>)</td>
+            <td>IN PROGRESS — §3.2, §3.3.1 and §3.3.2 landed (118 eqs); §3.3.3 + oracle outstanding, so not passed<strong>IN PROGRESS</strong> — §3.2 landed; not passed (<a href="#reimplementation">section 05</a>)</td>
             <td>2026-08-11</td>
             <td><code>tests/test_macro.py</code></td>
           </tr>
@@ -386,18 +386,10 @@ <h2>Reimplementing the published equations, milestone by milestone.</h2>
       </p>
       <p>
         <strong>Milestone 2: IN PROGRESS, not passed</strong> (2026-08-11).
-        §3.2 has landed: <code>model/sectors/macro.py</code> implements
-        Eqs.&nbsp;(21)&ndash;(43) — all <strong>23 equations</strong> — and
-        the section solves as a system
-        (<code>tests/test_macro.py</code>). §3.3 and the oracle comparison
-        against the baseline GDP path remain outstanding, so the milestone is
-        <em>not</em> passed. Milestones 3&ndash;5 — sectoral equations,
-        ecosystem block, policy scenarios — are pending, each validated
-        against the oracle and the published figures under the same standard.
-      </p>
+        §3.2, §3.3.1 and §3.3.2 have landed: <code>model/sectors/macro.py</code>, <code>production.py</code> and <code>power.py</code> implement Eqs.&nbsp;(21)&ndash;(138) &mdash; <strong>118 equations, contiguous</strong> &mdash; each solving as a system under its own test module. §3.3.3 input&ndash;output and the oracle baseline comparison remain outstanding, so milestone 2 is <strong>not passed</strong>.</p>
       <p>
         Reimplementation is already earning its keep as a reading of the
-        manual that the R source cannot give: <strong>three manual defects
+        manual that the R source cannot give: <strong>some thirty-five manual defects
         are now pinned in tests</strong> rather than papered over or absorbed
         into a tolerance.
       </p>
diff --git a/reveal.js b/reveal.js
index 01347386..abe08ebb 100644
--- a/reveal.js
+++ b/reveal.js
@@ -51,3 +51,28 @@
   if (document.fonts && document.fonts.ready) document.fonts.ready.then(set);
   set();
 })();
+
+/* A fragment that targets a <details> element scrolls to it but does not open
+   it — browsers auto-expand only when the fragment targets a DESCENDANT. Three
+   high-traffic links point at /models#score, where the id sits on the <details>
+   itself, so the primary "score a reform" journey landed on a closed summary
+   with the content hidden. Open it on load and on subsequent hash changes. */
+(function () {
+  "use strict";
+  function openTargetedDetails() {
+    if (!location.hash || location.hash.length < 2) return;
+    var target;
+    try { target = document.querySelector(location.hash); } catch (e) { return; }
+    if (!target) return;
+    if (target.tagName === "DETAILS") target.open = true;
+    var parent = target.closest && target.closest("details");
+    if (parent) parent.open = true;
+    target.scrollIntoView({ block: "start" });
+  }
+  window.addEventListener("hashchange", openTargetedDetails);
+  if (document.readyState === "loading") {
+    document.addEventListener("DOMContentLoaded", openTargetedDetails);
+  } else {
+    openTargetedDetails();
+  }
+})();
diff --git a/svar/code/index.html b/svar/code/index.html
index 01bbc3e8..c3679f45 100644
--- a/svar/code/index.html
+++ b/svar/code/index.html
@@ -101,7 +101,7 @@ <h4>Where to read the results</h4>
         <div class="meth-step-head"><span class="meth-step-num mono">2</span><h3>CLI mirrors</h3></div>
         <h4>Same tools from the terminal</h4>
         <p>The CLI mirrors the three MCP tools one for one: <code>pe-macro forecast</code>, <code>pe-macro shocks</code>, and <code>pe-macro summary</code>. Same outputs, no MCP client required.</p>
-        <div class="meth-inline" aria-hidden="true">
+        <div class="meth-inline">
         <div class="meth-sec-title">CLI commands</div>
         <div class="codeblock"><button class="cb-copy" data-copy>copy</button><pre><code>pe-macro forecast   <span class="cm"># fan charts</span>
 pe-macro shocks     <span class="cm"># latest-quarter shocks</span>
@@ -115,7 +115,7 @@ <h4>One pip install</h4>
         <p>The package installs straight from GitHub with the prepared quarterly dataset inside, so <code>boe_var.data.load_data()</code> works immediately.</p>
         <h4>All inputs are public</h4>
         <p>Every input series is public: ONS (CPI, CPI energy, real GDP), Bank of England (Bank Rate, sterling ERI), and FRED (oil, exchange rates, world components).</p>
-        <div class="meth-inline" aria-hidden="true">
+        <div class="meth-inline">
         <div class="meth-sec-title">install</div>
         <div class="codeblock"><button class="cb-copy" data-copy>copy</button><pre><code>pip install git+https://github.com/PolicyEngine/boe-var-model</code></pre></div>
         </div>
@@ -127,7 +127,7 @@ <h4>Clone to rebuild everything</h4>
         <p>The runner scripts are for development — clone the repository to rebuild the dataset or regenerate the paper outputs.</p>
         <h4>What lands where</h4>
         <p class="opt-note">outputs land in <code>results/</code> — the replication figures (fig2–fig6), the forecast figures (fig1, fig7–fig9), and two summary reports. runtime scales with draw count; the importance weights are the expensive part — one numerical Jacobian per accepted draw.</p>
-        <div class="meth-inline" aria-hidden="true">
+        <div class="meth-inline">
         <div class="meth-sec-title">runner scripts</div>
         <div class="codeblock"><button class="cb-copy" data-copy>copy</button><pre><code>python scripts/download_data.py     <span class="cm"># reassemble the public data</span>
 
@@ -144,19 +144,19 @@ <h4>What lands where</h4>
         <div class="meth-panel" data-step="1" hidden>
           <div class="meth-panel-head"><span class="meth-panel-title">CLI mirrors</span><span class="meth-panel-badge mono">Code 2</span></div>
           <div class="meth-sec-title">CLI commands</div>
-        <div class="codeblock"><button class="cb-copy" data-copy>copy</button><pre><code>pe-macro forecast   <span class="cm"># fan charts</span>
+        <div class="codeblock"><button class="cb-copy" data-copy tabindex="-1">copy</button><pre><code>pe-macro forecast   <span class="cm"># fan charts</span>
 pe-macro shocks     <span class="cm"># latest-quarter shocks</span>
 pe-macro summary    <span class="cm"># replication headlines</span></code></pre></div>
         </div>
         <div class="meth-panel" data-step="2" hidden>
           <div class="meth-panel-head"><span class="meth-panel-title">Install locally</span><span class="meth-panel-badge mono">Code 3</span></div>
           <div class="meth-sec-title">install</div>
-        <div class="codeblock"><button class="cb-copy" data-copy>copy</button><pre><code>pip install git+https://github.com/PolicyEngine/boe-var-model</code></pre></div>
+        <div class="codeblock"><button class="cb-copy" data-copy tabindex="-1">copy</button><pre><code>pip install git+https://github.com/PolicyEngine/boe-var-model</code></pre></div>
         </div>
         <div class="meth-panel" data-step="3" hidden>
           <div class="meth-panel-head"><span class="meth-panel-title">Development runner scripts</span><span class="meth-panel-badge mono">Code 4</span></div>
           <div class="meth-sec-title">runner scripts</div>
-        <div class="codeblock"><button class="cb-copy" data-copy>copy</button><pre><code>python scripts/download_data.py     <span class="cm"># reassemble the public data</span>
+        <div class="codeblock"><button class="cb-copy" data-copy tabindex="-1">copy</button><pre><code>python scripts/download_data.py     <span class="cm"># reassemble the public data</span>
 
 <span class="cm"># the paper's core outputs: IRFs, FEVDs, shocks, historical decompositions</span>
 python scripts/run_replication.py
@@ -213,7 +213,13 @@ <h4>What lands where</h4>
     });
     // Panels pair to steps by data-step. A step with no panel of its own
     // keeps the nearest earlier panel pinned rather than blanking the column.
-    var want = -1;
+    // Start at the FIRST panel, not -1: several pages have no panel for
+    // step 0, and seeding -1 left the sticky column empty for the whole of
+    // the opening step before popping in. Pinning the earliest panel means
+    // the column is populated from the top and only ever moves forward.
+    var want = Math.min.apply(null, panels.map(function (panel) {
+      return Number(panel.getAttribute("data-step"));
+    }));
     panels.forEach(function (panel) {
       var ds = Number(panel.getAttribute("data-step"));
       if (ds <= active && ds > want) want = ds;
diff --git a/svar/index.html b/svar/index.html
index 2ca6d903..2680c1d7 100644
--- a/svar/index.html
+++ b/svar/index.html
@@ -81,7 +81,7 @@ <h2>Its forecast prices CPI-linked benefits through the microsimulation.</h2>
       <div class="qa-facts">
         <div class="qa-fact"><span class="icon">&rarr;</span><span><strong>Connection.</strong> Forecast-to-incidence bridge: <code>svar_inflation_incidence</code> passes the gap between this model's CPI path and the official one through the April 2028 uprating of CPI-linked benefits in <a href="/pe">PolicyEngine</a>. Experimental, and labelled as such. No reform scoring — <code>score_reform</code> does not accept this model.</span></div>
         <div class="qa-fact"><span class="icon">&rarr;</span><span><strong>Worked example.</strong> 2027 median CPI runs 0.94pp above the March 2026 EFO; passed through Universal Credit and Child Benefit uprating, that costs &pound;0.09bn in 2028, with the largest relative gains in the second-to-fifth income deciles.</span></div>
-        <div class="qa-fact"><span class="icon">&rarr;</span><span><strong>Validation.</strong> <a href="/svar/validation">Replicates Brignone &amp; Piffer (2025)</a>, CI-gated; 49-origin rolling evaluation vs ONS outturns; public scored <a href="/forecasts">forecast record</a>.</span></div>
+        <div class="qa-fact"><span class="icon">&rarr;</span><span><strong>Validation.</strong> <a href="/svar/validation">Replicates Brignone &amp; Piffer (2025)</a>, CI-gated. But the 49-origin rolling evaluation is not a credential: against a random walk <em>with drift</em> it found UK GDP <strong>not statistically distinguishable from naive</strong> (p=0.38&ndash;0.67), and Bank Rate the only defensible forecasting claim. <a href="/svar/validation#forecasters">See the evaluation</a>; the scored record is on <a href="/forecasts">forecasts</a>.</span></div>
         <div class="qa-fact"><span class="icon">&rarr;</span><span><strong>Results.</strong> The live UK fan chart on the <a href="/">home page</a> and every archived round on <a href="/forecasts">forecasts</a>.</span></div>
       </div>
     </div>
diff --git a/svar/methodology/index.html b/svar/methodology/index.html
index 3e7aab18..7d629a50 100644
--- a/svar/methodology/index.html
+++ b/svar/methodology/index.html
@@ -134,7 +134,7 @@ <h4>Sample and specification</h4>
       <p>Estimation covers the inflation-targeting era, <strong>1992Q1–2025Q1</strong>, in both the paper replication and the hosted adapter — in levels, with \(p = 4\) lags and exogenous Covid dummies for 2020Q1–2021Q2.</p>
         <h4>Conditioning versus estimation</h4>
       <p>The hosted forecast conditions on observations through <strong>2026Q1</strong> without re-estimating coefficients on them.</p>
-        <div class="meth-inline" aria-hidden="true">
+        <div class="meth-inline">
         <div class="meth-sec-title">the eight variables</div>
         <div class="meth-eq"><div class="meth-eq-label">global block</div><div class="meth-eq-body">World GDP · World CPI (both UK-trade-weighted) · real oil price in sterling</div></div>
         <div class="meth-eq"><div class="meth-eq-label">UK block</div><div class="meth-eq-body">Bank Rate · sterling exchange-rate index · CPI · CPI energy · real GDP</div></div>
@@ -175,7 +175,7 @@ <h4>The prior</h4>
       <p>Estimation is Bayesian with a <strong>Minnesota (normal-inverse-Wishart) prior</strong> plus sum-of-coefficients and dummy-initial-observation priors — the framework of <a href="https://doi.org/10.1162/REST_a_00483">Giannone, Lenza &amp; Primiceri (2015)</a>.</p>
         <h4>The posterior</h4>
       <p>The output is a posterior over the reduced-form coefficients \(\Pi\) and error covariance \(\Sigma\); every downstream object — impulse responses, decompositions, fan charts — is a posterior distribution, not a point estimate, and every one is reported with credible bands.</p>
-        <div class="meth-inline" aria-hidden="true">
+        <div class="meth-inline">
         <div class="meth-sec-title">priors</div>
         <div class="meth-eq"><div class="meth-eq-label">reduced form</div><div class="meth-eq-body">\(y_t = c + \Pi_1 y_{t-1} + \cdots + \Pi_p y_{t-p} + u_t, \quad p = 4\)</div></div>
         <div class="meth-eq"><div class="meth-eq-label">main prior</div><div class="meth-eq-body">Minnesota (normal-inverse-Wishart) on \((\Pi, \Sigma)\)</div></div>
@@ -218,7 +218,7 @@ <h4>Zero and sign restrictions</h4>
       <p>Identification follows <strong>Arias–Rubio-Ramírez–Waggoner (2018)</strong>: zero restrictions make the UK a small open economy (UK shocks cannot move world variables on impact), and impact sign restrictions name six structural shocks — <strong>world demand, world energy, world supply, UK demand, UK supply, and UK monetary policy</strong>. Two unidentified shocks absorb residual volatility.</p>
         <h4>Corrections</h4>
       <p>Importance weights correct the sampler's non-uniformity; the <strong>Chan–Matthes–Yu (2025)</strong> permutation search cuts rejections.</p>
-        <div class="meth-inline" aria-hidden="true">
+        <div class="meth-inline">
         <div class="meth-sec-title">zero and sign restrictions</div>
         <div class="meth-eq"><div class="meth-eq-label">zero restrictions</div><div class="meth-eq-body">small open economy: UK shocks cannot move world variables on impact</div></div>
         <div class="meth-eq"><div class="meth-eq-label">sign restrictions</div><div class="meth-eq-body">impact signs name the shocks (Arias et al. 2018)</div></div>
@@ -250,7 +250,7 @@ <h4>Accept/reject sampling</h4>
       <p>This is accept/reject sampling. The production run drew <strong>10,000</strong> posterior \((\Pi, \Sigma)\) draws, of which <strong>751 passed</strong> the sign restrictions (7.5%), for an importance-weighted effective sample size of <strong>355.9</strong> — so medians and bands are correct but noisier than the paper's.</p>
         <h4>The hosted run</h4>
       <p>The hosted forecast conditions on the current vintage — data through 2026Q1 — with a <strong>2,000-draw</strong> default (135 accepted, ESS 65.3), while retaining the 2025Q1 coefficient-estimation endpoint. The published current-outlook chart on the <a href="/svar">overview</a> comes from a larger 5,600-draw run of the same pipeline (385 accepted, ESS 176.3).</p>
-        <div class="meth-inline" aria-hidden="true">
+        <div class="meth-inline">
         <div class="meth-sec-title">accept/reject sampling</div>
         <div class="meth-eq"><div class="meth-eq-label">production run</div><div class="meth-eq-body">10,000 draws &rarr; 751 accepted (7.5%)</div></div>
         <div class="meth-eq"><div class="meth-eq-label">effective sample size</div><div class="meth-eq-body">355.9 (importance-weighted)</div></div>
@@ -289,7 +289,7 @@ <h4>How the bands work</h4>
           capped at four quarters, where the satellite stops beating a
           no-change benchmark.
         </p>
-        <div class="meth-inline" aria-hidden="true">
+        <div class="meth-inline">
           <div class="meth-sec-title">the regression</div>
           <div class="meth-eq"><div class="meth-eq-label">the mapping (dynamic)</div><div class="meth-eq-body">\(\Delta u_t = \alpha + \beta\, g_t + \gamma\, \Delta u_{t-1} + \varepsilon_t\)</div></div>
         </div>
diff --git a/us-hank/code/index.html b/us-hank/code/index.html
index 4c579313..2eee6064 100644
--- a/us-hank/code/index.html
+++ b/us-hank/code/index.html
@@ -102,7 +102,7 @@ <h4>Cold vs warm</h4>
         <div class="meth-step-head"><span class="meth-step-num mono">2</span><h3>Or use the CLI</h3></div>
         <h4>Same two calls</h4>
         <p>The same two operations on the command line: <code>hank-summary</code> to see what the model accepts, <code>hank-shock</code> to run one — here a 25bp monetary easing with the distributional block on.</p>
-        <div class="meth-inline" aria-hidden="true">
+        <div class="meth-inline">
         <div class="meth-sec-title">pe-macro CLI</div>
         <div class="codeblock"><button class="cb-copy" data-copy>copy</button><pre><code>pe-macro hank-summary
 pe-macro hank-shock --kind monetary --size -0.0025 --persistence 0.6 --distribution</code></pre></div>
@@ -123,7 +123,7 @@ <h4>The Python API</h4>
         <p>Beyond the hosted surface, install the package and use the Python API directly: solve the steady state once, solve the Jacobian once, then draw impulse responses and distributional summaries instantly.</p>
         <h4>What sits behind it</h4>
         <p>The solve behind these calls — steady state, Jacobian, then instant IRFs — is on the <a href="/us-hank/methodology">methodology page</a>; what to trust the results for, on the <a href="/us-hank/validation">validation page</a>.</p>
-        <div class="meth-inline" aria-hidden="true">
+        <div class="meth-inline">
         <div class="meth-sec-title">install</div>
         <div class="codeblock"><button class="cb-copy" data-copy>copy</button><pre><code>pip install git+https://github.com/PolicyEngine/us-hank-model</code></pre></div>
         <div class="meth-sec-title">python api</div>
@@ -142,15 +142,15 @@ <h4>What sits behind it</h4>
         <div class="meth-panel" data-step="1" hidden>
           <div class="meth-panel-head"><span class="meth-panel-title">Or use the CLI</span><span class="meth-panel-badge mono">Step 2</span></div>
           <div class="meth-sec-title">pe-macro CLI</div>
-          <div class="codeblock"><button class="cb-copy" data-copy>copy</button><pre><code>pe-macro hank-summary
+          <div class="codeblock"><button class="cb-copy" data-copy tabindex="-1">copy</button><pre><code>pe-macro hank-summary
 pe-macro hank-shock --kind monetary --size -0.0025 --persistence 0.6 --distribution</code></pre></div>
         </div>
         <div class="meth-panel" data-step="3" hidden>
           <div class="meth-panel-head"><span class="meth-panel-title">Install and script it</span><span class="meth-panel-badge mono">Step 4</span></div>
           <div class="meth-sec-title">install</div>
-          <div class="codeblock"><button class="cb-copy" data-copy>copy</button><pre><code>pip install git+https://github.com/PolicyEngine/us-hank-model</code></pre></div>
+          <div class="codeblock"><button class="cb-copy" data-copy tabindex="-1">copy</button><pre><code>pip install git+https://github.com/PolicyEngine/us-hank-model</code></pre></div>
           <div class="meth-sec-title">python api</div>
-          <div class="codeblock"><button class="cb-copy" data-copy>copy</button><pre><code><span class="kw">from</span> us_hank <span class="kw">import</span> model, distributional
+          <div class="codeblock"><button class="cb-copy" data-copy tabindex="-1">copy</button><pre><code><span class="kw">from</span> us_hank <span class="kw">import</span> model, distributional
 
 ss = model.solve_steady_state()          <span class="cm"># ~13s, cached</span>
 G = model.solve_jacobian(ss, T=<span class="kw">300</span>)      <span class="cm"># ~5s, cached</span>
@@ -197,7 +197,13 @@ <h4>What sits behind it</h4>
     });
     // Panels pair to steps by data-step. A step with no panel of its own
     // keeps the nearest earlier panel pinned rather than blanking the column.
-    var want = -1;
+    // Start at the FIRST panel, not -1: several pages have no panel for
+    // step 0, and seeding -1 left the sticky column empty for the whole of
+    // the opening step before popping in. Pinning the earliest panel means
+    // the column is populated from the top and only ever moves forward.
+    var want = Math.min.apply(null, panels.map(function (panel) {
+      return Number(panel.getAttribute("data-step"));
+    }));
     panels.forEach(function (panel) {
       var ds = Number(panel.getAttribute("data-step"));
       if (ds <= active && ds > want) want = ds;
diff --git a/us-hank/methodology/index.html b/us-hank/methodology/index.html
index feb918da..f4d151cc 100644
--- a/us-hank/methodology/index.html
+++ b/us-hank/methodology/index.html
@@ -133,7 +133,7 @@ <h4>Households differ</h4>
       <p>This is a validated replication of the two-asset heterogeneous-agent New Keynesian (HANK) model of Auclert, Bardóczy, Rognlie and Straub (<em>Econometrica</em> 2021, "Using the Sequence-Space Jacobian to Solve and Estimate Heterogeneous-Agent Models"). Households differ in <strong>liquid and illiquid wealth</strong>, so their marginal propensities to consume differ, and the model carries that whole distribution through the general-equilibrium response — the thing a representative-agent model cannot say.</p>
       <h4>Production grids, not demo grids</h4>
       <p>The model runs at the paper's <strong>production grid sizes</strong> — 50 liquid × 70 illiquid × 50 capital points — rather than the packaged demo grids. It is a <strong>VAR-free, sequence-space HANK</strong>: no estimated reduced form, no expectations shortcut. A <code>variant</code> option selects the paper's <code>two_asset</code> model (the default) or a fast <code>one_asset</code> textbook variant — no capital, so no investment response, and monetary/productivity shocks only.</p>
-        <div class="meth-inline" aria-hidden="true">
+        <div class="meth-inline">
         <div class="meth-sec-title">household structure</div>
         <div class="meth-eq"><div class="meth-eq-label">assets</div><div class="meth-eq-body">liquid + illiquid, per household</div></div>
         <div class="meth-eq"><div class="meth-eq-label">grids</div><div class="meth-eq-body">50 liquid × 70 illiquid × 50 capital points</div></div>
@@ -171,7 +171,7 @@ <h4>Hit the paper's targets</h4>
       <p>The first stage solves the general-equilibrium steady state at the production grids, calibrating <strong>discounting and portfolio-adjustment costs</strong> to hit the paper's asset-market targets. The solve takes about <strong>13 seconds</strong>, is computed once, and is cached.</p>
       <h4>Checked value by value</h4>
       <p>The hosted steady state is set against the paper's published calibration targets, value by value, on the <a href="/us-hank/validation#targets">validation page</a>. The parameterisation is the published paper's throughout — there is no data vintage to refresh.</p>
-        <div class="meth-inline" aria-hidden="true">
+        <div class="meth-inline">
         <div class="meth-sec-title">steady-state solve</div>
         <div class="meth-eq"><div class="meth-eq-label">calibrated</div><div class="meth-eq-body">discounting + portfolio-adjustment costs</div></div>
         <div class="meth-eq"><div class="meth-eq-label">targets</div><div class="meth-eq-body">the paper's asset-market targets</div></div>
@@ -224,7 +224,7 @@ <h4>One linear map</h4>
       <p>The second stage builds the general-equilibrium <strong>sequence-space Jacobian</strong> — the \(T \times T\) linear map from each exogenous path to each outcome path, with <strong>\(T = 300\) quarters</strong>. It takes about <strong>5 seconds</strong> and is cached alongside the steady state.</p>
       <h4>Instant IRFs</h4>
       <p>With both caches warm, an impulse response is a <strong>matrix–vector product</strong> — effectively instant. The first call per variant pays the steady-state and Jacobian solves, roughly 18 seconds cold; every later call reuses them.</p>
-        <div class="meth-inline" aria-hidden="true">
+        <div class="meth-inline">
         <div class="meth-sec-title">Jacobian</div>
         <div class="meth-eq"><div class="meth-eq-label">object</div><div class="meth-eq-body">\(T \times T\) map: exogenous path &rarr; outcome path</div></div>
         <div class="meth-eq"><div class="meth-eq-label">horizon</div><div class="meth-eq-body">\(T = 300\) quarters</div></div>
@@ -266,7 +266,7 @@ <h4>AR(1) paths</h4>
       <p>A shock is an AR(1) path, \(\varepsilon_t = \text{size} \times \rho^t\), on one of three instruments: the Taylor-rule intercept (<code>monetary</code>), real government spending (<code>fiscal_spending</code>), or TFP (<code>productivity</code>).</p>
       <h4>The design point that matters</h4>
       <p><strong>The labor tax is endogenous.</strong> The fiscal block balances the government budget each period, so a spending shock is implicitly tax-financed — and there is <em>no</em> exogenous transfer or tax-rate instrument to shock. Requests for one are refused with that explanation rather than answered with an invented mapping; <code>score_reform</code> refuses <code>model="hank"</code> outright.</p>
-        <div class="meth-inline" aria-hidden="true">
+        <div class="meth-inline">
         <div class="meth-sec-title">shocks</div>
         <div class="meth-eq"><div class="meth-eq-label">path</div><div class="meth-eq-body">\(\varepsilon_t = \text{size} \times \rho^t\) (AR(1))</div></div>
         <div class="meth-eq"><div class="meth-eq-label">instruments</div><div class="meth-eq-body"><code>monetary</code> · <code>fiscal_spending</code> · <code>productivity</code></div></div>
@@ -385,7 +385,13 @@ <h4>What "first-order" means here</h4>
     });
     // Panels pair to steps by data-step. A step with no panel of its own
     // keeps the nearest earlier panel pinned rather than blanking the column.
-    var want = -1;
+    // Start at the FIRST panel, not -1: several pages have no panel for
+    // step 0, and seeding -1 left the sticky column empty for the whole of
+    // the opening step before popping in. Pinning the earliest panel means
+    // the column is populated from the top and only ever moves forward.
+    var want = Math.min.apply(null, panels.map(function (panel) {
+      return Number(panel.getAttribute("data-step"));
+    }));
     panels.forEach(function (panel) {
       var ds = Number(panel.getAttribute("data-step"));
       if (ds <= active && ds > want) want = ds;