Leaderboard: 95% CIs from the report's per-sample paired bootstrap (whiskers on bars, scatters and overview; CIs and paired deltas in tooltips)

README.md

@@ -103,6 +103,11 @@ causal-variant cohort and the tie-aware recall, so the Space and the article can
 diverge. The generator prints its checks (the S_bal of the four sizes, the cohort size
 and the headline recalls) so a mismatch with the manuscript is visible at build time.
 
+`data/leaderboard.json` also carries 95% confidence intervals (`s_bal_ci`, `families_ci`,
+`metrics[].ci`, `vs_plantcad2l`) from the report's per-sample paired bootstrap (20,000
+replicates, `figures/data/fig3_rerun/rerun_ci.json` in the report repository), added by
+`scripts/add_bootstrap_ci.py` and centred on the reported scores as in the report.
+
 `data/abi5-locus.json` is a separate extract of the original Botanic1-XL 4,096 bp
 scoring output for ABI5: 5,948 SNPs in a 100 kbp locus. Scores are the absolute
 values of the source `score` column. The validated SNP at TAIR10 Chr2:15,205,931

assets/figure1.js

@@ -22,7 +22,7 @@
   const M = lb.models, F = lb.families;
   const colour = (m) => m.ours ? css('--ours') : m.id.startsWith('Botanic0') ? css('--prior') : css('--other');
   const famTable = (m) => '<table>' + F.map(f => `<tr><td>${f.name}</td><td class="n">${fmt(m.families[f.key])}</td></tr>`).join('') + '</table>';
-  const modelTip = (m) => `<b>${esc(m.name)}</b> · ${esc(m.org)}<br>${m.params_str} parameters · ${m.objective}, ${m.domain} pre-training${m.bp_equivalent ? ' · ' + esc(m.bp_equivalent) + ' bp seen' : ''}<br>downstream task performance <b>${fmt(m.s_bal, 4)}</b>`;
+  const modelTip = (m) => `<b>${esc(m.name)}</b> · ${esc(m.org)}<br>${m.params_str} parameters · ${m.objective}, ${m.domain} pre-training${m.bp_equivalent ? ' · ' + esc(m.bp_equivalent) + ' bp seen' : ''}<br>downstream task performance <b>${fmt(m.s_bal, 4)}</b>${m.s_bal_ci ? ` <span class="ci">95% CI [${fmt(m.s_bal_ci[0], 4)}, ${fmt(m.s_bal_ci[1], 4)}]</span>` : ''}`;
 
   // ------------------------------------------------------------------ intro card
   const intro = document.getElementById('intro-chart');
@@ -40,6 +40,7 @@
     const left = new Set(['Evo2-40b', 'CARBON-8B', 'botanic1-S']);
     const dy = { 'botanic1-S': -20, 'botanic1-XL': -12, 'PlantCAD2-L': 18, 'PlantCaduceus': 14, GPN: -10 };
     const pt = svg.selectAll('g.pt').data(rows).join('g').attr('class', 'pt').attr('transform', d => `translate(${x(Math.min(Math.max(d.params, 5e7), 5e10))},${y(Math.max(d.s_bal, 0.55))})`);
+    pt.filter(d => d.s_bal_ci).append('line').attr('class', 'whisker').attr('x1', 0).attr('x2', 0).attr('y1', d => y(Math.max(d.s_bal_ci[0], 0.55)) - y(Math.max(d.s_bal, 0.55))).attr('y2', d => y(Math.max(d.s_bal_ci[1], 0.55)) - y(Math.max(d.s_bal, 0.55)));
     pt.append('circle').attr('r', d => d.ours ? 14 : 11).attr('fill', '#fff').attr('stroke', d => d.ours ? css('--ours') : css('--rule-2')).attr('stroke-width', d => d.ours ? 2.5 : 1);
     pt.append('image').attr('href', d => 'assets/logos/' + d.logo).attr('x', d => d.ours ? -9 : -7).attr('y', d => d.ours ? -9 : -7).attr('width', d => d.ours ? 18 : 14).attr('height', d => d.ours ? 18 : 14).attr('preserveAspectRatio', 'xMidYMid meet');
     pt.filter(d => labelled.has(d.id)).append('text').attr('class', 'lbl-2').attr('x', d => left.has(d.id) ? -15 : (d.ours ? 18 : 15)).attr('text-anchor', d => left.has(d.id) ? 'end' : 'start').attr('y', d => 4 + (dy[d.id] || 0)).text(d => d.name);
@@ -193,7 +194,7 @@
   }
 
   // ------------------------------------------------------------------ (d) ranking
-  const pd = panel('d', 'Model ranking', 'downstream task performance · hover a bar · click for the full leaderboard');
+  const pd = panel('d', 'Model ranking', 'downstream task performance · whiskers are 95% CIs · hover a bar · click for the full leaderboard');
   {
     const rows = M.filter(m => m.in_fig1).sort((a, b) => b.s_bal - a.s_bal);
     const W = 1100, H = 330, L = 44, R = 12, T = 56, B = 92;
@@ -207,6 +208,7 @@
     const bw = x.bandwidth();
     g.append('rect').attr('class', 'bar').attr('x', d => x(d.id)).attr('width', bw).attr('y', y(0.5)).attr('height', 0).attr('rx', 2)
       .attr('fill', d => colour(d)).attr('stroke', d => d.ours ? 'none' : css('--rule-2')).attr('stroke-width', 1);
+    g.filter(d => d.s_bal_ci).append('line').attr('class', 'whisker').attr('x1', d => x(d.id) + bw / 2).attr('x2', d => x(d.id) + bw / 2).attr('y1', d => y(d.s_bal_ci[0])).attr('y2', d => y(d.s_bal_ci[1]));
     const chip = g.append('g').attr('transform', d => `translate(${x(d.id) + bw / 2},${y(d.s_bal) - 16})`);
     chip.append('circle').attr('class', 'chip').attr('r', 11);
     chip.append('image').attr('href', d => 'assets/logos/' + d.logo).attr('x', -7).attr('y', -7).attr('width', 14).attr('height', 14).attr('preserveAspectRatio', 'xMidYMid meet');

assets/overview.js

@@ -73,7 +73,7 @@
   }
 
   function benchmark(host, data) {
-    const f = frame(host, '<span>Selected references</span><span class="plot-unit">S<sub>bal</sub><sup>test</sup> · axis starts at 0.5</span>');
+    const f = frame(host, '<span>Selected references</span><span class="plot-unit">S<sub>bal</sub><sup>test</sup> · axis starts at 0.5 · whiskers 95% CI</span>');
     const references = new Set(['PlantCAD2-L', 'GPN', 'CARBON-8B', 'NTv3-pre', 'Evo2-7b', 'AgroNT']);
     let selected = 'botanic1-XL';
     responsive(host, W => {
@@ -94,10 +94,11 @@
       marks.append('rect').attr('class', 'benchmark-bar').attr('x', L).attr('y', -6).attr('width', d => x(d.s_bal) - L).attr('height', 12).attr('rx', 2)
         .attr('fill', d => d.ours ? 'var(--lime)' : 'var(--plot-reference)')
         .attr('stroke', d => d.ours ? 'var(--accent)' : 'var(--rule-2)').attr('stroke-width', .6);
-      marks.append('text').attr('class', 'plot-value').attr('x', d => x(d.s_bal) + 5).attr('y', 4).text(d => d.s_bal.toFixed(4));
+      marks.filter(d => d.s_bal_ci).append('line').attr('class', 'benchmark-ci').attr('x1', d => x(d.s_bal_ci[0])).attr('x2', d => x(d.s_bal_ci[1])).attr('y1', 0).attr('y2', 0);
+      marks.append('text').attr('class', 'plot-value').attr('x', d => x(d.s_bal_ci ? d.s_bal_ci[1] : d.s_bal) + 5).attr('y', 4).text(d => d.s_bal.toFixed(4));
       const select = inspect(marks, d => `${d.name}, ${d.params_str} parameters, aggregate test score ${d.s_bal.toFixed(4)}`, d => {
         selected = d.id;
-        f.reading.innerHTML = `<b>${esc(d.name)}</b> · ${esc(d.params_str)} parameters · score ${d.s_bal.toFixed(4)}`;
+        f.reading.innerHTML = `<b>${esc(d.name)}</b> · ${esc(d.params_str)} parameters · score ${d.s_bal.toFixed(4)}${d.s_bal_ci ? ` · 95% CI [${d.s_bal_ci[0].toFixed(4)}, ${d.s_bal_ci[1].toFixed(4)}]` : ''}`;
       });
       select(rows.find(d => d.id === selected) || rows[0]);
     });

assets/site.css

@@ -330,6 +330,11 @@ input[type="search"] { min-width: 250px; }
 .scatter .pt.is-family .ptlabel { fill: var(--ink); }
 .scatter .pt { outline: none; }
 .scatter .point-dot { pointer-events: none; }
+.chart .whisker line { stroke: var(--ink); stroke-width: 1.4; pointer-events: none; }
+.scatter .errbar { stroke-width: 1.6; opacity: .9; pointer-events: none; }
+.scatter .pt.is-muted .errbar { opacity: .25; }
+.tip .ci, .tip td.ci { color: var(--ink-3); font-size: .92em; }
+.fig1 .whisker, .benchmark-ci { stroke: var(--ink); stroke-width: 1.4; pointer-events: none; }
 .scatter .point-halo { fill: none; stroke: var(--ink); stroke-width: 1.5; opacity: 0; pointer-events: none; }
 .scatter .pt:hover .point-halo, .scatter .pt:focus-visible .point-halo { opacity: 1; }
 .chart .ann { fill: var(--ink-3); font-size: 11.5px; }

index.html

@@ -51,7 +51,7 @@
         <h2>318M to 3.2B parameters</h2>
         <p>Botanic1-S scores 0.758 on the aggregate benchmark across nine task families. PlantCAD2-L scores 0.756, Carbon-8B 0.727, and Evo&nbsp;2-7B 0.695.</p>
         <figure class="overview-plot" id="overview-benchmark" aria-label="Model benchmark comparison" hidden></figure>
-        <p class="fact-note">Frozen models · aggregate test score (S<sub>bal</sub><sup>test</sup>)</p>
+        <p class="fact-note">Frozen models · aggregate test score (S<sub>bal</sub><sup>test</sup>) · whiskers are 95% CIs</p>
         <a class="text-link" href="leaderboard.html">Scores by task and species <span aria-hidden="true">↗</span></a>
       </article>
       <article class="fact">

leaderboard.html

@@ -43,7 +43,7 @@
 </div>
 
 <div class="chart" id="bars"></div>
-<p class="chartnote">Hover a bar for the nine family means. Botanic0-L is the previous generation, shown in olive. Parameter counts are trainable parameters as declared by each model's configuration and code; PlantBiMoE has 116M trainable parameters and 64M active per token.</p>
+<p class="chartnote">Hover a bar for the nine family means. Whiskers are 95% confidence intervals from the per-sample paired bootstrap of the technical report (20,000 replicates), centred on the reported score. Botanic0-L is the previous generation, shown in olive. Parameter counts are trainable parameters as declared by each model's configuration and code; PlantBiMoE has 116M trainable parameters and 64M active per token.</p>
 
 <h2 id="families">The nine task families</h2>
 <p>Zooming in per task shows that Botanic1 leads overall without dominating every capability: the flagship Botanic1-XL tops the benchmark and two task families, genomic-region classification (0.527 versus 0.525 for PlantCAD2-L) and causal-variant discovery (recall AUC 0.752), Botanic1-L leads three more (conservation, translation initiation and PRO-seq) and Botanic1-M leads splicing (donor and acceptor mean 0.984 versus 0.978 for NTv3-650M-post, which uses test annotations during post-training). The autoregressive Evo 2 leads variant-effect prediction, where its 7B model reaches an LLR AUROC of 0.724 against 0.721 for Botanic1-XL (its 40B model reaches 0.719); PlantCAD2-L leads on chromatin accessibility (0.483 versus 0.481 for Botanic1-L); and GPN leads translation termination (0.888 versus 0.887). Botanic1-XL is the only model that stays competitive across all nine families.</p>
@@ -55,7 +55,7 @@
       <button data-v="metrics" aria-pressed="false">22 metrics</button>
     </div>
   </div>
-  <span class="small">Click a column header to sort by it. Shading is relative within each column.</span>
+  <span class="small">Click a column header to sort by it. Shading is relative within each column. Hover a cell for its 95% confidence interval.</span>
 </div>
 <div class="tablewrap"><table class="heat" id="heat"></table></div>
 
@@ -65,7 +65,7 @@
   <div class="chart scatter" id="scatter-params"></div>
   <div class="chart scatter" id="scatter-tokens"></div>
 </div>
-<p class="chartnote">Tokens are converted to base pairs using approximations for 6-mer models. Models without a documented token count are omitted from the right panel.</p>
+<p class="chartnote">Vertical bars are 95% confidence intervals from the per-sample bootstrap. Tokens are converted to base pairs using approximations for 6-mer models. Models without a documented token count are omitted from the right panel.</p>
 
 <h2 id="uncertainty">Are leaderboard differences significant?</h2>
 <p>We assess leaderboard differences with two paired bootstrap tests, each using 20,000 replicates: one resampling the nine task families, and one resampling test examples within each benchmark cell. Together, they test whether observed margins are robust to variation across task families and to sampling noise in the benchmark. The technical report (v2) reports both tests: the table below reproduces its Supplementary Table on <span class="math">S<sub>bal</sub></span> uncertainty, and the per-sample intervals are drawn as error bars in Figure 1d and as paired differences to PlantCAD2-L in Figure 1e.</p>
@@ -156,10 +156,18 @@
     if (m.families[key] != null) return m.families[key];
     const mt = m.metrics.find(x => x.label === key); return mt ? mt.value : null;
   }
+  function ci(m, key) {
+    if (key === 's_bal') return m.s_bal_ci || null;
+    if (m.families_ci && m.families_ci[key]) return m.families_ci[key];
+    const mt = m.metrics.find(x => x.label === key); return mt && mt.ci ? mt.ci : null;
+  }
+  const fmtCI = (iv, d = 3) => iv ? `[${fmt(iv[0], d)}, ${fmt(iv[1], d)}]` : '';
+  const fmtDelta = (p) => p ? `${p.delta >= 0 ? '+' : '−'}${Math.abs(p.delta).toFixed(4)} [${p.lo >= 0 ? '+' : '−'}${Math.abs(p.lo).toFixed(4)}, ${p.hi >= 0 ? '+' : '−'}${Math.abs(p.hi).toFixed(4)}]${p.significant ? '<sup>*</sup>' : ' (ns)'}` : '';
+  const sbalTip = (m) => `S<sub>bal</sub><sup>test</sup> = <b>${fmt(m.s_bal, 4)}</b>${m.s_bal_ci ? ` <span class="ci">95% CI ${fmtCI(m.s_bal_ci, 4)}</span>` : ''}${m.vs_plantcad2l ? `<br>Δ to PlantCAD2-L ${fmtDelta(m.vs_plantcad2l)}` : ''}`;
   function sorted(rows) { return rows.slice().sort((a, b) => (value(b, state.sort) ?? -1) - (value(a, state.sort) ?? -1)); }
 
   function famTable(m) {
-    return '<table>' + F.map(f => `<tr><td>${f.name}</td><td class="n">${fmt(m.families[f.key])}</td></tr>`).join('') + '</table>';
+    return '<table>' + F.map(f => `<tr><td>${f.name}</td><td class="n">${fmt(m.families[f.key])}</td><td class="n ci">${fmtCI(m.families_ci && m.families_ci[f.key])}</td></tr>`).join('') + '</table>';
   }
 
   // ---- bars ----
@@ -193,7 +201,12 @@
       s += `<g data-id="${esc(m.id)}" class="row">`;
       s += `<text class="rowlabel" x="${left - 14}" y="${y + rowH / 2 + 4}" text-anchor="end">${esc(m.name)}<tspan class="rowsub"> ${sub}</tspan></text>`;
       s += `<rect x="${left}" y="${y + 6}" width="${Math.max(0, end - left)}" height="${rowH - 12}" rx="0" fill="${colour(m)}"/>`;
-      s += `<text class="value" x="${end + 8}" y="${y + rowH / 2 + 4}">${Number.isFinite(v) ? fmt(v) : 'n/a'}</text>`;
+      const iv = Number.isFinite(v) ? ci(m, key) : null;
+      if (iv) {
+        const lo = x(Math.max(iv[0], x0)), hi = x(Math.min(iv[1], x1)), cy = y + rowH / 2;
+        s += `<g class="whisker"><line x1="${lo}" x2="${hi}" y1="${cy}" y2="${cy}"/><line x1="${lo}" x2="${lo}" y1="${cy - 4}" y2="${cy + 4}"/><line x1="${hi}" x2="${hi}" y1="${cy - 4}" y2="${cy + 4}"/></g>`;
+      }
+      s += `<text class="value" x="${(iv ? x(Math.min(iv[1], x1)) : end) + 8}" y="${y + rowH / 2 + 4}">${Number.isFinite(v) ? fmt(v) : 'n/a'}</text>`;
       s += `<rect class="hit" x="0" y="${y}" width="${W}" height="${rowH}"/>`;
       s += '</g>';
     });
@@ -202,7 +215,7 @@
     el.innerHTML = s;
     el.querySelectorAll('g.row').forEach(g => {
       const m = M.find(mm => mm.id === g.dataset.id);
-      g.addEventListener('mousemove', (ev) => showTip(`<b>${esc(m.name)}</b> · ${esc(m.org)}<br>${m.params_str} parameters · ${m.objective}, ${m.domain} pre-training${m.bp_equivalent ? ' · ' + esc(m.bp_equivalent) + ' bp seen' : ''}<br>S<sub>bal</sub><sup>test</sup> = <b>${fmt(m.s_bal, 4)}</b>` + famTable(m), ev));
+      g.addEventListener('mousemove', (ev) => showTip(`<b>${esc(m.name)}</b> · ${esc(m.org)}<br>${m.params_str} parameters · ${m.objective}, ${m.domain} pre-training${m.bp_equivalent ? ' · ' + esc(m.bp_equivalent) + ' bp seen' : ''}<br>${sbalTip(m)}` + famTable(m), ev));
       g.addEventListener('mouseleave', hideTip);
     });
   }
@@ -222,8 +235,8 @@
     h += '</tr></thead><tbody>';
     rows.forEach(m => {
       h += `<tr class="${m.ours ? 'ours' : ''}"><td class="name"><span class="swatch" style="background:${colour(m)}"></span>${esc(m.name)}<span class="sub">${m.params_str}</span></td>`;
-      h += `<td class="cell" ${shade(m.s_bal, sbr)}>${fmt(m.s_bal)}</td>`;
-      cols.forEach(c => { const v = value(m, c.key); h += `<td class="cell" ${shade(v, colVals[c.key])}>${fmt(v)}</td>`; });
+      h += `<td class="cell" ${shade(m.s_bal, sbr)} title="${fmt(m.s_bal, 4)}${m.s_bal_ci ? ' · 95% CI ' + fmtCI(m.s_bal_ci, 4) : ''}">${fmt(m.s_bal)}</td>`;
+      cols.forEach(c => { const v = value(m, c.key); const iv = ci(m, c.key); h += `<td class="cell" ${shade(v, colVals[c.key])} title="${fmt(v, 4)}${iv ? ' · 95% CI ' + fmtCI(iv, 4) : ''}">${fmt(v)}</td>`; });
       h += '</tr>';
     });
     h += '</tbody>';
@@ -330,7 +343,7 @@
     s += '</g>';
     pts.forEach((pt, i) => {
       const m = pt.m;
-      s += `<g data-id="${esc(m.id)}" data-family="${esc(scatterFamily(m))}" class="pt" tabindex="0" role="img" aria-label="${esc(m.name)}, ${esc(m.params_str)} parameters, aggregate score ${fmt(m.s_bal, 4)}"><circle class="point-dot" cx="${pt.cx}" cy="${pt.cy}" r="${pt.r}" fill="${scatterColour(m)}" stroke="var(--paper)" stroke-width="2"/><circle class="point-halo" cx="${pt.cx}" cy="${pt.cy}" r="${pt.r + 4}"/>`;
+      s += `<g data-id="${esc(m.id)}" data-family="${esc(scatterFamily(m))}" class="pt" tabindex="0" role="img" aria-label="${esc(m.name)}, ${esc(m.params_str)} parameters, aggregate score ${fmt(m.s_bal, 4)}">${m.s_bal_ci ? `<line class="errbar" x1="${pt.cx}" x2="${pt.cx}" y1="${y(Math.min(Math.max(m.s_bal_ci[0], y0), y1))}" y2="${y(Math.min(Math.max(m.s_bal_ci[1], y0), y1))}" stroke="${scatterColour(m)}"/>` : ''}<circle class="point-dot" cx="${pt.cx}" cy="${pt.cy}" r="${pt.r}" fill="${scatterColour(m)}" stroke="var(--paper)" stroke-width="2"/><circle class="point-halo" cx="${pt.cx}" cy="${pt.cy}" r="${pt.r + 4}"/>`;
       const l = labels.find(x => x.i === i);
       if (l) {
         const anchor = l.dx > 0 ? 'start' : l.dx < 0 ? 'end' : 'middle';
@@ -343,7 +356,7 @@
     const el = document.getElementById(id); el.innerHTML = s;
     el.querySelectorAll('g.pt').forEach(g => {
       const m = M.find(mm => mm.id === g.dataset.id);
-      const showPointTip = ev => showTip(`<b>${esc(m.name)}</b> · ${esc(m.org)}<br>${m.params_str} parameters${m.bp_equivalent ? ' · ' + esc(m.bp_equivalent) + ' bp seen' : ''}<br>S<sub>bal</sub><sup>test</sup> = <b>${fmt(m.s_bal, 4)}</b>`, ev);
+      const showPointTip = ev => showTip(`<b>${esc(m.name)}</b> · ${esc(m.org)}<br>${m.params_str} parameters${m.bp_equivalent ? ' · ' + esc(m.bp_equivalent) + ' bp seen' : ''}<br>${sbalTip(m)}`, ev);
       g.addEventListener('mouseenter', ev => { hoveredPoint = g; highlightScatterFamily(); showPointTip(ev); });
       g.addEventListener('mousemove', showPointTip);
       g.addEventListener('mouseleave', () => { hoveredPoint = null; highlightScatterFamily(); hideTip(); });

scripts/add_bootstrap_ci.py

@@ -0,0 +1,108 @@
+#!/usr/bin/env python3
+"""Add per-sample bootstrap confidence intervals to data/leaderboard.json.
+
+Source: figures/data/fig3_rerun/rerun_ci.json in the technical report repository
+(dotomics/BOTANIC1-technical-report, branch overleaf, commit 6e420d15), the
+per-example paired bootstrap of the leaderboard cohort (20,000 replicates, draws
+shared across models within each cell, 95% percentile intervals). The same file
+feeds every interval in the v2 report (sec. S_bal uncertainty).
+
+Convention (figures/scripts/fig3_rerun_ci.py in that repository): the point
+estimates stay the reported scores; each interval is the rerun's percentile
+interval translated so that it sits about the reported value, i.e. the offsets
+(lo - mean, hi - mean) are added to the reported point.
+
+Fields added per model: s_bal_ci [lo, hi]; families_ci {family: [lo, hi]};
+metrics[i].ci [lo, hi]; vs_plantcad2l {delta, lo, hi, significant} (paired
+example-level delta to PlantCAD2-L, centred on the reported difference).
+
+Run: python3 scripts/add_bootstrap_ci.py path/to/rerun_ci.json
+"""
+import json
+import sys
+from pathlib import Path
+
+ROOT = Path(__file__).resolve().parents[1]
+LB = ROOT / "data" / "leaderboard.json"
+
+FAMILY = {"chromatin": "chromatin_access", "grc": "genomic_region_classification_v2",
+          "conservation": "plantcad_conservation", "splicing": "splicing", "tis": "plantcad_tis",
+          "tts": "plantcad_tts", "proseq": "pro_seq", "llr": "llr", "causal": "gwas"}
+D = "downstream_tasks."
+CELL = {
+    "GWAS · recall_auc": "gwas_eval_benchmark/recall_auc",
+    "Chromatin · arabidopsis": D + "chromatin_access/chromatin_access.arabidopis_thaliana/eval_aucpr_macro",
+    "Chromatin · brachypodium": D + "chromatin_access/chromatin_access.brachypodium_distachyon/eval_aucpr_macro",
+    "Chromatin · maize": D + "chromatin_access/chromatin_access.zea_mays/eval_aucpr_macro",
+    "Chromatin · rice MH63": D + "chromatin_access/chromatin_access.oryza_sativa_MH63_RS2/eval_aucpr_macro",
+    "Chromatin · rice ZS97": D + "chromatin_access/chromatin_access.oryza_sativa_ZS97_RS2/eval_aucpr_macro",
+    "Chromatin · setaria": D + "chromatin_access/chromatin_access.setaria_italica/eval_aucpr_macro",
+    "Chromatin · sorghum": D + "chromatin_access/chromatin_access.sorghum_bicolor/eval_aucpr_macro",
+    "Conservation · maize": D + "plantcad_conservation/plantcad_conservation.maize/eval_aucpr",
+    "Conservation · sorghum": D + "plantcad_conservation/plantcad_conservation.sorghum/eval_aucpr",
+    "GRC2 · arabidopsis": D + "genomic_region_classification_v2/genomic_region_classification_v2.arabidopsis_thaliana/eval_balanced_accuracy",
+    "GRC2 · maize": D + "genomic_region_classification_v2/genomic_region_classification_v2.zea_mays/eval_balanced_accuracy",
+    "GRC2 · rice": D + "genomic_region_classification_v2/genomic_region_classification_v2.oryza_sativa/eval_balanced_accuracy",
+    "GRC2 · tomato": D + "genomic_region_classification_v2/genomic_region_classification_v2.solanum_lycopersicum/eval_balanced_accuracy",
+    "LLR · arabidopsis": "llr_eval/arabidopsis_thaliana/auroc",
+    "LLR · maize": "llr_eval/zea_mays/auroc",
+    "LLR · tomato": "llr_eval/solanum_lycopersicum/auroc",
+    "PRO-seq · cassava": D + "pro_seq/pro_seq.m_esculenta/eval_aucpr",
+    "Splice acceptor · arabidopsis": D + "splicing/splicing.arabidopsis_thaliana_acceptor/eval_stratified_aucpr_corrected",
+    "Splice donor · arabidopsis": D + "splicing/splicing.arabidopsis_thaliana_donor/eval_stratified_aucpr_corrected",
+    "TIS · arabidopsis": D + "plantcad_tis/plantcad_tis.arabidopsis/eval_aucpr",
+    "TTS · arabidopsis": D + "plantcad_tts/plantcad_tts.arabidopsis/eval_aucpr",
+}
+
+
+def norm(model_id: str) -> str:
+    if model_id.startswith("botanic1-"):
+        return "Botanic1-" + model_id.split("-", 1)[1]
+    if model_id.startswith("CARBON-"):
+        return "Carbon-" + model_id.split("-", 1)[1]
+    return model_id
+
+
+def about(entry, published, key="mean"):
+    shift = published - entry[key]
+    return [round(entry["lo"] + shift, 4), round(entry["hi"] + shift, 4)]
+
+
+def main(rerun_path: str) -> None:
+    rerun = json.loads(Path(rerun_path).read_text())
+    lb = json.loads(LB.read_text())
+    ref = "PlantCAD2-L"
+    ref_sbal = next(m["s_bal"] for m in lb["models"] if m["id"] == ref)
+    n_ci = 0
+    for m in lb["models"]:
+        r = rerun["models"].get(norm(m["id"]))
+        if r is None:
+            print(f"  no rerun entry for {m['id']}")
+            continue
+        m["s_bal_ci"] = about(r["sbal"], m["s_bal"])
+        m["families_ci"] = {k: about(r["families"][FAMILY[k]], v) for k, v in m["families"].items() if FAMILY[k] in r["families"]}
+        for mt in m["metrics"]:
+            c = r["cells"].get(CELL[mt["label"]])
+            if c is not None:
+                mt["ci"] = about(c, mt["value"])
+        if m["id"] != ref:
+            p = rerun["pairs"].get(f"{norm(m['id'])} vs {ref}")
+            if p is not None:
+                d = round(m["s_bal"] - ref_sbal, 4)
+                lo, hi = about(p, d, key="delta")
+                m["vs_plantcad2l"] = {"delta": d, "lo": lo, "hi": hi, "significant": bool(lo > 0 or hi < 0)}
+        n_ci += 1
+        # sanity: rerun point within 0.02 of the reported score
+        gap = abs(r["sbal"]["mean"] - m["s_bal"])
+        flag = "  <-- check" if gap > 0.02 else ""
+        print(f"{m['id']:14s} s_bal={m['s_bal']:.4f} rerun={r['sbal']['mean']:.4f} ci={m['s_bal_ci']}{flag}")
+    lb["ci_note"] = ("95% confidence intervals from the per-sample paired bootstrap of the technical report "
+                     "(20,000 replicates, draws shared across models within each cell), centred on the reported score.")
+    lb["ci_meta"] = {"replicates": rerun["meta"]["replicates"], "level": 95,
+                     "source": "BOTANIC1-technical-report figures/data/fig3_rerun/rerun_ci.json @ 6e420d15"}
+    LB.write_text(json.dumps(lb, indent=1, ensure_ascii=False) + "\n")
+    print(f"wrote {LB} with intervals for {n_ci}/{len(lb['models'])} models")
+
+
+if __name__ == "__main__":
+    main(sys.argv[1])