docs(benchmarking): add coverage and review summary table

benchmarking/index.md

@@ -82,18 +82,8 @@ Many apparent disagreements come from the denominator, not from missing emission
 
 The goal is not an exact match to one external number. It is whether our estimate aligns with the **preponderance of the evidence**, after fair scope and dollar treatment.
 
-## Results for this release
-
-:::{figure} ../figures/diagrams/benchmark-2026-rc.png
-:label: fig-benchmark-2026-rc
-:alt: Donut chart titled How Cornerstone compares to third-party references. About 70 percent of country-sector pairs with a verified external benchmark fall within 25 percent of the benchmark; 21 percent have a 25 to 50 percent gap; 10 percent have a gap of 50 percent or more.
-Share of country–sector pairs with a verified external benchmark. About 70% fall within 25% of the benchmark; the rest have larger gaps.
-:::
-
-See [](../releases/rc.md) for the release summary.
-
 :::{seealso} Related
-Internal accounting checks live on [](../validation/index.md). Version-specific headlines will also appear in the [](../releases/index.md).
+Internal accounting checks live on [](../validation/index.md). Version-specific headlines appear in the [](../releases/index.md).
 :::
 
 +++

figures/figures.csv

@@ -18,4 +18,4 @@ ghg-pipeline,ee/index.md,diagram (SVG),hand-made,,,draft
 eeio-overview,intro/what-is-eeio.md,diagram (SVG),hand-made,,,draft
 user-guidance-ef-by-country,factors/user-guidance.md,chart,"Cornerstone GHG Emission Factors workbook, 2026-rc (2026PRO, sector 541610)",figures/scripts/user_guidance_figures.py,Yohanna Maldonado,draft
 user-guidance-worked-example,factors/user-guidance.md,diagram,"Worked example from the 2026-rc workbook (factor 0.0449; GBP rate 0.74283)",figures/scripts/user_guidance_figures.py,Yohanna Maldonado,draft
-benchmark-2026-rc,benchmarking/index.md,chart,"EF benchmark comparisons for 2026-rc (share of country-sector pairs vs third-party references)",,Ben Young,draft
+benchmark-2026-rc,releases/rc.md,chart,"EF benchmark comparisons for 2026-rc (61 / 18 / 9 of 88 sectors; model D from cornerstone_2026_v0_7, ceda 03facf41)",figures/scripts/benchmark_figures.py,Ben Young,draft

figures/scripts/benchmark_figures.py

@@ -0,0 +1,77 @@
+"""Write the benchmarking results donut to figures/diagrams/benchmark-2026-rc.png.
+
+Redraws the slide-deck donut with the release's numbers. Each of the 88 benchmarked sectors falls in
+one band by its gap, |model D - benchmark| / benchmark, with model D from the v1.0 release diagnostics
+(cornerstone_2026_v0_7, ceda commit 03facf41):
+
+- within 25% of the benchmark: 61 sectors
+- 25-50% gap: 18 sectors
+- 50% or more gap: 9 sectors
+
+Each percentage is rounded to the nearest whole percent, halves up (69.32%, 20.45% and 10.23% show
+as 69%, 20% and 10%), so the labels add to 99%.
+
+Run from the repo root (needs matplotlib):
+    uv run --with matplotlib python figures/scripts/benchmark_figures.py
+"""
+from decimal import ROUND_HALF_UP, Decimal
+from pathlib import Path
+
+import matplotlib
+
+matplotlib.use("Agg")
+import matplotlib.pyplot as plt  # noqa: E402
+from matplotlib.patches import Circle, Wedge  # noqa: E402
+
+OUT = Path(__file__).resolve().parents[1] / "diagrams"
+TITLE, INK, MUTED = "#2F6F67", "#2C3746", "#6A727C"
+BANDS = [  # (label, sectors, colour)
+    ("Within 25% of benchmark", 61, "#2F5D8A"),
+    ("25–50% gap", 18, "#D5D8DC"),
+    ("50% or more gap", 9, "#7D848D"),
+]
+plt.rcParams.update({"font.family": ["Helvetica", "Arial", "DejaVu Sans"]})
+
+
+def donut():
+    counts = [n for _, n, _ in BANDS]
+    total = sum(counts)
+    pcts = [int((Decimal(100 * n) / total).quantize(Decimal(1), rounding=ROUND_HALF_UP)) for n in counts]
+
+    fig = plt.figure(figsize=(6.2, 5.7), dpi=200)
+    ax = fig.add_axes([0, 0, 1, 1])
+    ax.set_xlim(0, 620)
+    ax.set_ylim(570, 0)
+    ax.axis("off")
+
+    ax.text(28, 26, "How Cornerstone compares to third-party references", color=TITLE, fontsize=14.5,
+            fontweight="bold", va="center")
+
+    cx, cy, r_out, r_in = 310, 231, 158, 103
+    # The y axis runs top-down, so data angles turn clockwise on the page; -90 is 12 o'clock.
+    start = -90.0
+    for (_, n, colour) in BANDS:
+        sweep = 360 * n / total
+        ax.add_patch(Wedge((cx, cy), r_out, start, start + sweep, width=r_out - r_in, facecolor=colour,
+                           edgecolor="white", linewidth=1.5))
+        start += sweep
+
+    ax.text(cx, cy - 18, f"{pcts[0]}%", color=BANDS[0][2], fontsize=34, fontweight="bold", ha="center",
+            va="center")
+    ax.text(cx, cy + 30, "within 25% of\nbenchmark", color=INK, fontsize=11, ha="center", va="center",
+            linespacing=1.4)
+
+    for i, ((label, _, colour), pct) in enumerate(zip(BANDS, pcts)):
+        y = 431 + 32 * i
+        ax.add_patch(Circle((37, y), 9, facecolor=colour, edgecolor="none"))
+        ax.annotate(label, (65, y), color=INK, fontsize=12.5, va="center")
+        ax.annotate(f" {pct}%", (1, 0.5), xycoords=ax.texts[-1], color=INK, fontsize=12.5,
+                    fontweight="bold", va="center")
+
+    ax.text(38, 531, f"Share of the {total} benchmarked sectors", color=MUTED, fontsize=10.5, va="center")
+    fig.savefig(OUT / "benchmark-2026-rc.png", facecolor="white")
+    plt.close(fig)
+
+
+if __name__ == "__main__":
+    donut()

releases/rc.md

@@ -42,6 +42,24 @@ Three groups of improvements sit behind the changes.
 
 Together these make the 2017–2026 time series possible.
 
+## Benchmarking results
+
+:::{figure} ../figures/diagrams/benchmark-2026-rc.png
+:label: fig-benchmark-2026-rc
+:alt: Donut chart titled How Cornerstone compares to third-party references. 69 percent of the 88 benchmarked sectors fall within 25 percent of the benchmark; 20 percent have a 25 to 50 percent gap; 10 percent have a gap of 50 percent or more.
+Share of the 88 benchmarked sectors. 69% fall within 25% of the benchmark; the rest have larger gaps.
+:::
+
+We benchmarked the direct emission factor (`D`) of 88 sectors, chosen because they drive a large share of emissions. Together they account for about 52% of global direct emissions in the model.
+
+| Measure | Sectors | Share of the 88 sectors | Share of global direct emissions |
+|---|---|---|---|
+| Within 25% of the benchmark | 61 | 69% | 46% |
+| Manually reviewed | 47 | 53% | 36% |
+| Manually reviewed, or within 25% of the benchmark | 78 | 89% | 50% |
+
+A sector counts as manually reviewed when someone investigated its gap and reached one of the three outcomes in [](../benchmarking/index.md): a sharper benchmark, a model improvement, or a documented difference. Sectors within 25% of the benchmark were not all reviewed in this depth, because the gap gave no reason to.
+
 ## Why do these improvements matter?
 
 They translate real world changes to corporate footprints. Three examples show how: