Replace benchmark timeline with monthly usage bar chart

app.py

@@ -408,15 +408,6 @@ def openness_chip(is_open: bool) -> str:
     return f'<span class="chip"><i style="background:{color}"></i>{text}</span>'
 
 
-def openness_legend() -> str:
-    return (
-        '<div class="legend">'
-        '<span class="key"><i class="round" style="background:var(--open)"></i>open-weight</span>'
-        '<span class="key"><i class="round" style="background:var(--closed)"></i>closed / API-only</span>'
-        "</div>"
-    )
-
-
 def nice_ceil(v: float) -> float:
     if v <= 0:
         return 1
@@ -593,55 +584,63 @@ def popularity_view(category_filter, openness_filter, min_models, search) -> str
 
 
 # ---------------------------------------------------------------------------
-# Tab 2: Benchmark -> Models (summary card, SVG timeline, table)
+# Tab 2: Benchmark -> Models (summary card, monthly-usage SVG bars, table)
 # ---------------------------------------------------------------------------
-def svg_benchmark_timeline(df: pd.DataFrame) -> str:
-    labs = df.groupby("lab")["release_date"].min().sort_values().index.tolist()
-    W, ML, MR, MT, MB, ROW = 920, 150, 24, 12, 34, 26
-    H = MT + MB + ROW * len(labs)
-    t0, t1 = df["release_date"].min(), df["release_date"].max()
-    pad = pd.Timedelta(days=max((t1 - t0).days * 0.05, 20))
-    t0p, t1p = t0 - pad, t1 + pad
-    span = max((t1p - t0p).days, 1)
-
-    def x(d):
-        return ML + (d - t0p).days / span * (W - ML - MR)
+def _rounded_top_rect(x, y, w, h, r) -> str:
+    r = min(r, h / 2, w / 2)
+    return (
+        f'M {x:.1f} {y + h:.1f} L {x:.1f} {y + r:.1f} Q {x:.1f} {y:.1f} {x + r:.1f} {y:.1f} '
+        f'L {x + w - r:.1f} {y:.1f} Q {x + w:.1f} {y:.1f} {x + w:.1f} {y + r:.1f} L {x + w:.1f} {y + h:.1f} Z'
+    )
+
+
+def svg_benchmark_monthly_usage(df: pd.DataFrame) -> str:
+    """Bars of how many model releases reported this benchmark each month, spanning
+    the whole dataset time range (empty months stay visible as gaps)."""
+    t0 = MODELS_DF["release_date"].min().to_period("M")
+    t1 = MODELS_DF["release_date"].max().to_period("M")
+    months = pd.period_range(t0, t1, freq="M")
+    counts = df["release_date"].dt.to_period("M").value_counts().reindex(months, fill_value=0)
+
+    W, H, ML, MR, MT, MB = 920, 300, 44, 10, 12, 36
+    plot_h = H - MT - MB
+    base = MT + plot_h
+    vmax = int(counts.max())
+    step = max(1, math.ceil(vmax / 4))
+    ymax = step * 4
+    scale = plot_h / ymax
+    band = (W - ML - MR) / len(months)
+    bw = max(band - 2, 1.5)  # 2px surface gap between adjacent bars
 
     parts = [
         f'<svg class="viz" viewBox="0 0 {W} {H}" role="img" '
-        f'aria-label="Timeline of model releases using this benchmark, one row per lab">'
+        f'aria-label="Monthly count of model releases reporting this benchmark, over the whole dataset period">'
     ]
-    for tick in month_ticks(t0p, t1p):
-        tx = x(tick)
-        parts.append(f'<line x1="{tx:.1f}" y1="{MT}" x2="{tx:.1f}" y2="{H - MB}" stroke="var(--grid)" stroke-width="1"/>')
-        parts.append(
-            f'<text x="{tx:.1f}" y="{H - MB + 18}" text-anchor="middle" font-size="11" fill="var(--muted)">'
-            f"{tick.strftime('%b %Y')}</text>"
-        )
-    for i, lab in enumerate(labs):
-        cy = MT + i * ROW + ROW / 2
-        parts.append(f'<line x1="{ML}" y1="{MT + (i + 1) * ROW}" x2="{W - MR}" y2="{MT + (i + 1) * ROW}" stroke="var(--grid)" stroke-width="1"/>')
-        name = lab if len(lab) <= 19 else lab[:18] + "…"
-        parts.append(f'<text x="{ML - 8}" y="{cy + 4}" text-anchor="end" font-size="11.5" fill="var(--ink)">{esc(name)}</text>')
-        sub = df[df["lab"] == lab].sort_values("release_date")
-        prev_cx, flip = None, 1
-        for r in sub.itertuples():
-            cx = x(r.release_date)
-            dy = 0.0
-            if prev_cx is not None and cx - prev_cx < 11:
-                dy = 5.0 * flip
-                flip = -flip
-            else:
-                flip = 1
-            prev_cx = cx
-            color = "var(--open)" if r.is_open else "var(--closed)"
-            kind = "open-weight" if r.is_open else "closed"
+    for k in range(5):
+        v = step * k
+        y = base - v * scale
+        parts.append(f'<line x1="{ML}" y1="{y:.1f}" x2="{W - MR}" y2="{y:.1f}" stroke="var(--grid)" stroke-width="1"/>')
+        parts.append(f'<text x="{ML - 7}" y="{y + 4:.1f}" text-anchor="end" font-size="11" fill="var(--muted)">{v}</text>')
+
+    tick_months = {t.to_period("M") for t in month_ticks(months[0].to_timestamp(), months[-1].to_timestamp(how="end"))}
+    for i, month in enumerate(months):
+        x0 = ML + i * band
+        if month in tick_months:
+            cx = x0 + band / 2
             parts.append(
-                f'<circle class="dot" cx="{cx:.1f}" cy="{cy + dy:.1f}" r="5.5" fill="{color}" '
-                f'stroke="var(--surf)" stroke-width="2">'
-                f"<title>{esc(r.model_id)} — {r.release_date.date()} ({kind})</title></circle>"
+                f'<text x="{cx:.1f}" y="{base + 20}" text-anchor="middle" font-size="11" fill="var(--muted)">'
+                f"{month.to_timestamp().strftime('%b %Y')}</text>"
             )
-    parts.append(f'<line x1="{ML}" y1="{H - MB}" x2="{W - MR}" y2="{H - MB}" stroke="var(--muted)" stroke-width="1"/>')
+        n = int(counts.loc[month])
+        if n == 0:
+            continue
+        h = n * scale
+        label = f"{month.to_timestamp().strftime('%b %Y')}: {n} model{'s' if n != 1 else ''}"
+        parts.append(
+            f'<path class="seg" d="{_rounded_top_rect(x0 + 1, base - h, bw, h, 4)}" fill="var(--accent)">'
+            f"<title>{esc(label)}</title></path>"
+        )
+    parts.append(f'<line x1="{ML}" y1="{base}" x2="{W - MR}" y2="{base}" stroke="var(--muted)" stroke-width="1"/>')
     parts.append("</svg>")
     return "".join(parts)
 
@@ -667,9 +666,8 @@ def benchmark_view(benchmark):
     )
 
     chart = (
-        f'<div class="viz-title">Who uses “{esc(benchmark)}”, and when did they release?</div>'
-        + openness_legend()
-        + svg_benchmark_timeline(df)
+        f'<div class="viz-title">How often is “{esc(benchmark)}” used? Model releases reporting it per month</div>'
+        + svg_benchmark_monthly_usage(df)
     )
 
     rows = []
@@ -738,14 +736,6 @@ def model_view(model_id):
 # ---------------------------------------------------------------------------
 # Tab 4: Category evolution (SVG stacked bars + table)
 # ---------------------------------------------------------------------------
-def _rounded_top_rect(x, y, w, h, r) -> str:
-    r = min(r, h / 2, w / 2)
-    return (
-        f'M {x:.1f} {y + h:.1f} L {x:.1f} {y + r:.1f} Q {x:.1f} {y:.1f} {x + r:.1f} {y:.1f} '
-        f'L {x + w - r:.1f} {y:.1f} Q {x + w:.1f} {y:.1f} {x + w:.1f} {y + r:.1f} L {x + w:.1f} {y + h:.1f} Z'
-    )
-
-
 def svg_category_bars(counts: pd.DataFrame, normalize: bool) -> str:
     periods = counts.index.tolist()
     plot = counts.copy()