fix(charts): readable cost axis and label spacing on cost-vs-accuracy

Show explicit dollar-formatted ticks instead of a single 10^0 mark,
include each model's cost next to its label, and place labels with
per-model offsets so the upper-right cluster (Gemini 3 Flash, GPT-5.5,
Grok 4.3, Gemini 3.1 Pro) no longer overlaps. Pareto-frontier points
get a darker outline and a soft green fill above the curve.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

scripts/make_neet_2026_charts.py

@@ -12,6 +12,7 @@ from collections import defaultdict
 from pathlib import Path
 
 import matplotlib.pyplot as plt
+import matplotlib.ticker as mticker
 import numpy as np
 
 # Display label and brand color per OpenRouter model id.
@@ -165,20 +166,28 @@ def chart_main_leaderboard(data: list[dict], out_path: Path):
 
 
 def chart_cost_vs_accuracy(data: list[dict], out_path: Path):
-    fig, ax = plt.subplots(figsize=(11, 7), dpi=200)
+    # Per-model label offsets (dx_log_factor, dy_pct, ha) to keep labels readable.
+    # dx_factor < 1 → label sits LEFT of point; > 1 → RIGHT. ha matches.
+    LABEL_OFFSETS = {
+        "google/gemini-3-flash-preview":     (1.07, +1.30, "left"),   # top of cluster, label up-right
+        "google/gemini-3.1-pro-preview":     (0.93, -0.20, "right"),  # rightmost point, label LEFT
+        "openai/gpt-5.5":                    (1.07, +1.30, "left"),   # label up-right (clear of Flash)
+        "openai/gpt-5.4":                    (1.07, +1.20, "left"),
+        "qwen/qwen3-vl-235b-a22b-thinking":  (0.93, +0.10, "right"),  # label LEFT (clear of Grok)
+        "anthropic/claude-sonnet-4.6":       (1.07, -1.30, "left"),   # label down-right (clear of Grok)
+        "anthropic/claude-haiku-4.5":        (1.07, +1.30, "left"),
+        "z-ai/glm-4.6v":                     (1.07, +1.30, "left"),
+        "x-ai/grok-4.3":                     (1.07, -1.40, "left"),   # label down-right (clear of Qwen + cluster)
+    }
+    DEFAULT_OFFSET = (1.07, +0.80, "left")
+
+    fig, ax = plt.subplots(figsize=(13, 8), dpi=200)
     fig.patch.set_facecolor("white")
 
-    for d in data:
-        if d["cost"] <= 0:
-            continue
-        pct = d["score"] / 720 * 100
-        ax.scatter(d["cost"], pct, s=320, color=color_for(d["model"]),
-                   edgecolors="white", linewidths=2, zorder=3, alpha=0.95)
-        ax.annotate(label_for(d["model"]), (d["cost"], pct),
-                    xytext=(d["cost"] * 1.05, pct + 0.6),
-                    fontsize=10.5, fontweight="bold", color="#222")
-
-    pareto_pts = sorted([(d["cost"], d["score"] / 720 * 100) for d in data if d["cost"] > 0])
+    points = [(d, d["score"] / 720 * 100) for d in data if d["cost"] > 0]
+
+    # Pareto frontier: cheapest cost for each accuracy level (upper envelope).
+    pareto_pts = sorted([(d["cost"], pct) for d, pct in points])
     frontier, best = [], -1
     for c, p in pareto_pts:
         if p > best:
@@ -186,16 +195,47 @@ def chart_cost_vs_accuracy(data: list[dict], out_path: Path):
             best = p
     if frontier:
         fx, fy = zip(*frontier)
-        ax.plot(fx, fy, "--", color="#888", linewidth=1.4, alpha=0.5, zorder=1, label="Pareto frontier")
+        ax.plot(fx, fy, "--", color="#888", linewidth=1.5, alpha=0.55, zorder=1,
+                label="Pareto frontier")
+        ax.fill_between(fx, fy, [102] * len(fx), color="#88CC88", alpha=0.06, zorder=0)
+
+    on_frontier = set(frontier)
+    for d, pct in points:
+        is_pareto = (d["cost"], pct) in on_frontier
+        ax.scatter(d["cost"], pct, s=360, color=color_for(d["model"]),
+                   edgecolors="#222" if is_pareto else "white",
+                   linewidths=2.2 if is_pareto else 2,
+                   zorder=4, alpha=0.97)
+
+        dx_factor, dy, ha = LABEL_OFFSETS.get(d["model"], DEFAULT_OFFSET)
+        label_x = d["cost"] * dx_factor
+        ax.annotate(
+            f"{label_for(d['model'])}\n${d['cost']:.2f}",
+            (d["cost"], pct),
+            xytext=(label_x, pct + dy),
+            fontsize=10, fontweight="bold", color="#222", ha=ha,
+            bbox=dict(boxstyle="round,pad=0.25", facecolor="white",
+                      edgecolor="none", alpha=0.85),
+            zorder=5,
+        )
 
     ax.set_xscale("log")
+    # Explicit log-spaced ticks with $ formatting so the cost axis is readable.
+    tick_locs = [0.1, 0.2, 0.3, 0.5, 0.7, 1.0, 1.5, 2.0, 3.0, 5.0]
+    ax.xaxis.set_major_locator(mticker.FixedLocator(tick_locs))
+    ax.xaxis.set_major_formatter(mticker.FuncFormatter(
+        lambda x, _: f"${x:.2f}" if x < 1 else f"${x:.1f}"))
+    ax.xaxis.set_minor_locator(mticker.NullLocator())
+    costs = [d["cost"] for d, _ in points]
+    ax.set_xlim(min(costs) * 0.7, max(costs) * 1.5)
+
     ax.set_xlabel("Total cost for full exam (USD, log scale)", fontsize=12, fontweight="bold")
     ax.set_ylabel("Accuracy (%)", fontsize=12, fontweight="bold")
     ax.set_title("NEET 2026 — Cost vs Accuracy (upper-left = best value)",
                  fontsize=14, pad=14)
-    pcts = [d["score"] / 720 * 100 for d in data]
+    pcts = [pct for _, pct in points]
     ax.set_ylim(min(pcts) - 5, 102)
-    ax.grid(alpha=0.25, linestyle="-", linewidth=0.5)
+    ax.grid(which="major", axis="both", alpha=0.28, linestyle="-", linewidth=0.6)
     ax.set_axisbelow(True)
     ax.legend(loc="lower right", frameon=False, fontsize=10)