Add fixed and exact-trace Frontier surfaces

runs/frontier-phase-factorial-v0/analyze_qwen30_prefill_fidelity.py

@@ -222,7 +222,11 @@ def parse_simulator(manifests: list[Path]) -> tuple[dict[str, Any], list[dict[st
     sources = []
     for path in manifests:
         payload = load_json(path)
-        if payload["status"] not in {"complete", "partial_not_decision_bearing"}:
+        if payload["status"] not in {
+            "complete",
+            "frozen_before_real",
+            "partial_not_decision_bearing",
+        }:
             raise RuntimeError(f"simulator manifest has invalid status: {path}")
         sources.append({"path": str(path.resolve()), "sha256": sha256(path)})
         result_by_name = {
@@ -305,12 +309,21 @@ def compare(real: dict[str, Any], simulated: dict[str, Any]) -> dict[str, Any]:
 
     confusion = {"real_pass_sim_pass": 0, "real_pass_sim_fail": 0,
                  "real_fail_sim_pass": 0, "real_fail_sim_fail": 0}
+    anchor_grid_coverage = {
+        "shared": 0,
+        "real_only": 0,
+        "simulator_only": 0,
+    }
     for name in names:
         real_anchors = {row["rate"]: row for row in real[name]["anchors"]}
         sim_anchors = {row["rate"]: row for row in simulated[name]["anchors"]}
-        if set(real_anchors) != set(sim_anchors):
-            raise RuntimeError(f"anchor grids differ for {name}")
-        for rate in real_anchors:
+        shared = set(real_anchors) & set(sim_anchors)
+        if not shared:
+            raise RuntimeError(f"anchor grids do not overlap for {name}")
+        anchor_grid_coverage["shared"] += len(shared)
+        anchor_grid_coverage["real_only"] += len(set(real_anchors) - shared)
+        anchor_grid_coverage["simulator_only"] += len(set(sim_anchors) - shared)
+        for rate in shared:
             real_pass = real_anchors[rate]["conservative_feasible"]
             sim_pass = sim_anchors[rate]["feasible"]
             key = f"real_{'pass' if real_pass else 'fail'}_sim_{'pass' if sim_pass else 'fail'}"
@@ -328,6 +341,7 @@ def compare(real: dict[str, Any], simulated: dict[str, Any]) -> dict[str, Any]:
         "kendall": tau,
         "pairwise_non_tied": pairwise,
         "anchor_confusion": confusion,
+        "anchor_grid_coverage": anchor_grid_coverage,
     }
 
 
@@ -356,7 +370,7 @@ def plot(path: Path, rows: list[dict[str, Any]], metrics: dict[str, Any]) -> Non
         x + width / 2,
         [row["simulator"] for row in rows],
         width,
-        label="Frontier profile-only",
+        label="Frontier simulator",
     )
     axis.set_xticks(x, labels, fontsize=8)
     axis.set_ylabel("Max tested SLO-feasible request rate / GPU")
@@ -374,7 +388,7 @@ def plot(path: Path, rows: list[dict[str, Any]], metrics: dict[str, Any]) -> Non
     )
     axis.text(
         0.01,
-        0.98,
+        0.82,
         annotation,
         transform=axis.transAxes,
         va="top",