spotlight: density × rarity rank (rare anchor wins over repeated common tokens)

Pure density ranking ("count distinct tokens within ±half-window")
loses on Sonic-shaped chunks: "Sonic" appears 20+ times, "1990"
appears once, so any Sonic-anchored slice scores 4 distinct tokens
while the one 1990-anchored slice scores 2. The 1990 slice IS the
load-bearing claim evidence; the Sonic slice is just where the
chunk happens to lead.

Fox flagged 2026-06-01: STRICT-verified claim "Sonic the Hedgehog was
created in 1990" anchored to chunk E5 (franchise timeline), but the
spotlit excerpt showed the AM8/Sega-Team paragraph, not the "April
1990" year entry where the date claim grounds.

Fix: multiply density by 1/freq(anchor_token) in the span. A token
appearing once gets full weight; a token appearing 20× gets 0.05.
Rare anchors (years, named entities, version numbers) now win even
with fewer neighbors in the window — the rare token marks the slice
the verifier actually verified.

Sample: synthetic Sonic-shaped chunk + claim "created in 1990" →
excerpt now centers on "In 1990, Sega began developing the Sonic
the Hedgehog character." (was the Sonic Team / AM8 paragraph).

Live: `make query Q="when was sonic the hedgehog created?"` →
spotlight now leads with "In April 1990, Sega commissioned its AM8
R&D department to create a character…"

Tie-break stays smallest-idx (deterministic). 175 claim_lattice +
verify + inspect tests pass.

arborist/qa/evidence.py

@@ -437,19 +437,30 @@ def _spotlight_excerpt(claim_text: str, span: str, *, window: int) -> str:
         suffix = "..." if new_end < len(span) else ""
         return f"{span[new_start:new_end]}{suffix}"
 
-    # Density rank: for each candidate position, count distinct
-    # tokens whose match positions fall within ±half. Pick the
-    # position with max distinct count; tie-break on smallest idx
-    # (deterministic, reproducible).
-    best_score = -1
+    # Density × rarity rank. For each candidate position, count
+    # distinct tokens whose match positions fall within ±half AND
+    # weight by 1/freq(anchor_token) in the span. A rare anchor
+    # (year, named entity, version number) appearing once in a
+    # Sonic-article-length chunk wins over a 20×-repeated "sonic"
+    # — the rare token marks the load-bearing slice for the claim.
+    # Without this, "Sonic the Hedgehog was created in 1990" cited
+    # to the franchise-timeline chunk spotlit a Sonic-Team paragraph
+    # instead of the "1990" year entry (2026-06-01 fox report).
+    # Tie-break on smallest idx (deterministic, reproducible).
+    freq: dict[str, int] = {}
+    for tok in tokens:
+        # Count overlapping occurrences using positions we already
+        # collected — same definition the density count uses.
+        freq[tok] = sum(1 for t, _ in positions if t == tok) or 1
+    best_score = -1.0
     match_idx = positions[0][1]
-    for _tok, idx in positions:
+    for tok, idx in positions:
         distinct: set[str] = set()
         lo, hi = idx - half, idx + half
         for tok2, idx2 in positions:
             if lo <= idx2 <= hi:
                 distinct.add(tok2)
-        score = len(distinct)
+        score = len(distinct) / freq[tok]
         if score > best_score or (score == best_score and idx < match_idx):
             best_score = score
             match_idx = idx