feat(skill): replace narrative verdict row with evidence-triggered prose beat

Review showed the "Setting the narrative?" verdict compared across
abstraction levels: a homepage tagline is deliberately broad
("financial infrastructure" covers a chargebacks thread), so
tagline-vs-thread alignment verdicts are unfalsifiable and carry no
information. The signal now ships as PROSE in the entity's narrative
section, fires only when the month's evidence directly bears on the
pitch (supports a specific claim, cuts against one, or is squarely
about the pitched ground), and stays SILENT when the pulse is
orthogonal - omission over a manufactured connection. Claims are
tested at matched altitude (specific claim vs specific thread) and
stay windowed (no trend verbs one 30-day window can't support). The
positioning fetch step survives unchanged and now also grounds the
"What it is" row and brand-noise rejection. All scope gating (people
never, ownerless topics excluded, no pitch from memory) carries over.

skills/last30days/scripts/lib/render.py

@@ -531,12 +531,10 @@ def _render_comparison_scaffold(topic: str) -> list[str]:
     Returns empty list if topic is not a comparison query. When present,
     the block is bracketed so the synthesizer can detect it and pass through.
 
-    Axes are the 9 from the April 9 launch-video exemplar (suited to AI-tool
-    comparisons) plus "Setting the narrative?", which asks whether each entity's
-    community conversation is about what the entity itself pitches, or about
-    something else (pricing, rivals, incidents). For comparisons where an axis
-    does not apply, the synthesizer writes N/A or a topic-appropriate substitute
-    in that row.
+    Axes match the April 9 launch-video exemplar (9 axes suited to AI-tool
+    comparisons). For non-AI-tool comparisons, the synthesizer writes N/A
+    or topic-appropriate substitutes in irrelevant rows. The "What it is" row
+    grounds in first-party positioning fetched during the run when available.
     """
     entities = _parse_comparison_entities(topic)
     if not entities:
@@ -546,11 +544,10 @@ def _render_comparison_scaffold(topic: str) -> list[str]:
     header = "| Dimension | " + " | ".join(entities) + " |"
     # Separator row matching column count
     separator = "|" + "|".join(["---"] * (len(entities) + 1)) + "|"
-    # 9 axes from the April 9 exemplar plus "Setting the narrative?" (pitch vs.
-    # what the community actually talks about). See the function docstring.
+    # 9 axes from the April 9 exemplar. Model fills with topic-appropriate
+    # content; irrelevant axes get "N/A" rather than invented data.
     axes = [
         "What it is",
-        "Setting the narrative?",
         "GitHub stars",
         "Philosophy",
         "Skills",
@@ -562,20 +559,11 @@ def _render_comparison_scaffold(topic: str) -> list[str]:
     ]
     body = [f"| {axis} | " + " | ".join([" "] * len(entities)) + " |" for axis in axes]
 
-    # Generic fill rules plus guidance for "Setting the narrative?" - the one
-    # axis that needs a judgement, not a lookup.
     fill_instructions = (
         "Fill each cell based on the research above. Keep cells short (5-15 words). "
         "Use ' - ' (hyphen with spaces) not em-dashes. Write N/A for axes that do not apply to this topic class. "
-        "For the \"Setting the narrative?\" row, judge whether each entity's community conversation is about "
-        "what the entity itself pitches: start the cell with Yes / Partly / No / Unclear, then name the topic "
-        "the community is ACTUALLY on, anchored to a real item (e.g. \"No - pitches uptime, but the top thread "
-        "is friendly-fraud (323pt HN)\"). Use Unclear when evidence is thin or polluted with unrelated "
-        "brand-name matches; do NOT infer a verdict from vibes. Write N/A for entities with no public pitch: "
-        "people (always, even famous founders - the lens applies to companies and products, never persons), "
-        "events, abstract concepts, and ownerless topics with no authoritative first party (e.g. Bitcoin). "
-        "Only fill a verdict when the entity's first-party positioning was actually fetched during this "
-        "run's research; if it was not, write Unclear - never supply the pitch from memory. "
+        "Ground the \"What it is\" row in first-party positioning fetched during this run's research when "
+        "available - describe each entity as it pitches itself today, never from memory. "
         "This scaffold matches the April 9 launch-video exemplar shape."
     )