answer --seed, and runs/spread.sh: a score comes with its spread

Every score so far was one run of a small visitor. `answer --seed n`
answers the same tasks with other shuffles and sampling, and
runs/spread.sh prints a run's score per seed, so a change can be told
from the visitor's mood.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

runs/spread.sh

@@ -0,0 +1,19 @@
+#!/bin/sh
+# The noise floor: answer one run's tasks with several seeds and print
+# each seed's score, so a change can be told from the visitor's mood.
+#   runs/spread.sh runs/<run> [seeds...]   (default seeds: 1 2 3)
+set -eu
+run=${1:?usage: spread.sh runs/<run> [seeds...]}; shift
+seeds=${*:-1 2 3}
+IRIS=/var/local/cargo-target/release/iris
+T=/var/local/cargo-target/release/portal-trainer
+LLM="MODEL=qwen3:8b $(dirname "$0")/../engines/ollama.sh"
+cd "$run"
+for mode in full skim; do
+  skim=""; [ "$mode" = skim ] && skim="--skim"
+  for s in $seeds; do
+    $T answer --tasks tasks-$mode.jsonl --repo site --llm "$LLM" --seed "$s" > answers-$mode-seed$s.jsonl
+    printf "%s seed %s: " "$mode" "$s"
+    $IRIS check --path site/questions --needs-tasks tasks-$mode.jsonl $skim --needs-score answers-$mode-seed$s.jsonl 2>&1 | grep -E "^wording:"
+  done
+done