Skip to content

AI agent skill that researches any topic across Reddit, X, YouTube, HN, Polymarket, and the web - then synthesizes a grounded summary

fix(reddit): clamp engagement before log10 so downvoted posts don't crash ranking (#1119)

Ilia Alshanetsky586ba1deacef
diff --git a/changelog.d/+reddit-negative-engagement.fixed.md b/changelog.d/+reddit-negative-engagement.fixed.md
new file mode 100644
index 00000000..06c66828
--- /dev/null
+++ b/changelog.d/+reddit-negative-engagement.fixed.md
@@ -0,0 +1 @@
+Reddit ranking no longer crashes with a `math domain error` when a downvoted post has a negative score; such posts now receive the minimum engagement bonus.
diff --git a/skills/last30days/scripts/lib/reddit.py b/skills/last30days/scripts/lib/reddit.py
index ddb37b03..9476b90e 100644
--- a/skills/last30days/scripts/lib/reddit.py
+++ b/skills/last30days/scripts/lib/reddit.py
@@ -264,7 +264,7 @@ def _relevance_rank_key(item: Dict[str, Any]) -> float:
~0) above an on-topic one (relevance >= RELEVANCE_FLOOR).
"""
rel = item.get("relevance") or 0.0
- eng_bonus = min(0.25, math.log10(_total_engagement(item) + 1) / 20.0)
+ eng_bonus = min(0.25, math.log10(max(0, _total_engagement(item)) + 1) / 20.0)
return rel + eng_bonus
diff --git a/skills/last30days/scripts/lib/reddit_keyless.py b/skills/last30days/scripts/lib/reddit_keyless.py
index db7ea23d..9b6f75af 100644
--- a/skills/last30days/scripts/lib/reddit_keyless.py
+++ b/skills/last30days/scripts/lib/reddit_keyless.py
@@ -51,7 +51,7 @@ def _relevance_rank_key(post: Dict[str, Any]) -> float:
"""
eng = post.get("engagement", {})
total = (eng.get("score", 0) or 0) + (eng.get("num_comments", 0) or 0)
- return (post.get("relevance") or 0.0) + min(0.25, math.log10(total + 1) / 20.0)
+ return (post.get("relevance") or 0.0) + min(0.25, math.log10(max(0, total) + 1) / 20.0)
def _log(msg: str) -> None:
diff --git a/tests/test_reddit_relevance_ranking.py b/tests/test_reddit_relevance_ranking.py
index 8bf792dc..7d2c2f5b 100644
--- a/tests/test_reddit_relevance_ranking.py
+++ b/tests/test_reddit_relevance_ranking.py
@@ -8,8 +8,11 @@ on-topic posts rank first and pure zero-overlap posts are dropped when anything
relevant remains.
"""
+import math
from unittest import mock
+import pytest
+
from lib import reddit, reddit_keyless
@@ -36,6 +39,25 @@ class TestRelevanceRankKey:
off_topic = {"relevance": 0.0, "engagement": {"score": 99999, "num_comments": 4000}}
assert reddit_keyless._relevance_rank_key(on_topic) > reddit_keyless._relevance_rank_key(off_topic)
+ @pytest.mark.parametrize("key", [reddit._relevance_rank_key, reddit_keyless._relevance_rank_key])
+ def test_negative_engagement_scores_the_floor(self, key):
+ # Downvoted posts report a negative score; log10 of a non-positive total
+ # must not raise, and the bonus bottoms out at zero.
+ zero_total = {"relevance": 0.3, "engagement": {"score": -1, "num_comments": 0}}
+ negative_total = {"relevance": 0.3, "engagement": {"score": -10, "num_comments": 3}}
+ assert key(zero_total) == pytest.approx(0.3)
+ assert key(negative_total) == pytest.approx(0.3)
+
+ @pytest.mark.parametrize("key", [reddit._relevance_rank_key, reddit_keyless._relevance_rank_key])
+ def test_zero_engagement_adds_no_bonus(self, key):
+ item = {"relevance": 0.3, "engagement": {"score": 0, "num_comments": 0}}
+ assert key(item) == pytest.approx(0.3)
+
+ @pytest.mark.parametrize("key", [reddit._relevance_rank_key, reddit_keyless._relevance_rank_key])
+ def test_non_negative_engagement_bonus_unchanged(self, key):
+ item = {"relevance": 0.3, "engagement": {"score": 5, "num_comments": 2}}
+ assert key(item) == pytest.approx(0.3 + math.log10(8) / 20.0)
+
# --------------------------------------------------------------------------- #
# Keyed path (reddit.search_reddit)