events: bursts of news items become one aggregate communication event (titles kept as evidence)

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

Simon-Pierre Boucher committed 1 mo ago (Sep 13, 2026) parent 75ba9ab Browse files

2 changed files +36 −3

modified src/companyatlas/services/events.py +23 −3

				`@@ -43,6 +43,7 @@ MAX_ENTITY_ITEMS = 50 # bounded entity lists on aggregate events`
43	PER_JOB_EVENT_MAX = 5 # NEW_JOB per job only when ≤ 5 jobs added	43	PER_JOB_EVENT_MAX = 5 # NEW_JOB per job only when ≤ 5 jobs added
44	PER_PERSON_EVENT_MAX = 10	44	PER_PERSON_EVENT_MAX = 10
45	NEWS_ITEMS_MAX = 20	45	NEWS_ITEMS_MAX = 20
		46	+NEWS_ITEMS_PER_EVENT_MAX = 3 # more items than this in one observation → one aggregate communication event
46	LEADERSHIP_AGGREGATE_MIN = 3	47	LEADERSHIP_AGGREGATE_MIN = 3
47	SURGE_MIN_JOBS = 10 # fallback thresholds when no baseline is available yet	48	SURGE_MIN_JOBS = 10 # fallback thresholds when no baseline is available yet
48	SURGE_MIN_RATIO = 0.5	49	SURGE_MIN_RATIO = 0.5
				`@@ -525,10 +526,29 @@ def _news_rules(delta: dict[str, Any], surface: str, evidence: str) -> list[Even`
525	prefix = {"NEWS_RELEASE": "News release", "BLOG_POST": "Blog post", "CHANGELOG_ENTRY": "Changelog entry", "INVESTOR_UPDATE": "Investor update",	526	prefix = {"NEWS_RELEASE": "News release", "BLOG_POST": "Blog post", "CHANGELOG_ENTRY": "Changelog entry", "INVESTOR_UPDATE": "Investor update",
526	"EARNINGS_RELEASE": "Earnings release"}	527	"EARNINGS_RELEASE": "Earnings release"}
527	out: list[EventDraft] = []	528	out: list[EventDraft] = []
528	− for item in list(news.get("added") or [])[:NEWS_ITEMS_MAX]:	529	+ added = [i for i in list(news.get("added") or []) if (i.get("title") or "").strip()]
		530	+ if len(added) > NEWS_ITEMS_PER_EVENT_MAX:
		531	+ # A burst of items in one observation (catalogue re-listing, archive page, many posts at once) is one communication event,
		532	+ # not a flood: individual titles stay in `entities.news` for the evidence drawer.
		533	+ subtypes = [_news_subtype(i, surface) for i in added]
		534	+ subtype = max(set(subtypes), key=subtypes.count)
		535	+ label = {"NEWS_RELEASE": "news releases", "BLOG_POST": "blog posts", "CHANGELOG_ENTRY": "changelog entries", "INVESTOR_UPDATE": "investor updates",
		536	+ "EARNINGS_RELEASE": "earnings releases"}[subtype]
		537	+ titles = [(i.get("title") or "").strip() for i in added]
		538	+ tags = ["news", subtype.lower()]
		539	+ if any(_is_ai(t) for t in titles):
		540	+ tags.append("ai")
		541	+ if any(_LAUNCH_RE.search(t) for t in titles):
		542	+ tags.append("launch")
		543	+ out.append(EventDraft(
		544	+ subtype=subtype, title=f"{len(added)} new {label} published on {_label(surface)}",
		545	+ entity_key=f"news_batch:{normalize_entity_key(titles[0])}:{len(added)}", evidence=evidence,
		546	+ summary="Latest: " + " · ".join(t[:80] for t in titles[:3]) + (" …" if len(titles) > 3 else ""),
		547	+ entities={"news": [{"title": i.get("title"), "url": i.get("url"), "published_at": i.get("published_at")} for i in added[:NEWS_ITEMS_MAX]]},
		548	+ payload={"count": len(added), "category": added[0].get("category")}, tags=tags, magnitude=min(1.0, len(added) / 20)))
		549	+ return out
		550	+ for item in added[:NEWS_ITEMS_MAX]:
529	title = (item.get("title") or "").strip()	551	title = (item.get("title") or "").strip()
530	− if not title:
531	− continue
532	subtype = _news_subtype(item, surface)	552	subtype = _news_subtype(item, surface)
533	tags = ["news", subtype.lower()]	553	tags = ["news", subtype.lower()]
534	if _is_ai(title):	554	if _is_ai(title):

modified tests/test_events_rules.py +13 −0

+ def test_kind_enum_values_used_by_rules():
+     assert ChangeKind.NOISE == "noise" and ChangeKind.CRITICAL == "critical"
++
++
++def test_news_bursts_become_one_aggregate_event() -> None:
++    from companyatlas.services.events import derive_events
++
++    items = [{"title": f"Release {i}: quarterly update", "url": f"https://x.com/news/{i}", "published_at": None, "category": "press"} for i in range(9)]
++    change = {"id": "chg_x", "sensor_id": "sen_x", "surface": "newsroom", "significance": 0.55, "kind": "meaningful", "diff": {}, "structured_delta": {"news": {"added": items}}}
++    drafts = derive_events(change, {"id": "co_x", "display_name": "X"}, {"id": "sen_x", "surface": "newsroom", "connector_id": "generic-html-v1", "url": "https://x.com/news"}).events
++    news = [d for d in drafts if d.subtype == "NEWS_RELEASE"]
++    assert len(news) == 1 and news[0].title.startswith("9 new news releases published on") and len(news[0].entities["news"]) == 9
++    small = dict(change, structured_delta={"news": {"added": items[:2]}})
++    drafts = derive_events(small, {"id": "co_x", "display_name": "X"}, {"id": "sen_x", "surface": "newsroom", "connector_id": "generic-html-v1", "url": "https://x.com/news"}).events
++    assert len([d for d in drafts if d.subtype == "NEWS_RELEASE"]) == 2

spb/company-atlas

events: bursts of news items become one aggregate communication event (titles kept as evidence)

Theme