sentimentstream-worker / scripts /benchmark_major_update.py
GitHub Action
deploy: worker release from GitHub
8ff1b66
#!/usr/bin/env python3
"""
Benchmark script for the major update detection heuristic.
Evaluates UpdateDetectionService._is_update_related, _collect_update_candidates,
and _is_major_update against a curated set of Steam games.
Three modes:
--discover Fetch news for all games (count=20 by default, matches
production) and display all items with classification
details. Use this to identify ground truth.
--evaluate Item-level evaluation: for each ItemCase, find the item
by gid and check if _is_update_related / _is_major_update
match expectations.
--evaluate-service Service-level evaluation: for each ServiceCase, run the
full selection pipeline and compare the outcome.
Both --evaluate and --evaluate-service run by default when no mode is specified.
Examples:
python scripts/benchmark_major_update.py --discover
python scripts/benchmark_major_update.py --discover --count 50
python scripts/benchmark_major_update.py --evaluate
python scripts/benchmark_major_update.py --evaluate-service
python scripts/benchmark_major_update.py # runs both evaluate modes
"""
from __future__ import annotations
import argparse
import sys
from dataclasses import dataclass
from datetime import datetime, timezone
from pathlib import Path
from typing import Literal
import httpx
# ── import project service ────────────────────────────────────────────────────
sys.path.insert(0, str(Path(__file__).parent.parent / "backend"))
from app.services.update_detection_service import UpdateDetectionService # noqa: E402
STEAM_NEWS_API_URL = "https://api.steampowered.com/ISteamNews/GetNewsForApp/v2/"
# ── benchmark games ───────────────────────────────────────────────────────────
GAMES: list[tuple[str, str]] = [
("Going Medieval", "1029780"),
("Timberborn", "1062090"),
("Hades II", "1145350"),
("Against the Storm", "1336490"),
("Valheim", "892970"),
("Manor Lords", "1363080"),
("Project Zomboid", "108600"),
("Dwarf Fortress", "975370"),
("Helldivers 2", "553850"),
("Deep Rock Galactic", "548430"),
("Lethal Company", "1966720"),
("Factorio", "427520"),
("Satisfactory", "526870"),
]
# ── ground truth structures ───────────────────────────────────────────────────
@dataclass
class ItemCase:
"""Per-item ground truth: is this specific event major?"""
game_name: str
appid: str
gid: str
title: str # for display
expected: Literal["major", "not_major", "ambiguous"]
reasoning: str
@dataclass
class ServiceCase:
"""Per-game ground truth: what should the production code do?"""
game_name: str
appid: str
expected_major: bool | None # True / False / None = ambiguous
reasoning: str
# ── item-level ground truth ───────────────────────────────────────────────────
# Populated from --discover run on 2026-03-19.
ITEM_CASES: list[ItemCase] = [
# ── Going Medieval ────────────────────────────────────────────────────────
ItemCase(
game_name="Going Medieval",
appid="1029780",
gid="1826992588604105",
title="Going Medieval is out now in 1.0!",
expected="major",
reasoning=(
"1.0 full release out of Early Access — unambiguously major. "
"Phase 1: RELEASE_PHRASE_RE matches 'is out now' → update-related. "
"ONE_ZERO_RE matches '1.0' → major."
),
),
ItemCase(
game_name="Going Medieval",
appid="1029780",
gid="1827626365751261",
title="Experimental Branch Patch (1.0.48)",
expected="not_major",
reasoning=(
"Experimental branch incremental patch. Three-segment version (1.0.48) "
"excluded by VERSION_RE. BRANCH_RE blocks major classification."
),
),
ItemCase(
game_name="Going Medieval",
appid="1029780",
gid="1827626365750723",
title="Patch Notes (1.0.47)",
expected="not_major",
reasoning="Incremental stable patch, three-segment version. not_major is correct.",
),
# ── Timberborn ────────────────────────────────────────────────────────────
ItemCase(
game_name="Timberborn",
appid="1062090",
gid="1826992588592887",
title="Timberborn 1.0 is live!",
expected="major",
reasoning=(
"1.0 full release out of Early Access — unambiguously major. "
"Phase 1: RELEASE_PHRASE_RE matches 'is live' → update-related. "
"ONE_ZERO_RE matches '1.0' → major."
),
),
ItemCase(
game_name="Timberborn",
appid="1062090",
gid="1826992588603124",
title="Patch notes 2026-03-17 (experimental)",
expected="not_major",
reasoning="Experimental branch date-based patch notes. No version number. not_major is correct.",
),
# ── Hades II ──────────────────────────────────────────────────────────────
ItemCase(
game_name="Hades II",
appid="1145350",
gid="1816215235360707",
title="Hades II v1.0 Hotfix 3",
expected="not_major",
reasoning=(
"A bugfix hotfix on top of the v1.0 launch — not a content update. "
"Phase 1: HOTFIX_RE blocks major classification. Correct: not_major."
),
),
ItemCase(
game_name="Hades II",
appid="1145350",
gid="1811772772516846",
title="Hades II v1.0 Hotfix 2",
expected="not_major",
reasoning="Same pattern: HOTFIX_RE blocks 'v1.0 Hotfix N' from being classified as major.",
),
ItemCase(
game_name="Hades II",
appid="1145350",
gid="1811772772248738",
title="Hades II v1.0 Is Now Available!",
expected="major",
reasoning=(
"v1.0 full launch — unambiguously major. "
"Phase 1: RELEASE_PHRASE_RE matches 'Is Now Available' → update-related. "
"No hotfix/branch blocker. VERSION_RE matches 'v1.0' → major."
),
),
# ── Against the Storm ─────────────────────────────────────────────────────
ItemCase(
game_name="Against the Storm",
appid="1336490",
gid="1818752592135840",
title="Demo Update 1.9.6",
expected="not_major",
reasoning=(
"Demo game update, three-segment version 1.9.6. "
"Service correctly classifies as not_major."
),
),
ItemCase(
game_name="Against the Storm",
appid="1336490",
gid="1816849002010836",
title="Brineworks Update (1.9) available!",
expected="major",
reasoning=(
"Named major content update with version 1.9. "
"Phase 1: VERSION_RE matches '1.9' + ACTION_WORD_RE matches 'Update'/'available' "
"→ update-related. VERSION_RE → major."
),
),
# ── Valheim ───────────────────────────────────────────────────────────────
ItemCase(
game_name="Valheim",
appid="892970",
gid="1825093633184197",
title="Patch 0.221.12",
expected="not_major",
reasoning="Three-segment maintenance patch. Correctly classified as not_major.",
),
ItemCase(
game_name="Valheim",
appid="892970",
gid="1809869179994587",
title="Patch 0.221.4 (Public Test)",
expected="not_major",
reasoning="Public test branch three-segment patch. Correctly classified as not_major.",
),
# ── Manor Lords ───────────────────────────────────────────────────────────
ItemCase(
game_name="Manor Lords",
appid="1363080",
gid="1827626365750540",
title="Major Update #6: Battlefield Changes, New Map, and Family Based Progression",
expected="major",
reasoning=(
"Developer-declared major content drop. "
"Phase 1: CONTENT_UPDATE_RE matches 'Major Update' → update-related and major."
),
),
ItemCase(
game_name="Manor Lords",
appid="1363080",
gid="1826992588603500",
title="New BETA version is available for testing (0.8.065)",
expected="not_major",
reasoning=(
"Beta/testing build announcement, not a production major update. "
"Current heuristic misses it entirely, which is acceptable for this benchmark case."
),
),
# ── Project Zomboid ───────────────────────────────────────────────────────
ItemCase(
game_name="Project Zomboid",
appid="108600",
gid="1826992588590120",
title="42.15.2 UNSTABLE HOTFIX Released",
expected="not_major",
reasoning=(
"Unstable-branch hotfix. patchnotes tag makes it update-related, "
"but HOTFIX_RE correctly blocks major classification."
),
),
ItemCase(
game_name="Project Zomboid",
appid="108600",
gid="1826362059930323",
title="Build 42.15.0 Unstable Released",
expected="not_major",
reasoning=(
"Unstable build release, not a production major update. "
"Current heuristic does not classify it as update-related because the three-segment "
"build number fails VERSION_RE."
),
),
# ── Dwarf Fortress ────────────────────────────────────────────────────────
ItemCase(
game_name="Dwarf Fortress",
appid="975370",
gid="1826362059918689",
title="Food fixes, AMA, community spotlight and more! Dwarf Fortress Patch 53.11",
expected="not_major",
reasoning=(
"Maintenance patch with Dwarf Fortress' two-segment numbering scheme. "
"Phase 2: PATCH_WORD_RE matches 'Patch'; MAINT_LANGUAGE_RE matches 'fixes' "
"→ maintenance blocker fires before VERSION_RE → not_major."
),
),
ItemCase(
game_name="Dwarf Fortress",
appid="975370",
gid="1821288646585998",
title="Aquatic portraits, Naked dwarf fix and more Dwarf Fortress Patch 53.10",
expected="not_major",
reasoning=(
"Another maintenance patch under the same numbering scheme. "
"Phase 2: PATCH_WORD_RE matches 'Patch'; MAINT_LANGUAGE_RE matches 'fix' "
"→ maintenance blocker fires → not_major."
),
),
# ── Helldivers 2 ──────────────────────────────────────────────────────────
ItemCase(
game_name="Helldivers 2",
appid="553850",
gid="1826992588603352",
title="Machinery of Oppression: 6.1.0",
expected="major",
reasoning=(
"Named content drop with new missions/enemies. This should count as a major update. "
"Useful to test whether named major drops with three-segment versions are still found."
),
),
ItemCase(
game_name="Helldivers 2",
appid="553850",
gid="1826992588603981",
title="Revealing our Machinery of Oppression Content Roadmap!",
expected="not_major",
reasoning=(
"Roadmap/announcement post, not the update itself. Should not be treated as major."
),
),
# ── Deep Rock Galactic ────────────────────────────────────────────────────
ItemCase(
game_name="Deep Rock Galactic",
appid="548430",
gid="1825727806720055",
title="'Eight Years in Orbit' Anniversary Event is live now!",
expected="not_major",
reasoning=(
"Live event announcement, not a game patch. "
"Phase 2: EVENT_FESTIVAL_RE matches 'anniversary event'; no 'update'/'patch' in title "
"→ UPDATE_OR_PATCH_RE guard fails → event blocker fires → not_major."
),
),
ItemCase(
game_name="Deep Rock Galactic",
appid="548430",
gid="1824644522847377",
title="Lunar Festival 2026 is now live!",
expected="not_major",
reasoning=(
"Seasonal event announcement, not a major patch/update. "
"Phase 2: EVENT_FESTIVAL_RE matches 'festival'; no 'update'/'patch' → event blocker fires → not_major."
),
),
# ── Lethal Company ────────────────────────────────────────────────────────
ItemCase(
game_name="Lethal Company",
appid="1966720",
gid="1800991756395986",
title="V70 - The Incubating Update",
expected="major",
reasoning=(
"Named major content update. "
"Phase 2: NAMED_VERSION_RE matches 'V70'; UPDATE_WORD_RE matches 'Update' "
"→ condition F makes it update-related; named version positive signal → major."
),
),
ItemCase(
game_name="Lethal Company",
appid="1966720",
gid="1801617199407807",
title="V72 Bug fix patch",
expected="not_major",
reasoning=(
"Small bug-fix patch. patchnotes tag makes it update-related. "
"Phase 2: PATCH_WORD_RE matches 'patch'; MAINT_LANGUAGE_RE matches 'bug fix' "
"→ maintenance blocker fires → not_major."
),
),
# ── Factorio ──────────────────────────────────────────────────────────────
ItemCase(
game_name="Factorio",
appid="427520",
gid="1827626365752749",
title="Version 2.0.76 released as stable",
expected="not_major",
reasoning=(
"Stable maintenance patch under a three-segment versioning scheme. "
"Useful as a clean true negative."
),
),
# ── Satisfactory ──────────────────────────────────────────────────────────
ItemCase(
game_name="Satisfactory",
appid="526870",
gid="1826992588604352",
title="Update 1.2 is out now on Experimental!",
expected="not_major",
reasoning=(
"Experimental-branch release, not a production major update. "
"Phase 2: extended BRANCH_RE matches 'on Experimental' → branch blocker fires → not_major."
),
),
ItemCase(
game_name="Satisfactory",
appid="526870",
gid="1825093633185794",
title="Experimental Hotfix v1.1.3.1",
expected="not_major",
reasoning=(
"Experimental hotfix on a three-segment version. Correct behavior is not_major."
),
),
]
# ── service-level ground truth ────────────────────────────────────────────────
# What SHOULD the production code do for this game given the current news window?
# Populated from --discover run on 2026-03-19.
# Phase 1 semantics: verdict based on is_major (major_date is not None), not on selected item title.
SERVICE_CASES: list[ServiceCase] = [
ServiceCase(
game_name="Going Medieval",
appid="1029780",
expected_major=True,
reasoning=(
"Game released 1.0 on 2026-03-17. Phase 1: 'is out now in 1.0!' matches "
"RELEASE_PHRASE_RE → update-related. ONE_ZERO_RE → major. "
"Expected: major_date is not None (TP)."
),
),
ServiceCase(
game_name="Timberborn",
appid="1062090",
expected_major=True,
reasoning=(
"Game reached 1.0 on 2026-03-12. Phase 1: '1.0 is live!' matches "
"RELEASE_PHRASE_RE → update-related. ONE_ZERO_RE → major. "
"Expected: major_date is not None (TP)."
),
),
ServiceCase(
game_name="Hades II",
appid="1145350",
expected_major=True,
reasoning=(
"Game launched v1.0 on 2025-09-25. Phase 1: 'v1.0 Is Now Available!' matches "
"RELEASE_PHRASE_RE → update-related (developer feed). VERSION_RE matches 'v1.0' → major. "
"Subsequent hotfixes (v1.0 Hotfix 2, 3) are correctly blocked by HOTFIX_RE. "
"major_date = v1.0 launch date, latest_update_date = most recent hotfix date. "
"Expected: major_date is not None (TP)."
),
),
ServiceCase(
game_name="Against the Storm",
appid="1336490",
expected_major=True,
reasoning=(
"'Brineworks Update (1.9) available!' is a named major content update. "
"Phase 1: VERSION_RE matches '1.9' + ACTION_WORD_RE matches 'Update'/'available' "
"→ update-related (developer feed). VERSION_RE → major. "
"Expected: major_date is not None (TP)."
),
),
ServiceCase(
game_name="Valheim",
appid="892970",
expected_major=False,
reasoning=(
"Top items are three-segment maintenance patches. "
"Correctly classified as not_major. TN."
),
),
ServiceCase(
game_name="Manor Lords",
appid="1363080",
expected_major=True,
reasoning=(
"Current window contains a clearly labeled 'Major Update #6' post. "
"Expected: major_date is not None."
),
),
ServiceCase(
game_name="Project Zomboid",
appid="108600",
expected_major=False,
reasoning=(
"Current window is dominated by unstable builds and hotfixes. "
"These should update activity, but should not count as major releases."
),
),
ServiceCase(
game_name="Dwarf Fortress",
appid="975370",
expected_major=False,
reasoning=(
"Current window contains only maintenance patches (53.11/53.10/53.09 plus hotfixes). "
"Phase 2: maintenance blocker (patch + fix language) correctly blocks all of them → no major_date."
),
),
ServiceCase(
game_name="Helldivers 2",
appid="553850",
expected_major=True,
reasoning=(
"Current window contains 'Machinery of Oppression: 6.1.0', a named content update. "
"Expected: major_date is not None."
),
),
ServiceCase(
game_name="Lethal Company",
appid="1966720",
expected_major=True,
reasoning=(
"Current window contains 'V70 - The Incubating Update', a named major content drop, "
"plus newer bug-fix patches. Phase 2: NAMED_VERSION_RE + UPDATE_WORD_RE detects V70 → major_date set."
),
),
ServiceCase(
game_name="Factorio",
appid="427520",
expected_major=False,
reasoning=(
"Current window contains only three-segment stable maintenance releases (2.0.x). "
"Expected: not_major."
),
),
ServiceCase(
game_name="Satisfactory",
appid="526870",
expected_major=False,
reasoning=(
"Current window contains an experimental 1.2 rollout and experimental hotfixes. "
"Phase 2: extended BRANCH_RE ('on Experimental') blocks the 1.2 rollout → no major_date."
),
),
]
# ── helpers ───────────────────────────────────────────────────────────────────
def _fmt_ts(ts: int | None) -> str:
if not ts:
return "—"
try:
return datetime.fromtimestamp(ts, tz=timezone.utc).strftime("%Y-%m-%d")
except (OSError, ValueError):
return "—"
def _fmt_dt(dt: datetime | None) -> str:
if dt is None:
return "—"
return dt.strftime("%Y-%m-%d")
def _trunc(s: str, n: int) -> str:
return (s[:n] + "…") if len(s) > n else s
def _fetch_news(client: httpx.Client, appid: str, count: int) -> list[dict]:
try:
resp = client.get(
STEAM_NEWS_API_URL,
params={"appid": appid, "count": count, "maxlength": 0},
)
if resp.status_code != 200:
print(f" [WARN] HTTP {resp.status_code} for appid {appid}", file=sys.stderr)
return []
data = resp.json()
return data.get("appnews", {}).get("newsitems", []) or []
except Exception as exc:
print(f" [WARN] Request failed for appid {appid}: {exc}", file=sys.stderr)
return []
# ── Mode 1: discover ──────────────────────────────────────────────────────────
def run_discover(count: int) -> None:
if count != 20:
print(f"NOTE: count={count} — beyond production window (prod uses count=20)\n")
col_idx = 4
col_gid = 20
col_date = 10
col_title = 40
col_fl = 16
col_tags = 24
col_ur = 9
col_maj = 7
header = (
f"{'#':<{col_idx}} "
f"{'gid':<{col_gid}} "
f"{'date':<{col_date}} "
f"{'title':<{col_title}} "
f"{'feedlabel':<{col_fl}} "
f"{'tags':<{col_tags}} "
f"{'upd_rel?':<{col_ur}} "
f"{'major?':<{col_maj}}"
)
sep = "-" * len(header)
with httpx.Client(timeout=30.0) as client:
for game_name, appid in GAMES:
print(f"\n{'=' * len(header)}")
print(f" {game_name} (appid={appid})")
print(f"{'=' * len(header)}")
print(header)
print(sep)
items = _fetch_news(client, appid, count)
if not items:
print(" (no items returned)")
continue
for idx, item in enumerate(items, start=1):
gid = str(item.get("gid") or "")[:col_gid]
date_str = _fmt_ts(item.get("date"))
title = _trunc(item.get("title", ""), col_title)
feedlabel = _trunc(item.get("feedlabel") or "", col_fl)
tags = _trunc(str(item.get("tags") or ""), col_tags)
is_ur = UpdateDetectionService._is_update_related(item)
is_maj = UpdateDetectionService._is_major_update(item)
ur_str = "Yes" if is_ur else "No"
maj_str = "Yes" if is_maj else "No"
print(
f"{idx:<{col_idx}} "
f"{gid:<{col_gid}} "
f"{date_str:<{col_date}} "
f"{title:<{col_title}} "
f"{feedlabel:<{col_fl}} "
f"{tags:<{col_tags}} "
f"{ur_str:<{col_ur}} "
f"{maj_str:<{col_maj}}"
)
latest_update_date, major_date = UpdateDetectionService._collect_update_candidates(items)
print(f"\n >> latest_update_date: {_fmt_dt(latest_update_date)} | major_date: {_fmt_dt(major_date)}")
verdict = "MAJOR" if major_date is not None else "not_major"
print(f" >> Service result: {verdict}")
# ── Mode 2: evaluate (item-level) ─────────────────────────────────────────────
def run_evaluate() -> None:
if not ITEM_CASES:
print("[evaluate] No item-level ground truth defined yet.")
print(" Run --discover first, then populate ITEM_CASES in this script.")
return
# Build lookup: appid → {gid → item}
gid_index: dict[str, dict[str, dict]] = {}
needed_appids = {case.appid for case in ITEM_CASES}
with httpx.Client(timeout=30.0) as client:
for appid in needed_appids:
items = _fetch_news(client, appid, count=20)
gid_index[appid] = {str(item.get("gid", "")): item for item in items}
tp = tn = fp = fn = amb = not_found = 0
rows: list[tuple] = []
for case in ITEM_CASES:
item = gid_index.get(case.appid, {}).get(case.gid)
if item is None:
not_found += 1
rows.append((case.game_name, case.title, "—", "—", "—", case.expected, "NOT FOUND"))
continue
is_ur = UpdateDetectionService._is_update_related(item)
is_maj = UpdateDetectionService._is_major_update(item)
predicted = "major" if (is_ur and is_maj) else "not_major"
expected = case.expected
if expected == "ambiguous":
verdict = "ambiguous"
amb += 1
elif predicted == expected:
verdict = "PASS"
if expected == "major":
tp += 1
else:
tn += 1
else:
if predicted == "major" and expected == "not_major":
verdict = "FAIL (FP)"
fp += 1
else:
verdict = "FAIL (FN)"
fn += 1
rows.append((
case.game_name,
_trunc(case.title, 30),
_fmt_ts(item.get("date")),
str(item.get("tags", ""))[:20],
item.get("feedlabel", "")[:16],
expected,
"Yes" if is_ur else "No",
"Yes" if is_maj else "No",
verdict,
))
# Print report
print("\n" + "=" * 110)
print("REPORT A — Item-level classification")
print("=" * 110)
hdr = f"{'Game':<18} {'Title':<30} {'Date':<10} {'Tags':<20} {'FeedLabel':<16} {'Expected':<10} {'UpdRel?':<8} {'Major?':<7} Verdict"
print(hdr)
print("-" * 110)
for row in rows:
if len(row) == 7:
print(f"{row[0]:<18} {row[1]:<30} {row[2]:<10} {'—':<20} {'—':<16} {row[5]:<10} {'—':<8} {'—':<7} {row[6]}")
else:
print(f"{row[0]:<18} {row[1]:<30} {row[2]:<10} {row[3]:<20} {row[4]:<16} {row[5]:<10} {row[6]:<8} {row[7]:<7} {row[8]}")
total = tp + tn + fp + fn
print("\nSummary:")
print(f" Total cases : {len(ITEM_CASES)} | not found: {not_found} | ambiguous: {amb}")
print(f" TP={tp} TN={tn} FP={fp} FN={fn}")
if total > 0:
prec = tp / (tp + fp) if (tp + fp) else float("nan")
recall = tp / (tp + fn) if (tp + fn) else float("nan")
acc = (tp + tn) / total
print(f" Precision={prec:.2f} Recall={recall:.2f} Accuracy={acc:.2f}")
fps = [c for c in ITEM_CASES if "FAIL (FP)" in str(rows[ITEM_CASES.index(c)])]
fns = [c for c in ITEM_CASES if "FAIL (FN)" in str(rows[ITEM_CASES.index(c)])]
if fps:
print("\nFalse Positives:")
for c in fps:
print(f" [{c.game_name}] {c.title!r}{c.reasoning}")
if fns:
print("\nFalse Negatives:")
for c in fns:
print(f" [{c.game_name}] {c.title!r}{c.reasoning}")
# ── Mode 3: evaluate-service (end-to-end) ─────────────────────────────────────
def run_evaluate_service() -> None:
if not SERVICE_CASES:
print("[evaluate-service] No service-level ground truth defined yet.")
print(" Run --discover first, then populate SERVICE_CASES in this script.")
return
tp = tn = fp = fn = amb = 0
rows: list[tuple] = []
with httpx.Client(timeout=30.0) as client:
for case in SERVICE_CASES:
items = _fetch_news(client, case.appid, count=20)
latest_update_date, major_date = UpdateDetectionService._collect_update_candidates(items)
is_maj = major_date is not None
latest_str = _fmt_dt(latest_update_date)
major_str = _fmt_dt(major_date)
maj_label = "Yes" if is_maj else "No"
if case.expected_major is None:
verdict = "ambiguous"
amb += 1
elif is_maj == case.expected_major:
verdict = "PASS"
if case.expected_major:
tp += 1
else:
tn += 1
else:
if is_maj and not case.expected_major:
verdict = "FAIL (FP)"
fp += 1
else:
verdict = "FAIL (FN)"
fn += 1
rows.append((
case.game_name,
latest_str,
major_str,
maj_label,
"True" if case.expected_major else ("None" if case.expected_major is None else "False"),
verdict,
))
print("\n" + "=" * 100)
print("REPORT B — Service-level (end-to-end)")
print("=" * 100)
hdr = f"{'Game':<18} {'LatestUpdate':<13} {'MajorDate':<11} {'Major?':<7} {'Expected':<9} Verdict"
print(hdr)
print("-" * 100)
for row in rows:
print(f"{row[0]:<18} {row[1]:<13} {row[2]:<11} {row[3]:<7} {row[4]:<9} {row[5]}")
total = tp + tn + fp + fn
print("\nSummary:")
print(f" Total games : {len(SERVICE_CASES)} | ambiguous: {amb}")
print(f" TP={tp} TN={tn} FP={fp} FN={fn}")
if total > 0:
prec = tp / (tp + fp) if (tp + fp) else float("nan")
recall = tp / (tp + fn) if (tp + fn) else float("nan")
acc = (tp + tn) / total
print(f" Precision={prec:.2f} Recall={recall:.2f} Accuracy={acc:.2f}")
for idx, case in enumerate(SERVICE_CASES):
verdict = rows[idx][5]
if verdict.startswith("FAIL"):
print(f"\n [{case.game_name}] {verdict}{case.reasoning}")
# ── main ──────────────────────────────────────────────────────────────────────
def _parse_args() -> argparse.Namespace:
p = argparse.ArgumentParser(
description="Benchmark the major update detection heuristic against real Steam games."
)
p.add_argument(
"--discover",
action="store_true",
help="Fetch news for all games and display per-item classification details.",
)
p.add_argument(
"--evaluate",
action="store_true",
help="Run item-level evaluation against ITEM_CASES ground truth.",
)
p.add_argument(
"--evaluate-service",
action="store_true",
dest="evaluate_service",
help="Run service-level end-to-end evaluation against SERVICE_CASES ground truth.",
)
p.add_argument(
"--count",
type=int,
default=20,
help="Number of news items to fetch (default: 20, matches production). "
"Values > 20 are beyond the production window.",
)
return p.parse_args()
def main() -> int:
args = _parse_args()
discover = args.discover
evaluate = args.evaluate
eval_svc = args.evaluate_service
# Default: run both evaluate modes when nothing is specified
if not discover and not evaluate and not eval_svc:
evaluate = True
eval_svc = True
if discover:
run_discover(count=args.count)
if evaluate:
run_evaluate()
if eval_svc:
run_evaluate_service()
return 0
if __name__ == "__main__":
raise SystemExit(main())