wassname/moral-maps
1
0
Fork 0
mirror of https://github.com/wassname/moral-maps.git synced 2026-09-23 13:30:23 +08:00
Code Issues Packages Projects Releases Wiki Activity

Publish score-all-options map refresh and HLE data

Browse Source 
Co-Authored-By: PI[openai-codex] <288921227+claudypoo@users.noreply.github.com>
This commit is contained in:
wassname
2026-09-17 21:13:46 +08:00
co-authored by PI[openai-codex]
parent b444e1fcf7
commit bdf593e0de
14 changed files with 2272 additions and 314 deletions
Download Patch File Download Diff File Expand all files Collapse all files
scripts/wvs_react/uat.py
+47 -13
View File
@@ -51,10 +51,12 @@ def main() -> None:
if chrome is None:
raise RuntimeError("google-chrome is required for the Playwright UAT")
data = json.loads((ROOT / "docs/wvs/wvs_map_data.json").read_text())
assert data["schema"] == 2
assert data["capability_x"]["status"] == "blocked"
assert data["capability_x"]["source_url"] == "https://artificialanalysis.ai/models"
assert len(data["models"]) == 64
assert data["schema"] == 3
assert data["capability_x"]["status"] == "available"
assert data["capability_x"]["source_url"] == "https://artificialanalysis.ai/evaluations/humanitys-last-exam"
assert data["capability_x"]["fetched_utc"] == "2026-09-17T10:13:20Z"
capability_matched = [model for model in data["models"] if model["provenance"]["hle_score"] is not None]
assert len(capability_matched) == data["capability_x"]["matched_models"] == 46
assert len(data["countries"]) == 90
assert len(data["zone_hulls"]) == 4
assert len(data["latest_by_family"]) == 13
@@ -79,20 +81,23 @@ def main() -> None:
screenshot(page, f"{BASE}/", "wvs_react_root_playwright_default.png")
page.wait_for_selector("svg .model-mark")
assert page.title() == "Frontier LLMs on the World Values Survey"
assert page.title() == "Moral Maps: Where Do Frontier Models' Cultural Values Lie?"
assert page.locator("html").get_attribute("lang") == "en"
assert page.locator('meta[name="viewport"]').count() == 1
assert page.locator("h1").inner_text().startswith("How do AI models score on human values surveys?")
assert page.locator("h1").inner_text() == "Moral Maps: Where Do Frontier Models' Cultural Values Lie?"
assert page.locator(".lede").inner_text() == "We start with the World Values Survey, a map of human values across about ninety countries."
visible_copy = page.locator("main").inner_text()
for absent in ("React/SVG rendering", "Historical coordinates", "Dated releases only"):
assert absent not in visible_copy
assert "Artificial Analysis Intelligence Index, unavailable pending redistribution permission" in visible_copy
assert data["capability_x"]["blocker"] in visible_copy
assert "Artificial Analysis HLE score" in visible_copy
assert "redistribution permission" not in visible_copy
capability_option = page.get_by_label("Release-panel x axis").locator('option[value="capability"]')
assert capability_option.get_attribute("disabled") is not None
assert capability_option.get_attribute("disabled") is None
assert page.locator('.release-axis-selector a[href="https://artificialanalysis.ai/evaluations/humanitys-last-exam"]').count() == 1
assert "saved 2026-09-17" in page.locator(".release-axis-selector").inner_text()
layout_order = page.locator("main > *").evaluate_all("nodes => nodes.map(node => node.className.baseVal || node.className || node.tagName)")
assert layout_order.index("chart-shell") < layout_order.index("map-explanation") < layout_order.index("release-panels") < layout_order.index("caption")
assert page.locator(".release-panels .release-axis-selector").count() == 1
assert "weak descriptive correlations" in page.locator(".caption").inner_text()
assert "code and records" in page.locator(".caption").inner_text()
@@ -100,8 +105,9 @@ def main() -> None:
svg_description(page, "svg[data-median-x]", "map-svg-title", "map-svg-desc")
assert page.locator("path.zone").count() == 4
assert page.locator("polygon.zone").count() == 0
assert page.locator(".map-note").get_attribute("text-anchor") == "end"
assert float(page.locator(".map-note").get_attribute("x")) > float(page.locator(".map-title").get_attribute("x"))
assert page.locator(".map-note").get_attribute("text-anchor") == "start"
assert float(page.locator(".map-note").get_attribute("x")) == float(page.locator(".map-title").get_attribute("x"))
assert float(page.locator(".map-note").get_attribute("y")) > float(page.locator(".map-title").get_attribute("y"))
assert float(map_svg.get_attribute("data-median-x")) == data["median"]["x"]
assert float(map_svg.get_attribute("data-median-y")) == data["median"]["y"]
models_dom = page.locator(".model-mark").evaluate_all(
@@ -123,6 +129,10 @@ def main() -> None:
svg_description(page, '.release-panel[data-coordinate="y"] svg', "release-y-svg-title", "release-y-svg-desc")
svg_description(page, '.release-panel[data-coordinate="x"] svg', "release-x-svg-title", "release-x-svg-desc")
assert page.locator(".release-mark").count() == len(dated) * 2
assert page.locator(".frontier-label").count() > 0
frontier_boxes = page.locator('.frontier-label text').evaluate_all("nodes => nodes.map(node => { const b = node.getBBox(); return [b.x, b.y, b.width, b.height]; })")
assert all(x >= 96 and y >= 42 and x + width <= 1165 and y + height <= 272 for x, y, width, height in frontier_boxes)
assert all(a[0] + a[2] <= b[0] or b[0] + b[2] <= a[0] or a[1] + a[3] <= b[1] or b[1] + b[3] <= a[1] for index, a in enumerate(frontier_boxes) for b in frontier_boxes[index + 1:])
for grok_name in grok_dates:
assert page.locator(f'[data-release-model="{grok_name}"]').count() == 2
self_expression_panel = page.locator('.release-panel[data-coordinate="x"]')
@@ -144,6 +154,27 @@ def main() -> None:
assert panel.locator(".release-fit line").count() == 1
page.screenshot(path=OUT / "wvs_react_playwright_release_fit_all_families.png", full_page=True)
page.get_by_label("Release-panel x axis").select_option("capability")
assert page.locator('.release-panel[data-coordinate="y"] svg').get_attribute("data-axis-mode") == "capability"
assert page.locator('.release-mark').count() == len(capability_matched) * 2
assert all(int(panel.locator("svg").get_attribute("data-fit-n")) == len(capability_matched) for panel in page.locator(".release-panel").all())
assert all(int(panel.locator("svg").get_attribute("data-omitted-model-count")) == len(data["models"]) - len(capability_matched) for panel in page.locator(".release-panel").all())
page.screenshot(path=OUT / "wvs_react_playwright_capability_all_families.png", full_page=True)
page.get_by_role("button", name="qwen", exact=True).click()
assert all(int(panel.locator("svg").get_attribute("data-fit-n")) == sum(model["family"] != "qwen" for model in capability_matched) for panel in page.locator(".release-panel").all())
page.screenshot(path=OUT / "wvs_react_playwright_capability_qwen_hidden.png", full_page=True)
page.get_by_role("button", name="qwen", exact=True).click()
for family in sorted(families - {"gpt"}):
page.get_by_role("button", name=family, exact=True).click()
page.wait_for_timeout(100)
gpt_capability = sum(model["family"] == "gpt" for model in capability_matched)
gpt_fit_n = [int(panel.locator("svg").get_attribute("data-fit-n")) for panel in page.locator(".release-panel").all()]
assert gpt_fit_n == [gpt_capability, gpt_capability], (gpt_fit_n, gpt_capability)
page.screenshot(path=OUT / "wvs_react_playwright_capability_gpt_only.png", full_page=True)
for family in sorted(families - {"gpt"}):
page.get_by_role("button", name=family, exact=True).click()
page.get_by_label("Release-panel x axis").select_option("release-date")
page.get_by_role("button", name="qwen", exact=True).click()
page.wait_for_timeout(100)
assert map_svg.locator(':scope > g[data-family="qwen"]').get_attribute("display") == "none"
@@ -243,8 +274,11 @@ def main() -> None:
"qwen_toggle": "clicked, map and dated-panel family groups hidden, country coordinates invariant",
"tooltip": "map and release-panel pointer hover and focus show panel-specific model name, coordinate, and release date only",
"release_panels": {"panels": 2, "dated_models": len(dated), "date_order": "DOM order verified; both catalog-dated Grok models rendered in each panel",
"ols": "line, n, and R squared recompute from currently visible dated models; the Self-expression panel renders negative stored x, so upward means Self-expression; fewer than two distinct dates hide the fit"},
"copy": "one short linked WVS sentence above the map; history and axes below it; weak descriptive-fit method and code link below the release panels; capability selector remains visible but scores are absent pending redistribution permission",
"ols": "line, n, and R squared recompute from currently visible matched models; the Self-expression panel renders negative stored x, so upward means Self-expression; fewer than two distinct x values hide the fit"},
"capability_panels": {"matched_models": len(capability_matched), "omitted_models": len(data["models"]) - len(capability_matched),
"source": data["capability_x"]["source_url"], "fetched_utc": data["capability_x"]["fetched_utc"],
"selector": "enabled; all-family and Qwen-hidden fits checked"},
"copy": "one short linked WVS sentence above the map; history and axes below it; source credit beside the enabled lower-panel selector; weak descriptive-fit method and code link below the release panels",
"hashes": hashes,
}, indent=2))