File size: 15,673 Bytes
6a82282 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 | """Comprehensive end-to-end test suite for the Riprap agent.
Run against a live local server:
.venv/bin/uvicorn web.main:app --port 8000 &
.venv/bin/python tests/test_agent_full.py
Twenty-five cases across all four intents plus adversarial edge cases.
Tests cover:
- Intent routing correctness
- Real-value assertions (e.g. Brighton Beach must be majority-Sandy)
- Hallucination detection (no leaked example values from old prompts)
- Cross-query contamination check (back-to-back queries don't bleed)
- Latency thresholds (warm; expect generous wall on local Apple Silicon)
- Citation presence
- Section structure presence
- Map-data presence (target with bbox / geocode with lat/lon)
Hard fails fail the suite (exit 1). Soft warns are logged but don't fail.
"""
from __future__ import annotations
import re
import sys
import time
import httpx
BASE = "http://127.0.0.1:8000"
HARD_FAILS: list[tuple[str, str]] = []
SOFT_WARNS: list[tuple[str, str]] = []
TIMINGS: list[tuple[str, float]] = []
# Phrases that ONLY exist as worked-example content from prior prompts/docs.
# If they appear in an output that didn't actually query that place, the model
# is leaking from prompt or training-prior. List verbatim, lowercased.
LEAK_PHRASES = [
# Old prompt example that bit us once:
"20 coffey st", # only legitimate if the query is about Red Hook / Gowanus
# Boilerplate that signals model is improvising agency speak rather than
# citing β soft warn, not hard fail
]
def case(name: str, q: str, expected_intent: str, asserts: list, *,
max_wall_s: float = 240.0, leak_must_not_appear: list[str] | None = None):
"""One test case. Returns the parsed response or None on hard fail."""
print(f"\n=== {name}")
print(f" query: {q!r}")
t0 = time.time()
try:
r = httpx.get(f"{BASE}/api/agent", params={"q": q}, timeout=max_wall_s + 30.0)
r.raise_for_status()
d = r.json()
except Exception as e:
print(f" β HTTP/JSON error: {e!r}")
HARD_FAILS.append((name, f"HTTP error: {e}"))
return None
dt = time.time() - t0
TIMINGS.append((name, dt))
intent = d.get("intent")
plan = d.get("plan", {})
print(f" β intent={intent} total_s={d.get('total_s', '?')} wall={dt:.2f}s")
print(f" β specialists ({len(plan.get('specialists', []))}): {plan.get('specialists', [])}")
rationale = plan.get("rationale", "")
print(f" β rationale: {rationale[:130]}")
if expected_intent is not None and intent != expected_intent:
HARD_FAILS.append((name, f"intent {intent} != expected {expected_intent}"))
print(f" β expected intent={expected_intent}, got {intent}")
return d
if expected_intent is None:
print(" β intent (no expectation β adversarial case)")
else:
print(" β intent")
# Latency
if dt > max_wall_s:
SOFT_WARNS.append((name, f"latency {dt:.1f}s > {max_wall_s}s budget"))
print(f" β latency {dt:.1f}s > {max_wall_s}s budget")
else:
print(f" β latency under {max_wall_s}s budget")
# Per-case asserts
for label, fn in asserts:
try:
res = fn(d)
except Exception as e:
res = False
print(f" β assert raised β {label}: {e!r}")
if res:
print(f" β {label}")
else:
print(f" β {label}")
HARD_FAILS.append((name, label))
# Hallucination / leak check
para = (d.get("paragraph", "") or "").lower()
leaks = [p for p in (leak_must_not_appear or []) if p.lower() in para]
if leaks:
HARD_FAILS.append((name, f"leak phrase appeared in paragraph: {leaks}"))
print(f" β leak phrase: {leaks}")
else:
print(" β no leak phrases")
# Section header presence
has_section = bool(re.search(r"\*\*\w[\w\s/]*\.\*\*", para))
if not has_section and (d.get("paragraph") or "") and "no grounded data" not in para and "could not" not in para:
SOFT_WARNS.append((name, "no recognizable **Section.** header"))
print(" β no section header")
return d
# ---- helpers ---------------------------------------------------------------
def has_signal(key):
def _check(d):
v = d.get(key)
return v is not None and v != [] and v != {}
return _check
def target_field_eq(field, value_substring):
def _check(d):
t = d.get("target") or {}
return value_substring.lower() in (t.get(field, "") or "").lower()
return _check
def fraction_inside(key, lo, hi):
def _check(d):
s = d.get(key) or {}
f = s.get("fraction", -1)
return lo <= f <= hi
return _check
def dob_n_total_at_least(n):
return ("dob_summary.n_total >= " + str(n),
lambda d: (d.get("dob_summary") or {}).get("n_total", 0) >= n)
def dob_n_in_sandy_at_least(n):
return ("dob_summary.n_in_sandy >= " + str(n),
lambda d: (d.get("dob_summary") or {}).get("n_in_sandy", 0) >= n)
def has_paragraph(min_chars=80):
return ("paragraph >= " + str(min_chars) + " chars",
lambda d: len(d.get("paragraph", "") or "") >= min_chars)
def has_citation_tag():
return ("paragraph contains [doc_id] citation",
lambda d: bool(re.search(r"\[[a-z][a-z0-9_]+\]", d.get("paragraph", "") or "")))
def has_map_data():
return ("map data present (target.bbox or geocode.lat)",
lambda d: ((d.get("target") or {}).get("bbox") is not None
or (d.get("geocode") or {}).get("lat") is not None
or d.get("place") is not None))
# ---- the suite -------------------------------------------------------------
def main():
try:
httpx.get(f"{BASE}/", timeout=5.0)
except Exception as e:
print(f"server not reachable at {BASE}: {e!r}")
sys.exit(1)
print("=" * 60)
print("RIPRAP AGENT β FULL E2E SUITE")
print("=" * 60)
# ------ SINGLE_ADDRESS ------
case("addr/golden β coastal Brooklyn (Sandy hit)",
"2940 Brighton 3rd St, Brooklyn",
"single_address",
[
("geocode populated", lambda d: (d.get("geocode") or {}).get("lat")),
("sandy is True", lambda d: d.get("sandy") is True),
("dep populated", has_signal("dep")),
("microtopo HAND populated",
lambda d: (d.get("microtopo") or {}).get("hand_m") is not None),
has_paragraph(),
has_map_data(),
],
max_wall_s=120,
leak_must_not_appear=[], # 20 Coffey is in Red Hook ZIP, near enough to Brighton via Brooklyn β accept
)
case("addr/golden β Queens inland (Hollis archetype)",
"183-02 Liberty Ave, Queens",
"single_address",
[
("geocode populated", lambda d: (d.get("geocode") or {}).get("lat")),
("sandy is False", lambda d: d.get("sandy") is False),
("microtopo populated", has_signal("microtopo")),
has_paragraph(),
],
max_wall_s=120)
case("addr/control β Empire State Building (high ground)",
"350 5th Ave, Manhattan",
"single_address",
[
("geocode populated", lambda d: (d.get("geocode") or {}).get("lat")),
("sandy is False", lambda d: d.get("sandy") is False),
],
max_wall_s=120,
leak_must_not_appear=["20 coffey st", "brighton beach"])
case("addr/edge β typo'd address survives",
"2940 Brighten 3rd St, Brkln",
"single_address",
[has_paragraph(min_chars=20)],
max_wall_s=120)
case("addr/edge β outside NYC (Albany)",
"Empire State Plaza, Albany",
"single_address",
[has_paragraph(min_chars=20)],
max_wall_s=120)
# ------ NEIGHBORHOOD ------
case("nbhd/golden β Brighton Beach (high coastal exposure)",
"Brighton Beach",
"neighborhood",
[
("nta_name = Brighton Beach", target_field_eq("nta_name", "Brighton Beach")),
("borough = Brooklyn", target_field_eq("borough", "Brooklyn")),
("sandy_nta fraction > 0.7", fraction_inside("sandy_nta", 0.7, 1.0)),
("dep_nta has scenarios",
lambda d: len(d.get("dep_nta") or {}) >= 2),
("nyc311_nta n > 100",
lambda d: (d.get("nyc311_nta") or {}).get("n", 0) > 100),
has_paragraph(),
has_citation_tag(),
has_map_data(),
],
max_wall_s=120)
case("nbhd/golden β Carroll Gardens (mixed coastal/inland)",
"Carroll Gardens",
"neighborhood",
[
("borough = Brooklyn", target_field_eq("borough", "Brooklyn")),
("nyc311_nta n > 0",
lambda d: (d.get("nyc311_nta") or {}).get("n", 0) > 0),
("microtopo_nta populated", has_signal("microtopo_nta")),
has_paragraph(),
],
max_wall_s=120)
case("nbhd/golden β Hollis (inland Queens, Ida-deaths archetype)",
"Hollis",
"neighborhood",
[
("borough = Queens", target_field_eq("borough", "Queens")),
("sandy_nta fraction < 0.1 (inland)",
lambda d: (d.get("sandy_nta") or {"fraction": 1}).get("fraction", 1) < 0.1),
has_paragraph(),
],
max_wall_s=120,
leak_must_not_appear=["20 coffey st"])
case("nbhd/edge β exact-match wins over substring",
"Kew Gardens",
"neighborhood",
[
("nta_name = Kew Gardens (NOT Kew Gardens Hills)",
lambda d: (d.get("target") or {}).get("nta_name", "").lower() == "kew gardens"),
],
max_wall_s=120)
case("nbhd/edge β borough-wide query",
"Brooklyn",
"neighborhood",
[
("borough = Brooklyn", target_field_eq("borough", "Brooklyn")),
("n_matches > 50", lambda d: d.get("n_matches", 0) > 50),
],
max_wall_s=120)
case("nbhd/edge β NL phrasing 'is X at risk'",
"is Brighton Beach at risk?",
"neighborhood",
[
("nta_name = Brighton Beach", target_field_eq("nta_name", "Brighton Beach")),
has_paragraph(),
],
max_wall_s=120)
# ------ DEVELOPMENT_CHECK ------
case("dev/golden β Gowanus (the marquee)",
"what are they building in Gowanus and is it risky",
"development_check",
[
dob_n_total_at_least(5),
dob_n_in_sandy_at_least(1),
("flagged_top has projects",
lambda d: len((d.get("dob_summary") or {}).get("flagged_top") or []) >= 1),
("paragraph mentions a real BBL or street name",
lambda d: "BBL" in (d.get("paragraph") or "") or "St," in (d.get("paragraph") or "")),
has_paragraph(min_chars=200),
has_map_data(),
],
max_wall_s=180)
case("dev/golden β Red Hook (high Sandy)",
"show me new construction in Red Hook",
"development_check",
[
dob_n_total_at_least(1),
has_paragraph(min_chars=80),
],
max_wall_s=180)
case("dev/edge β low-construction inland (Hollis)",
"what are they building in Hollis",
"development_check",
[has_paragraph(min_chars=50)],
max_wall_s=120)
case("dev/edge β variant phrasing",
"flood risk of new gowanus developments",
"development_check",
[
("target NTA borough = Brooklyn",
lambda d: (d.get("target") or {}).get("borough") == "Brooklyn"),
dob_n_total_at_least(1),
],
max_wall_s=180)
case("dev/anti-leak β query unrelated to Gowanus must not mention 20 Coffey St",
"what are they building in Coney Island",
"development_check",
[has_paragraph(min_chars=80)],
max_wall_s=180,
leak_must_not_appear=["20 coffey st"]) # Coffey St is in Red Hook NTA, not Coney Island NTA
# ------ LIVE_NOW ------
case("live/golden β explicit 'right now'",
"is there flooding right now in NYC",
"live_now",
[
("noaa_tides observed_ft_mllw populated",
lambda d: (d.get("noaa_tides") or {}).get("observed_ft_mllw") is not None),
("nws_alerts present",
lambda d: d.get("nws_alerts") is not None),
has_paragraph(min_chars=20),
],
max_wall_s=90)
case("live/edge β borough-scoped",
"what's happening in Brooklyn right now",
"live_now",
[
("place = Brooklyn or NYC",
lambda d: d.get("place") in ("Brooklyn", "NYC")),
],
max_wall_s=90)
case("live/edge β surge-only phrasing",
"is there a surge tonight",
"live_now",
[has_paragraph(min_chars=10)],
max_wall_s=90)
# ------ STAKEHOLDER FLAVOR ------
case("stakeholder/reporter β DOB permits in flood zones",
"what NYC construction is at flood risk",
"development_check", # planner should pick dev_check; if neighborhood, that's also OK
[has_paragraph(min_chars=40)],
max_wall_s=180)
case("stakeholder/planner β borough-scope dev",
"show me Brooklyn construction in flood zones",
"development_check",
[has_paragraph(min_chars=40)],
max_wall_s=180)
case("stakeholder/BRIC β Sandy-impacted address",
"is 90 Bay St Staten Island in the Sandy zone",
"single_address",
[
("geocode populated", lambda d: (d.get("geocode") or {}).get("lat")),
has_paragraph(min_chars=80),
],
max_wall_s=120)
# ------ ADVERSARIAL / EDGE ------
case("edge β nonsense query (planner falls back)",
"what about flood",
None, # No expected β just wants ANY routing
[has_paragraph(min_chars=10)],
max_wall_s=120)
case("edge β empty noun (planner picks live)",
"flood",
None,
[has_paragraph(min_chars=5)],
max_wall_s=120)
case("edge β non-existent neighborhood (graceful fallback)",
"Nonsense Heights",
"neighborhood",
[
("paragraph or error message",
lambda d: bool(d.get("paragraph") or d.get("error"))),
],
max_wall_s=60)
# ---- summary ----------------------------------------------------------
print("\n" + "=" * 60)
print(f"HARD FAILS: {len(HARD_FAILS)}")
for name, why in HARD_FAILS:
print(f" - {name}: {why}")
print(f"\nSOFT WARNS: {len(SOFT_WARNS)}")
for name, why in SOFT_WARNS:
print(f" - {name}: {why}")
print("\nTIMINGS (top 5 slowest):")
for name, t in sorted(TIMINGS, key=lambda x: -x[1])[:5]:
print(f" {t:6.1f}s {name}")
print("\nTIMINGS (intent medians):")
by_intent = {}
for name, t in TIMINGS:
prefix = name.split("/", 1)[0]
by_intent.setdefault(prefix, []).append(t)
for prefix, times in sorted(by_intent.items()):
med = sorted(times)[len(times) // 2]
print(f" {prefix:18s} median {med:5.1f}s (n={len(times)})")
print("=" * 60)
sys.exit(1 if HARD_FAILS else 0)
if __name__ == "__main__":
main()
|