Spaces:
Sleeping
Sleeping
File size: 11,716 Bytes
89b6a21 127bfda 89b6a21 127bfda 89b6a21 127bfda 89b6a21 127bfda 89b6a21 127bfda 89b6a21 127bfda 89b6a21 127bfda 89b6a21 127bfda 89b6a21 127bfda 89b6a21 127bfda 89b6a21 127bfda 89b6a21 127bfda 89b6a21 127bfda 89b6a21 127bfda 89b6a21 127bfda 89b6a21 | 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 | import re
from datetime import UTC, datetime
from urllib.parse import urlparse
from bot.byte_semantics_base import (
compact_message,
is_current_events_prompt,
is_high_risk_current_events_prompt,
)
from bot.byte_semantics_constants import (
CONFIDENCE_LABEL_TERMS,
CURRENT_EVENTS_DEFAULT_SOURCE,
CURRENT_EVENTS_PENDING_SOURCE,
MAX_CHAT_MESSAGE_LENGTH,
MAX_REPLY_LINES,
QUALITY_SAFE_FALLBACK,
SOURCE_ANCHOR_TERMS,
TEMPORAL_ANCHOR_TERMS,
UNCERTAINTY_HINT_TERMS,
)
from bot.logic import enforce_reply_limits, has_grounding_signal
def build_server_time_anchor_instruction(reference_utc_iso: str | None = None) -> str:
now_utc_iso = (reference_utc_iso or "").strip()
if not now_utc_iso:
now_utc = datetime.now(UTC)
now_utc_iso = now_utc.isoformat(timespec="seconds").replace("+00:00", "Z")
return f"Timestamp de referencia do servidor (UTC): {now_utc_iso}. Use esse horario para interpretar hoje/agora/nesta semana."
def build_verifiable_prompt(
prompt: str, concise_mode: bool = True, server_time_instruction: str | None = None
) -> str:
clean_prompt = (prompt or "").strip()
if not clean_prompt:
return clean_prompt
is_current_events = is_current_events_prompt(clean_prompt)
active_server_time_instruction = (
server_time_instruction or build_server_time_anchor_instruction()
)
if not concise_mode:
return (
f"{clean_prompt}\n"
f"{active_server_time_instruction} Instrucoes obrigatorias de confiabilidade: priorize fatos recentes e verificaveis. "
f"{'Considere prioridade alta para acontecimentos de hoje/agora. ' if is_current_events else ''}"
"Diferencie o que esta confirmado do que ainda e hipotese. "
"Se houver ambiguidade no nome (ex.: Push/Punch/Posh), avise explicitamente e peca 1 link para confirmar. "
"Se nao houver confirmacao forte, diga isso com clareza e solicite fonte."
)
if not is_current_events:
return clean_prompt
return (
f"{clean_prompt}\n"
f"{active_server_time_instruction} Instrucoes obrigatorias de confiabilidade: responda em ate 4 linhas curtas, densas e sem enrolacao. "
"Use apenas fatos com confirmacao recente em fontes confiaveis. "
"Inclua ancora temporal explicita (ex.: hoje, nesta semana, mes/ano) quando falar de status atual. "
"Para noticia/anuncio atual, inclua no final: 'Confianca: alta|media|baixa' e 'Fonte: <origem confiavel>'. "
"Diferencie o que esta confirmado do que ainda e rumor/hipotese. "
"Se houver ambiguidade no nome (ex.: Push/Punch/Posh), avise explicitamente e peca 1 link para confirmar. "
"Se nao houver confirmacao forte, responda exatamente: "
f"'{QUALITY_SAFE_FALLBACK}'"
)
def _extract_reference_utc_iso(server_time_instruction: str | None) -> str:
text = (server_time_instruction or "").strip()
if not text:
return ""
match = re.search(r"\b(\d{4}-\d{2}-\d{2}T\d{2}:\d{2}:\d{2}Z)\b", text)
return match.group(1) if match else ""
def _extract_grounding_queries(grounding_metadata: dict | None) -> list[str]:
if not isinstance(grounding_metadata, dict):
return []
queries = grounding_metadata.get("web_search_queries", [])
if not isinstance(queries, list):
return []
return [str(query).strip() for query in queries if str(query or "").strip()]
def _extract_grounding_source_urls(grounding_metadata: dict | None) -> list[str]:
if not isinstance(grounding_metadata, dict):
return []
source_urls = grounding_metadata.get("source_urls", [])
if not isinstance(source_urls, list):
return []
return [str(source_url).strip() for source_url in source_urls if str(source_url or "").strip()]
def _build_grounding_source_line(grounding_metadata: dict | None) -> str:
source_urls = _extract_grounding_source_urls(grounding_metadata)
if source_urls:
hosts: list[str] = []
for source_url in source_urls:
parsed_url = urlparse(source_url)
host = parsed_url.netloc.lower().removeprefix("www.")
if host and host not in hosts:
hosts.append(host)
if len(hosts) >= 2:
break
if hosts:
return f"Fonte: DuckDuckGo em {', '.join(hosts)}."
queries = _extract_grounding_queries(grounding_metadata)
if queries:
query_preview = compact_message(queries[0], max_len=90).strip().rstrip(" ,;:.")
if query_preview:
return f"Fonte: DuckDuckGo query: {query_preview}."
return CURRENT_EVENTS_DEFAULT_SOURCE
def _is_confidence_line(line: str) -> bool:
lowered = (line or "").strip().lower()
return lowered.startswith(("confianca:", "confiança:"))
def _is_source_line(line: str) -> bool:
lowered = (line or "").strip().lower()
return lowered.startswith("fonte:")
def _is_temporal_line(line: str) -> bool:
lowered = (line or "").strip().lower()
return lowered.startswith("recorte temporal:")
def _split_non_empty_lines(text: str) -> list[str]:
normalized = (text or "").replace("\r\n", "\n").replace("\r", "\n")
return [line.strip() for line in normalized.split("\n") if line.strip()]
def _is_canonical_high_risk_fallback(answer: str) -> bool:
lines = _split_non_empty_lines(answer)
if not lines or lines[0].lower() != QUALITY_SAFE_FALLBACK.lower():
return False
lowered_lines = [line.lower() for line in lines]
has_low_confidence = any(
line.startswith(("confianca: baixa", "confiança: baixa")) for line in lowered_lines
)
has_pending_source = any(
line == CURRENT_EVENTS_PENDING_SOURCE.lower() for line in lowered_lines
)
if not has_low_confidence or not has_pending_source:
return False
for line in lowered_lines[1:]:
if _is_temporal_line(line) or _is_confidence_line(line) or _is_source_line(line):
continue
return False
return True
def _fit_high_risk_current_events_reply(
body_lines: list[str], confidence_line: str, source_line: str, temporal_line: str = ""
) -> str:
normalized_body = [line.strip() for line in body_lines if line and line.strip()]
tail_lines = [
line.strip()
for line in (temporal_line, confidence_line, source_line)
if line and line.strip()
]
tail_text = "\n".join(tail_lines)
if not tail_text:
return enforce_reply_limits("\n".join(normalized_body))
if not normalized_body:
return tail_text
max_body_lines = max(1, MAX_REPLY_LINES - len(tail_lines))
max_body_length = MAX_CHAT_MESSAGE_LENGTH - len(tail_text) - 1
if max_body_length <= 0:
return tail_text[:MAX_CHAT_MESSAGE_LENGTH].strip()
body_text = enforce_reply_limits(
"\n".join(normalized_body), max_lines=max_body_lines, max_length=max_body_length
).strip()
combined = f"{body_text}\n{tail_text}" if body_text else tail_text
if len(combined) <= MAX_CHAT_MESSAGE_LENGTH:
return combined
compact_body = compact_message(" ".join(normalized_body), max_len=max_body_length).strip()
combined = f"{compact_body}\n{tail_text}" if compact_body else tail_text
if len(combined) <= MAX_CHAT_MESSAGE_LENGTH:
return combined
if max_body_length <= 0:
return tail_text[:MAX_CHAT_MESSAGE_LENGTH].strip()
cropped_body = compact_body[:max_body_length].rstrip(" ,;:")
return f"{cropped_body}\n{tail_text}" if cropped_body else tail_text
def normalize_current_events_reply_contract(
prompt: str,
answer: str,
server_time_instruction: str | None = None,
grounding_metadata: dict | None = None,
) -> str:
clean_prompt = (prompt or "").strip()
clean_answer = (answer or "").strip()
if not clean_prompt or not clean_answer or not is_current_events_prompt(clean_prompt):
return clean_answer
lines = _split_non_empty_lines(clean_answer)
if not lines:
lines = [" ".join(clean_answer.split())]
merged_answer = "\n".join(lines)
lowered_answer = merged_answer.lower()
has_uncertainty = any(term in lowered_answer for term in UNCERTAINTY_HINT_TERMS)
has_temporal_anchor = any(term in lowered_answer for term in TEMPORAL_ANCHOR_TERMS) or bool(
re.search(r"\b20(2[5-9]|3[0-9])\b", lowered_answer)
)
reference_utc_iso = _extract_reference_utc_iso(server_time_instruction)
added_temporal_line = ""
high_risk_current_events = is_high_risk_current_events_prompt(clean_prompt)
must_validate_grounding = grounding_metadata is not None
grounding_signal_present = has_grounding_signal(grounding_metadata)
if high_risk_current_events and has_uncertainty:
return build_current_events_safe_fallback_reply(
clean_prompt, server_time_instruction=server_time_instruction
)
if high_risk_current_events and must_validate_grounding and not grounding_signal_present:
return build_current_events_safe_fallback_reply(
clean_prompt, server_time_instruction=server_time_instruction
)
if not has_temporal_anchor and not has_uncertainty:
added_temporal_line = (
f"Recorte temporal: {reference_utc_iso} UTC."
if reference_utc_iso
else "Recorte temporal: agora (UTC)."
)
lines.append(added_temporal_line)
if not high_risk_current_events:
return "\n".join(lines)
body_lines: list[str] = []
confidence_line = ""
source_line = ""
for line in lines:
if added_temporal_line and line.strip() == added_temporal_line:
continue
if _is_confidence_line(line):
if not confidence_line:
confidence_line = line.strip()
continue
if _is_source_line(line):
if not source_line:
source_line = line.strip()
continue
body_lines.append(line.strip())
confidence_line = "Confianca: media"
source_line = _build_grounding_source_line(grounding_metadata)
return _fit_high_risk_current_events_reply(
body_lines, confidence_line, source_line, temporal_line=added_temporal_line
)
def build_current_events_safe_fallback_reply(
prompt: str, server_time_instruction: str | None = None
) -> str:
clean_prompt = (prompt or "").strip()
if not is_high_risk_current_events_prompt(clean_prompt):
return QUALITY_SAFE_FALLBACK
reference_utc_iso = _extract_reference_utc_iso(server_time_instruction)
temporal_line = (
f"Recorte temporal: {reference_utc_iso} UTC."
if reference_utc_iso
else "Recorte temporal: agora (UTC)."
)
return "\n".join(
[QUALITY_SAFE_FALLBACK, temporal_line, "Confianca: baixa", CURRENT_EVENTS_PENDING_SOURCE]
)
def has_current_events_source_anchor(text: str) -> bool:
lowered = (text or "").lower()
return any(term in lowered for term in SOURCE_ANCHOR_TERMS)
def has_current_events_confidence_label(text: str) -> bool:
lowered = (text or "").lower()
return any(term in lowered for term in CONFIDENCE_LABEL_TERMS)
def has_current_events_temporal_anchor(text: str) -> bool:
lowered = (text or "").lower()
return any(term in lowered for term in TEMPORAL_ANCHOR_TERMS) or bool(
re.search(r"\b20(2[5-9]|3[0-9])\b", lowered)
)
def has_current_events_uncertainty(text: str) -> bool:
lowered = (text or "").lower()
return any(term in lowered for term in UNCERTAINTY_HINT_TERMS)
def is_canonical_high_risk_fallback(answer: str) -> bool:
return _is_canonical_high_risk_fallback(answer)
|