phase1/scripts/serve_agent.py

"""Moonley API service. Loads the legal tool registry once and streams grounded research over SSE.
The React interface is hosted only on Vercel; this process is the private HF backend. Run:
  MOONLEY_DATA=.../thor_artifacts MOONLEY_STATUTE=".../statute corpus" \
  .venv/bin/uvicorn --app-dir phase1/scripts serve_agent:app --host 127.0.0.1 --port 8001
"""
import os, sys, re, json, time, hashlib
from urllib.parse import unquote

def _promote_moonley_environment() -> None:
    """Let unchanged corpus internals consume canonical Moonley configuration."""
    for name, value in list(os.environ.items()):
        if name.startswith("MOONLEY_"):
            os.environ.setdefault("THEMIS_" + name[len("MOONLEY_"):], value)


def _load_env(path: str) -> None:
    if os.path.exists(path):
        for line in open(path):
            line = line.strip()
            if line and not line.startswith("#") and "=" in line:
                key, value = line.split("=", 1)
                os.environ.setdefault(key.strip(), value.strip().strip('"').strip("'"))


HERE = os.path.dirname(os.path.abspath(__file__))
_load_env(os.path.join(HERE, ".env"))
_promote_moonley_environment()

sys.path.insert(0, HERE)
from tools import Corpus as LegacyCorpus
from corpus_v5 import CorpusV5
import agent as A
import requests
from pdf_sources import PdfSourceResolver
from bharat_courts_source import BharatCourtsPdfError, resolve_and_fetch_pdf
from graph_view import graph_node_card
from project_store import ProjectStore, ProjectStoreError, QuotaExceeded
from drafting_service import (
    DRAFT_PROFILES,
    DraftingError,
    TemplateRegistry,
    apply_drafting_intake,
    draft_docx,
    draft_pdf,
    draft_profile,
    drafting_intake_messages,
    drafting_messages,
    finalization_messages,
    infer_draft_profile,
    missing_draft_fields,
    public_draft_profile,
    revision_messages,
    extract_uploaded_template,
)
from knowledge_service import KnowledgeService, KnowledgeServiceError
from research_release import build_research_release
from statute_crosswalk import ACT_NAMES, normalise_act, normalise_section
from fastapi import BackgroundTasks, FastAPI, Request
from fastapi.responses import FileResponse, StreamingResponse, JSONResponse, RedirectResponse, Response
from clerk_auth import (  # noqa: E402 - the local .env must be loaded first
    PUBLIC_PATHS,
    authenticate_clerk_request,
    clerk_settings,
    cors_origins,
    frontend_auth_config,
)
HDR = {"Authorization": f"Bearer {os.environ.get('DEEPSEEK_API_KEY','')}", "Content-Type": "application/json"}

def llm_fn(msgs):
    for _ in range(2):
        try:
            r = requests.post("https://api.deepseek.com/chat/completions", headers=HDR, timeout=60,
                json={"model": "deepseek-v4-flash", "temperature": 0, "max_tokens": 700,
                      "thinking": {"type": "disabled"}, "messages": msgs})
            if r.status_code == 200: return r.json()["choices"][0]["message"]["content"]
        except Exception: time.sleep(1)
    return "{}"

def fast_llm_fn(msgs):
    """Fast user-facing turn: fail closed instead of holding the interface through retries."""
    try:
        r = requests.post("https://api.deepseek.com/chat/completions", headers=HDR, timeout=30,
            json={"model": "deepseek-v4-flash", "temperature": 0, "max_tokens": 600,
                  "thinking": {"type": "disabled"}, "messages": msgs})
        if r.status_code == 200:
            return r.json()["choices"][0]["message"]["content"]
    except Exception:
        pass
    return "{}"

def ds_call(messages, tools):
    """DeepSeek function-calling turn -> the assistant message (with tool_calls or content)."""
    r = requests.post("https://api.deepseek.com/chat/completions", headers=HDR, timeout=90,
        json={"model": "deepseek-v4-flash", "temperature": 0, "max_tokens": 800,
              "thinking": {"type": "disabled"},
              "messages": messages, "tools": tools, "tool_choice": "auto"})
    return r.json()["choices"][0]["message"]

DATA_DIR = os.environ.get("THEMIS_DATA", ".")
STATUTE_DIR = os.environ.get("THEMIS_STATUTE", ".")
if os.path.exists(os.path.join(DATA_DIR, "release_manifest.json")):
    C = CorpusV5(DATA_DIR, STATUTE_DIR, device=os.environ.get("THEMIS_DEVICE", "cpu"))
    RUNTIME_KIND = "schema-v5-qwen"
else:
    C = LegacyCorpus(DATA_DIR, STATUTE_DIR, device=os.environ.get("THEMIS_DEVICE", "cpu"))
    RUNTIME_KIND = "legacy-bge"
RESEARCH_RELEASE = build_research_release(C, A)

# citation resolution for the judgment view (neutral + equivalent -> doc)
def norm_cite(c): return re.sub(r"\s+", " ", (c or "").replace(".", "")).strip().upper()
cite_resolver = {}; nc2doc = {}
for _d, _m in C.meta.items():
    if _m.get("neutral_citation"): nc2doc[_m["neutral_citation"]] = _d
    for _k in [_m.get("neutral_citation")] + (_m.get("equivalent_citations") or []):
        if _k: cite_resolver.setdefault(norm_cite(_k), _d)
CITE_RE = re.compile(r"\[\d{4}\]\s*\d+\s*S\.?C\.?R\.?\s*\d+|\(\d{4}\)\s*\d+\s*SCC\s*\d+|\d{4}\s+INSC\s+\d+|AIR\s+\d{4}\s+SC\s+\d+")

def doc_links(text, self_id):
    out = {}
    for c in CITE_RE.findall(text or ""):
        rid = cite_resolver.get(norm_cite(c))
        if rid and rid != self_id and C.is_retrieval_eligible(rid) and c not in out: out[c] = rid
    return [{"cite": k, "id": v} for k, v in out.items()]

def resolve_cited(cases_cited, self_id):
    out = []
    for c in (cases_cited or []):
        rid = None
        for cstr in (c.get("citations") or []):
            for part in re.split(r"\s*[:;]\s*", cstr):
                rid = cite_resolver.get(norm_cite(part))
                if rid and rid != self_id: break
                rid = None
            if rid: break
        if not rid and c.get("name"):
            hits = C.name_lookup(c["name"], 1)
            if hits and hits[0]["doc_id"] != self_id: rid = hits[0]["doc_id"]
        if rid:
            card = graph_node_card(
                rid,
                C.meta.get(rid, {}),
                treatment=c.get("treatment"),
                cited_by=C.cite_indeg.get(rid, 0),
                good_law_status=C.goodlaw.get(rid, {}).get(
                    "good_law_status", "unknown"
                ),
            )
            if not card["hover"]["case_name"] and c.get("name"):
                card["display_name"] = c["name"]
                card["name"] = c["name"]
                card["hover"]["case_name"] = c["name"]
            out.append(card)
        else:
            out.append(
                {
                    "name": c.get("name"),
                    "display_name": c.get("name") or "Unresolved cited case",
                    "citation": ((c.get("citations") or [""])[0]),
                    "treatment": c.get("treatment"),
                    "id": None,
                    "node_id": None,
                    "judgment_id": None,
                    "label": None,
                }
            )
    return out

# --- verified source PDFs -----------------------------------------------------
# Map membership is not availability: the upstream bucket contains a small number
# of application/pdf objects whose payload is actually an HTML error page. Probe a
# bounded byte range, expose the PDF only after its payload is verified, and let the
# browser load the public source directly so large PDFs retain byte-range support.
PDF_SOURCES = PdfSourceResolver(os.path.join(DATA_DIR, "escr_pdfmap.jsonl"))
print(f"[serve_agent] pdfmap: {PDF_SOURCES.mapped_count} unique judgments have mapped source candidates", flush=True)

app = FastAPI(title="Moonley API", description="Private grounded Indian legal research API", version="2")
RUNTIME_WARM = RUNTIME_KIND != "schema-v5-qwen"
PROJECTS = ProjectStore.from_env()
DRAFTING = TemplateRegistry(os.path.join(HERE, "..", "drafting"))
KNOWLEDGE = KnowledgeService(PROJECTS, C)

@app.on_event("startup")
def _warm_runtime():
    global RUNTIME_WARM
    if RUNTIME_KIND == "schema-v5-qwen" and os.environ.get("THEMIS_WARM_QUERY_MODEL", "1") == "1":
        C.warmup()
    RUNTIME_WARM = True
    print(f"[serve_agent] READY runtime={RUNTIME_KIND} accepted={len(C.eligible_doc_ids)}", flush=True)

# --- Clerk access gate (public hosting) ---
@app.middleware("http")
async def _clerk_gate(request: Request, call_next):
    # CORS preflights and the boot/config endpoints must be reachable before sign-in.
    if request.method != "OPTIONS" and request.url.path not in PUBLIC_PATHS:
        rejection = authenticate_clerk_request(request)
        if rejection is not None:
            return rejection
    return await call_next(request)

# CORS and Clerk's authorized-parties check share one explicit origin allow-list.
from fastapi.middleware.cors import CORSMiddleware
app.add_middleware(CORSMiddleware, allow_origins=cors_origins(), allow_methods=["*"],
                   allow_headers=["*"], expose_headers=["*"])

def sse(o): return "data: " + json.dumps(o, ensure_ascii=False) + "\n\n"

# --- SESSION CAPTURE (the pooled-verification machine for daily lawyer sessions) ---
# Every search + every 👍/👎 lands in append-only JSONL; each graded result is a future qrel row.
LOG_DIR = os.environ.get("THEMIS_LOG_DIR") or os.path.join(HERE, "..", "logs")
os.makedirs(LOG_DIR, exist_ok=True)
def _log(name, obj):
    try:
        obj = {"ts": time.strftime("%Y-%m-%dT%H:%M:%S"), **obj}
        with open(os.path.join(LOG_DIR, f"{name}.jsonl"), "a", encoding="utf-8") as fh:
            fh.write(json.dumps(obj, ensure_ascii=False) + "\n")
    except Exception:
        pass

LOG_RAW_QUERIES = os.environ.get("THEMIS_LOG_RAW_QUERIES", "0") == "1"
def _query_log_fields(query: str) -> dict:
    normalized = re.sub(r"\s+", " ", query or "").strip()
    fields = {
        "query_sha256": hashlib.sha256(normalized.encode("utf-8")).hexdigest(),
        "query_chars": len(normalized),
    }
    if LOG_RAW_QUERIES:
        fields["q"] = normalized[:2000]
    return fields

from pydantic import BaseModel, Field

class CaseChatTurn(BaseModel):
    role: str
    content: str

class QueryBriefRequest(BaseModel):
    query: str
    refinements: list[str] = Field(default_factory=list)
    history: list[CaseChatTurn] = Field(default_factory=list)
    active_case_id: str | None = None
    recent_case_ids: list[str] = Field(default_factory=list)

class CaseChatRequest(BaseModel):
    doc_id: str
    question: str
    history: list[CaseChatTurn] = Field(default_factory=list)

class SearchRequest(BaseModel):
    q: str
    original_q: str | None = None
    approved: bool = True
    search_frame: dict | None = None
    brief_revision: int | None = None
    route: str = "legal_research"
    retrieval_scope: str = "global"
    active_case_id: str | None = None
    recent_case_ids: list[str] = Field(default_factory=list)
    history: list[CaseChatTurn] = Field(default_factory=list)
    case_question: str | None = None

class ProjectRequest(BaseModel):
    name: str

class DraftChatSource(BaseModel):
    id: str = Field(default="", max_length=200)
    title: str = Field(default="Saved chat", max_length=120)
    content: str = Field(max_length=16_000)

class DraftMatterDetails(BaseModel):
    matter_title: str = Field(default="", max_length=500)
    parties: str = Field(default="", max_length=4_000)
    lower_court: str = Field(default="", max_length=500)
    case_number: str = Field(default="", max_length=300)
    impugned_order_date: str = Field(default="", max_length=100)
    synopsis: str = Field(default="", max_length=8_000)
    list_of_dates: str = Field(default="", max_length=8_000)
    questions_of_law: str = Field(default="", max_length=8_000)
    grounds: str = Field(default="", max_length=8_000)
    relief: str = Field(default="", max_length=8_000)
    advocate: str = Field(default="", max_length=500)

class DraftRequest(BaseModel):
    template_id: str = Field(default="", max_length=100)
    template_text: str = Field(default="", max_length=60_000)
    document_type: str = Field(default="", max_length=100)
    project_id: str | None = None
    document_ids: list[str] = Field(default_factory=list)
    chat_sources: list[DraftChatSource] = Field(default_factory=list)
    matter_details: DraftMatterDetails = Field(default_factory=DraftMatterDetails)
    intake_details: dict[str, str] = Field(default_factory=dict)
    instructions: str = Field(default="", max_length=6_000)

class DraftExportRequest(BaseModel):
    title: str = Field(default="Moonley working draft", max_length=180)
    draft: str = Field(max_length=80_000)

class DraftIntakeTurn(BaseModel):
    role: str = Field(max_length=20)
    content: str = Field(max_length=2_000)

class DraftIntakeRequest(BaseModel):
    message: str = Field(max_length=4_000)
    document_type: str = Field(default="", max_length=100)
    details: dict[str, str] = Field(default_factory=dict)
    history: list[DraftIntakeTurn] = Field(default_factory=list)

class DraftFinalizeRequest(DraftExportRequest):
    document_type: str = Field(default="", max_length=100)

class DraftRevisionRequest(DraftFinalizeRequest):
    instruction: str = Field(max_length=4_000)

def _project_owner(request: Request) -> str:
    return str(getattr(request.state, "clerk_user_id", ""))

def _project_error(exc: ProjectStoreError) -> JSONResponse:
    return JSONResponse(
        {"error": exc.code, "message": exc.message},
        status_code=exc.status_code,
        headers={"Cache-Control": "no-store"},
    )

@app.get("/api/v2/auth/config")
def auth_config():
    return frontend_auth_config()

@app.get("/api/v2/projects")
def list_projects(request: Request):
    try:
        projects = PROJECTS.list_projects(_project_owner(request))
        return JSONResponse(
            {"projects": projects, "storage": PROJECTS.status()},
            headers={"Cache-Control": "no-store"},
        )
    except ProjectStoreError as exc:
        return _project_error(exc)

@app.post("/api/v2/projects")
def create_project(request: Request, body: ProjectRequest):
    try:
        project = PROJECTS.create_project(_project_owner(request), body.name)
        return JSONResponse(
            {"project": project, "storage": PROJECTS.status()},
            status_code=201,
            headers={"Cache-Control": "no-store"},
        )
    except ProjectStoreError as exc:
        return _project_error(exc)

@app.get("/api/v2/projects/{project_id}")
def get_project(project_id: str, request: Request):
    try:
        return JSONResponse(
            {"project": PROJECTS.get_project(_project_owner(request), project_id)},
            headers={"Cache-Control": "no-store"},
        )
    except ProjectStoreError as exc:
        return _project_error(exc)

@app.patch("/api/v2/projects/{project_id}")
def rename_project(project_id: str, request: Request, body: ProjectRequest):
    try:
        return JSONResponse(
            {"project": PROJECTS.rename_project(_project_owner(request), project_id, body.name)},
            headers={"Cache-Control": "no-store"},
        )
    except ProjectStoreError as exc:
        return _project_error(exc)

@app.delete("/api/v2/projects/{project_id}")
def delete_project(project_id: str, request: Request):
    try:
        KNOWLEDGE.delete_project(_project_owner(request), project_id)
        PROJECTS.delete_project(_project_owner(request), project_id)
        return Response(status_code=204, headers={"Cache-Control": "no-store"})
    except ProjectStoreError as exc:
        return _project_error(exc)
    except KnowledgeServiceError as exc:
        return JSONResponse(
            {"error": "knowledge_delete_failed", "message": str(exc)},
            status_code=502,
            headers={"Cache-Control": "no-store"},
        )

@app.post("/api/v2/projects/{project_id}/documents")
async def upload_project_document(project_id: str, request: Request, background_tasks: BackgroundTasks):
    try:
        content_length = request.headers.get("content-length", "").strip()
        if content_length and int(content_length) > PROJECTS.limits.max_file_bytes:
            raise QuotaExceeded(
                f"Each document must be {PROJECTS.limits.max_file_bytes // (1024 * 1024)} MiB or smaller."
            )
        filename = unquote(request.headers.get("x-document-name", ""))
        content = bytearray()
        async for chunk in request.stream():
            content.extend(chunk)
            if len(content) > PROJECTS.limits.max_file_bytes:
                raise QuotaExceeded(
                    f"Each document must be {PROJECTS.limits.max_file_bytes // (1024 * 1024)} MiB or smaller."
                )
        document = PROJECTS.add_document(_project_owner(request), project_id, filename, bytes(content))
        background_tasks.add_task(
            KNOWLEDGE.ingest, _project_owner(request), project_id, document["id"]
        )
        return JSONResponse(
            {"document": document, "project": PROJECTS.get_project(_project_owner(request), project_id)},
            status_code=201,
            headers={"Cache-Control": "no-store"},
        )
    except (ValueError, ProjectStoreError) as exc:
        if isinstance(exc, ProjectStoreError):
            return _project_error(exc)
        return JSONResponse({"error": "invalid_content_length"}, status_code=400)

@app.delete("/api/v2/projects/{project_id}/documents/{document_id}")
def delete_project_document(project_id: str, document_id: str, request: Request):
    try:
        KNOWLEDGE.delete(_project_owner(request), project_id, document_id)
        PROJECTS.delete_document(_project_owner(request), project_id, document_id)
        return Response(status_code=204, headers={"Cache-Control": "no-store"})
    except ProjectStoreError as exc:
        return _project_error(exc)
    except KnowledgeServiceError as exc:
        return JSONResponse(
            {"error": "knowledge_delete_failed", "message": str(exc)},
            status_code=502,
            headers={"Cache-Control": "no-store"},
        )

@app.get("/api/v2/projects/{project_id}/documents/{document_id}")
def download_project_document(project_id: str, document_id: str, request: Request):
    try:
        document = PROJECTS.document_record(_project_owner(request), project_id, document_id)
        path = PROJECTS.document_path(_project_owner(request), project_id, document_id)
        return FileResponse(
            path,
            media_type=document.get("media_type") or "application/octet-stream",
            filename=document.get("name") or "document",
            headers={"Cache-Control": "private, no-store"},
        )
    except ProjectStoreError as exc:
        return _project_error(exc)

@app.post("/api/v2/projects/{project_id}/documents/{document_id}/ingest", status_code=202)
def ingest_project_document(
    project_id: str, document_id: str, request: Request, background_tasks: BackgroundTasks
):
    try:
        PROJECTS.document_record(_project_owner(request), project_id, document_id)
        background_tasks.add_task(KNOWLEDGE.ingest, _project_owner(request), project_id, document_id)
        return JSONResponse(
            {"status": "queued", "document_id": document_id},
            status_code=202,
            headers={"Cache-Control": "no-store"},
        )
    except ProjectStoreError as exc:
        return _project_error(exc)

@app.get("/api/v2/statute-crosswalk")
def statute_crosswalk(act: str, section: str):
    code, number = normalise_act(act), normalise_section(section)
    if not code or not number:
        return JSONResponse(
            {
                "error": "invalid_provision",
                "message": "Choose IPC, BNS, CrPC, BNSS, IEA, or BSA and enter a section number.",
            },
            status_code=400,
        )
    result = C.statute_crosswalk(code, number)
    result["provision"] = C.statute_provision(code, number)
    for item in result.get("corresponding") or []:
        item["provision"] = C.statute_provision(item["act"], item["section"])
    result["supported_acts"] = [
        {"act": value, "name": ACT_NAMES[value]} for value in ("IPC", "BNS", "CRPC", "BNSS", "IEA", "BSA")
    ]
    return JSONResponse(result, headers={"Cache-Control": "private, max-age=3600"})

@app.get("/api/v2/statute-lookup")
def statute_lookup(act: str, section: str):
    """Exact statutory-text lookup; never infer or substitute another section."""
    code, number = normalise_act(act), normalise_section(section)
    if not code or not number:
        return JSONResponse(
            {"found": False, "error": "invalid_provision", "message": "Use IPC, BNS, CrPC, BNSS, IEA, or BSA with an exact section number."},
            status_code=400,
        )
    provision = C.statute_provision(code, number)
    if not provision:
        return JSONResponse(
            {"found": False, "act": code, "section": number, "message": "The exact provision is not available in the private statute source; do not guess it."},
            headers={"Cache-Control": "private, max-age=300"},
        )
    return JSONResponse(
        {"found": True, "act": code, "act_name": ACT_NAMES.get(code), "section": number, "provision": provision},
        headers={"Cache-Control": "private, max-age=3600"},
    )

@app.get("/api/v2/drafting/templates")
def drafting_templates():
    return JSONResponse(
        {
            "version": DRAFTING.version,
            "templates": DRAFTING.list(),
            "document_types": [
                public_draft_profile(profile)
                for profile in DRAFT_PROFILES.values()
            ],
            "knowledge": KNOWLEDGE.status(),
        },
        headers={"Cache-Control": "private, max-age=300"},
    )

@app.get("/api/v2/drafting/templates/{template_id}/pdf")
def drafting_template_pdf(template_id: str):
    try:
        template = DRAFTING.get(template_id)
        return FileResponse(
            template["path"],
            media_type="application/pdf",
            filename=template["filename"],
            headers={"Cache-Control": "private, max-age=3600"},
        )
    except DraftingError as exc:
        return JSONResponse({"error": "template_not_found", "message": str(exc)}, status_code=404)

@app.get("/api/v2/drafting/templates/{template_id}/text")
def drafting_template_text(template_id: str):
    try:
        template = DRAFTING.get(template_id)
        return JSONResponse(
            {
                "template": {
                    key: value
                    for key, value in template.items()
                    if key not in {"path", "filename"}
                },
                "text": DRAFTING.text(template_id),
                "editable": True,
            },
            headers={"Cache-Control": "private, no-store"},
        )
    except DraftingError as exc:
        return JSONResponse({"error": "template_not_found", "message": str(exc)}, status_code=404)

@app.post("/api/v2/drafting/templates/extract")
async def extract_private_drafting_template(request: Request):
    """Extract one authenticated user's template without retaining the uploaded file."""
    max_bytes = 10 * 1024 * 1024
    content_length = request.headers.get("content-length", "").strip()
    if content_length and int(content_length) > max_bytes:
        return JSONResponse(
            {"error": "template_too_large", "message": "Template must be 10 MiB or smaller."},
            status_code=413,
            headers={"Cache-Control": "no-store"},
        )
    filename = unquote(request.headers.get("x-template-name", ""))
    content = bytearray()
    async for chunk in request.stream():
        content.extend(chunk)
        if len(content) > max_bytes:
            return JSONResponse(
                {"error": "template_too_large", "message": "Template must be 10 MiB or smaller."},
                status_code=413,
                headers={"Cache-Control": "no-store"},
            )
    try:
        extracted = extract_uploaded_template(filename, bytes(content), request.headers.get("content-type", ""))
        return JSONResponse(
            extracted,
            headers={"Cache-Control": "no-store"},
        )
    except DraftingError as exc:
        return JSONResponse(
            {"error": "template_extraction_failed", "message": str(exc)},
            status_code=400,
            headers={"Cache-Control": "no-store"},
        )

def draft_llm_fn(messages, *, max_tokens: int = 5000, timeout: int = 150):
    try:
        response = requests.post(
            "https://api.deepseek.com/chat/completions",
            headers=HDR,
            timeout=timeout,
            json={
                "model": "deepseek-v4-flash",
                "temperature": 0,
                "max_tokens": max_tokens,
                "thinking": {"type": "disabled"},
                "messages": messages,
            },
        )
        if response.status_code == 200:
            return str(response.json()["choices"][0]["message"]["content"] or "").strip()
    except Exception:
        pass
    return ""

@app.post("/api/v2/drafting/intake")
def drafting_intake(request: Request, body: DraftIntakeRequest):
    message = re.sub(r"\x00", "", body.message or "").strip()[:4_000]
    if not message:
        return JSONResponse(
            {"error": "empty_message", "message": "Tell Moonley what you want drafted."},
            status_code=400,
        )
    current_profile = draft_profile(body.document_type)
    prompt_profile = current_profile or draft_profile(infer_draft_profile(message))
    messages = drafting_intake_messages(
        message,
        prompt_profile,
        body.details,
        [turn.dict() for turn in body.history[-8:]],
    )
    output = draft_llm_fn(messages, max_tokens=800, timeout=60)
    model_returned = bool(output)
    if not output and current_profile:
        missing = missing_draft_fields(current_profile, body.details)
        if missing:
            output = json.dumps(
                {
                    "document_type": current_profile["id"],
                    "updates": {missing[0]["key"]: message},
                    "acknowledgement": "Noted.",
                }
            )
    state = apply_drafting_intake(
        message,
        body.document_type,
        body.details,
        output,
    )
    _log(
        "drafting_intake",
        {
            "owner_sha256": hashlib.sha256(_project_owner(request).encode("utf-8")).hexdigest(),
            "document_type": state.get("document_type"),
            "detail_count": len(state.get("details") or {}),
            "ready": bool(state.get("ready")),
            "model_returned": model_returned,
        },
    )
    return JSONResponse(
        {
            **state,
            "model_call": {
                "provider": "deepseek",
                "attempted": True,
                "succeeded": model_returned,
            },
        },
        headers={"Cache-Control": "no-store"},
    )

@app.post("/api/v2/drafting/generate")
def generate_draft(request: Request, body: DraftRequest):
    owner = _project_owner(request)
    try:
        profile = draft_profile(body.document_type)
        if profile:
            missing = missing_draft_fields(profile, body.intake_details)
            if missing:
                return JSONResponse(
                    {
                        "error": "draft_intake_incomplete",
                        "message": f"Complete the drafting chat first: {missing[0]['label']} is still required.",
                        "missing_fields": [field["key"] for field in missing],
                    },
                    status_code=409,
                )
            template_id = body.template_id or str(profile.get("template_id") or "")
        else:
            template_id = body.template_id
        if template_id:
            template = DRAFTING.get(template_id)
            edited_template = re.sub(r"\x00", "", body.template_text or "").strip()[:60_000]
            template_text = edited_template or DRAFTING.text(template_id)
        elif profile:
            template = {
                "id": profile["id"],
                "title": profile["title"],
                "description": profile["description"],
                "category": "Chat-led",
            }
            edited_template = re.sub(r"\x00", "", body.template_text or "").strip()[:60_000]
            template_text = edited_template or str(profile.get("structure") or "")
        else:
            raise DraftingError("Tell Moonley what document to draft first.")
        sources = []
        document_ids = list(dict.fromkeys(body.document_ids))[:8]
        if document_ids and not body.project_id:
            raise DraftingError("Choose the project that owns the selected documents.")
        for document_id in document_ids:
            document = PROJECTS.document_record(owner, body.project_id or "", document_id)
            text, extraction = KNOWLEDGE.source_text(owner, body.project_id or "", document_id)
            if text:
                sources.append(
                    {
                        "label": f"Project document: {document.get('name')}",
                        "text": text,
                        "kind": "document",
                        "document_id": document_id,
                        "extraction": extraction,
                    }
                )
        for chat in body.chat_sources[:5]:
            text = re.sub(r"\x00", "", chat.content or "").strip()[:16_000]
            if text:
                sources.append({"label": f"Selected chat: {chat.title[:120]}", "text": text, "kind": "chat"})
        messages = drafting_messages(
            template,
            template_text,
            body.instructions,
            sources,
            body.matter_details.dict(),
            intake_details=body.intake_details,
            profile=profile,
        )
        draft = draft_llm_fn(messages)
        if not draft:
            return JSONResponse(
                {"error": "draft_generation_unavailable", "message": "The drafting model did not return a draft. Try again."},
                status_code=503,
            )
        _log(
            "drafting",
            {
                "owner_sha256": hashlib.sha256(owner.encode("utf-8")).hexdigest(),
                "template_id": template_id or profile.get("id"),
                "document_type": body.document_type,
                "document_count": len(document_ids),
                "chat_count": len(body.chat_sources[:5]),
            },
        )
        return JSONResponse(
            {
                "draft": draft[:80_000],
                "template": {key: value for key, value in template.items() if key not in {"path", "filename"}},
                "sources": [
                    {key: value for key, value in source.items() if key not in {"text"}}
                    for source in sources
                ],
                "notice": "Working draft only. Verify every fact, authority, annexure and filing requirement before use.",
            },
            headers={"Cache-Control": "no-store"},
        )
    except ProjectStoreError as exc:
        return _project_error(exc)
    except DraftingError as exc:
        return JSONResponse({"error": "invalid_draft_request", "message": str(exc)}, status_code=400)

@app.post("/api/v2/drafting/finalize")
def finalize_draft(request: Request, body: DraftFinalizeRequest):
    draft = re.sub(r"\x00", "", body.draft or "").strip()[:80_000]
    if not draft:
        return JSONResponse(
            {"error": "empty_draft", "message": "Generate or enter a draft before finalizing."},
            status_code=400,
        )
    profile = draft_profile(body.document_type)
    final = draft_llm_fn(
        finalization_messages(body.title, draft, profile),
        max_tokens=6_000,
        timeout=150,
    )
    if not final:
        return JSONResponse(
            {"error": "finalization_unavailable", "message": "The drafting model did not return a final version. Your editable draft is unchanged."},
            status_code=503,
        )
    _log(
        "drafting_finalize",
        {
            "owner_sha256": hashlib.sha256(_project_owner(request).encode("utf-8")).hexdigest(),
            "document_type": body.document_type,
            "input_chars": len(draft),
        },
    )
    return JSONResponse(
        {
            "draft": final[:80_000],
            "notice": "Finalized working draft only. Counsel must verify the record, law and filing requirements.",
        },
        headers={"Cache-Control": "no-store"},
    )

@app.post("/api/v2/drafting/revise")
def revise_draft(request: Request, body: DraftRevisionRequest):
    draft = re.sub(r"\x00", "", body.draft or "").strip()[:80_000]
    instruction = re.sub(r"\x00", "", body.instruction or "").strip()[:4_000]
    if not draft:
        return JSONResponse(
            {"error": "empty_draft", "message": "Generate or enter a draft before asking for changes."},
            status_code=400,
        )
    if not instruction:
        return JSONResponse(
            {"error": "empty_instruction", "message": "Tell Moonley what to change or what new draft to prepare."},
            status_code=400,
        )
    profile = draft_profile(body.document_type)
    revised = draft_llm_fn(
        revision_messages(body.title, draft, instruction, profile),
        max_tokens=6_000,
        timeout=150,
    )
    if not revised:
        return JSONResponse(
            {"error": "revision_unavailable", "message": "The drafting model did not return an update. Your editable draft is unchanged."},
            status_code=503,
        )
    _log(
        "drafting_revision",
        {
            "owner_sha256": hashlib.sha256(_project_owner(request).encode("utf-8")).hexdigest(),
            "document_type": body.document_type,
            "input_chars": len(draft),
            "instruction_chars": len(instruction),
        },
    )
    return JSONResponse(
        {
            "draft": revised[:80_000],
            "model_call": {"provider": "deepseek", "attempted": True, "succeeded": True},
            "notice": "AI-updated working draft only. Review every change before finalizing.",
        },
        headers={"Cache-Control": "no-store"},
    )

@app.post("/api/v2/drafting/export/docx")
def export_draft_docx(body: DraftExportRequest):
    draft = re.sub(r"\x00", "", body.draft or "").strip()[:80_000]
    if not draft:
        return JSONResponse(
            {"error": "empty_draft", "message": "Generate or enter a draft before exporting."},
            status_code=400,
        )
    title = re.sub(r"\s+", " ", body.title or "").strip()[:180] or "Moonley working draft"
    filename = re.sub(r"[^A-Za-z0-9._-]+", "-", title).strip("-.")[:80] or "moonley-working-draft"
    return Response(
        content=draft_docx(title, draft),
        media_type="application/vnd.openxmlformats-officedocument.wordprocessingml.document",
        headers={
            "Cache-Control": "private, no-store",
            "Content-Disposition": f'attachment; filename="{filename}.docx"',
        },
    )

@app.post("/api/v2/drafting/export/pdf")
def export_draft_pdf(body: DraftExportRequest):
    draft = re.sub(r"\x00", "", body.draft or "").strip()[:80_000]
    if not draft:
        return JSONResponse(
            {"error": "empty_draft", "message": "Generate or enter a draft before exporting."},
            status_code=400,
        )
    title = re.sub(r"\s+", " ", body.title or "").strip()[:180] or "Moonley working draft"
    filename = re.sub(r"[^A-Za-z0-9._-]+", "-", title).strip("-.")[:80] or "moonley-working-draft"
    try:
        payload = draft_pdf(title, draft)
    except DraftingError as exc:
        return JSONResponse({"error": "pdf_export_failed", "message": str(exc)}, status_code=400)
    return Response(
        content=payload,
        media_type="application/pdf",
        headers={
            "Cache-Control": "private, no-store",
            "Content-Disposition": f'attachment; filename="{filename}.pdf"',
        },
    )

@app.post("/api/v2/query_brief")
@app.post("/api/query_brief")
def query_brief(req: QueryBriefRequest):
    query = re.sub(r"\s+", " ", req.query or "").strip()
    if not query:
        return JSONResponse({"error": "query is required"}, status_code=400)
    history = [
        turn.model_dump() if hasattr(turn, "model_dump") else turn.dict()
        for turn in req.history[-8:]
    ]
    active_doc = _eligible_doc(req.active_case_id or "")
    active_case = C._card(active_doc) if active_doc else None
    brief = A.query_brief(
        query,
        req.refinements,
        fast_llm_fn,
        history=history,
        active_case=active_case,
    )
    route = str(brief.get("route") or "legal_research")
    scope = str(brief.get("retrieval_scope") or "global")
    if route.startswith("case_") or scope == "case_plus_global":
        resolution = A.resolve_case_reference(
            C,
            brief.get("case_reference") or query,
            active_case_id=active_doc,
            recent_case_ids=req.recent_case_ids,
        )
        brief["case_resolution"] = resolution
        if resolution.get("status") == "resolved":
            brief["active_case"] = resolution.get("case")
        elif resolution.get("status") == "ambiguous":
            brief["case_message"] = (
                "I found more than one plausible case-title match in the corpus. "
                "Choose the intended judgment; Moonley will not silently substitute one case for another."
            )
        else:
            reference = re.sub(r"\s+", " ", str(brief.get("case_reference") or query)).strip()
            brief["case_message"] = (
                f"I could not find an exact or reliable close match for {reference!r} in the Supreme Court corpus. "
                "Add a citation, year, another party name, or subject if you want me to search differently."
            )
    crosswalks = []
    for mention in A.extract_statute_mentions(" ".join([query, *req.refinements])):
        result = C.statute_crosswalk(mention["act"], mention["section"])
        if result.get("found"):
            result["provision"] = C.statute_provision(
                mention["act"], mention["section"]
            )
            for item in result.get("corresponding") or []:
                item["provision"] = C.statute_provision(
                    item["act"], item["section"]
                )
            crosswalks.append(result)
    if crosswalks:
        brief["statute_crosswalks"] = crosswalks
        if brief.get("mode") == "research":
            provisions = list(brief.get("provisions") or [])
            for item in crosswalks:
                label = f"{item['from']} corresponds directly to {item['to']}"
                if label not in provisions:
                    provisions.append(label)
            brief["provisions"] = provisions[:8]
    _log("query_briefs", {
        **_query_log_fields(query),
        "refinement_count": len([x for x in req.refinements if str(x).strip()]),
        "history_turn_count": len(history),
        "route": brief.get("route"),
        "case_resolution": (brief.get("case_resolution") or {}).get("status"),
    })
    return JSONResponse(brief)

def _eligible_results(rows):
    out, seen = [], set()
    for card in rows or []:
        if not isinstance(card, dict):
            continue
        d = str(card.get("judgment_id") or card.get("doc_id") or "")
        if not d or d in seen or not C.is_retrieval_eligible(d):
            continue
        copy = dict(card)
        copy["doc_id"] = d
        copy["judgment_id"] = d
        out.append(copy)
        seen.add(d)
    return out

def _search_response(
    q: str,
    *,
    original_q: str | None = None,
    approved_frame: dict | None = None,
    route: str = "legal_research",
    retrieval_scope: str = "global",
    active_case_id: str | None = None,
    case_question: str | None = None,
    history: list[dict] | None = None,
    primary_limit: int = 6,
    more_limit: int = 14,
):
    q = re.sub(r"\s+", " ", q or "").strip()
    original_q = re.sub(r"\s+", " ", original_q or q).strip()
    if not q:
        return JSONResponse({"error": "query is required"}, status_code=400)
    case_doc = _eligible_doc(active_case_id or "")
    requested_scope = retrieval_scope if retrieval_scope in {"case", "graph", "case_plus_global", "global"} else "global"
    if route in {"case_lookup", "case_question"}:
        scope = "case"
    elif route == "case_lineage":
        scope = "graph"
    else:
        scope = requested_scope if requested_scope in {"global", "case_plus_global"} else "global"
    frame = dict(approved_frame or {}) if approved_frame else None
    if case_doc and scope == "case_plus_global":
        frame = dict(frame or {})
        known = list(frame.get("known_citations") or [])
        case_name = str(C.meta.get(case_doc, {}).get("case_name") or "").strip()
        if case_name and case_name not in known:
            known.insert(0, case_name)
        frame["known_citations"] = known[:4]
    t0 = time.time()
    def gen():
        yield sse({"t": "meta", "corpus": C.coverage(), "grounding": "stored-source-only", "research_release": RESEARCH_RELEASE})
        if case_doc:
            card = C._card(case_doc)
            yield sse({
                "t": "case_context",
                "route": route,
                "retrieval_scope": scope,
                "source": "verified_doc_id",
                "case": card,
            })
        final, pending_more, more_sent = [], [], False
        try:
            if case_doc and scope == "case":
                events = A.case_context_stream(
                    C,
                    re.sub(r"\s+", " ", str(case_question or q)).strip()[:1200],
                    case_doc,
                    history or [],
                    fast_llm_fn,
                )
            elif case_doc and scope == "graph":
                events = A.case_lineage_stream(C, case_question or q, case_doc)
            else:
                events = A.structured_search_stream(
                    C, q, llm_fn, approved_frame=frame, identity_query=original_q
                )
            for ev in events:
                if ev.get("t") == "_trace":                        # stage-level instrumentation -> log only
                    _log("trace", {**_query_log_fields(q), "stage": ev.get("stage"), "data": ev.get("data")})
                    continue
                if ev.get("t") == "results":
                    final = _eligible_results(ev.get("results"))
                    pending_more = final[primary_limit:]
                    ev = {**ev, "results": final[:primary_limit]}
                elif ev.get("t") == "more_results":
                    combined = _eligible_results(pending_more + list(ev.get("results") or []))
                    pending_more = []
                    more_sent = True
                    ev = {**ev, "results": combined[:more_limit]}
                    if not ev["results"]:
                        continue
                elif ev.get("t") == "done" and pending_more and not more_sent:
                    yield sse({"t": "more_results", "results": pending_more[:more_limit]})
                    pending_more = []
                yield sse(ev)
        except Exception as e:
            yield sse({"t": "error", "message": str(e)[:200]}); yield sse({"t": "done"})
        _log("searches", {**_query_log_fields(q), "latency_s": round(time.time() - t0, 1),
                          "result_ids": [c.get("doc_id") for c in final[:20]]})
        query_log = _query_log_fields(q)
        print(
            f"[agent] query_sha256={query_log['query_sha256']} "
            f"query_chars={query_log['query_chars']} {time.time()-t0:.1f}s",
            flush=True,
        )
    return StreamingResponse(gen(), media_type="text/event-stream",
                             headers={"Cache-Control": "no-cache", "X-Accel-Buffering": "no", "Connection": "keep-alive"})

@app.get("/api/search_stream")
def search_stream(q: str, request: Request):
    return _search_response(q)

@app.post("/api/v2/search_stream")
def search_stream_v2(req: SearchRequest, request: Request):
    direct_case = req.route in {"case_lookup", "case_question", "case_lineage"}
    if not req.approved and not direct_case:
        return JSONResponse({"error": "query understanding must be approved"}, status_code=409)
    if direct_case and not _eligible_doc(req.active_case_id or ""):
        return JSONResponse({"error": "selected judgment is required"}, status_code=409)
    return _search_response(
        req.q,
        original_q=req.original_q,
        approved_frame=req.search_frame,
        route=req.route,
        retrieval_scope=req.retrieval_scope,
        active_case_id=req.active_case_id,
        case_question=req.case_question,
        history=[
            turn.model_dump() if hasattr(turn, "model_dump") else turn.dict()
            for turn in req.history[-6:]
        ],
    )

def _graph_card(t, src_dst):
    tm = C.meta.get(t, {})
    return graph_node_card(
        t,
        tm,
        treatment=C.edge_meta.get(src_dst, {}).get("treatment"),
        cited_by=C.cite_indeg.get(t, 0),
        good_law_status=C.goodlaw.get(t, {}).get("good_law_status", "unknown"),
    )

@app.get("/api/deep_search_stream")
def deep_search_stream(q: str, request: Request):
    # the agent IS the unified deep pipeline — alias so the frontend's 'deep' toggle never 404s
    return _search_response(q)

def _eligible_doc(value: str):
    d = value if value in C.meta else nc2doc.get(value)
    return d if d and C.is_retrieval_eligible(d) else None

def _pdf_aliases(metadata: dict) -> list[str]:
    return [
        value
        for value in [
            metadata.get("neutral_citation"),
            *(metadata.get("equivalent_citations") or []),
        ]
        if value
    ]


@app.get("/api/v2/judgment")
@app.get("/api/judgment")
def judgment(id: str, q: str = ""):
    d = _eligible_doc(id)
    if not d: return JSONResponse({"error": "not found"}, status_code=404)
    m = C.meta.get(d, {}); jv = C.judgment_view(d)
    jv["judgment_id"] = str(d)
    jv["bench"] = m.get("bench"); jv["author_judge"] = m.get("author_judge"); jv["acts"] = m.get("acts")
    jv["case_number"] = m.get("case_number"); jv["year"] = m.get("year")
    jv["text_raw"] = jv.get("text", "")
    aliases = _pdf_aliases(m)
    pdf_status = PDF_SOURCES.probe(d, aliases=aliases)
    pdf_public = pdf_status.public_dict()
    # A mapped individual object is the fast path.  Bharat Courts can still
    # resolve the same public archive by year and identity if that map misses.
    pdf_public["fallback_available"] = bool(
        pdf_public.get("fallback_available")
        or (m.get("year") and (m.get("neutral_citation") or m.get("case_name")))
    )
    pdf_public["fallback_provider"] = "bharat_courts"
    pdf_public["route"] = f"/api/v2/pdf?id={d}"
    jv["pdf"] = pdf_public
    jv["has_pdf"] = pdf_status.verified                         # compatibility with cached/older frontends
    text_provider = m.get("source_provider") or m.get("provider") or "Supreme Court Reports open registry"
    jv["grounding"] = {
        "text_available": bool((jv.get("text") or "").strip()),
        "retrieval_eligible": True,
        "text_origin": f"judgment text extracted from {text_provider}",
        "pdf_status": pdf_status.status,
        "source_name": text_provider,
        "source_url": m.get("source_url"),
    }
    clean_query = re.sub(r"\s+", " ", q or "").strip()[:4000]
    if clean_query:
        if hasattr(C, "relevant_passages"):
            highlights = C.relevant_passages(clean_query, d, k=6)
        else:
            highlights = C.case_chat_passages(clean_query, d, k=6)
        jv["relevance_highlights"] = highlights
        jv["highlighting"] = {
            "query_specific": True,
            "method": "case-local semantic retrieval resolved to stored paragraphs",
            "grounding": "stored_paragraph_ids_only",
        }
    else:
        jv["relevance_highlights"] = []
        jv["highlighting"] = {"query_specific": False, "grounding": "stored_paragraph_ids_only"}
    jv["corpus_notice"] = (
        f"Searched {C.coverage()['accepted_judgments']:,} accepted Supreme Court judgments. "
        "Unavailable or unmapped judgments were not evaluated."
    )
    if not pdf_status.verified:
        _log("pdf_sources", {"doc_id": d, "status": pdf_status.status, "reason": pdf_status.reason})
    # CITATOR from the citation GRAPH (meta.cases_cited is only ~3% populated): note-up + note-down
    jv["cited_cases"] = [
        _graph_card(t, (d, t))
        for t in list(dict.fromkeys(C.out_edges.get(d, [])))
        if C.is_retrieval_eligible(t)
    ][:20]
    jv["citing_cases"] = [
        _graph_card(s, (s, d))
        for s in sorted(set(C.in_edges.get(d, [])), key=lambda x: -C.cite_indeg.get(x, 0))
        if C.is_retrieval_eligible(s)
    ][:20]
    if not jv["cited_cases"]:                                  # fallback to the sparse metadata if the graph has nothing
        jv["cited_cases"] = resolve_cited(m.get("cases_cited"), d)
    jv["links"] = doc_links(jv.get("text"), d)
    return JSONResponse(jv)

@app.get("/api/v2/judgment/{judgment_id}/paragraphs")
def judgment_paragraphs(judgment_id: str, offset: int = 0, limit: int = 50):
    d = _eligible_doc(judgment_id)
    if not d:
        return JSONResponse({"error": "judgment not found"}, status_code=404)
    limit = max(1, min(int(limit), 100))
    offset = max(0, int(offset))
    if hasattr(C, "judgment_paragraphs"):
        return JSONResponse(C.judgment_paragraphs(d, offset=offset, limit=limit))
    cis = C.doc_chunks.get(d, [])
    rows = [
        {
            "paragraph_id": f"{d}:chunk:{ci}",
            "label": f"Indexed passage {position + 1}",
            "sequence": position + 1,
            "text": C.texts[ci],
            "html_anchor": f"paragraph-{d}-chunk-{ci}",
            "source_kind": "legacy_chunk",
        }
        for position, ci in enumerate(cis[offset:offset + limit], start=offset)
        if str(C.texts[ci]).strip()
    ]
    return JSONResponse({
        "judgment_id": str(d),
        "paragraphs": rows,
        "offset": offset,
        "limit": limit,
        "total": len(cis),
        "next_offset": offset + len(rows) if offset + len(rows) < len(cis) else None,
    })

@app.get("/api/v2/graph")
def graph(id: str, direction: str = "both", limit: int = 50):
    d = _eligible_doc(id)
    if not d:
        return JSONResponse({"error": "judgment not found"}, status_code=404)
    direction = direction if direction in {"incoming", "outgoing", "both"} else "both"
    limit = max(1, min(int(limit), 100))
    root = graph_node_card(
        d,
        C.meta.get(d, {}),
        cited_by=C.cite_indeg.get(d, 0),
        good_law_status=C.goodlaw.get(d, {}).get("good_law_status", "unknown"),
    )
    nodes = {d: root}
    edges = []
    if direction in {"outgoing", "both"}:
        for target in list(dict.fromkeys(C.out_edges.get(d, []))):
            if len(edges) >= limit or not C.is_retrieval_eligible(target):
                continue
            card = _graph_card(target, (d, target)); nodes[target] = card
            edge = C.edge_meta.get((d, target), {})
            edges.append({
                "source_id": str(d), "target_id": str(target),
                "relation": edge.get("treatment") or "referred_to",
                "scope": edge.get("scope") or "unknown",
                "confidence": edge.get("confidence"),
                "direction": "outgoing", "evidence": edge.get("evidence") or [],
            })
    if direction in {"incoming", "both"}:
        for source in sorted(set(C.in_edges.get(d, [])), key=lambda x: -C.cite_indeg.get(x, 0)):
            if len(edges) >= limit or not C.is_retrieval_eligible(source):
                continue
            card = _graph_card(source, (source, d)); nodes[source] = card
            edge = C.edge_meta.get((source, d), {})
            edges.append({
                "source_id": str(source), "target_id": str(d),
                "relation": edge.get("treatment") or "referred_to",
                "scope": edge.get("scope") or "unknown",
                "confidence": edge.get("confidence"),
                "direction": "incoming", "evidence": edge.get("evidence") or [],
            })
    return JSONResponse({
        "judgment_id": str(d), "root": root,
        "nodes": list(nodes.values()), "edges": edges,
        "unresolved_edges_hidden": True,
    })

@app.post("/api/v2/judgment_chat")
@app.post("/api/judgment_chat")
def judgment_chat(req: CaseChatRequest):
    d = _eligible_doc(req.doc_id)
    if not d:
        return JSONResponse({"error": "judgment not found"}, status_code=404)
    question = re.sub(r"\s+", " ", req.question or "").strip()
    if not question:
        return JSONResponse({"error": "question is required"}, status_code=400)
    jv = C.judgment_view(d)
    summary = jv.get("summary") or {}
    if not summary.get("available") or not summary.get("text"):
        return JSONResponse(
            {"error": "case summary unavailable", "code": "summary_unavailable"},
            status_code=409,
        )
    passages = C.case_chat_passages(question, d, k=5)
    response = A.case_chat_grounded_response(
        summary["text"],
        passages,
        question,
        [turn.dict() for turn in req.history],
        jv.get("case_name"),
        jv.get("neutral_citation"),
        fast_llm_fn,
    )
    if not response.get("answer"):
        return JSONResponse({"error": "case chat unavailable"}, status_code=502)
    _log("judgment_chats", {"doc_id": d, **_query_log_fields(question), "summary_source": summary.get("source")})
    return JSONResponse({
        "doc_id": d,
        "judgment_id": str(d),
        "answer": response["answer"],
        "evidence": response.get("evidence") or [],
        "supported": bool(response.get("supported")),
        "grounded_in": "case_summary_and_stored_passages",
        "summary_source": summary.get("source"),
    })

@app.get("/api/v2/pdf/status")
@app.get("/api/pdf/status")
def pdf_status(id: str, refresh: int = 0):
    d = _eligible_doc(id)
    if not d: return JSONResponse({"error": "not found"}, status_code=404)
    m = C.meta.get(d, {})
    status = PDF_SOURCES.probe(d, aliases=_pdf_aliases(m), force=bool(refresh))
    public = status.public_dict()
    public["fallback_available"] = bool(
        public.get("fallback_available")
        or (m.get("year") and (m.get("neutral_citation") or m.get("case_name")))
    )
    public["fallback_provider"] = "bharat_courts"
    return JSONResponse({"doc_id": d, "judgment_id": str(d), **public})

@app.get("/api/v2/pdf")
@app.get("/api/pdf")
async def pdf(id: str, dl: int = 0):
    d = _eligible_doc(id)
    if not d: return JSONResponse({"error": "not found"}, status_code=404)
    m = C.meta.get(d, {})
    aliases = _pdf_aliases(m)
    status = PDF_SOURCES.probe(d, aliases=aliases)
    if status.verified:
        # Redirect instead of downloading the complete document into the Space.
        # The public object supports byte ranges used by native PDF viewers.
        return RedirectResponse(
            status.url,
            status_code=307,
            headers={"Cache-Control": "private, max-age=3600", "X-PDF-Provider": "aws_open_data"},
        )
    archive = PDF_SOURCES.archive_candidate(d, aliases)
    try:
        data, provenance = await resolve_and_fetch_pdf(
            year=(archive or {}).get("year") or m.get("year"),
            path=(archive or {}).get("path"),
            case_name=m.get("case_name") or "",
            neutral_citation=m.get("neutral_citation") or "",
            equivalent_citations=m.get("equivalent_citations") or [],
            decision_date=m.get("date") or "",
        )
    except BharatCourtsPdfError as exc:
        _log("pdf_sources", {"doc_id": d, "status": "bharat_courts_unavailable", "reason": str(exc)[:200]})
        return JSONResponse(
            {
                "error": "pdf unavailable",
                "reason": str(exc)[:300],
                "pdf_status": "bharat_courts_unavailable",
                "official_search_url": status.public_dict()["official_search_url"],
            },
            status_code=503 if status.status == "temporarily_unavailable" else 422,
        )
    citation = re.sub(r"[^A-Za-z0-9._-]+", "-", m.get("neutral_citation") or "judgment").strip("-")
    disposition = "attachment" if dl else "inline"
    _log("pdf_sources", {"doc_id": d, "status": "verified", "provider": provenance["provider"]})
    return Response(
        content=data,
        media_type="application/pdf",
        headers={
            "Cache-Control": "private, max-age=86400",
            "Content-Disposition": f'{disposition}; filename="{citation or "judgment"}.pdf"',
            "X-PDF-Provider": "bharat_courts",
        },
    )

@app.get("/api/v2/health")
def health():
    return JSONResponse({
        "service": "Moonley API",
        "status": "ok",
        "auth_configured": clerk_settings().configured,
        "api_version": "v2-grounded-preview",
        "runtime": RUNTIME_KIND,
        "warm": RUNTIME_WARM,
        "corpus": C.coverage(),
        "research_release": RESEARCH_RELEASE,
        "device": C.device,
        "pdf_sources": {
            "mapped_identities": PDF_SOURCES.mapped_count,
            "primary": "aws_open_data",
            "fallback": "bharat_courts",
        },
        "project_storage": PROJECTS.status(),
        "knowledge": KNOWLEDGE.status(),
        "statute_crosswalk": {
            "loaded": True,
            "indexed_directions": C.crosswalk.mapping_count,
        },
        "statute_library": C.statute_library.status(),
        "drafting_templates": len(DRAFTING.list()),
    }, headers={"Cache-Control": "no-store"})

@app.get("/api/v2/ready")
def ready():
    return JSONResponse({
        "ready": bool(C.eligible_doc_ids) and RUNTIME_WARM,
        "api_version": "v2-grounded-preview",
        "runtime": RUNTIME_KIND,
        "accepted_judgments": len(C.eligible_doc_ids),
        "research_release": RESEARCH_RELEASE,
    }, headers={"Cache-Control": "no-store"})

@app.get("/")
def home():
    return JSONResponse(
        {"service": "Moonley API", "status": "ok", "ui": "https://moonley-pilot.vercel.app"},
        headers={"Cache-Control": "no-store"},
    )

print(f"[serve_agent] boot configured runtime={RUNTIME_KIND}", flush=True)