Add the ingest pipeline and CLI, replacing the ad-hoc scripts

`python -m ingest.cli fetch|load|sync` now exists — README, PORTING.md, the systemd
example and pyproject's console script have all been referencing it.

This replaces six scripts that the rename had left genuinely broken: the speech
loader's INSERT listed `year, year` as two columns, and both loaders read the source
JSON using plenum's column names rather than the Riksdag's field names, so a sync
would have written empty rows without erroring. Rewriting on top of the adapter was
safer than repairing them, because the adapter makes the source-name/column-name
boundary explicit rather than leaving it to be remembered.

Two source quirks found by running it over real archives rather than reasoning about
it: a few records carry a bare string where a child element belongs, which took the
whole load down on an AttributeError; and party codes appear in both cases, so
`parties` held {S,s} and any filter on 'S' would have silently missed half the
documents. Both are handled in the adapter, where source messiness belongs.

Also adds doc_type to schema.sql. It was only in the migration, so a fresh install
lacked a column the loader writes.

Verified end to end against real archives: 400 speeches and 400 documents ingested
into a clean database, producing 3,742 authors and 2,229 proposals, with search
vectors built by the triggers and full-text queries returning hits.

Post-load steps (debate ids, embeddings, summaries) still run separately via
scripts/debates.py and scripts/make_embeddings.py; wiring them into `sync` is
tracked.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

_postgres/schema.sql

@@ -177,6 +177,12 @@ CREATE INDEX IF NOT EXISTS debates_summary_embedding_idx ON debates
 CREATE TABLE IF NOT EXISTS documents (
     -- Primary key: doc_id (e.g. "HD02846")
     doc_id          TEXT PRIMARY KEY,
+
+    -- Which kind of member-submitted document this is. Everything is 'motion'
+    -- today; the column exists so bills, written questions and committee reports
+    -- can share this table rather than each needing their own.
+    doc_type        TEXT NOT NULL DEFAULT 'motion',
+
     source_record_id       TEXT,
 
     -- Identity / classification

bootstrap.py

@@ -1,6 +1,6 @@
 """Locate the project root and make it importable.
 
-Scripts under ``scripts/`` are run directly (``python scripts/sync_talks.py``), so
+Scripts under ``scripts/`` are run directly (``python scripts/make_embeddings.py``), so
 the repository root is not on ``sys.path`` and relative paths would resolve against
 whatever directory the caller happened to be in. Importing this module fixes both,
 deriving the root from this file's own location.

ingest/adapters/riksdagen.py

@@ -22,11 +22,17 @@ from typing import Any, Iterable, Iterator, Optional
 # ── source quirks ─────────────────────────────────────────────────────────────
 
 
-def _as_list(value: Any) -> list:
-    """One child arrives as an object, several as a list, none as null."""
+def _as_list(value: Any) -> list[dict]:
+    """One child arrives as an object, several as a list, none as null.
+
+    Non-object entries are dropped rather than raising: a handful of records in
+    the archive carry a bare string where a child element is expected, and one
+    malformed row should not abort a multi-hour load.
+    """
     if value is None:
         return []
-    return value if isinstance(value, list) else [value]
+    items = value if isinstance(value, list) else [value]
+    return [i for i in items if isinstance(i, dict)]
 
 
 def _clean(value: Any) -> Any:
@@ -166,7 +172,9 @@ def adapt_document(payload: dict) -> Optional[dict]:
             "ordinal": i,
             "person_id": _clean(person.get("intressent_id")),
             "name": _clean(person.get("namn")),
-            "party": _clean(person.get("partibet")),
+            # The archive carries both "S" and "s" for the same party. Left as-is,
+            # a filter on 'S' silently misses half the documents.
+            "party": (_clean(person.get("partibet")) or "").upper() or None,
             "role": _clean(person.get("roll")),
         })
 

ingest/cli.py

@@ -0,0 +1,78 @@
+"""Command-line entry point for ingesting a parliament's open data.
+
+    python -m ingest.cli fetch --source documents --range 2022-2025
+    python -m ingest.cli load  --source documents
+    python -m ingest.cli sync  --source all
+
+`fetch` downloads, `load` adapts and upserts what is on disk, `sync` does both.
+Everything is resumable: archives already unpacked are skipped, and inserts use
+ON CONFLICT DO NOTHING, so re-running after an interruption is safe.
+"""
+from __future__ import annotations
+
+import argparse
+import sys
+from pathlib import Path
+
+sys.path.insert(0, str(Path(__file__).resolve().parents[1]))
+
+from ingest import pipeline  # noqa: E402
+from ingest.adapters import load_adapter  # noqa: E402
+from parliament import PARLIAMENT  # noqa: E402
+
+SOURCES = ("speeches", "documents", "people")
+
+
+def _adapter():
+    module_path = PARLIAMENT.sources.get("adapter")
+    if not module_path:
+        raise SystemExit("parliament.yaml has no `sources.adapter`. See docs/PORTING.md.")
+    return load_adapter(module_path)
+
+
+def _resolve(source: str) -> list[str]:
+    if source == "all":
+        return [s for s in SOURCES if s in PARLIAMENT.sources]
+    if source not in SOURCES:
+        raise SystemExit(f"Unknown source {source!r}; expected one of {SOURCES + ('all',)}")
+    return [source]
+
+
+def main(argv: list[str] | None = None) -> int:
+    parser = argparse.ArgumentParser(prog="ingest", description=__doc__)
+    parser.add_argument("command", choices=("fetch", "load", "sync"))
+    parser.add_argument("--source", default="all",
+                        help=f"one of {', '.join(SOURCES)}, or 'all' (default)")
+    parser.add_argument("--range", dest="ranges", action="append",
+                        help="archive range to fetch, e.g. 2022-2025. Repeatable. "
+                             "Defaults to the ranges in parliament.yaml.")
+    parser.add_argument("--limit", type=int,
+                        help="stop after this many records — useful for a smoke test")
+    args = parser.parse_args(argv)
+
+    adapter = _adapter()
+    print(f"parliament: {PARLIAMENT.meta.get('name')}  adapter: {PARLIAMENT.sources['adapter']}")
+
+    exit_code = 0
+    for source in _resolve(args.source):
+        print(f"\n── {source} ──")
+        try:
+            if args.command in ("fetch", "sync"):
+                pipeline.fetch(source, args.ranges)
+            if args.command in ("load", "sync"):
+                counts = pipeline.load(source, adapter.adapt, limit=args.limit)
+                print(f"  read {counts['read']}, wrote {counts['written']}, "
+                      f"skipped {counts['skipped']}")
+        except Exception as exc:
+            # One failing source should not abandon the others; a daily sync that
+            # aborts because one archive moved is worse than a partial one.
+            print(f"  FAILED: {type(exc).__name__}: {exc}")
+            exit_code = 1
+
+    if args.command in ("load", "sync"):
+        print("\nNext: python scripts/make_embeddings.py")
+    return exit_code
+
+
+if __name__ == "__main__":
+    sys.exit(main())

ingest/pipeline.py

@@ -0,0 +1,209 @@
+"""Generic fetch → adapt → upsert pipeline.
+
+Country-neutral: every source-specific decision comes from `sources:` in
+parliament.yaml and from the adapter named there.
+"""
+from __future__ import annotations
+
+import io
+import json
+import zipfile
+from pathlib import Path
+from typing import Callable, Iterator, Optional
+
+import requests
+
+from bootstrap import DATA_DIR
+from parliament import PARLIAMENT
+from postgres_client import pg
+
+# Rows per INSERT. Large enough that round-trips are not the bottleneck, small
+# enough that one bad batch is cheap to retry.
+BATCH = 500
+
+
+# ── fetching ──────────────────────────────────────────────────────────────────
+
+
+def source_config(name: str) -> dict:
+    sources = PARLIAMENT.sources
+    if name not in sources:
+        available = [k for k in sources if k != "adapter"]
+        raise ValueError(f"No source {name!r} in parliament.yaml; have {available}")
+    return sources[name]
+
+
+def dest_dir(name: str) -> Path:
+    return DATA_DIR / source_config(name).get("dest_dir", name)
+
+
+def fetch(name: str, ranges: Optional[list[str]] = None) -> list[Path]:
+    """Download a source's bulk archives and unpack them under DATA_DIR.
+
+    Archives already unpacked are skipped, so this is safe to re-run and cheap to
+    resume after an interruption — which matters when a full download is tens of GB.
+    """
+    cfg = source_config(name)
+    kind = cfg.get("kind", "zip-dataset")
+    target = dest_dir(name)
+    target.mkdir(parents=True, exist_ok=True)
+
+    if kind == "json":
+        out = target / f"{name}.json"
+        out.write_bytes(requests.get(cfg["url"], timeout=120).content)
+        print(f"  fetched {out.relative_to(DATA_DIR)}")
+        return [out]
+
+    if kind != "zip-dataset":
+        raise ValueError(f"Unsupported source kind {kind!r} for {name}")
+
+    wanted = ranges or cfg.get("ranges")
+    if not wanted:
+        raise ValueError(
+            f"Source {name!r} has no `ranges:` in parliament.yaml and none was given. "
+            f"Pass --range, or list them in the config."
+        )
+
+    written = []
+    for rng in wanted:
+        folder = target / rng
+        if folder.exists() and any(folder.iterdir()):
+            print(f"  {rng}: already present, skipping")
+            continue
+        url = cfg["url_template"].format(range=rng)
+        print(f"  {rng}: downloading {url}")
+        resp = requests.get(url, timeout=1800)
+        if resp.status_code != 200:
+            print(f"  {rng}: HTTP {resp.status_code}, skipping")
+            continue
+        folder.mkdir(parents=True, exist_ok=True)
+        with zipfile.ZipFile(io.BytesIO(resp.content)) as zf:
+            zf.extractall(folder)
+        written.append(folder)
+        print(f"  {rng}: unpacked {len(list(folder.rglob('*.json')))} files")
+    return written
+
+
+def read_records(name: str) -> Iterator[dict]:
+    """Yield every source record on disk for a source."""
+    root = dest_dir(name)
+    if not root.exists():
+        raise FileNotFoundError(f"No data at {root}. Run `ingest.cli fetch --source {name}` first.")
+    for path in sorted(root.rglob("*.json")):
+        try:
+            # utf-8-sig: the archives carry a byte-order mark.
+            with open(path, encoding="utf-8-sig") as fh:
+                payload = json.load(fh)
+        except (json.JSONDecodeError, OSError) as exc:
+            print(f"  skipping {path.name}: {exc}")
+            continue
+        # A people listing is one file holding many records.
+        if isinstance(payload, dict) and "personlista" in payload:
+            for person in payload["personlista"].get("person", []):
+                yield person
+        else:
+            yield payload
+
+
+# ── loading ───────────────────────────────────────────────────────────────────
+
+
+def _insert(table: str, columns: list[str], rows: list[tuple], conflict: str) -> int:
+    if not rows:
+        return 0
+    sql = (
+        f"INSERT INTO {table} ({', '.join(columns)}) VALUES %s "
+        f"ON CONFLICT ({conflict}) DO NOTHING"
+    )
+    pg.execute_values(sql, rows)
+    return len(rows)
+
+
+def _flush(table: str, columns: list[str], conflict: str, buffer: list[dict]) -> int:
+    """Insert and clear a buffer. Argument order matches `_target()` unpacking."""
+    rows = [tuple(r.get(c) for c in columns) for r in buffer]
+    n = _insert(table, columns, rows, conflict)
+    buffer.clear()
+    return n
+
+
+SPEECH_COLUMNS = [
+    "id", "source_speech_id", "source_doc_id", "text", "section_title", "sequence",
+    "activity_type", "speaker_name", "party", "person_id", "date", "source_datetime",
+    "year", "session_year", "related_doc_id", "source_doc_number", "source_record_id",
+    "title", "is_reply",
+]
+DOCUMENT_COLUMNS = [
+    "doc_id", "doc_type", "source_record_id", "session_label", "designation", "subtype",
+    "committee", "status", "date", "source_updated_at", "published_at", "session_year",
+    "title", "subtitle", "text", "proposals_text", "has_text", "url_text", "url_html",
+    "url_pdf", "parties", "author_names", "proposals_raw", "attachments", "num_proposals",
+]
+AUTHOR_COLUMNS = ["doc_id", "ordinal", "person_id", "name", "party", "role"]
+PROPOSAL_COLUMNS = [
+    "id", "doc_id", "ordinal", "number", "text",
+    "committee_recommendation", "chamber_decision", "handled_in",
+]
+PERSON_COLUMNS = [
+    "person_id", "source_record_id", "source_record_guid", "source_id", "birth_year",
+    "gender", "last_name", "first_name", "sort_name", "home_town", "party",
+    "constituency", "status", "source_url", "image_url_small", "image_url_medium",
+    "image_url_large", "assignments", "contact_details", "name", "active",
+]
+
+
+def _json_columns(rows: list[dict], columns: Iterator[str]) -> None:
+    """psycopg2 cannot adapt dicts/lists destined for JSONB; serialise them."""
+    for row in rows:
+        for col in columns:
+            if isinstance(row.get(col), (dict, list)):
+                row[col] = json.dumps(row[col], ensure_ascii=False)
+
+
+def load(name: str, adapt: Callable[[str, dict], Optional[dict]], limit: Optional[int] = None) -> dict:
+    """Adapt every record on disk for a source and upsert it."""
+    counts = {"read": 0, "written": 0, "skipped": 0}
+    buf: list[dict] = []
+    authors: list[dict] = []
+    proposals: list[dict] = []
+
+    for payload in read_records(name):
+        if limit and counts["read"] >= limit:
+            break
+        counts["read"] += 1
+        row = adapt(name, payload)
+        if row is None:
+            counts["skipped"] += 1
+            continue
+
+        if name == "documents":
+            _json_columns([row["document"]], ("proposals_raw", "attachments"))
+            buf.append(row["document"])
+            authors.extend(row["authors"])
+            proposals.extend(row["proposals"])
+        else:
+            if name == "people":
+                _json_columns([row], ("assignments", "contact_details"))
+            buf.append(row)
+
+        if len(buf) >= BATCH:
+            counts["written"] += _flush(*_target(name), buf)
+            if authors:
+                _flush("document_authors", AUTHOR_COLUMNS, "doc_id, ordinal", authors)
+            if proposals:
+                _flush("document_proposals", PROPOSAL_COLUMNS, "id", proposals)
+
+    counts["written"] += _flush(*_target(name), buf)
+    if authors:
+        _flush("document_authors", AUTHOR_COLUMNS, "doc_id, ordinal", authors)
+    if proposals:
+        _flush("document_proposals", PROPOSAL_COLUMNS, "id", proposals)
+    return counts
+
+
+def _target(name: str) -> tuple[str, list[str], str]:
+    return {
+        "speeches": ("speeches", SPEECH_COLUMNS, "id"),
+        "documents": ("documents", DOCUMENT_COLUMNS, "doc_id"),
+        "people": ("people", PERSON_COLUMNS, "person_id"),
+    }[name]

recovery.sh

@@ -7,7 +7,7 @@
 #   bash recovery.sh 2>&1 | tee logs/recovery.log
 #
 # Steps (ordered, each depends on the previous):
-#   1. Load ~450k raw talks from disk JSON files  (documents_to_postgres.py)
+#   1. Load raw speeches from disk JSON files  (ingest.cli load)
 #   2. Enrich with summaries+tags from talks_training in ArangoDB
 #   3. Assign debate IDs to talks that don't have one
 #   4. Migrate debates from ArangoDB
@@ -48,9 +48,9 @@ ok "Pre-flight passed"
 # =============================================================================
 # STEP 1 — Load all talks from disk JSON files
 # =============================================================================
-step "Step 1 — Load talks from disk (documents_to_postgres.py)"
+step "Step 1 — Load speeches from disk (ingest.cli load)"
 
-$PYTHON scripts/documents_to_postgres.py \
+$PYTHON -m ingest.cli load --source speeches \
     2>&1 | tee "$LOG_DIR/step1_documents_to_postgres.log"
 
 ok "Step 1 done"

scripts/backfill_motions.sh

@@ -1,29 +0,0 @@
-#!/bin/bash
-# One-time backfill of motioner 1990→idag.
-#
-# Kör i tre steg (alla resumérbara — avbrott är ofarligt, kör bara om):
-#   1. Ladda ned alla mot-arkiv som saknas (hoppar över ifyllda mappar)
-#   2. Parsa alla mappar i motioner/ till Postgres (ON CONFLICT DO NOTHING)
-#   3. Bygg embeddings år för år, nyast först
-#
-# Starta frikopplat:  nohup scripts/backfill_motions.sh > backfill_motions.log 2>&1 &
-set -e
-cd "$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)"
-PY=.venv/bin/python
-
-echo "=== Step 1: download all ranges ==="
-$PY scripts/download_motions.py
-
-echo "=== Step 2: parse all folders into Postgres ==="
-$PY scripts/motions_to_postgres.py
-
-echo "=== Step 3: full-text embeddings, newest year first ==="
-for year in $(seq 2025 -1 1990); do
-    echo "--- embeddings for year $year ---"
-    $PY scripts/make_embeddings.py motions "$year"
-done
-
-echo "=== Step 4: yrkande (condensed proposal) embeddings ==="
-$PY scripts/make_embeddings.py yrkanden
-
-echo "=== Backfill complete ==="

scripts/documents_to_postgres.py

@@ -1,167 +0,0 @@
-"""
-Läser in anföranden från JSON-filer till PostgreSQL.
-
-Ersätter scripts/documents_to_arango.py.
-
-Används av sync_talks.py (update_folder) och kan köras direkt för att
-(om)ladda alla mappar i speeches/:
-
-    python scripts/documents_to_postgres.py
-"""
-from pathlib import Path
-
-import json
-import logging
-import os
-import re
-import sys
-
-sys.path.insert(0, str(Path(__file__).resolve().parents[1]))
-import bootstrap  # noqa: E402,F401  — sets cwd and sys.path to the project root
-
-from postgres_client import pg
-
-logging.basicConfig(level=logging.WARNING, format="%(asctime)s - %(levelname)s - %(message)s")
-
-
-def clean_text(text: str) -> str:
-    if text is None:
-        return ""
-    text = text.replace("<br>", "\n").replace("<br/>", "\n").replace("<br />", "\n")
-    text = text.replace("</p>", "\n")
-    text = re.sub(r"\n+", "\n", text)
-    text = text.strip()
-    text = re.sub(r"<.*?>", "", text)
-    return text
-
-
-def clean_speaker_name(name: str) -> str:
-    if name is None:
-        return ""
-    name = name.strip()
-    name = re.sub(r"\s*\(.*?\)\s*$", "", name)
-    return name.strip()
-
-
-def _parse_date(s: str) -> str | None:
-    if not s:
-        return None
-    return str(s).split(" ")[0][:10] or None
-
-
-def process_folder(folder_path: str, already_processed: set[str] = frozenset()) -> list[dict]:
-    """Parse JSON files in folder_path, return list of talk dicts (skipping known IDs)."""
-    docs = []
-    for file in os.listdir(folder_path):
-        if not file.endswith(".json"):
-            continue
-        try:
-            with open(os.path.join(folder_path, file), "r", encoding="utf-8-sig") as f:
-                data = json.load(f)
-            doc = data["anforande"]
-            source_doc_id = doc.get("dok_id", "")
-            sequence = doc.get("sequence", "")
-            speech_id = f"{source_doc_id}-{sequence}"  # unique per speech, e.g. "GH09116-16"
-            if speech_id in already_processed:
-                continue
-            doc["year"] = int(doc.get("dok_rm", "0000")[:4])
-            doc.pop("dok_rm", None)
-            doc["text"] = clean_text(doc.get("text", ""))
-            doc["speaker_name"] = clean_speaker_name(doc.get("speaker_name", ""))
-            doc["id"] = speech_id
-            doc["dok_id"] = source_doc_id
-            doc["source_speech_id"] = doc.get("source_speech_id", "")  # UUID, kept for reference
-            doc["date"] = _parse_date(doc.get("source_datetime", ""))
-            doc["source_datetime"] = doc.get("source_datetime", "")
-            doc["title"] = doc.get("dok_titel", "")
-            doc.pop("dok_titel", None)
-            doc["sequence"] = int(sequence) if sequence else 0
-            doc["source_record_id"] = doc.get("dok_hangar_id", "")
-            doc.pop("dok_hangar_id", None)
-            doc["is_reply"] = doc.get("is_reply", "N") == "Y"
-            doc.pop("source_updated_at", None)
-            doc.pop("underrubrik", None)
-            year = doc.get("year") or (int(doc["date"][:4]) if doc.get("date") else None)
-            doc["year"] = year
-            docs.append(doc)
-        except (json.JSONDecodeError, KeyError, ValueError) as e:
-            logging.warning(f"Skipping {file}: {e}")
-    return docs
-
-
-_UPSERT_SQL = """
-INSERT INTO speeches (
-    id, source_speech_id, source_doc_id,
-    text, section_title,
-    sequence, activity_type,
-    speaker_name, party, person_id,
-    date, source_datetime, year, year,
-    related_doc_id, source_doc_number, source_record_id, title,
-    is_reply
-) VALUES %s
-ON CONFLICT (id) DO NOTHING
-"""
-
-
-def _doc_to_row(doc: dict) -> tuple:
-    return (
-        doc.get("id"),           # source_speech_id UUID — primary key
-        doc.get("source_speech_id"), # same value, kept for reference
-        doc.get("dok_id"),       # debate/protocol document id
-        doc.get("text"),
-        doc.get("section_title"),
-        doc.get("sequence"),
-        doc.get("activity_type"),
-        doc.get("speaker_name"),
-        doc.get("party"),
-        doc.get("person_id"),
-        doc.get("date"),
-        doc.get("source_datetime"),
-        doc.get("year"),
-        doc.get("year"),
-        doc.get("related_doc_id"),
-        doc.get("source_doc_number"),
-        doc.get("source_record_id"),
-        doc.get("title"),
-        doc.get("is_reply", False),
-    )
-
-
-def insert_docs(docs: list[dict]) -> None:
-    if not docs:
-        return
-    rows = [_doc_to_row(d) for d in docs if d.get("id")]
-    if rows:
-        pg.execute_values(_UPSERT_SQL, rows)
-
-
-def update_folder(path: str, already_processed: set[str] = None) -> int:
-    """
-    Upsert talk documents from JSON files in path into PostgreSQL.
-    Returns the number of new speeches inserted.
-    """
-    if already_processed is None:
-        rows = pg.execute("SELECT id FROM speeches")
-        already_processed = {row["id"] for row in rows}
-
-    docs = process_folder(path, already_processed)
-    insert_docs(docs)
-    return len(docs)
-
-
-if __name__ == "__main__":
-    # Load all folders in speeches/ into PostgreSQL
-    existing = {row["id"] for row in pg.execute("SELECT id FROM speeches")}
-    total = 0
-    for folder in sorted(os.listdir("speeches")):
-        path = str(bootstrap.DATA_DIR / 'speeches' / folder)
-        if not os.path.isdir(path):
-            continue
-        print(f"Processing {folder} …", end=" ", flush=True)
-        docs = process_folder(path, already_processed=existing)
-        insert_docs(docs)
-        new_ids = {d["id"] for d in docs if d.get("id")}
-        existing |= new_ids
-        total += len(new_ids)
-        print(f"{len(new_ids)} inserted")
-    print(f"\nTotal: {total} new speeches inserted")

scripts/download_motions.py

@@ -1,91 +0,0 @@
-"""
-Laddar ned motioner från riksdagens öppna data som ZIP-arkiv med JSON-filer.
-
-Arkiven är uppdelade i fyraårsperioder (förankrade vid 1998; allt före ligger
-i 1990-1997) och uppdateras dagligen:
-
-    https://data.riksdagen.se/dataset/dokument/mot-{range}.json.zip
-
-Varje arkiv innehåller en JSON-fil per motion (dokumentstatus-kuvert med
-metadata + fulltext som HTML). Extraheras till motioner/mot-{range}/.
-
-Backfill (alla perioder 1990→):  python scripts/download_motions.py
-Används av sync_motions.py för daglig uppdatering av aktuell year.
-"""
-from pathlib import Path
-
-import logging
-import os
-import sys
-from io import BytesIO
-from urllib.request import urlopen
-from zipfile import ZipFile
-
-sys.path.insert(0, str(Path(__file__).resolve().parents[1]))
-import bootstrap  # noqa: E402,F401  — sets cwd and sys.path to the project root
-
-logging.basicConfig(
-    level=logging.INFO,
-    format="%(asctime)s [%(levelname)s] %(message)s",
-)
-logger = logging.getLogger(__name__)
-
-RANGES = [
-    "1990-1997",
-    "1998-2001",
-    "2002-2005",
-    "2006-2009",
-    "2010-2013",
-    "2014-2017",
-    "2018-2021",
-    "2022-2025",
-]
-URL_TEMPLATE = "https://data.riksdagen.se/dataset/dokument/mot-{r}.json.zip"
-MOTIONS_DIR = "motioner"
-
-
-def get_current_range(year: int) -> str:
-    """Fyraårsperiod förankrad vid 1998; allt före 1998 ligger i 1990-1997."""
-    if year < 1998:
-        return "1990-1997"
-    start = year - ((year - 1998) % 4)
-    return f"{start}-{start + 3}"
-
-
-def download_range(r: str, force: bool = False) -> str:
-    """
-    Laddar ned och extraherar arkivet för en year till motioner/mot-{r}/.
-    Hoppar över om mappen redan är ifylld, om inte force=True (töms först).
-    """
-    dir_path = os.path.join(MOTIONS_DIR, f"mot-{r}")
-
-    if os.path.exists(dir_path) and os.listdir(dir_path):
-        if not force:
-            logger.info(f"Skipping {dir_path}, already exists and is not empty.")
-            return dir_path
-        for f in os.listdir(dir_path):
-            os.remove(os.path.join(dir_path, f))
-
-    os.makedirs(dir_path, exist_ok=True)
-    url = URL_TEMPLATE.format(r=r)
-    logger.info(f"Downloading {url} → {dir_path}")
-
-    with urlopen(url) as resp:
-        with ZipFile(BytesIO(resp.read())) as zf:
-            zf.extractall(dir_path)
-
-    count = len(os.listdir(dir_path))
-    logger.info(f"Extracted {count} files to {dir_path}")
-    return dir_path
-
-
-def download_all(force: bool = False) -> None:
-    for r in RANGES:
-        download_range(r, force=force)
-
-
-if __name__ == "__main__":
-    if len(sys.argv) > 1:
-        download_range(sys.argv[1], force="--force" in sys.argv)
-    else:
-        download_all()

scripts/download_talks.py

@@ -1,63 +0,0 @@
-import requests
-from bs4 import BeautifulSoup
-from io import BytesIO
-from urllib.request import urlopen
-from zipfile import ZipFile
-import os
-from time import sleep
-
-
-def download(all=False, year=None):
-    if all:
-        for year in range(1999, 2026):
-            first_part = str(year)
-            second_part = str(year + 1)[2:]
-            if first_part == '1999':
-                 url = 'https://data.riksdagen.se/dataset/anforande/anforande-19992000.json.zip'
-            else:
-                url = f"https://data.riksdagen.se/dataset/anforande/anforande-{first_part}{second_part}.json.zip"
-            print(url)
-
-            # Ensure the 'speeches' directory exists
-            talks_dir = "speeches"
-            os.makedirs(talks_dir, exist_ok=True)
-
-            # Create a subdirectory for the current year range
-            dir_name = os.path.join(talks_dir, f"anforande-20{first_part}{second_part}")
-            if os.path.exists(dir_name) and os.listdir(dir_name):
-                print(f"Skipping {dir_name}, already exists and is not empty.")
-                continue
-
-            os.makedirs(dir_name, exist_ok=True)
-
-            # Download and extract the zip file directly into the subdirectory
-            with urlopen(url) as zipresp:
-                with ZipFile(BytesIO(zipresp.read())) as zfile:
-                    zfile.extractall(dir_name)
-    elif year:
-        first_part = str(year)
-        second_part = str(year + 1)[2:]
-        url = f"https://data.riksdagen.se/dataset/anforande/anforande-{first_part}{second_part}.json.zip"
-        print(url)
-
-        # Ensure the 'speeches' directory exists
-        talks_dir = "speeches"
-        os.makedirs(talks_dir, exist_ok=True)
-
-        # Create a subdirectory for the current year range
-        dir_name = os.path.join(talks_dir, f"anforande-20{first_part}{second_part}")
-        if os.path.exists(dir_name) and os.listdir(dir_name):
-            print(f"Skipping {dir_name}, already exists and is not empty.")
-            return
-
-        os.makedirs(dir_name, exist_ok=True)
-
-        # Download and extract the zip file directly into the subdirectory
-        with urlopen(url) as zipresp:
-            with ZipFile(BytesIO(zipresp.read())) as zfile:
-                zfile.extractall(dir_name)
-
-if __name__ == "__main__":
-    while True:
-        new_files = download()
-

scripts/fix_replik_from_json.py

@@ -1,182 +0,0 @@
-"""
-Fix speeches.is_reply by reading the ground-truth value from the JSON source files.
-
-Background:
-  documents_to_postgres.py uses ON CONFLICT DO NOTHING, so speeches ingested before
-  is_reply was handled correctly were never updated. This script does the corrective
-  pass: reads every JSON file, finds speeches where the stored is_reply differs from the
-  file, updates the DB, then re-assigns debate IDs for the affected dates so that
-  proper multi-speaker debates are created.
-
-Steps:
-  1. Scan all speeches/ subfolders and build {speech_id → replik_bool} from JSON files.
-  2. Query DB for speeches where is_reply is currently False.
-  3. Update any talk whose JSON says True.
-  4. For each affected date, clear speeches.debate and re-run make_debate_ids().
-  5. Delete stale single-talk debate rows whose IDs no longer exist in speeches.debate.
-
-Usage:
-  python scripts/fix_replik_from_json.py
-"""
-from pathlib import Path
-
-import logging
-import os
-import sys
-from collections import defaultdict
-
-sys.path.insert(0, str(Path(__file__).resolve().parents[1]))
-import bootstrap  # noqa: E402,F401  — sets cwd and sys.path to the project root
-
-from postgres_client import pg
-from scripts.documents_to_postgres import process_folder
-from scripts.debates import assign_debate_ids
-
-logging.basicConfig(
-    level=logging.INFO,
-    format="%(asctime)s [%(levelname)s] %(message)s",
-)
-logger = logging.getLogger(__name__)
-
-TALKS_DIR = str(bootstrap.DATA_DIR / "speeches")
-BATCH_SIZE = 500
-
-
-def build_replik_map() -> dict[str, bool]:
-    """Return {speech_id: is_reply} for every JSON file in speeches/."""
-    replik_map: dict[str, bool] = {}
-    folders = sorted(os.listdir(TALKS_DIR))
-    for i, folder in enumerate(folders):
-        path = os.path.join(TALKS_DIR, folder)
-        if not os.path.isdir(path):
-            continue
-        docs = process_folder(path)
-        for doc in docs:
-            tid = doc.get("id")
-            if tid:
-                replik_map[tid] = bool(doc.get("is_reply", False))
-        if (i + 1) % 5 == 0 or (i + 1) == len(folders):
-            logger.info(f"  Scanned {i+1}/{len(folders)} folders ({len(replik_map):,} speeches)")
-    return replik_map
-
-
-def fix_replik(replik_map: dict[str, bool]) -> set[str]:
-    """
-    Update speeches where DB is_reply=False but JSON says True.
-    Returns the set of affected talk IDs.
-    """
-    # Only need to check speeches where DB currently says False
-    db_false = pg.execute("SELECT id FROM speeches WHERE is_reply = false")
-    db_false_ids = {row["id"] for row in db_false}
-    logger.info(f"Talks with is_reply=False in DB: {len(db_false_ids):,}")
-
-    to_fix = [tid for tid in db_false_ids if replik_map.get(tid) is True]
-    logger.info(f"Talks to correct (JSON says True, DB says False): {len(to_fix):,}")
-
-    if not to_fix:
-        return set()
-
-    for i in range(0, len(to_fix), BATCH_SIZE):
-        batch = to_fix[i : i + BATCH_SIZE]
-        pg.execute_many(
-            "UPDATE speeches SET is_reply = true WHERE id = %s",
-            [(tid,) for tid in batch],
-        )
-        if (i + BATCH_SIZE) % 5000 == 0 or i + BATCH_SIZE >= len(to_fix):
-            logger.info(f"  Updated {min(i+BATCH_SIZE, len(to_fix)):,}/{len(to_fix):,}")
-
-    return set(to_fix)
-
-
-def get_affected_dates(fixed_ids: set[str]) -> list[str]:
-    """Return distinct dates for the speeches whose is_reply was corrected."""
-    if not fixed_ids:
-        return []
-    rows = pg.execute(
-        "SELECT DISTINCT date::text AS date FROM speeches WHERE id = ANY(%s::text[])",
-        (list(fixed_ids),),
-    )
-    return [row["date"] for row in rows if row.get("date")]
-
-
-def reassign_debate_ids(dates: list[str]) -> None:
-    """
-    For each date, clear speeches.debate and re-assign using assign_debate_ids().
-    This is the same logic as make_debate_ids() but scoped to the given dates.
-    """
-    logger.info(f"Re-assigning debate IDs for {len(dates):,} dates …")
-    for i, date in enumerate(sorted(dates)):
-        # Clear existing debate assignments for this date
-        pg.execute_void(
-            "UPDATE speeches SET debate = NULL WHERE date = %s::date",
-            (date,),
-        )
-
-        speeches = pg.execute(
-            """
-            SELECT id, is_reply
-            FROM speeches
-            WHERE date = %s::date
-            ORDER BY sequence ASC
-            """,
-            (date,),
-        )
-        if not speeches:
-            continue
-
-        updated = assign_debate_ids(list(speeches), date)
-        pg.execute_many(
-            "UPDATE speeches SET debate = %s WHERE id = %s",
-            [(doc["debate"], doc["id"]) for doc in updated],
-        )
-
-        if (i + 1) % 100 == 0 or (i + 1) == len(dates):
-            logger.info(f"  {i+1}/{len(dates)} dates processed")
-
-
-def remove_stale_debate_rows() -> int:
-    """
-    Delete debate rows whose debate ID no longer appears in speeches.debate.
-    This cleans up the old one-per-talk debate rows that became invalid after
-    debate IDs were re-assigned.
-    """
-    result = pg.execute(
-        """
-        DELETE FROM debates
-        WHERE debate NOT IN (SELECT DISTINCT debate FROM speeches WHERE debate IS NOT NULL)
-        RETURNING debate
-        """
-    )
-    count = len(result) if result else 0
-    logger.info(f"Removed {count:,} stale debate rows")
-    return count
-
-
-def main() -> None:
-    logger.info("=== Step 1: Build is_reply map from JSON files ===")
-    replik_map = build_replik_map()
-    logger.info(f"Total speeches in JSON files: {len(replik_map):,}")
-
-    logger.info("=== Step 2: Fix is_reply in DB ===")
-    fixed_ids = fix_replik(replik_map)
-    if not fixed_ids:
-        logger.info("Nothing to fix — is_reply is already correct.")
-        return
-
-    logger.info("=== Step 3: Find affected dates ===")
-    dates = get_affected_dates(fixed_ids)
-    logger.info(f"Affected dates: {len(dates):,}")
-
-    logger.info("=== Step 4: Re-assign debate IDs ===")
-    reassign_debate_ids(dates)
-
-    logger.info("=== Step 5: Remove stale debate rows ===")
-    remove_stale_debate_rows()
-
-    logger.info("=== Done! ===")
-    logger.info("Next: run 'python scripts/debates.py' (or deploy_debate_embeddings.sh)")
-    logger.info("to generate summaries for the newly grouped multi-talk debates.")
-
-
-if __name__ == "__main__":
-    main()

scripts/motions_progress.sh

@@ -1,29 +0,0 @@
-#!/bin/bash
-# Progress of the motion embedding backfills. Run anytime:
-#   bash scripts/motions_progress.sh
-set -e
-cd "$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)"
-export PGPASSWORD=$(grep '^PG_PASSWORD=' .env | cut -d= -f2)
-PGDB=$(grep '^PG_DB=' .env | cut -d= -f2)
-PGUSER=$(grep '^PG_USER=' .env | cut -d= -f2)
-PGHOST=$(grep '^PG_HOST=' .env | cut -d= -f2)
-
-echo "=== Motion embedding coverage ($(date '+%H:%M:%S')) ==="
-psql -h "$PGHOST" -U "$PGUSER" -d "$PGDB" -tA -F $'\t' <<'SQL' | column -t -s $'\t'
-SELECT 'full-text chunks (motions)' AS what,
-       count(*) FILTER (WHERE c.id IS NOT NULL) AS done,
-       count(*) AS total,
-       round(100.0*count(*) FILTER (WHERE c.id IS NOT NULL)/nullif(count(*),0),1)||'%' AS pct
-FROM (SELECT dok_id FROM motions WHERE has_text) m
-LEFT JOIN LATERAL (SELECT 1 AS id FROM motion_chunks c WHERE c.motion_id = m.dok_id LIMIT 1) c ON true
-UNION ALL
-SELECT 'yrkande embeddings',
-       count(*) FILTER (WHERE embedding IS NOT NULL),
-       count(*),
-       round(100.0*count(*) FILTER (WHERE embedding IS NOT NULL)/nullif(count(*),0),1)||'%'
-FROM motion_yrkanden;
-SQL
-
-echo
-echo "=== Backfill processes ==="
-pgrep -af "make_embeddings.py|backfill_motions.sh" | grep -v pgrep || echo "  (none running — backfills finished)"

scripts/motions_to_postgres.py

@@ -1,268 +0,0 @@
-"""
-Läser in motioner från JSON-filer (dokumentstatus-kuvert) till PostgreSQL.
-
-Används av sync_motions.py (update_folder) och kan köras direkt för att
-(om)ladda alla mappar i motioner/:
-
-    python scripts/motions_to_postgres.py
-"""
-from pathlib import Path
-
-import json
-import logging
-import os
-import sys
-
-sys.path.insert(0, str(Path(__file__).resolve().parents[1]))
-import bootstrap  # noqa: E402,F401  — sets cwd and sys.path to the project root
-
-from bs4 import BeautifulSoup
-from postgres_client import pg
-from scripts.documents_to_postgres import _parse_date, clean_speaker_name
-
-logging.basicConfig(level=logging.WARNING, format="%(asctime)s - %(levelname)s - %(message)s")
-
-# Rader kortare än så räknas som textlösa (inskannad PDF med stubb-HTML)
-MIN_TEXT_LEN = 200
-
-
-def _as_list(v) -> list:
-    """Riksdagens XML→JSON: ett barn blir dict, flera blir list, inget blir None."""
-    if v is None:
-        return []
-    return v if isinstance(v, list) else [v]
-
-
-def _clean(v):
-    """API:et serialiserar null som strängen 'None'."""
-    if v is None or v == "None":
-        return None
-    return v
-
-
-def html_to_text(html: str) -> str:
-    """Extraherar ren text ur dokumentets html-fält (inleds med stort <style>-block)."""
-    if not html:
-        return ""
-    soup = BeautifulSoup(html, "html.parser")
-    for tag in soup(["style", "script"]):
-        tag.decompose()
-    text = soup.get_text("\n")
-    lines = [line.strip() for line in text.split("\n")]
-    out = []
-    for line in lines:
-        if line:
-            out.append(line)
-        elif out and out[-1] != "":
-            out.append("")
-    return "\n".join(out).strip()
-
-
-def parse_file(path: str) -> dict | None:
-    """Parsar en documents-JSON till en dict med rader för documents + document_authors."""
-    with open(path, "r", encoding="utf-8-sig") as f:
-        data = json.load(f)
-    ds = data["dokumentstatus"]
-    dok = ds["dokument"]
-
-    doc_id = _clean(dok.get("doc_id"))
-    if not doc_id:
-        return None
-    if _clean(dok.get("doktyp")) not in (None, "mot"):
-        return None  # defensivt; mot-arkiven ska bara innehålla motioner
-
-    session_label = _clean(dok.get("session_label")) or ""
-    try:
-        year = int(session_label[:4])
-    except (ValueError, TypeError):
-        year = None
-
-    text = html_to_text(_clean(dok.get("html")) or "")
-    has_text = len(text) >= MIN_TEXT_LEN
-
-    authors = []
-    parties: list[str] = []
-    author_names: list[str] = []
-    intressenter = _as_list((ds.get("dokintressent") or {}).get("intressent"))
-    for i, x in enumerate(intressenter):
-        if not isinstance(x, dict):
-            continue
-        name = clean_speaker_name(_clean(x.get("name")) or "")
-        party = _clean(x.get("party"))
-        authors.append(
-            (doc_id, i, _clean(x.get("person_id")), name, party, _clean(x.get("role")))
-        )
-        if name:
-            author_names.append(name)
-        if party and party not in parties:
-            parties.append(party)
-
-    proposals_raw = _as_list((ds.get("dokforslag") or {}).get("proposals_raw"))
-    # Condensed, high-signal yrkanden: one row per <proposals_raw> for FTS + embeddings.
-    yrkanden = []
-    lydelser = []
-    for i, f in enumerate(proposals_raw):
-        if not isinstance(f, dict):
-            continue
-        text = _clean(f.get("text"))
-        if not text:
-            continue
-        lydelser.append(text)
-        yrkanden.append((
-            f"{doc_id}:{i}", doc_id, i, _clean(f.get("number")), text,
-            _clean(f.get("committee_recommendation")), _clean(f.get("chamber_decision")), _clean(f.get("handled_in")),
-        ))
-    proposals_text = " ".join(lydelser) or None
-
-    attachments = _as_list((ds.get("dokbilaga") or {}).get("bilaga"))
-    url_pdf = None
-    for b in attachments:
-        if isinstance(b, dict) and (b.get("filtyp") or "").lower() == "pdf":
-            url_pdf = _clean(b.get("fil_url"))
-            break
-
-    return {
-        "doc_id": doc_id,
-        "source_record_id": _clean(dok.get("source_record_id")),
-        "session_label": session_label or None,
-        "designation": _clean(dok.get("designation")),
-        "subtype": _clean(dok.get("subtype")),
-        "committee": _clean(dok.get("committee")),
-        "status": _clean(dok.get("status")),
-        "date": _parse_date(_clean(dok.get("date")) or ""),
-        "source_updated_at": _clean(dok.get("source_updated_at")),
-        "published_at": _clean(dok.get("published_at")),
-        "year": year,
-        "title": _clean(dok.get("title")),
-        "subtitle": _clean(dok.get("subtitel")) or _clean(dok.get("subtitle")),
-        "text": text,
-        "proposals_text": proposals_text,
-        "has_text": has_text,
-        "url_text": _clean(dok.get("url_text")),
-        "url_html": _clean(dok.get("url_html")),
-        "url_pdf": url_pdf,
-        "parties": parties,
-        "author_names": author_names,
-        "proposals_raw": json.dumps(proposals_raw, ensure_ascii=False) if proposals_raw else None,
-        "attachments": json.dumps(attachments, ensure_ascii=False) if attachments else None,
-        "num_proposals": len(proposals_raw),
-        "authors": authors,
-        "yrkanden": yrkanden,
-    }
-
-
-def process_folder(folder_path: str, already_processed: set[str] = frozenset()) -> list[dict]:
-    """Parsar JSON-filer i folder_path, returnerar documents-dicts (hoppar över kända ID:n)."""
-    docs = []
-    for file in os.listdir(folder_path):
-        if not file.endswith(".json"):
-            continue
-        # Filnamnet är doc_id (gemener) — hoppa över kända utan att parsa
-        if file[:-5].upper() in already_processed:
-            continue
-        try:
-            doc = parse_file(os.path.join(folder_path, file))
-            if doc is None or doc["doc_id"] in already_processed:
-                continue
-            docs.append(doc)
-        except (json.JSONDecodeError, KeyError, ValueError, TypeError) as e:
-            logging.warning(f"Skipping {file}: {e}")
-    return docs
-
-
-_UPSERT_SQL = """
-INSERT INTO documents (
-    doc_id, source_record_id, session_label, designation, subtype, committee, status,
-    date, source_updated_at, published_at, session_year,
-    title, subtitle, text, proposals_text, has_text,
-    url_text, url_html, url_pdf,
-    parties, author_names, proposals_raw, attachments, num_proposals
-) VALUES %s
-ON CONFLICT (doc_id) DO NOTHING
-"""
-# Framtida förbättring: DO UPDATE ... WHERE EXCLUDED.source_updated_at > documents.source_updated_at
-# för att plocka upp utskottsutfall på redan inlästa motioner.
-
-_AUTHORS_SQL = """
-INSERT INTO document_authors (doc_id, ordinal, person_id, name, party, role)
-VALUES %s
-ON CONFLICT (doc_id, ordinal) DO NOTHING
-"""
-
-_YRKANDEN_SQL = """
-INSERT INTO document_proposals (id, doc_id, ordinal, number, text, committee_recommendation, chamber_decision, handled_in)
-VALUES %s
-ON CONFLICT (id) DO NOTHING
-"""
-
-
-def _doc_to_row(doc: dict) -> tuple:
-    return (
-        doc["doc_id"],
-        doc["source_record_id"],
-        doc["session_label"],
-        doc["designation"],
-        doc["subtype"],
-        doc["committee"],
-        doc["status"],
-        doc["date"],
-        doc["source_updated_at"],
-        doc["published_at"],
-        doc["year"],
-        doc["title"],
-        doc["subtitle"],
-        doc["text"],
-        doc["proposals_text"],
-        doc["has_text"],
-        doc["url_text"],
-        doc["url_html"],
-        doc["url_pdf"],
-        doc["parties"],
-        doc["author_names"],
-        doc["proposals_raw"],
-        doc["attachments"],
-        doc["num_proposals"],
-    )
-
-
-def insert_docs(docs: list[dict]) -> None:
-    if not docs:
-        return
-    rows = [_doc_to_row(d) for d in docs]
-    pg.execute_values(_UPSERT_SQL, rows)
-    author_rows = [row for d in docs for row in d["authors"]]
-    if author_rows:
-        pg.execute_values(_AUTHORS_SQL, author_rows)
-    yrkande_rows = [row for d in docs for row in d["yrkanden"]]
-    if yrkande_rows:
-        pg.execute_values(_YRKANDEN_SQL, yrkande_rows)
-
-
-def update_folder(path: str, already_processed: set[str] = None) -> int:
-    """
-    Upsertar motioner från JSON-filer i path till PostgreSQL.
-    Returnerar antalet nya motioner.
-    """
-    if already_processed is None:
-        rows = pg.execute("SELECT doc_id FROM documents")
-        already_processed = {row["doc_id"] for row in rows}
-
-    docs = process_folder(path, already_processed)
-    insert_docs(docs)
-    return len(docs)
-
-
-if __name__ == "__main__":
-    existing = {row["doc_id"] for row in pg.execute("SELECT doc_id FROM documents")}
-    total = 0
-    for folder in sorted(os.listdir("motioner")):
-        path = str(bootstrap.DATA_DIR / 'motioner' / folder)
-        if not os.path.isdir(path):
-            continue
-        print(f"Processing {folder} …", end=" ", flush=True)
-        docs = process_folder(path, already_processed=existing)
-        insert_docs(docs)
-        existing |= {d["doc_id"] for d in docs}
-        total += len(docs)
-        print(f"{len(docs)} inserted")
-    print(f"\nTotal: {total} new documents inserted")

scripts/sync_motions.py

@@ -1,62 +0,0 @@
-"""
-Synkroniserar nya motioner från riksdagen.se till PostgreSQL-databasen.
-
-Pipeline (körs dagligen via systemd timer, se etc/riksdagen-documents-sync.*):
-  1. Ladda ned arkivet för aktuell fyraårsperiod (ersätter tidigare nerladdning)
-  2. Infoga nya motioner i PostgreSQL (hoppar över redan existerande via ON CONFLICT)
-  3. Bygg embeddings för motioner som saknar speech_chunks
-
-Alternativ till steg 1 om den dagliga nerladdningen (~50-110 MB) blir ett problem:
-dokumentlista-API:et (https://data.riksdagen.se/dokumentlista/?doktyp=mot&sort=source_updated_at
-&sortorder=desc&utformat=json) kan pagineras tills redan inlästa doc_id påträffas.
-
-Kör manuellt: python scripts/sync_motions.py
-"""
-from pathlib import Path
-
-import logging
-import os
-import sys
-
-sys.path.insert(0, str(Path(__file__).resolve().parents[1]))
-import bootstrap  # noqa: E402,F401  — sets cwd and sys.path to the project root
-
-logging.basicConfig(
-    level=logging.INFO,
-    format="%(asctime)s [%(levelname)s] %(message)s",
-)
-logger = logging.getLogger(__name__)
-
-
-def sync() -> None:
-    """Kör hela sync-pipelinen för motioner."""
-    logger.info("=== Starting daily documents sync ===")
-
-    # --- Steg 1: Ladda ned aktuell year ---
-    from scripts.download_motions import download_range, get_current_range
-    from scripts.sync_talks import get_current_session_year
-
-    current_range = get_current_range(get_current_session_year())
-    logger.info(f"Current range: {current_range}")
-    dir_path = download_range(current_range, force=True)
-
-    # --- Steg 2: Infoga nya motioner ---
-    logger.info("Stage 2: Inserting new documents into PostgreSQL...")
-    from scripts.motions_to_postgres import update_folder
-
-    new_motions = update_folder(os.path.abspath(dir_path))
-    logger.info(f"Stage 2 complete: {new_motions} new documents inserted")
-
-    # --- Steg 3: Chunk + bygg embeddings (fulltext + yrkanden) ---
-    logger.info("Stage 3: Chunking and embedding new documents...")
-    from scripts.make_embeddings import make_motion_embeddings, make_yrkande_embeddings
-
-    total_chunks = make_motion_embeddings()
-    total_yrkanden = make_yrkande_embeddings()
-    logger.info(f"Stage 3 complete: {total_chunks} speech_chunks + {total_yrkanden} yrkanden embedded")
-
-    logger.info("=== Motions sync complete ===")
-
-
-if __name__ == "__main__":
-    sync()

scripts/sync_talks.py

@@ -1,130 +0,0 @@
-"""
-Synkroniserar nya anföranden från riksdagen.se till PostgreSQL-databasen.
-
-Ersätter den ArangoDB-baserade versionen.
-
-Pipeline (körs dagligen via systemd timer):
-  1. Ladda ned årets anföranden från riksdagen.se (ersätter tidigare nerladdning)
-  2. Infoga nya anföranden i PostgreSQL (hoppar över redan existerande via ON CONFLICT)
-  3. Tilldela debatt-ID:n till anföranden som saknar det
-  4. Bygg embeddings för anföranden som saknar speech_chunks
-  5. Generera sammanfattningar för date som saknar summary
-
-Kör manuellt: python scripts/sync_talks.py
-"""
-from pathlib import Path
-
-import logging
-import os
-import sys
-from datetime import datetime
-from io import BytesIO
-from urllib.request import urlopen
-from zipfile import ZipFile
-
-sys.path.insert(0, str(Path(__file__).resolve().parents[1]))
-import bootstrap  # noqa: E402,F401  — sets cwd and sys.path to the project root
-
-from postgres_client import pg
-
-logging.basicConfig(
-    level=logging.INFO,
-    format="%(asctime)s [%(levelname)s] %(message)s",
-)
-logger = logging.getLogger(__name__)
-
-SYSTEM_MESSAGE = """Din uppgift är att sammanfatta debatter i Sveriges riksdag.
-Du kommer först att få enskilda tal som du ska sammanfatta var för sig, efter det ska du sammanfatta hela debatten.
-Sammanfattningarna ska vara på svenska och vara koncisa och informativa.
-Det är viktigt att du förstår vad som är kärnan i varje tal och debatt, fokusera därför på de argument och sakförhållanden som framförs.
-"""
-
-
-def get_current_session_year() -> int:
-    """
-    Returnerar startåret för aktuell riksdagssession.
-    Riksdagssessionen löper september–augusti.
-    """
-    now = datetime.now()
-    return now.year if now.month >= 9 else now.year - 1
-
-
-def download_current_year(year: int) -> str:
-    """Laddar ned och extraherar ZIP-arkivet för angiven riksdagssession."""
-    second_part = str(year + 1)[2:]
-    url = f"https://data.riksdagen.se/dataset/anforande/anforande-{year}{second_part}.json.zip"
-    folder_name = f"anforande-{year}{second_part}"
-    dir_path = os.path.join("speeches", folder_name)
-
-    logger.info(f"Downloading {url} → {dir_path}")
-    os.makedirs(dir_path, exist_ok=True)
-
-    for f in os.listdir(dir_path):
-        os.remove(os.path.join(dir_path, f))
-
-    with urlopen(url) as resp:
-        with ZipFile(BytesIO(resp.read())) as zf:
-            zf.extractall(dir_path)
-
-    count = len(os.listdir(dir_path))
-    logger.info(f"Extracted {count} files to {dir_path}")
-    return dir_path
-
-
-def get_unsummarized_dates() -> list[str]:
-    """Hämtar date som har anföranden utan sammanfattning."""
-    rows = pg.execute(
-        "SELECT DISTINCT date::text AS date FROM speeches WHERE summary IS NULL ORDER BY date"
-    )
-    dates = sorted(row["date"] for row in rows if row.get("date"))
-    logger.info(f"Found {len(dates)} dates with unsummarized speeches")
-    return dates
-
-
-def sync() -> None:
-    """Kör hela sync-pipelinen."""
-    logger.info("=== Starting daily riksdagen sync ===")
-
-    # --- Steg 1: Ladda ned ---
-    year = get_current_session_year()
-    logger.info(f"Current session year: {year}/{year + 1}")
-    dir_path = download_current_year(year)
-
-    # --- Steg 2: Infoga nya anföranden ---
-    logger.info("Stage 2: Inserting new speeches into PostgreSQL...")
-    from scripts.documents_to_postgres import update_folder
-
-    new_talks = update_folder(os.path.abspath(dir_path))
-    logger.info(f"Stage 2 complete: {new_talks} new speeches inserted")
-
-    # --- Steg 3: Tilldela debatt-ID:n ---
-    logger.info("Stage 3: Assigning debate IDs to speeches missing them...")
-    from scripts.debates import make_debate_ids
-
-    make_debate_ids()
-    logger.info("Stage 3 complete")
-
-    # --- Steg 4: Chunk + bygg embeddings ---
-    logger.info("Stage 4: Chunking and embedding new speeches...")
-    from scripts.make_embeddings import make_embeddings
-
-    total_chunks = make_embeddings()
-    logger.info(f"Stage 4 complete: {total_chunks} speech_chunks created")
-
-    # --- Steg 5: Generera sammanfattningar ---
-    new_dates = get_unsummarized_dates()
-    if new_dates:
-        logger.info(f"Stage 5: Generating summaries for {len(new_dates)} dates...")
-        from scripts.debates import process_debate_date
-
-        for date in new_dates:
-            process_debate_date(date, SYSTEM_MESSAGE)
-        logger.info(f"Stage 5 complete: summaries generated for {len(new_dates)} dates")
-    else:
-        logger.info("Stage 5: No unsummarized dates, skipping")
-
-    logger.info("=== Sync complete ===")
-
-
-if __name__ == "__main__":
-    sync()