Model/scripts/backfill_site_notes_epc_property_uprn.py

"""Backfill ``epc_property.uprn`` on PasHub site-notes rows written before the
UPRN-persistence fix (#1538). Targets have ``uprn IS NULL`` and a non-null
``uploaded_file_id`` (the site-notes signature; these rows carry no ``property_id``
/ ``portfolio_id``).

For each target the UPRN is recovered, in priority order, from (1)
``uploaded_files.uprn``, then when that's null (2) ``property.landlord_property_id``
or (3) ``hubspot_deal_data.uprn`` (via ``uploaded_files.hubspot_deal_id``). Every
method is validated against the target portfolio's ``property`` table — the only
handle for scoping these portfolio-less rows — and the written value is that
property row's ``uprn``. Rows matching no property, or ambiguously matching several
UPRNs, are skipped and reported. ``uprn_source`` is left untouched, as in #1538.

Set ``PORTFOLIO_ID`` below before running. DRY-RUN by default (writes nothing);
pass ``--apply`` to execute in a transaction. Idempotent — re-running is a no-op.

Run with:  ``python -m scripts.backfill_site_notes_epc_property_uprn [--apply]``
"""

from __future__ import annotations

import argparse

from sqlalchemy import Connection, text

from scripts.e2e_common import build_engine, load_env

# ---------------------------------------------------------------------------
# Set this to the portfolio you are backfilling before running the script.
# ---------------------------------------------------------------------------
PORTFOLIO_ID: int = 0  # TODO: set the target portfolio id


# Target rows: site-notes-mapped epc_property rows still missing a UPRN. The
# non-null uploaded_file_id is the site-notes signature (predicted / lodged-EPC
# rows don't have one), and uprn IS NULL is the pre-#1538 gap.
#
# For each we join uploaded_files (the only link to a property), the deal row
# (hubspot_deal_id -> hubspot_deal_data.deal_id), and the target portfolio's
# property table, resolving the UPRN from uploaded_files.uprn or, when that's null,
# via landlord_property_id or the deal's uprn. The value written is always the
# property row's uprn. GROUP BY ... HAVING count(DISTINCT p.uprn) = 1 drops any row
# that resolves to more than one portfolio UPRN (ambiguous — left for a human).
# hubspot_deal_data.uprn is text, so it's numeric-guarded (^[0-9]+$) before casting.
_RESOLVED_CTE = """
    WITH target AS (
        SELECT e.id AS epc_property_id, e.uploaded_file_id
          FROM epc_property e
         WHERE e.uploaded_file_id IS NOT NULL
           AND e.uprn IS NULL
    ),
    resolved AS (
        SELECT t.epc_property_id,
               min(p.uprn) AS resolved_uprn,
               bool_or(uf.uprn IS NOT NULL AND p.uprn = uf.uprn) AS via_uploaded_file,
               bool_or(uf.uprn IS NULL
                       AND uf.landlord_property_id IS NOT NULL
                       AND p.landlord_property_id = uf.landlord_property_id
                       ) AS via_property_lpi,
               bool_or(uf.uprn IS NULL
                       AND btrim(hdd.uprn) ~ '^[0-9]+$'
                       AND p.uprn = btrim(hdd.uprn)::bigint
                       ) AS via_hubspot_deal,
               count(DISTINCT p.uprn) AS distinct_uprns
          FROM target t
          JOIN uploaded_files uf ON uf.id = t.uploaded_file_id
          LEFT JOIN hubspot_deal_data hdd ON hdd.deal_id = uf.hubspot_deal_id
          JOIN property p
            ON p.portfolio_id = :portfolio_id
           AND p.uprn IS NOT NULL
           AND (
                 (uf.uprn IS NOT NULL AND p.uprn = uf.uprn)
              OR (uf.uprn IS NULL
                  AND uf.landlord_property_id IS NOT NULL
                  AND p.landlord_property_id = uf.landlord_property_id)
              OR (uf.uprn IS NULL
                  AND btrim(hdd.uprn) ~ '^[0-9]+$'
                  AND p.uprn = btrim(hdd.uprn)::bigint)
               )
         GROUP BY t.epc_property_id
    )
"""

_REPORT = text(
    _RESOLVED_CTE
    + """
    SELECT
        (SELECT count(*) FROM target)                                       AS targets,
        count(*) FILTER (WHERE distinct_uprns = 1)                          AS resolvable,
        count(*) FILTER (WHERE distinct_uprns = 1
                         AND via_uploaded_file)                             AS via_uploaded_file,
        count(*) FILTER (WHERE distinct_uprns = 1
                         AND NOT via_uploaded_file
                         AND via_property_lpi)                              AS via_property_lpi,
        count(*) FILTER (WHERE distinct_uprns = 1
                         AND NOT via_uploaded_file
                         AND NOT via_property_lpi
                         AND via_hubspot_deal)                              AS via_hubspot_deal,
        count(*) FILTER (WHERE distinct_uprns > 1)                          AS ambiguous
      FROM resolved
    """
)

_UPDATE = text(
    _RESOLVED_CTE
    + """
    UPDATE epc_property e
       SET uprn = r.resolved_uprn
      FROM resolved r
     WHERE e.id = r.epc_property_id
       AND r.distinct_uprns = 1
       AND e.uprn IS NULL
    """
)


def backfill(conn: Connection, *, portfolio_id: int, apply: bool) -> dict[str, int]:
    """Report (and, when ``apply``, execute) the UPRN backfill for ``portfolio_id``.

    Returns the diagnostic counts: total ``targets`` (site-notes rows missing a
    UPRN), how many are ``resolvable`` within the portfolio (split by the method
    that resolved them — ``via_uploaded_file`` / ``via_property_lpi`` /
    ``via_hubspot_deal``, in that priority order), how many are ``ambiguous``
    (skipped), and — when applied — how many rows were ``updated``."""
    row = conn.execute(_REPORT, {"portfolio_id": portfolio_id}).one()
    counts = {
        "targets": row.targets,
        "resolvable": row.resolvable,
        "via_uploaded_file": row.via_uploaded_file,
        "via_property_lpi": row.via_property_lpi,
        "via_hubspot_deal": row.via_hubspot_deal,
        "ambiguous": row.ambiguous,
    }
    if apply:
        result = conn.execute(_UPDATE, {"portfolio_id": portfolio_id})
        counts["updated"] = result.rowcount
    return counts


def main() -> None:
    load_env()
    parser = argparse.ArgumentParser(description=__doc__)
    parser.add_argument(
        "--apply",
        action="store_true",
        help="execute the updates (default: dry-run, writes nothing)",
    )
    args = parser.parse_args()

    if PORTFOLIO_ID <= 0:
        raise SystemExit(
            "Set PORTFOLIO_ID at the top of this script to the target portfolio "
            "before running."
        )

    engine = build_engine()
    with engine.begin() as conn:
        conn.execute(text("SET statement_timeout = 120000"))
        counts = backfill(conn, portfolio_id=PORTFOLIO_ID, apply=args.apply)

    unresolved = counts["targets"] - counts["resolvable"] - counts["ambiguous"]
    verb = "backfilled" if args.apply else "would backfill"
    print(f"Portfolio {PORTFOLIO_ID}: {counts['targets']} site-notes epc_property "
          f"row(s) missing a UPRN.")
    print(
        f"  {verb} {counts['resolvable']} row(s) "
        f"({counts['via_uploaded_file']} via uploaded_files.uprn, "
        f"{counts['via_property_lpi']} via property.landlord_property_id, "
        f"{counts['via_hubspot_deal']} via hubspot_deal_data.uprn)."
    )
    if counts["ambiguous"]:
        print(
            f"  SKIPPED {counts['ambiguous']} row(s) that matched more than one "
            "portfolio UPRN (ambiguous — resolve by hand)."
        )
    if unresolved:
        print(
            f"  SKIPPED {unresolved} row(s) with no matching property in this "
            "portfolio (no uploaded_files.uprn, landlord_property_id or deal match)."
        )
    if args.apply:
        print(f"\nApplied. {counts.get('updated', 0)} row(s) updated.")
    else:
        print("\nDRY-RUN — nothing written. Re-run with --apply to execute.")


if __name__ == "__main__":
    main()