FabledCurator/tests/test_process_auto_apply.py

"""Process-group auto-apply sweep (#1464): wip / editor screenshot auto-tag at a
flat threshold with a PROVISIONAL source (`process_auto`) so the head never trains
on its own output, and stay VISIBLE (unlike chrome). Mirrors the chrome guards.
numpy-only (no sklearn), tested directly via the sync session."""
import pytest
from sqlalchemy import select

from backend.app.models import (
    ImageRecord,
    MLSettings,
    PresentationReview,
    Tag,
    TagHead,
    TagKind,
)
from backend.app.models.tag import image_tag
from backend.app.services.ml.heads import system_tag_auto_apply_sweep
from backend.app.services.ml.training_data import _ids_with_tag

pytestmark = pytest.mark.integration


def _emb(slot: int) -> list[float]:
    v = [0.0] * 1152
    v[slot] = 3.0
    return v


def _img(db, sha: str, emb) -> ImageRecord:
    img = ImageRecord(
        path=f"/images/{sha}.jpg", sha256=sha, size_bytes=1, mime="image/jpeg",
        width=1, height=1, origin="imported_filesystem",
        integrity_status="unknown", siglip_embedding=emb,
    )
    db.add(img)
    db.flush()
    return img


def _head(db, tag_id: int, slot: int, *, weight=1.0):
    s = db.execute(select(MLSettings).where(MLSettings.id == 1)).scalar_one()
    w = [0.0] * 1152
    w[slot] = weight
    db.add(TagHead(
        tag_id=tag_id, embedding_version=s.embedder_model_version,
        weights=w, bias=0.0, suggest_threshold=0.5, auto_apply_threshold=0.5,
        n_pos=60, n_neg=90, ap=0.9, precision_cv=0.98, recall=0.7,
    ))


def _system_tag(db, name):
    return db.execute(
        select(Tag).where(Tag.is_system.is_(True), Tag.name == name)
    ).scalar_one()


def _enable_process(db):
    # process auto-apply is opt-in (default False) — turn it on for these tests.
    db.execute(select(MLSettings).where(MLSettings.id == 1)).scalar_one().process_auto_apply_enabled = True


def _source(db, image_id, tag_id):
    return db.execute(
        select(image_tag.c.source)
        .where(image_tag.c.image_record_id == image_id)
        .where(image_tag.c.tag_id == tag_id)
    ).scalar_one_or_none()


def test_process_sweep_applies_wip_and_editor(db_sync):
    _enable_process(db_sync)
    wip = _system_tag(db_sync, "wip")
    editor = _system_tag(db_sync, "editor screenshot")
    _head(db_sync, wip.id, 0, weight=3.0)
    _head(db_sync, editor.id, 1, weight=3.0)
    w_img = _img(db_sync, "a" * 64, _emb(0))
    e_img = _img(db_sync, "b" * 64, _emb(1))
    db_sync.commit()
    res = system_tag_auto_apply_sweep(db_sync, mode="process")
    assert res["n_applied"] == 2
    assert _source(db_sync, w_img.id, wip.id) == "process_auto"
    assert _source(db_sync, e_img.id, editor.id) == "process_auto"


def test_process_sweep_disabled_by_default_is_noop(db_sync):
    # process_auto_apply_enabled defaults False (opt-in) — no enable = no-op.
    wip = _system_tag(db_sync, "wip")
    _head(db_sync, wip.id, 0, weight=3.0)
    img = _img(db_sync, "c" * 64, _emb(0))
    db_sync.commit()
    res = system_tag_auto_apply_sweep(db_sync, mode="process")
    assert res["n_applied"] == 0
    assert _source(db_sync, img.id, wip.id) is None


def test_process_sweep_skips_valued_image(db_sync):
    # Guard 1: never auto-apply to an image the operator already content-tagged.
    _enable_process(db_sync)
    wip = _system_tag(db_sync, "wip")
    _head(db_sync, wip.id, 0, weight=3.0)
    content = Tag(name="mychar", kind=TagKind.character)
    db_sync.add(content)
    db_sync.flush()
    img = _img(db_sync, "d" * 64, _emb(0))
    db_sync.execute(image_tag.insert().values(
        image_record_id=img.id, tag_id=content.id, source="manual"))
    db_sync.commit()
    res = system_tag_auto_apply_sweep(db_sync, mode="process")
    assert res["n_applied"] == 0
    assert _source(db_sync, img.id, wip.id) is None


def test_process_sweep_flags_conflict_with_process_mode(db_sync):
    # Guard 2: also scores high on a content head → still applied, but flagged
    # for review with mode='process' (the ring-loud guard).
    _enable_process(db_sync)
    wip = _system_tag(db_sync, "wip")
    _head(db_sync, wip.id, 0, weight=3.0)
    content = Tag(name="looksreal", kind=TagKind.general)
    db_sync.add(content)
    db_sync.flush()
    _head(db_sync, content.id, 0, weight=1.0)
    img = _img(db_sync, "e" * 64, _emb(0))
    db_sync.commit()
    res = system_tag_auto_apply_sweep(db_sync, mode="process")
    assert res["n_applied"] == 1
    assert res["n_flagged"] == 1
    flag = db_sync.execute(
        select(PresentationReview).where(
            PresentationReview.image_record_id == img.id,
            PresentationReview.tag_id == wip.id,
        )
    ).scalar_one()
    assert flag.mode == "process"
    assert flag.conflict_tag_id == content.id


def test_process_auto_source_never_trains_head(db_sync):
    # The runaway break: provisional wip tags (process sweep 'process_auto', soft
    # title 'wip_title_soft') are NOT training positives; a HARD title-heuristic /
    # manual one IS. So the head learns only from trusted labels, never its own
    # output or the low-precision sketch/doodle tier (#1464 + #1474).
    wip = _system_tag(db_sync, "wip")
    auto_img = _img(db_sync, "f" * 64, _emb(0))
    soft_img = _img(db_sync, "9" * 64, _emb(2))
    title_img = _img(db_sync, "0" * 64, _emb(1))
    db_sync.execute(image_tag.insert().values(
        image_record_id=auto_img.id, tag_id=wip.id, source="process_auto"))
    db_sync.execute(image_tag.insert().values(
        image_record_id=soft_img.id, tag_id=wip.id, source="wip_title_soft"))
    db_sync.execute(image_tag.insert().values(
        image_record_id=title_img.id, tag_id=wip.id, source="wip_title"))
    db_sync.commit()
    positives = set(_ids_with_tag(db_sync, wip.id))
    assert title_img.id in positives       # trusted HARD label trains the head
    assert auto_img.id not in positives    # its own auto-applied output does NOT
    assert soft_img.id not in positives    # low-precision soft tier does NOT


def test_soft_wip_conflict_audit_flags_ring_loud(db_sync):
    # A soft-tagged image (sketch/doodle title) that ALSO scores high on a content
    # head is probably finished art mis-tagged — flagged for review; a quiet one is not.
    from backend.app.services.ml.heads import soft_wip_conflict_audit

    s = db_sync.execute(select(MLSettings).where(MLSettings.id == 1)).scalar_one()
    s.process_conflict_threshold = 0.6
    wip = _system_tag(db_sync, "wip")
    content = Tag(name="looksreal", kind=TagKind.general)
    db_sync.add(content)
    db_sync.flush()
    _head(db_sync, content.id, 0, weight=1.0)   # sigmoid(1)=0.73 > 0.6 conflict
    ring = _img(db_sync, "1" * 64, _emb(0))      # scores on the content head
    quiet = _img(db_sync, "2" * 64, _emb(5))     # orthogonal → 0.5 < 0.6
    for img in (ring, quiet):
        db_sync.execute(image_tag.insert().values(
            image_record_id=img.id, tag_id=wip.id, source="wip_title_soft"))
    db_sync.commit()

    res = soft_wip_conflict_audit(db_sync)
    assert res["n_flagged"] == 1
    flag = db_sync.execute(
        select(PresentationReview).where(PresentationReview.image_record_id == ring.id)
    ).scalar_one()
    assert flag.mode == "process"
    assert flag.conflict_tag_id == content.id
    assert db_sync.execute(
        select(PresentationReview).where(PresentationReview.image_record_id == quiet.id)
    ).scalar_one_or_none() is None