feat(translation): tunable acceptance floor (0.90) + per-post sticky override (#155)

The gate at a fixed 0.80 couldn't catch the real pain: Interpreter (fresh ==
cached, verified by probe) confidently mis-detects short ASCII English like
"... WIP Part 1" as German at 0.86 — above the floor — so it was accepted and a
re-translate reproduced it. Confidence alone can't separate the 0.86 collision
(genuine German lands there too), and single-word mis-flags sit at a confident
1.0 no floor catches.

Two operator-approved levers:

- Acceptance floor is now a live Settings value (ImportSettings.
  translation_min_confidence, default 0.90; surfaced in the Translation card), so
  it's tunable without a redeploy. _accept takes the threshold as a parameter.

- Per-post sticky override (Post.translation_override: auto/force/original).
  'force' stores a translation even below the floor (rescue a skipped
  legit-foreign title); 'original' keeps the original and clears any stored
  translation (kill a confident mis-flag no floor catches). The sweep honors it
  on every run and _reset_translations skips 'original', so the choice survives a
  Re-translate-all. POST /api/posts/<id>/translation-override applies it
  immediately (translate now when the service is up, else queue for the sweep).
  UI: PostTranslationControl on the posts-feed card.

Migration 0084 (both columns + a CHECK on the override). The feed + provenance
serializers expose translation_override.

With a stricter floor the rollback finally works: raise it -> Re-translate all ->
the 0.86 mis-flags are rejected and restored to the original; force /
keep-original handle the residual either way.

Tests: gate thresholds against the param (0.86 rejected at 0.90, explicit-floor
cases); sweep force/original + re-translate-skips-original; override endpoint
(validation, original clears, force queues when disabled, feed exposes it);
settings min_confidence default/save/validate.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CgZP9v2otxVJymiYsnVuMy

tests/test_api_posts.py

@@ -184,6 +184,68 @@ async def test_rejects_bad_direction(client):
     assert (await resp.get_json())["error"] == "invalid_direction"
 
 
+@pytest.mark.asyncio
+async def test_translation_override_rejects_bad_value(client, seeded_post):
+    _, _, post = seeded_post
+    resp = await client.post(
+        f"/api/posts/{post.id}/translation-override", json={"override": "nope"}
+    )
+    assert resp.status_code == 400
+    assert (await resp.get_json())["error"] == "invalid_override"
+
+
+@pytest.mark.asyncio
+async def test_translation_override_404_for_unknown(client):
+    resp = await client.post(
+        "/api/posts/999999/translation-override", json={"override": "original"}
+    )
+    assert resp.status_code == 404
+
+
+@pytest.mark.asyncio
+async def test_translation_override_original_clears_and_keeps(client, db, seeded_post):
+    # 'keep original' clears a stored translation immediately (no Interpreter) and
+    # marks the post handled (== target). Asserts on the response, which reflects
+    # the endpoint's own committed session.
+    _, _, post = seeded_post
+    post.post_title_translated = "STALE"
+    post.description_translated = "STALE"
+    post.translated_source_lang = "de"
+    await db.commit()
+    resp = await client.post(
+        f"/api/posts/{post.id}/translation-override", json={"override": "original"}
+    )
+    assert resp.status_code == 200
+    body = await resp.get_json()
+    assert body["translation_override"] == "original"
+    assert body["applied"] == "cleared"
+    assert body["post_title_translated"] is None
+    assert body["translated_source_lang"] == "en"
+
+
+@pytest.mark.asyncio
+async def test_translation_override_force_queues_when_disabled(client, seeded_post):
+    # Translation disabled (default) → can't translate inline; the override is
+    # saved and the post is queued (columns NULLed) for the next sweep.
+    _, _, post = seeded_post
+    resp = await client.post(
+        f"/api/posts/{post.id}/translation-override", json={"override": "force"}
+    )
+    assert resp.status_code == 200
+    body = await resp.get_json()
+    assert body["translation_override"] == "force"
+    assert body["applied"] == "queued"
+    assert body["translated_source_lang"] is None
+
+
+@pytest.mark.asyncio
+async def test_list_exposes_translation_override(client, seeded_post):
+    resp = await client.get("/api/posts")
+    assert resp.status_code == 200
+    body = await resp.get_json()
+    assert body["items"][0]["translation_override"] == "auto"   # default
+
+
 @pytest.mark.asyncio
 async def test_detail_returns_uncapped_thumbnails(client, db):
     """Feed query caps thumbnails at 6 for previews; detail endpoint

tests/test_api_settings.py

@@ -53,6 +53,26 @@ async def test_translation_settings_defaults_and_patch(client):
         "/api/settings/import", json={"interpreter_base_url": 123})).status_code == 400
 
 
+@pytest.mark.asyncio
+async def test_translation_min_confidence_default_and_patch(client):
+    # #155: the latin-script acceptance floor is an operator-tunable setting,
+    # default 0.9 (stricter than the old hardcoded 0.8).
+    body = await (await client.get("/api/settings/import")).get_json()
+    assert body["translation_min_confidence"] == 0.9
+
+    ok = await client.patch(
+        "/api/settings/import", json={"translation_min_confidence": 0.95}
+    )
+    assert ok.status_code == 200
+    assert (await ok.get_json())["translation_min_confidence"] == 0.95
+
+    # Out-of-range + wrong-type are rejected.
+    assert (await client.patch(
+        "/api/settings/import", json={"translation_min_confidence": 1.5})).status_code == 400
+    assert (await client.patch(
+        "/api/settings/import", json={"translation_min_confidence": "high"})).status_code == 400
+
+
 @pytest.mark.asyncio
 async def test_translation_status_defaults(client):
     # Off + no URL → no network health call, healthy False (#143).

tests/test_translate_posts.py

@@ -314,6 +314,41 @@ def test_translate_posts_accepts_low_confidence_cjk(db_sync, monkeypatch):
     assert p.translated_source_lang == "zh"
 
 
+def test_translate_posts_force_override_translates_below_floor(db_sync, monkeypatch):
+    # override='force' stores the translation even below the confidence floor —
+    # rescuing a legitimately-foreign title the gate would otherwise skip.
+    _patch(monkeypatch, db_sync)
+    _mock_lang(monkeypatch, lang="de", conf=0.42)     # below the 0.90 default floor
+    a = _artist(db_sync)
+    p = _post(db_sync, a.id, title="Kurz", ext="p1")
+    p.translation_override = "force"
+    _enable(db_sync)
+    db_sync.commit()
+
+    translate_posts()
+    db_sync.refresh(p)
+    assert p.post_title_translated == "EN:Kurz"        # forced through despite 0.42
+    assert p.translated_source_lang == "de"
+
+
+def test_translate_posts_original_override_keeps_original(db_sync, monkeypatch):
+    # override='original' never translates — even a confident (0.98) detection is
+    # kept as the original. The sweep marks it handled (== target) and stores
+    # nothing; Interpreter isn't consulted for the decision.
+    _patch(monkeypatch, db_sync)
+    _mock_lang(monkeypatch, lang="de", conf=0.98)     # would be accepted under auto
+    a = _artist(db_sync)
+    p = _post(db_sync, a.id, title="Das Mädchen", ext="p1")
+    p.translation_override = "original"
+    _enable(db_sync)
+    db_sync.commit()
+
+    translate_posts()
+    db_sync.refresh(p)
+    assert p.post_title_translated is None             # kept original
+    assert p.translated_source_lang == "en"            # marked handled
+
+
 def _mock_new_model(monkeypatch, *, ver="v2"):
     """Interpreter is up and translates with a NEW engine version."""
     monkeypatch.setattr("backend.app.tasks.translation.ic.health", lambda *a, **k: True)
@@ -344,6 +379,25 @@ def test_retranslate_resets_and_reruns(db_sync, monkeypatch):
     assert p.translation_engine_version == "v2"
 
 
+def test_retranslate_leaves_keep_original_posts(db_sync, monkeypatch):
+    # A 'keep original' post survives a Re-translate-all: _reset_translations skips
+    # override='original', so it stays handled + untranslated instead of being
+    # re-run (which would re-introduce the mis-flag the operator killed).
+    _patch(monkeypatch, db_sync)
+    _mock_new_model(monkeypatch)
+    a = _artist(db_sync)
+    p = _post(db_sync, a.id, title="ねこ", ext="p1")
+    p.translation_override = "original"
+    p.translated_source_lang = "en"       # handled by a prior 'keep original' apply
+    _enable(db_sync)
+    db_sync.commit()
+
+    retranslate_posts(artist_ids=[a.id])
+    db_sync.refresh(p)
+    assert p.translated_source_lang == "en"     # not reset
+    assert p.post_title_translated is None       # not re-translated
+
+
 def test_retranslate_artist_scope_leaves_others(db_sync, monkeypatch):
     # Scoping to one artist must not touch another artist's translations.
     _patch(monkeypatch, db_sync)

tests/test_translation_gate.py

@@ -1,26 +1,39 @@
-"""Translation acceptance gate (#1376) — pure unit tests for ``_accept``, which
-decides whether curator stores Interpreter's translation or keeps the original.
-Curator does no detection of its own; it only thresholds Interpreter's reported
-detected language + confidence (Scribe rule 133). No DB, no network."""
-from backend.app.tasks.translation import _MIN_LATIN_CONFIDENCE, _accept
+"""Translation acceptance gate (#1376, #155) — pure unit tests for ``_accept``,
+which decides whether curator stores Interpreter's translation or keeps the
+original. Curator does no detection of its own; it only thresholds Interpreter's
+reported detected language + confidence against the operator-tunable floor
+(Scribe rule 133). No DB, no network."""
+from backend.app.tasks.translation import _DEFAULT_MIN_LATIN_CONFIDENCE, _accept
 
 
-def test_accept_trusts_cjk_regardless_of_confidence():
-    # ja/ko/zh are script-detected — accepted even at a low reported number
-    # (pure-kanji Japanese can come back labelled zh ~0.75, still translated).
+def test_accept_trusts_cjk_regardless_of_confidence_or_floor():
+    # ja/ko/zh are script-detected — accepted even at a low reported number and
+    # even against a high floor (pure-kanji Japanese can come back labelled zh
+    # ~0.75, still translated).
     assert _accept("ja", None) is True
-    assert _accept("ja", 0.10) is True
+    assert _accept("ja", 0.10, 0.99) is True
     assert _accept("ko", 0.99) is True
     assert _accept("zh", 0.75) is True
     assert _accept("zh-CN", 0.20) is True      # region suffix is still CJK
 
 
-def test_accept_requires_floor_for_latin():
+def test_accept_default_floor_is_strict():
+    # Default floor is 0.90 (#155): the ~0.86 band — where genuine German AND
+    # mis-flagged short English collide — is now REJECTED (the whole point of the
+    # stricter default). Genuine high-confidence non-English (~1.0) still passes.
     assert _accept("de", 0.98) is True
-    assert _accept("de", 0.86) is True    # calibrated: ambiguous-but-correct German stays
-    assert _accept("de", _MIN_LATIN_CONFIDENCE) is True          # exactly at floor
-    assert _accept("de", _MIN_LATIN_CONFIDENCE - 0.01) is False
-    assert _accept("fr", 0.40) is False                          # short mis-flag
+    assert _accept("de", 0.86) is False            # the collision band, now cut
+    assert _accept("de", _DEFAULT_MIN_LATIN_CONFIDENCE) is True          # at floor
+    assert _accept("de", _DEFAULT_MIN_LATIN_CONFIDENCE - 0.01) is False
+    assert _accept("fr", 0.40) is False            # short mis-flag
+
+
+def test_accept_honors_an_explicit_floor():
+    # The floor is a parameter now (fed from ImportSettings). A looser floor keeps
+    # the 0.86 band; a stricter one rejects even a fairly confident detection.
+    assert _accept("de", 0.86, 0.80) is True
+    assert _accept("de", 0.86, 0.95) is False
+    assert _accept("de", 0.96, 0.95) is True
 
 
 def test_accept_is_case_insensitive():
@@ -32,3 +45,4 @@ def test_accept_missing_confidence_fails_open():
     # No confidence reported → don't silently drop a translation.
     assert _accept("de", None) is True
     assert _accept("", None) is True
+    assert _accept("de", None, 0.99) is True       # fail-open beats a high floor