feat(translation): tunable acceptance floor (0.90) + per-post sticky override (#155)
CI / lint (push) Successful in 3s
CI / frontend-build (push) Successful in 18s
CI / backend-lint-and-test (push) Successful in 27s
CI / integration (push) Successful in 3m52s

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
This commit is contained in:
2026-07-10 22:38:25 -04:00
parent a444cf82d1
commit aea2701c28
16 changed files with 579 additions and 52 deletions
+54
View File
@@ -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)