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_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