Add naturalization pass — 9,025 sayings, 36K training pairs

New pipeline step: naturalize_corpus.py runs Prompt A ("dialect coach")
over both polished and previously-discarded sayings, recovering material
the first polish pass was too aggressive with.

Results:
- 9,468 usable from naturalization (vs 5,499 from initial polish)
- After dedup: 9,025 unique sayings (was 2,312)
- 36,079 training pairs (was 9,257)
- 100% vocab coverage, avg 10.1 words (punchier than 13.1)
- Relaxed quality filter: drops artifacts/nonsense, not noun presence

New scripts:
- naturalize_corpus.py: gentle LLM naturalization pass, resume-safe
- rebuild_training_pairs.py: combined filter + dedup + training pair
  generation from naturalized corpus, replaces separate steps

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

corpus/corpus_stats.json

@@ -1,91 +1,31 @@
 {
-  "raw_count": 9835,
-  "raw_by_template": {
-    "deconstruction": 1500,
-    "denial_of_consequences": 1500,
-    "false_equivalence": 1500,
-    "futile_preparation": 1500,
-    "hypocritical_complaint": 1500,
-    "ironic_deficiency": 1500,
-    "tautological_wisdom": 835
+  "naturalization_input": 19540,
+  "naturalization_status": {
+    "skipped": 436,
+    "naturalized": 18578,
+    "unchanged": 453,
+    "filtered": 73
   },
-  "polished_count": 5499,
-  "discarded_during_polish": 4336,
-  "errors_during_polish": 0,
-  "polish_discard_rate": "44.1%",
-  "polished_by_template": {
-    "deconstruction": 1105,
-    "denial_of_consequences": 733,
-    "false_equivalence": 590,
-    "futile_preparation": 882,
-    "hypocritical_complaint": 573,
-    "ironic_deficiency": 831,
-    "tautological_wisdom": 785
+  "usable_before_dedup": 19031,
+  "duplicates_removed": 10006,
+  "final_filtered": 9025,
+  "training_pairs": 36079,
+  "by_template": {
+    "deconstruction": 1544,
+    "denial_of_consequences": 750,
+    "false_equivalence": 1897,
+    "futile_preparation": 1735,
+    "hypocritical_complaint": 811,
+    "ironic_deficiency": 1563,
+    "tautological_wisdom": 725
   },
-  "discarded_by_template": {
-    "deconstruction": 395,
-    "denial_of_consequences": 767,
-    "false_equivalence": 910,
-    "futile_preparation": 618,
-    "hypocritical_complaint": 927,
-    "ironic_deficiency": 669,
-    "tautological_wisdom": 50
+  "by_input_type": {
+    "category_seeded": 9025,
+    "open_ended": 2146,
+    "persona_seeded": 9025,
+    "template_seeded": 6858,
+    "word_seeded": 9025
   },
-  "filtered_count": 2312,
-  "filtered_by_template": {
-    "deconstruction": 619,
-    "denial_of_consequences": 159,
-    "false_equivalence": 517,
-    "futile_preparation": 284,
-    "hypocritical_complaint": 168,
-    "ironic_deficiency": 358,
-    "tautological_wisdom": 207
-  },
-  "discarded_during_filter": 3187,
-  "training_pair_count": 9257,
-  "training_by_template": {
-    "deconstruction": 2488,
-    "denial_of_consequences": 630,
-    "false_equivalence": 2059,
-    "futile_preparation": 1146,
-    "hypocritical_complaint": 681,
-    "ironic_deficiency": 1429,
-    "tautological_wisdom": 824
-  },
-  "training_by_input_type": {
-    "category_seeded": 2312,
-    "open_ended": 562,
-    "persona_seeded": 2312,
-    "template_seeded": 1759,
-    "word_seeded": 2312
-  },
-  "unique_slot_words_used": 609,
-  "total_vocab_words": 624,
-  "vocab_coverage": "97.6%",
-  "words_never_used": [
-    "agate",
-    "alabaster",
-    "anise",
-    "azalea",
-    "bee",
-    "blowfish",
-    "cattail",
-    "cypress",
-    "emerald",
-    "gem",
-    "grebe",
-    "juniper",
-    "lyre",
-    "spear",
-    "theater"
-  ],
-  "words_never_used_count": 15,
-  "avg_saying_length_words": 13.1,
-  "min_saying_length_words": 6,
-  "max_saying_length_words": 23,
-  "balance_warnings": [
-    "WARNING: denial_of_consequences has only 159 entries (6.9%) — below 10% threshold. Generate more raw sayings for this family.",
-    "WARNING: hypocritical_complaint has only 168 entries (7.3%) — below 10% threshold. Generate more raw sayings for this family.",
-    "WARNING: tautological_wisdom has only 207 entries (9.0%) — below 10% threshold. Generate more raw sayings for this family."
-  ]
+  "vocab_coverage": "624/624 (100.0%)",
+  "avg_length_words": 10.1
 }