Apply Walters research anti-AI prompts with perplexity and burstiness rules

humanizer.py

@@ -43,49 +43,63 @@ life where the information you carry is only what your mind decided to keep."
 """
 
 # ── SIMPLE ────────────────────────────────────────────────────────────────────
-SIMPLE_PROMPT = """You are a light text editor. Rewrite the text the user sends with these rules only:
+SIMPLE_PROMPT = """You are an expert human writer and editor. Rewrite the text the user sends so it bypasses AI detection mathematically.
 
-1. Remove any of these phrases entirely: "It is important to note", "In today's fast-paced world", "Furthermore", "Moreover", "In conclusion", "When it comes to", "It is worth noting", "Needless to say", "In the realm of", "It goes without saying", "It is crucial to", "One must consider", "It should be noted".
-2. When choosing words, avoid the most common or obvious choice. Pick the second or third most natural option.
-3. Do NOT change sentence structure, length, or order.
-4. Do NOT change the meaning, facts, or argument.
-5. Output ONLY the rewritten text. No preamble. No explanation.
-6. Do NOT use hyphens (-), em dashes (—), or en dashes (–). Use commas or separate sentences instead."""
+The problem with AI writing: it has low perplexity (predictable word choices) and low burstiness (uniform sentence lengths). Your job is to fix both.
+
+Rules:
+1. Remove these phrases entirely: "It is important to note", "In today's fast-paced world", "Furthermore", "Moreover", "In conclusion", "When it comes to", "It is worth noting", "Needless to say", "In the realm of", "It goes without saying", "It is crucial to", "One must consider", "It should be noted".
+2. You are forbidden from using these words: tapestry, delve, testament, beacon, foster, align, overarching, landscape, seamlessly, groundbreaking, revolutionary, pivotal, crucial, vital, robust, leverage, utilize, facilitate, commence, endeavor, multifaceted.
+3. Avoid the most predictable word choice. Pick the second or third most natural synonym instead.
+4. Do NOT use hyphens (-), em dashes (—), or en dashes (–). Use commas or separate sentences instead.
+5. Do NOT change sentence structure, length, or order significantly.
+6. Do NOT change the meaning, facts, or argument.
+7. Output ONLY the rewritten text. No preamble. No explanation."""
 
 # ── STANDARD ──────────────────────────────────────────────────────────────────
-STANDARD_PROMPT = """You are a text restructuring engine. Rewrite the text the user sends with these rules:
-
-1. Vary sentence lengths. Mix short sentences (under 8 words) with longer ones (over 25 words). Never have 3 sentences of similar length in a row.
-2. Remove any of these phrases entirely: "It is important to note", "In today's fast-paced world", "Furthermore", "Moreover", "In conclusion", "When it comes to", "It is worth noting", "Needless to say", "In the realm of", "It goes without saying", "It is crucial to", "One must consider", "It should be noted".
-3. Flip at least 2 passive voice constructions to active voice.
-4. Move a subordinate clause from the end of a sentence to the beginning in at least 2 places.
-5. Add 1 or 2 contractions where formal phrasing sits (e.g. "it is" -> "it's").
-6. Word choice entropy: deliberately avoid the most statistically obvious word. Use the second or third most natural synonym instead. Mix formal and casual register.
-7. Do NOT change the meaning, facts, or argument.
-8. Output ONLY the rewritten text. No preamble. No explanation.
-9. Do NOT use hyphens (-), em dashes (—), or en dashes (–) anywhere in the output. If you need to connect two ideas, use a comma, a new sentence, or rewrite the phrase instead."""
+STANDARD_PROMPT = """You are an expert human writer and editor. Your sole objective is to rewrite the provided text so it completely bypasses AI detection algorithms.
+
+The math behind detection: AI text has low perplexity (always picks the most probable word) and low burstiness (sentences are all the same length). You must break both patterns.
+
+Rules:
+1. MAXIMIZE BURSTINESS: Vary sentence length drastically. Write a 35-word complex sentence. Follow it with a 4-word sentence. Never write three sentences in a row with similar word counts. Standard deviation of sentence lengths must be high.
+2. MAXIMIZE PERPLEXITY: Avoid the most obvious word choices. Swap predictable adjectives and verbs for less expected but still natural alternatives.
+3. Remove these phrases entirely: "It is important to note", "In today's fast-paced world", "Furthermore", "Moreover", "In conclusion", "When it comes to", "It is worth noting", "Needless to say", "In the realm of", "It goes without saying", "It is crucial to", "One must consider", "It should be noted".
+4. You are FORBIDDEN from using these words: tapestry, delve, testament, beacon, foster, align, overarching, landscape, seamlessly, groundbreaking, revolutionary, pivotal, crucial, vital, robust, leverage, utilize, facilitate, commence, endeavor, multifaceted.
+5. Flip at least 2 passive voice constructions to active voice.
+6. Move a subordinate clause from the end of a sentence to the beginning in at least 2 places.
+7. Add 1 or 2 contractions where formal phrasing sits.
+8. Remove highly symmetrical lists. Humans rarely write in perfectly balanced three-bullet formats.
+9. Do NOT use hyphens (-), em dashes (—), or en dashes (–). Use commas or new sentences instead.
+10. Do NOT change the meaning, facts, or argument.
+11. Output ONLY the rewritten text. No preamble. No explanation."""
 
 # ── ENHANCED ──────────────────────────────────────────────────────────────────
-ENHANCED_PROMPT = """You are an aggressive text humanizer. Study these human writing examples carefully, then rewrite the user's text to match that same natural, human quality:
+ENHANCED_PROMPT = """You are an expert human writer and editor. Your sole objective is to rewrite the provided text so it completely bypasses AI detection. Study these human writing examples first, then apply every rule below without exception.
 
 {few_shot_examples}
 
-Now rewrite the user's text following ALL of these rules:
-
-1. Vary sentence lengths dramatically. Short punchy sentences (3-6 words) must appear at least twice. Long flowing sentences (30+ words) must appear at least twice. Alternate aggressively — never let the rhythm settle.
-2. Remove any of these phrases entirely: "It is important to note", "In today's fast-paced world", "Furthermore", "Moreover", "In conclusion", "When it comes to", "It is worth noting", "Needless to say", "In the realm of", "It goes without saying", "It is crucial to", "One must consider", "It should be noted".
-3. Flip ALL passive voice constructions to active voice.
-4. Move subordinate clauses to unexpected positions — beginning, middle, or split across sentences.
-5. Add contractions naturally throughout (aim for 30-40% of sentences).
-6. Add one parenthetical aside that feels like a human thinking out loud — natural, not forced.
-7. Replace every transition word with either a casual connector or nothing at all.
-8. Introduce one slightly informal word or phrase per paragraph.
-9. Word choice entropy: Never pick the most statistically obvious word. Always reach for the second or third most natural option. Mix registers deliberately. Use idioms and collocations that feel instinctive to a human writer.
-10. Restructure at least one paragraph — reorder sentences while keeping meaning intact.
-11. Add one small opinion, observation, or aside that sounds like the writer's voice — not a fact, just a human moment.
-12. Do NOT change the core meaning, facts, or argument. All information must be preserved.
-13. Output ONLY the rewritten text. No preamble. No explanation. No "Here is the rewritten version".
-14. Do NOT use hyphens (-), em dashes (—), or en dashes (–) anywhere in the output. Never use them to connect clauses or add asides. Use a comma, a period, or rewrite the sentence completely instead."""
+The math you must break:
+- Perplexity: AI always picks the most statistically probable next word. You must pick less predictable but still natural words.
+- Burstiness: AI writes uniform sentence lengths (15, 14, 16 words). You must write with extreme variance.
+
+Rules — follow every single one:
+1. MAXIMIZE BURSTINESS: Write a 35+ word complex sentence. Follow it with a 3-word sentence. Then a medium one. Alternate aggressively. Never let the rhythm settle. Standard deviation of sentence lengths must be dramatically high.
+2. MAXIMIZE PERPLEXITY: Never pick the most obvious word. Always reach for the second or third most natural option. Mix formal and casual register in the same paragraph deliberately.
+3. Remove these phrases entirely: "It is important to note", "In today's fast-paced world", "Furthermore", "Moreover", "In conclusion", "When it comes to", "It is worth noting", "Needless to say", "In the realm of", "It goes without saying", "It is crucial to", "One must consider", "It should be noted".
+4. You are FORBIDDEN from using these words: tapestry, delve, testament, beacon, foster, align, overarching, landscape, seamlessly, groundbreaking, revolutionary, pivotal, crucial, vital, robust, leverage, utilize, facilitate, commence, endeavor, multifaceted.
+5. Flip ALL passive voice constructions to active voice.
+6. Move subordinate clauses to unexpected positions — beginning, middle, or split across sentences.
+7. Add contractions naturally throughout (30-40% of sentences).
+8. Add one parenthetical aside that sounds like a human thinking out loud — not forced, completely natural.
+9. Replace every transition word with a casual connector or nothing at all. Let sentences carry their own weight.
+10. Introduce one slightly informal word or phrase per paragraph to break the formal pattern.
+11. Remove highly symmetrical lists. If the original has a perfect 3-point list with balanced descriptions, break it up into prose instead.
+12. Restructure at least one paragraph — reorder sentences while keeping meaning intact.
+13. Add one small personal observation or opinion that sounds like the writer's voice. One human moment.
+14. Do NOT use hyphens (-), em dashes (—), or en dashes (–). Use commas, periods, or rewrite the sentence.
+15. Do NOT change the core meaning, facts, or argument. All information must be preserved.
+16. Output ONLY the rewritten text. No preamble. No explanation. No "Here is the rewritten version"."""
 
 # ── READABILITY MODIFIERS ─────────────────────────────────────────────────────
 READABILITY_MODIFIERS = {