patina-cli 3.11.0

package/patterns/en-filler.md

@@ -0,0 +1,113 @@
+---
+pack: en-filler
+language: en
+name: Filler & Hedging Patterns
+version: 1.1.0
+patterns: 4
+---
+
+# Filler & Hedging Patterns
+
+### 22. Filler Phrases
+
+**Watch words:** it's important to note that, it's worth mentioning that, it should be noted that, in order to, due to the fact that, at the end of the day, when it comes to, in terms of, the fact of the matter is, it goes without saying, needless to say, as a matter of fact, for all intents and purposes, in light of the fact that
+
+**Fire condition:** 2+ filler phrases appear in the same paragraph, or a single phrase that can be deleted without any loss of meaning.
+
+**Exclusion:** "In order to" is occasionally necessary for clarity in legal or formal regulatory writing where "to" could be misread as part of an infinitive chain. Use judgment in those contexts.
+
+**Problem:** AI pads sentences with unnecessary filler that adds length but no meaning. These phrases can almost always be deleted or replaced with a shorter alternative.
+
+**Semantic Risk:** LOW
+
+**Substitutions:**
+- "it's important to note that" --> delete, just state the fact
+- "in order to" --> "to"
+- "due to the fact that" --> "because"
+- "at the end of the day" --> delete or rephrase
+- "when it comes to" --> delete or use the subject directly
+- "in terms of" --> delete or rephrase
+- "in light of the fact that" --> "because" or "since"
+
+**Before:**
+> It's important to note that, in order to achieve sustainable growth, companies need to invest in R&D. Due to the fact that competition is increasing, it's worth mentioning that firms must also focus on talent retention. At the end of the day, when it comes to long-term success, innovation is key.
+
+**After:**
+> Companies that want to grow need to invest in R&D. Competition for talent is rising just as fast — Gartner reports that tech attrition hit 20% last year.
+
+---
+
+### 23. Excessive Hedging
+
+**Watch words:** it could potentially be argued that perhaps, may or may not, it is possible that, to some extent, in certain cases, it could be said that, there is a chance that, one might argue, it seems likely that, arguably, it is conceivable that, tends to suggest
+
+**Fire condition:** 3+ qualifiers stacked on a single claim, or a statement so hedged it makes no falsifiable assertion.
+
+**Exclusion:** A single hedge on a genuinely uncertain claim is appropriate and not this pattern ("this approach may reduce costs" when the evidence is preliminary). Only trigger when qualifiers are stacked or when the combined hedging evacuates all meaning.
+
+**Problem:** AI over-qualifies every statement to avoid being wrong, producing text so hedged that it says almost nothing. A single hedge is fine; stacking three or four qualifiers on one claim is not.
+
+**Semantic Risk:** MEDIUM
+**Preservation Note:** Removing hedges from a genuinely uncertain claim may produce an over-confident assertion; ensure the corrected version retains at least one qualifier if the underlying evidence is preliminary or contested.
+
+**Before:**
+> It could potentially be argued that this approach may, to some extent, offer certain advantages in some cases, though it is possible that the results might vary depending on a number of factors that arguably remain to be fully understood.
+
+**After:**
+> This approach has advantages, but results vary. The biggest variable is team size: teams under five people saw a 30% improvement, while larger teams saw little change.
+
+---
+
+### 24. Generic Positive Conclusions
+
+**Watch words:** a bright future lies ahead, exciting journey, the possibilities are endless, stands as a testament, the sky is the limit, promising outlook, exciting times ahead, the best is yet to come, transformative potential, a new chapter
+
+**Fire condition:** 2+ vague optimism phrases appear in the same paragraph or conclusion, or a standalone final sentence that is entirely optimism filler with no specific claim.
+
+**Exclusion:** A specific optimistic statement backed by evidence ("given the trial data, FDA approval looks likely by Q3") is not this pattern — specificity makes it acceptable.
+
+**Problem:** AI ends with vague optimism instead of specific conclusions. This pattern often appears alongside Pattern 6 (Formulaic Challenges and Prospects) but can also stand alone as a final sentence.
+
+**Semantic Risk:** LOW
+
+**Before:**
+> As the company enters this exciting new chapter, the possibilities are truly endless. With its talented team and innovative vision, a bright future undoubtedly lies ahead. The best is yet to come.
+
+**After:**
+> The company plans to open two more locations by June and is hiring 15 engineers for its new climate-tech division.
+
+**Closing formula before:**
+> In conclusion, this topic is one that will continue to generate discussion and debate for years to come. One thing is certain: the future of this field is full of promise and potential.
+
+**Closing formula after:**
+> The bill goes to committee vote on March 12. If it passes, the new rules take effect in January.
+
+**Not this pattern:** The "challenges, but bright future" two-step structure --> Pattern 6 (Formulaic "Challenges and Prospects," en-content.md) handles that.
+
+---
+
+### 31. Conclusion Signal Words
+
+**Watch words:** In conclusion, Ultimately, In summary, To conclude, To sum up, Finally, All in all, In the end, Overall, On the whole
+
+**Fire condition:** The document's final paragraph (or second-to-last) opens with one of the watch words. Or 2+ such conclusion signals appear across paragraphs.
+
+**Exclusion:**
+- Academic papers with an explicit "Conclusion" section header (the header already signals; in-prose markers are redundant)
+- Multi-step arguments where "Therefore" or "Thus" follows from a clear preceding chain (logic markers, not stylistic markers)
+- Speech or lecture transcripts where the speaker explicitly signposts wrap-up
+
+**Problem:** AI essays announce their conclusions ("In conclusion, X is Y") instead of just stating them. Human writers rarely flag the conclusion paragraph — they just write the conclusion. AI feels obligated to scaffold document structure for the reader.
+
+**Semantic Risk:** LOW
+**Preservation:** Removing only the signal word leaves a hollow paragraph. Rewrite the actual conclusion to be more concrete. The paragraph should still read as a conclusion without the marker.
+
+**Burstiness guidance:** When removing conclusion signals, vary the replacement. Don't replace every "In conclusion, X is Y" with "X is Y, then" — that creates a new mechanical pattern. Mix short verdicts with longer explanatory wrap-ups. This pattern often co-fires with #24 (Generic Positive Conclusions); handle both together.
+
+**Before:**
+> In conclusion, work-life balance is not a luxury but a necessity. It is a critical foundation upon which sustainable success — both personal and professional — must be built.
+
+**After:**
+> Work-life balance isn't optional. Without it, the rest of the structure cracks.
+
+**Not this pattern:** When the *content* of the conclusion is vague optimism --> Pattern 24 (Generic Positive Conclusions) handles that. Pattern 31 is about the signal *word itself* regardless of what follows.

package/patterns/en-language.md

@@ -0,0 +1,163 @@
+---
+pack: en-language
+language: en
+name: Language Patterns
+version: 1.0.0
+patterns: 7
+dedupe-with:
+  - source: "en-language:7"
+    target: "en-filler:24"
+    owner: "en-filler:24"
+    reason: "The phrase 'stands as a testament' belongs to filler #24; language #7 keeps 'testament' only as one word in a broader AI-vocabulary cluster."
+  - source: "en-language:7"
+    target: "en-language:8"
+    owner: "en-language:8"
+    reason: "When 'testament' appears inside a copula-avoidance construction, pattern #8 owns the rewrite; pattern #7 only fires on a broader AI-vocabulary cluster."
+  - source: "en-language:8"
+    target: "en-filler:24"
+    owner: "en-filler:24"
+    reason: "The exact phrase 'stands as a testament' is a generic positive-conclusion marker when used as vague praise; pattern #8 still owns non-conclusion copula avoidance such as 'serves as'."
+---
+
+# Language Patterns
+
+### 7. AI Vocabulary Words
+
+**Watch words:** delve, tapestry, landscape, multifaceted, comprehensive, pivotal, testament, intricate, nuanced, leverage, foster, crucial, moreover, furthermore, realm, robust, facilitate, endeavor, resonate, underscore, embark, myriad, paramount, encompass, holistic, synergy, elucidate, culminate, juxtapose, burgeoning
+
+**Fire condition:** 3+ watch words appear in one paragraph.
+
+**Semantic Risk:** MEDIUM
+**Preservation Note:** Replacing AI vocabulary words changes the specific connotations carried by words like "nuanced" or "multifaceted"; ensure the substituted terms preserve the intended meaning and register.
+
+**Problem:** These words appear at dramatically higher rates in AI-generated text than in human writing. Individually they are normal English words; clustered together they are a telltale fingerprint.
+
+**Before:**
+> This comprehensive report delves into the multifaceted landscape of renewable energy, leveraging nuanced insights to foster a robust understanding. Moreover, the intricate tapestry of stakeholder interests underscores the pivotal role of policy in this crucial endeavor.
+
+**After:**
+> This report covers renewable energy policy from three angles: cost, grid reliability, and public opinion. Each section uses data from the EIA and interviews with six state utility commissioners.
+
+---
+
+### 8. Copula Avoidance ("serves as")
+
+**Watch words:** serves as, acts as, functions as, stands as, operates as, works as, remains as, exists as
+
+**Fire condition:** 2+ copula-avoidance constructions in the same paragraph, or a single instance where "is" would be shorter and clearer.
+
+**Exclusion:** "Serves as" is acceptable when the subject has a formally designated role function (e.g., "the committee serves as an advisory board" — institutional role, not decorative usage).
+
+**Semantic Risk:** MEDIUM
+**Preservation Note:** "Serves as" sometimes implies a functional or designated role distinct from mere identity; replacing with "is" may flatten a meaningful distinction about purpose or function.
+
+**Problem:** AI avoids the simple verb "is" by using unnecessarily complex copula constructions. "The park serves as a gathering place" instead of "The park is a gathering place."
+
+**Before:**
+> The library serves as a vital community hub, functioning as both an educational resource and a social gathering space. It also acts as a testament to the city's commitment to public access.
+
+**After:**
+> The library is the neighborhood's main public space. People use it for study groups, ESL classes, and the Saturday morning story hour, which draws about 40 kids a week.
+
+---
+
+### 9. Negative Parallelisms
+
+**Watch words:** not just...but, not merely...but also, not only...but, goes beyond...to, more than just...it is, transcends...to become
+
+**Fire condition:** 2+ "not X but Y" structures in the same document, or a single instance where the positive statement alone would be simpler and equally clear.
+
+**Exclusion:** Genuine contrastive clarification correcting a misconception ("the event is not a conference but a workshop — no keynotes, only hands-on sessions") — the negative frame is doing real work here.
+
+**Semantic Risk:** MEDIUM
+**Preservation Note:** The negative framing sometimes carries a genuine contrast or correction; collapsing "not X but Y" to just "Y" may lose the implicit rejection of a common assumption the author intended to address.
+
+**Problem:** AI uses negative-then-positive constructions to make simple points sound profound. Instead of stating what something is, it first states what it is not.
+
+**Before:**
+> This initiative is not just a policy change but a fundamental reimagining of urban planning. It goes beyond mere infrastructure investment to become a statement about the kind of city residents want to live in.
+
+**After:**
+> The initiative rezones 12 blocks for mixed-use development and adds a bike lane network connecting the east and west sides of the river.
+
+---
+
+### 10. Rule of Three Overuse
+
+**Fire condition:** 3-item lists appear 2+ times in the same document, or a triple-part sentence where the count is arbitrary and another count would be equally valid.
+
+**Exclusion:** Naturally occurring triads in genuinely three-part processes (past/present/future; input/process/output; beginning/middle/end) are not this pattern.
+
+**Burstiness note:** When rewriting, use the naturally correct count — one strong point if that is all there is, two contrasting items if that is the real structure, four if there are actually four. Varying list counts across a document signals human authorship.
+
+**Problem:** AI defaults to triple-item lists and three-part structures when the natural count might be two, four, or one. This creates a rhythm that feels rehearsed rather than spontaneous.
+
+**Semantic Risk:** LOW
+
+**Before:**
+> The program fosters creativity, innovation, and collaboration. Participants gain inspiration, practical skills, and lasting connections. The result is a more dynamic, inclusive, and forward-thinking community.
+
+**After:**
+> The program pairs early-career designers with experienced mentors for a 10-week project. Last year's cohort produced two apps that are still in active use.
+
+---
+
+### 11. Elegant Variation (Synonym Cycling)
+
+**Fire condition:** The same entity referred to by 3+ different names or synonyms within a single paragraph.
+
+**Exclusion:** Legitimate disambiguation (using "the company" vs. "Microsoft" when distinguishing a parent from a subsidiary, or "the study" vs. "the 2023 Stanford paper" for precision) is not this pattern.
+
+**Problem:** AI rotates synonyms for the same entity to avoid repeating a word, producing text that feels strangely evasive. A city becomes "the metropolis," then "the urban center," then "the municipality" — all within the same paragraph.
+
+**Semantic Risk:** LOW
+
+**Before:**
+> Tokyo is the most populous city in Japan. The metropolis is known for its blend of tradition and modernity. The urban center attracts millions of tourists each year. The Japanese capital continues to grow.
+
+**After:**
+> Tokyo is the most populous city in Japan, with about 14 million residents. It draws roughly 20 million foreign tourists a year, a number that has doubled since 2015.
+
+---
+
+### 12. False Ranges
+
+**Watch words:** from X to Y, ranging from...to, spanning...to, everything from...to, whether...or
+
+**Fire condition:** A "from X to Y" construction appears where the two poles do not meaningfully bound a spectrum — the range is decorative rather than informative.
+
+**Exclusion:** Genuine numeric or temporal ranges ("from 10 to 100 employees", "from January to March", "from $5 to $50") are not this pattern.
+
+**Problem:** AI creates artificial ranges to sound comprehensive, often pairing two extremes that do not meaningfully define a spectrum.
+
+**Semantic Risk:** MEDIUM
+**Preservation Note:** Removing a range construction may drop a genuine scope claim about audience or coverage; verify whether the range was intended to convey actual breadth before eliminating it.
+
+**Before:**
+> The festival offers something for everyone, from young children to seasoned professionals, spanning everything from traditional folk music to cutting-edge electronic performances.
+
+**After:**
+> The festival has three stages. The main stage books established acts (this year: Yo-Yo Ma, Thundercat). The tent stage is for local bands. There is also a children's area with instrument demos.
+
+---
+
+### 32. Comparison Adverb Overuse ("more" without target)
+
+**Watch words:** more specific, more concrete, more efficient, more effective, more comprehensive, more robust, more seamless, more meaningful, more strategic, more impactful, more nuanced, more proactive, more sustainable, more scalable
+
+**Fire condition:** 2+ "more + adjective/adverb" comparative phrases appear in one document without a clear target, baseline, or metric. A single instance can fire when it appears in the same paragraph as other formal AI markers such as "comprehensive", "strategic", "framework", "stakeholder", or "in-depth".
+
+**Exclusion:** Do not fire when the comparison target is explicit ("more efficient than the old process"), when a before/after metric is provided ("reduced latency from 240 ms to 130 ms"), in ordinary quantifiers ("more than 10 users"), or in fixed phrases such as "more or less".
+
+**Semantic Risk:** LOW
+**Preservation Note:** Comparative language can carry a real improvement claim. Preserve the target, metric, or priority if one exists; otherwise replace the vague comparative with the concrete requirement or result the sentence is trying to name.
+
+**Problem:** AI text often implies progress by stacking "more X" phrases while never saying compared with what. Human prose either names the baseline ("faster than last quarter's build") or drops the decorative comparison and states the concrete change.
+
+**Burstiness note:** Do not convert every "more X" phrase into the same alternative. Mix direct adjectives, named metrics, shorter verbs, and one retained comparative when the contrast is real.
+
+**Before:**
+> The initiative will enable more specific milestones, more efficient resource allocation, and more comprehensive stakeholder alignment. Moving forward, the team should develop a more strategic framework for more meaningful collaboration.
+
+**After:**
+> The initiative needs dated milestones, named owners, and a budget review. The team also needs one shared plan for partner handoffs, because the current spreadsheet has three conflicting owners for the same launch tasks.

package/patterns/en-structure.md

@@ -0,0 +1,173 @@
+---
+pack: en-structure
+language: en
+name: Structure Patterns
+version: 1.2.0
+patterns: 5
+phase: structure
+---
+
+# Structure Patterns
+
+---
+
+### 25. Metronomic Paragraph Structure
+
+**Fire condition:** 3+ consecutive paragraphs follow the identical internal template — e.g., claim → evidence → significance, or problem → solution → benefit, or intro → development → conclusion.
+
+**Exclusion:** Academic abstracts, legal briefs, and comparative reviews (product A vs. product B on the same criteria) where the repeated structure is the required format, not an AI artifact. Also exclude texts with 2 or fewer paragraphs.
+
+**Burstiness note:** After breaking the structure, vary paragraph function deliberately. Some paragraphs should present pure evidence with no stated implication. Some should be a single-sentence observation. Some should open with a question, a contradiction, or a concrete detail rather than a topic claim.
+
+**Problem:** AI stamps every paragraph with the same internal template regardless of what the content actually requires. Human writing varies paragraph shape: some paragraphs open with a question, some lead with a specific detail, some end without resolving anything. Metronomic regularity is one of the strongest structural signals of AI generation.
+
+**Semantic Risk:** MEDIUM
+**Preservation Note:** Restructuring paragraph order or shape may alter the logical flow of an argument; verify that each paragraph's core claim survives the restructuring and that no evidence is dropped.
+
+**Before:**
+> The electric vehicle market has experienced unprecedented growth in recent years. Sales doubled between 2021 and 2023, driven by falling battery costs and expanded charging infrastructure. This trend signals a fundamental shift in how consumers relate to personal transportation.
+>
+> Battery technology has improved considerably over the same period. Energy density has increased by 40% since 2018, while costs have fallen from $140 to $90 per kWh. These advances have made electric vehicles competitive with internal combustion alternatives on a total cost of ownership basis.
+>
+> Government policy has played a significant supporting role. Tax credits of up to $7,500 are available in the United States, and the EU has mandated a ban on new petrol vehicle sales by 2035. Such regulatory tailwinds are expected to sustain the sector's momentum into the next decade.
+
+**After:**
+> Electric vehicle sales doubled between 2021 and 2023. The main driver isn't policy — it's battery prices, which fell from $140 to $90 per kWh.
+>
+> The US $7,500 tax credit helps, but it phases out once a manufacturer hits 200,000 vehicles sold. Tesla already passed that cap years ago.
+>
+> Whether the EU's 2035 petrol ban survives is genuinely unclear. Germany's coalition nearly collapsed over it in 2023, and three major automakers are lobbying for a 2040 date instead.
+
+---
+
+### 26. Passive Nominalization Chains
+
+**Watch words:** was conducted, was performed, was developed, was established, were identified, is required, has been shown, is provided, are considered, is achieved, is determined, is utilized, is noted
+
+**Fire condition:** 2+ passive nominalization phrases in the same paragraph — e.g., "an analysis was conducted", "consideration was given to", "a decision was made".
+
+**Exclusion:** Scientific methods sections and formal regulatory documents where passive voice is a disciplinary norm. Also acceptable when the actor is genuinely unknown, irrelevant, or appropriately omitted.
+
+**Burstiness note:** Replace passives with active verbs — but vary the subjects. Avoid making every sentence "We did X." Mix "The team found Y" with "Data showed Z" with bare declaratives that start with the finding itself.
+
+**Problem:** AI chains passive nominalizations that convert active situations into abstract bureaucratic prose. "An analysis was conducted" means "we analyzed." The passive form removes agency and adds syllables without adding meaning.
+
+**Semantic Risk:** MEDIUM
+**Preservation Note:** Converting passive to active assigns a specific agent; if the original intentionally omitted the actor (unknown, irrelevant, or sensitive), the active rewrite may introduce an unintended attribution.
+
+**Before:**
+> An extensive review of the literature was conducted in order to identify key patterns. Consideration was given to a range of methodological approaches, and a decision was made to adopt a mixed-methods design. Data collection was performed over a six-month period, after which an analysis was undertaken to identify recurring themes.
+
+**After:**
+> We reviewed 47 papers published between 2018 and 2024. Most used surveys; we chose interviews instead, since the population was small enough to go deep. Six months of interviews, then we coded the transcripts.
+
+---
+
+### 27. Zombie Nouns (Excessive Nominalization)
+
+**Watch words:** make an improvement to, provide a description of, conduct an analysis of, give consideration to, offer an explanation of, achieve a reduction in, have an impact on, make a decision about, reach a conclusion regarding, perform an evaluation of, carry out an investigation into, give an indication of
+
+**Fire condition:** 3+ nominalized verb phrases in the same paragraph — e.g., "make improvements" instead of "improve", "conduct an analysis" instead of "analyze".
+
+**Exclusion:** Noun forms that carry meaning unavailable in the verb ("the investigation" as a noun refers to an ongoing process; "investigate" does not). Also acceptable in legal contexts where nominalized forms have specific technical definitions.
+
+**Burstiness note:** Replace nominalized phrases with verbs, but not every instance. Keep one or two where the noun form fits naturally — avoid creating staccato prose by over-correcting.
+
+**Problem:** AI converts verbs into bloated noun phrases, a pattern called "zombie nouns." Every action becomes an abstract event: "improve" becomes "make an improvement to," "analyze" becomes "conduct an analysis of." The result is wordy, impersonal, and slow.
+
+**Semantic Risk:** MEDIUM
+**Preservation Note:** Some nominalized forms carry distinct meaning from their verb counterparts ("the investigation" vs. "investigate"); verify that converting to a verb does not collapse a reference to an ongoing process or formal procedure.
+
+**Before:**
+> The committee made a decision to conduct an analysis of the current situation and to provide a recommendation regarding future strategy. An evaluation of the available options was performed, and after careful consideration was given to each alternative, an agreement was reached on the most viable approach.
+
+**After:**
+> The committee analyzed the situation and agreed on a strategy. They evaluated three options and chose the one with the lowest implementation risk.
+
+---
+
+### 28. Stacked Subordinate Clauses
+
+**Fire condition:** A single sentence contains 3+ embedded relative clauses, appositives, or participial phrases before the main verb reaches its object — or the reader must parse 4+ commas before understanding the sentence's main claim.
+
+**Exclusion:** Legal definitions and technical specifications occasionally require nested qualification. Academic writing may also stack clauses when the embedded information is non-negotiable to the sentence's meaning and cannot be separated without losing precision.
+
+**Burstiness note:** Break stacked sentences into two or three shorter ones. Start one sentence with the main clause, then add a separate sentence for the embedded detail. Vary lengths — don't make every resulting sentence the same.
+
+**Problem:** AI builds sentences with cascading embedded clauses that delay the main verb and obscure the subject-verb relationship. The result is technically grammatical but exhausting to read and often contains no insight beyond what a shorter sentence would convey.
+
+**Semantic Risk:** MEDIUM
+**Preservation Note:** Breaking stacked clauses into separate sentences may lose precision embedded in the nested qualifications; ensure every embedded condition or qualification appears in the rewritten version.
+
+**Before:**
+> The initiative, which was developed in response to increasing concerns about, and growing awareness of, the urgent need to address the challenges facing both urban and rural communities in the context of rapid technological transformation, aims to bridge the gap between innovation and equitable access.
+
+**After:**
+> Rural and urban communities are getting left behind as technology changes fast. The initiative targets that gap directly — access to tools, not just the tools themselves.
+
+---
+
+### 30. Rhetorical Question Openers
+
+**Problem:** AI essays open paragraphs with rhetorical questions ("So what does this mean?", "Why is this important?", "But how did we get here?") to scaffold structure. The writer already knows the answer; the question is filler that buys time before the actual claim. Real writers state observations directly. Rhetorical questions belong in speeches and editorials, not in informational essays.
+
+**Trigger conditions:** A paragraph opens with an interrogative sentence ("Why...?", "How...?", "What...?", "But what if...?") AND the answer follows in the same or next sentence within the same paragraph. Or: the document contains 2+ rhetorical questions used as paragraph openers across paragraphs.
+
+**Exceptions:**
+- Interview, FAQ, or Q&A format documents
+- Genuine open questions where the author truly doesn't know the answer (research questions, philosophical inquiry)
+- Speech or lecture transcripts where audience engagement is the goal
+- Academic papers' research question sections
+
+**Semantic risk:** LOW
+**Preservation:** Distinguish structural filler from genuinely emphatic questioning. If converting to declarative, verify the new sentence still carries the original emphasis.
+
+**Burstiness guidance:** When converting rhetorical questions to declaratives, vary the result. Don't replace every "Why X?" with "X is because Y" — that just substitutes one mechanical pattern for another. Mix short emphatic statements with longer explanatory clauses.
+
+**Before:**
+> So why did Korean coffee culture grow so fast? The answer is surprisingly simple. Cafés moved beyond being places to buy drinks and became social hubs.
+>
+> But what about the future? Experts believe the trend will continue.
+
+**After:**
+> Korean coffee culture grew fast for a surprisingly simple reason: cafés moved beyond being places to buy drinks and became social hubs.
+>
+> The trend is likely to continue. Experts agree.
+
+---
+
+## English Burstiness Guidelines
+
+Even without dedicated structure patterns, apply these burstiness principles when rewriting English text in Phase 1 (structure pass) or Phase 2 (sentence/lexical pass):
+
+#### What is burstiness?
+
+Human writing varies sentence length and structure unpredictably. AI writing is uniform: similar sentence length, consistent clause structure, even paragraph size. Burstiness means introducing deliberate irregularity.
+
+#### How to apply
+
+**Sentence length variation:**
+- Mix short punchy sentences (4–8 words) with longer ones (20–35 words).
+- A paragraph that opens with a one-sentence fragment and then unpacks in the next sentence reads as human.
+- Avoid stretches of 4+ consecutive sentences with similar length.
+
+**Paragraph length variation:**
+- Paragraphs do not need to be 3–5 sentences. A single-sentence paragraph is fine for emphasis.
+- Vary: short/long/medium, not medium/medium/medium.
+
+**List count variation:**
+- Do not default to 3-item lists. Use 2 items if there are 2, or 4 if there are 4.
+- See Pattern #10 (Rule of Three) for correction details.
+
+**Structure variation:**
+- Mix declarative sentences, rhetorical questions, and sentence fragments.
+- Not every sentence needs a subject-verb-object structure.
+
+#### When to apply
+
+Apply burstiness corrections when:
+- All paragraphs are approximately the same length.
+- Most sentences follow the same syntactic pattern (Subject + Verb + Object + subordinate clause).
+- The overall rhythm feels metronomic (consistently even pacing).
+
+Do **not** manufacture burstiness if the source text is already varied. Check before correcting.

package/patterns/en-style.md

@@ -0,0 +1,139 @@
+---
+pack: en-style
+language: en
+name: Style Patterns
+version: 1.0.0
+patterns: 6
+---
+
+# Style Patterns
+
+### 13. Em Dash Overuse
+
+**Fire condition:** 3+ em dashes appear in a single text.
+
+**Exclusion:** A single em dash for a sharp parenthetical or abrupt sentence break is stylistically valid in informal or literary writing. Only trigger on clusters.
+
+**Burstiness note:** When rewriting em-dash-heavy sentences, vary the alternatives — use commas for some, parentheses for others, or break into separate sentences. Uniform substitution (replacing every em dash with a comma) produces a new kind of rhythm problem.
+
+**Problem:** AI uses em dashes for emphasis, parenthetical asides, and clause breaks far more frequently than human writers. A text peppered with em dashes reads as distinctly AI-generated.
+
+**Semantic Risk:** LOW
+
+**Before:**
+> The new policy — which was announced last Tuesday — aims to address a long-standing issue — affordable housing. Critics — including several council members — argue that the plan lacks specifics — a concern that the mayor dismissed as premature.
+
+**After:**
+> The new policy, announced last Tuesday, aims to address affordable housing. Several council members say the plan lacks specifics. The mayor called the criticism premature.
+
+---
+
+### 14. Boldface Overuse
+
+**Fire condition:** 5+ bolded terms in a single document, or 3+ in a single paragraph.
+
+**Exclusion:** Bold is appropriate in UI documentation (button names, field labels), safety warnings ("**Do not** delete this file"), and genuine reference material where bold signals terminology. Only trigger in flowing prose.
+
+**Problem:** AI bolds key terms as if writing a study guide or corporate slide deck. In normal prose, bolding should be rare or absent.
+
+**Semantic Risk:** LOW
+
+**Before:**
+> **Machine learning** is a subset of **artificial intelligence** that uses **statistical methods** to enable computers to **learn from data**. The most common approaches include **supervised learning**, **unsupervised learning**, and **reinforcement learning**.
+
+**After:**
+> Machine learning is a subset of artificial intelligence that uses statistical methods to let computers learn from data. The three main approaches are supervised learning, unsupervised learning, and reinforcement learning.
+
+---
+
+### 15. Inline-Header Vertical Lists
+
+**Fire condition:** 2+ "**Label:** explanation" bullets in the same list.
+
+**Exclusion:** Legitimate reference content — API parameter tables, changelog entries, feature comparison grids — where label-and-description is the correct and expected format.
+
+**Burstiness note:** When rewriting, avoid converting every bullet into a single run-on sentence. Vary the approach: some items can merge into prose, others can become numbered steps if order matters, and short items can stay as a plain list without bold labels.
+
+**Problem:** AI converts prose into pseudo-heading bullet points with a bold label and colon on every item. This format is appropriate for reference docs but not for articles, essays, or narrative text.
+
+**Semantic Risk:** LOW
+
+**Before:**
+> - **Accessibility:** The platform supports screen readers and keyboard navigation.
+> - **Performance:** Load times have been reduced by 40%.
+> - **Security:** All data is encrypted end-to-end.
+
+**After:**
+> The platform now supports screen readers and keyboard navigation, load times are down 40%, and all data is encrypted end-to-end.
+
+---
+
+### 16. Title Case in Headings
+
+**Fire condition:** 3+ content words (non-articles, non-prepositions) capitalized in the same heading when sentence case would be correct.
+
+**Exclusion:** Proper nouns, brand names, product names, and acronyms within headings always require capitalization regardless of style guide.
+
+**Problem:** AI capitalizes every word in headings (Title Case) instead of using sentence case. In most modern style guides (AP, APA, Google developer docs), sentence case is preferred for readability.
+
+**Semantic Risk:** LOW
+
+**Before:**
+> ## The Impact Of Remote Work On Urban Development And Housing Markets
+
+**After:**
+> ## The impact of remote work on urban development and housing markets
+
+---
+
+### 17. Emojis
+
+**Watch words:** emojis used as section markers or emphasis
+
+**Fire condition:** Any emoji in professional, academic, or editorial text.
+
+**Exclusion:** Personal blogs, social media copy, and casual newsletters where emojis are a deliberate stylistic choice. Also exclude emoji in quoted material being analyzed.
+
+**Problem:** AI inserts emojis to add visual interest to text where they are inappropriate. This is especially common in listicles, summaries, and instructional content.
+
+**Semantic Risk:** LOW
+
+**Before:**
+> Here are five tips for better sleep:
+> 1. Set a consistent bedtime
+> 2. Avoid screens before bed
+> 3. Keep your room cool
+> 4. Limit caffeine after noon
+> 5. Try reading instead of scrolling
+
+**After:**
+> Five tips for better sleep:
+> 1. Set a consistent bedtime.
+> 2. Avoid screens before bed.
+> 3. Keep your room cool.
+> 4. Limit caffeine after noon.
+> 5. Read instead of scrolling.
+
+---
+
+### 18. Curly Quotation Marks
+
+**Watch words:** curly/smart quotes used in contexts where straight quotes are standard
+
+**Fire condition:** Curly quotes appear inside code blocks, inline code spans, configuration file examples, or plain-text technical documents where straight quotes are required.
+
+**Exclusion:** Curly quotes in narrative prose are typographically correct and not this pattern. Only trigger in technical or code contexts where curly quotes cause syntax errors or compatibility issues.
+
+**Problem:** AI generates curly quotes in contexts where straight quotes are standard, such as code blocks, plain-text documents, configuration files, and technical documentation. This can cause syntax errors and compatibility issues.
+
+**Semantic Risk:** LOW
+
+**Before:**
+> Set the variable: `name = "hello"`
+>
+> In your config file, add: `timeout = '30'`
+
+**After:**
+> Set the variable: `name = "hello"`
+>
+> In your config file, add: `timeout = '30'`