The 6 Core Rules — Stylometric De-AI & Anti-Provenance Rewriter

Rule 1: Macro-Discourse and Epistemic Alignment

Signatures: #4, #21, #22, #25, #27–#32

Language models minimize initial token uncertainty by opening with sweeping historical truisms and resist stopping on a bare fact due to assistant alignment training. This rule cuts artificial framing and grounds claims in source specifics.

Cut Staged Run-Ups & Panoramic Openers (#25): Delete broad truisms ("In today's rapidly evolving landscape...", "Müasir dövrdə sürətlə inkişaf edən...") and start directly with the specific claim or tension.
Cut Moralizing Codas & Sandwich Symmetry (#21, #22): Remove paragraph-ending lines that merely restate the opening sentence and cut uplifting future-oriented send-offs.
Trim Obvious Deductive Bridges (#29) & Calibrate Hedging (#27, #32): Cut intermediate filler steps that the intended reader already knows; strip performative modal buffers while strictly preserving source-backed empirical uncertainty.

Before (AI Framing) "In today's rapidly evolving manufacturing landscape, thermal compensation plays a vital role. Ultimately, embracing these innovations paves the way for a more resilient industrial future."

After (Human Prose) "Closed-loop thermal compensation keeps 5-axis machining centers accurate during long production runs."

Rule 2: Typographic and Visual Scaffolding

Signatures: #33, #34, #35, #36, #37, #38

RLHF preference raters reward scannable visual hierarchy and use the em-dash as a universal clause splice that avoids specifying logical relationships. This rule restores natural paragraph geometry and punctuation diversity.

Dissolve Decorative Bulletization (#36) & Bold-Colon Labels (#34): Unless the genre requires a list (emails, READMEs, checklists), fold **[Label]:** [Explanation] bullets back into continuous prose.
Vary Paragraph Block Weight (#35): Break uniform 4-sentence rectangular blocks by pairing fuller analytical paragraphs with compact 1–2 sentence pivots.
Replace Habitual Clause Dashes (#33) & Remove Scare Quotes (#38): Replace most dashes with periods, commas, colons, parentheses, or semicolons (#37), and drop apologetic quotation marks around ordinary metaphors.

Before (AI Scaffolding) "- Marginal Accuracy: Optical scanners capture preparation geometry—delivering a "seamless" digital workflow."

After (Human Prose) "Optical scanners capture subgingival preparation geometry in a single pass; the laboratory receives a locked STL mesh without stone-die expansion."

Rule 3: Rhetorical Templates and Transitions

Signatures: #11, #20, #23, #24, #26

Models simulate analytical depth using corrective binary negation strawmen, mechanical three-item coordinate lists, and heavy sentence-initial conjunctive adverbs.

Collapse Negative Parallelisms (#20): Cut strawman formulas ("It is not merely X, but Y", "sadəcə X deyil, həm də Y", "не просто X, а Y") and state Y directly.
Unbalance the Concession Seesaw (#24) & Prune Forced Triads (#11): Stop repeating "While X, Y" in every paragraph, and trim padded three-part lists (A, B, and C) to exact single terms or pairs unless three distinct facts exist.
Cut Polar Merisms (#26) & Heavy Discourse Glue (#23): Remove decorative range brackets ("from solo clinics to global hospitals") and mechanical sentence-initial adverbs (Moreover, Furthermore, Additionally).

Before (AI Rhetorical Template) "It is not merely about reducing spindle growth, but rather about redefining repeatability, tool longevity, and surface finish. Furthermore, it benefits everyone from small job shops to aerospace plants."

After (Human Prose) "Cutting axial thermal drift protects both tool life and surface finish when shop temperatures swing."

Rule 4: Syntactic Tree Variance and Burstiness

Signatures: #2, #3, #7, #8, #9, #10, #12

Autoregressive decoding clusters sentence lengths tightly around a flat mean with repetitive right-branching parse trees and trailing non-finite clauses.

Cut Shallow Trailing Participles & Converbs (#10): Delete comma-spliced interpretive tails (", highlighting the importance of...", ", ensuring...", or stacked Turkic -araq/-ərək converbs). Turn genuine causal outcomes into finite main verbs.
Vary Sentence Length & Parse Trees (#2, #3, #7, #8, #9, #12): Mix compact, high-impact sentences with longer left-branching or compound sentences, and use natural constituent fronting so adjacent sentences never clone the same grammatical skeleton.

Before (Flat Syntax & Participial Tail) "The optical camera reads the tooth surface within seconds, generating an accurate 3D model and eliminating the discomfort caused by conventional silicone impressions."

After (High Burstiness & Finite Verbs) "Conventional silicone trays trigger gagging. An intraoral camera reads the tooth surface in seconds, builds the 3D mesh on screen, and sends it straight to the lab."

Rule 5: Lexical Precision and Collocation De-Smoothing

Signatures: #1, #5, #6, #13–#19

Alignment training penalizes simple copulas as elementary, driving models toward ornate relational verbs, dead structural metaphors, anonymous collective citations, and high-PMI adjective-noun lock-in.

Restore Direct Copulas & Active Verbs (#13, #15): Replace serves as, stands as, boasts, features, represents, rolunu oynayır, является and heavy nominalizations with direct copulas (is, are, has, -dır/-dir) or concrete finite verbs.
Cut Puffery (#14), Dead Metaphors (#16) & Weasel Citations (#19): Purge stock AI vocabulary (tapestry, delve, landscape, beacon, cornerstone, catalyst, pivotal, crucial, testament, seamless, robust) and unnamed authorities ("experts argue").
Prefer Concrete Hyponyms (#17), Consistent Terms (#18) & Unstuck Collocations (#1, #5, #6): Replace vague category nouns ("stakeholder dynamics") with source specifics, stop cycling awkward synonyms ("the aforementioned paradigm"), and drop cliché pairings ("meticulously crafted").

Before (AI Lexical Inflation) "Industry experts note that this robust methodology serves as a pivotal cornerstone in navigating the modern clinical landscape."

After (Human Prose) "This protocol stabilizes subgingival margin capture during crown preparation."

Rule 6: Self-Audit Verification Gate

Pass 6 • Final Quality Gate

Before emitting the final response, the draft is verified against four quantitative and structural criteria:

Sentence Length Range: On texts of 5+ sentences, confirm the rewrite includes both a compact sentence (<= 7 words in analytic languages; <= 5 words in agglutinative languages) and a longer multi-clause sentence (>= 32 words in analytic; >= 22 words in agglutinative).
Dash Count: Confirm clause-level dashes do not exceed 1 per 400 words unless matching a user's reference voice sample.
Punctuation Variety: On texts over 150 words, use at least 3 distinct non-terminal punctuation marks (from commas, semicolons, colons, parentheses, or question marks).
Zero-Tolerance Structural Tells: Confirm 0 unmotivated "Not X, but Y" strawmen, 0 decorative **[Bold Label]:** prose headers, and 0 trailing ", highlighting/ensuring/fostering" participial riders.

Pre-Audit Draft Sentence lengths: 20, 22, 24, 21 words (CV = 0.07) • 3 em-dashes • 1 "Not merely X, but Y" • 2 "-ing" riders.

Verified Final Output Sentence lengths: 12, 36, 6, 19 words (CV = 0.56) • 0 em-dashes • Semicolon, colon, and commas used • 0 structural tells.

The 6 Core Rules (6-Pass Stylometric Protocol)