The patterns Radar Six writes against.

The em-dash punctuation mark is banned across published surfaces.

Em-dashes have become a tell of LLM-assisted prose smoothing. The Radar Six convention uses a sentence break, a comma, or parentheses where an em-dash would otherwise appear. A short hyphen inside a compound word is fine. The U+2014 character itself fails the voice scan.

Use straight ASCII quotes everywhere. Curly quotes fail the scan.

Curly single quotes (U+2018, U+2019) and curly double quotes (U+201C, U+201D) appear in autoformatted CMS output and in pasted copy from word processors. Straight quotes (") and apostrophes (') render identically across feed readers and search snippets, and they keep the copy auditable against the canonical text the operator typed.

Avoid sentence shapes built on the X, not Y contrast.

This shape reads as marketing rhetoric. A claim that something is X, not Y forces the reader to walk a comparison the author has already foreclosed. Replace with two short declarative sentences. State what the thing is. If the foil matters, name it in a separate sentence and explain why.

The two-step rhetorical pivot is a tell of AI-smoothed copy.

Variants include "this isn't a leak, it's a signal" and "Radar Six isn't a dashboard, it's an audit trail." The pivot manufactures contrast where the underlying claim is best stated directly. Write the second half on its own. If the negation adds information, write it as a separate sentence with the actual context that justifies it.

Parallel triplets for rhetorical effect are banned in marketing copy.

Rule-of-three rhythm reads as machine-tuned cadence. The pattern appears as three parallel noun phrases, three parallel verb phrases, or three short clauses ending in matched syllables. Inventory lists (banned vocabulary, required patterns, gate categories) are not rule-of-three; those are reference material. The ban applies to rhetorical prose that builds toward emphasis through triplet structure.

Banned words: game-changer, leverage, robust, exclusive, unlock, seamless, intuitive, empowers, revolutionary.

These words inflate description without adding signal. The voice scan rejects them on sight. Replace with the specific behaviour the word claims to summarise. If a feature is fast, state the latency in milliseconds. If a surface is simple, describe the concrete affordance rather than asserting simplicity.

Banned words and tokens: breaking, huge, JUST IN, mind-blowing, massive, EXCLUSIVE, MUST READ.

Hype tokens read as engagement bait and damage the audit-trail framing Radar Six runs on. A claim earns attention by being correctly anchored and clearly labelled, not by shouting. Capitals-only emphasis tokens (JUST IN, BREAKING, EXCLUSIVE) are rejected even in social copy.

Banned words: best, leading, unique, premier, world-class, best-in-class, industry-leading.

Self-praise vocabulary is rejected because Radar Six does not classify its own output as the best version of anything. The published surface describes what was observed, where it sat in the source chain, and what verdict the operator applied. Readers form their own classification.

The third-person "the desk" device appears at most once per paragraph.

The device is useful as a single grounding reference inside a paragraph. Stacking it (every sentence opens with "The desk...") reads as artificial and obscures the actual subject of each sentence. Use a concrete subject (the source chain, the registry row, the silence counter) where one is available; reserve the third-person device for the single sentence where it actually grounds the framing.

Memorable-sounding short phrases ("chases the spike, charts the slope") are banned in marketing copy.

Aphoristic pull-quotes read as taglines designed to be screenshot-shared. They sacrifice precision for memorability and they invite misquotation in the reverse direction. Radar Six surfaces communicate by walking the source chain in full, not by compressing a position into a poster. The pattern is forbidden on hero blocks, in the manifesto, in claim cards, and in social copy.

Every source named in published copy carries its tier classification in parentheses.

Example: "CharlieIntel (tier 2) amplified the Best Buy screenshot at 06:20 UTC." The parenthetical lets a reader anchor evidence weight without leaving the sentence. Tier labels follow the four-tier vocabulary defined in /sources and /glossary.

A claim mentioned by reference carries its verdict and confidence band on first reference.

Example: "The November 19, 2026 launch-date claim (confirmed, confidence 88) anchors the related-claims graph." Without verdict and confidence on first reference, the reader has no way to weigh the claim against the surrounding argument.

Watching-verdict claims with a fixed expiry name the date the kill clock expires.

Example: "The Best Buy preorder window kill clock expires 2026-05-22." A claim that is watching pending an event resolution must name the event and the date. Open-ended watching verdicts without a kill clock are an audit-trail failure mode and should not appear in published copy.

When tier 1 evidence is absent, the absence is named directly.

Example: "No Rockstar Newswire post has confirmed the date. The Take-Two IR feed is the sole tier 1 anchor on the register today." Absence carries information weight. The voice profile requires that absence be stated rather than smoothed over with implication.

At most one direct quote per source, at most 15 words per quote.

A direct quote is preserved when the exact wording carries weight. Beyond 15 words the desk paraphrases with the source named and the URL linked. Over-quotation inflates the published surface without adding signal, and it makes corrections harder to apply when the source updates its language.

Sentences default to declarative. Volatility on the topic does not surface as volatility in the prose.

A trailer drop, a leak window, a stock surge, and a Rockstar Newswire silence are all written in the same calm register. The topic can be volatile. The published voice stays at a calm register so the reader can weigh content against a consistent baseline; cadence shifts read as editorial bias and break that contract.