# Voice Dimensions — Extraction Reference
Detail for the seven voice dimensions in [`../SKILL.md`](../SKILL.md) Step 2. Read this before extracting. Each dimension below has: what you are looking for, how to find it in a sample, and the anti-pattern (the lazy version that records an adjective instead of evidence).
The governing law from the skill: **every trait you record must be quotable from a real sample.** These notes tell you *what to quote* and *what the quote proves*.
---
## 1. Diction & lexicon
**Looking for:** the words this person reaches for and the ones they avoid. Plain vs. ornate. Latinate vs. Anglo-Saxon. Jargon tolerance (do they explain terms, assume them, or avoid them?). Recurring metaphor domains (war? gardening? infrastructure? sports?). Pet words and pet constructions.
**How to find it:** scan for words that repeat across samples and aren't topic-forced. Note the *register range* — do they swing between very plain and suddenly precise/technical, or stay level? Note any word they use where a more common word existed (a deliberate choice = voice).
**Records as:** "Plain Anglo-Saxon verbs, technical nouns left unexplained for a peer audience (sample 1: *'the query regressed'* — no gloss). Reaches for infrastructure metaphors. Avoids intensifiers."
**Anti-pattern:** "Uses clear, accessible language." (No evidence, true of everyone.)
---
## 2. Sentence rhythm
**Looking for:** typical sentence length and — more important — **variance**. The signature is usually in the *pattern*, not the average: long-long-long-short. Fragments for emphasis. Lists of three. Semicolon chains. One-sentence paragraphs.
**How to find it:** read a paragraph aloud (or count). Mark where they go long and where they snap short. The snap-short moments are often the voice's punctuation — where the point lands. Klinkenborg's lens: every sentence is a unit; how do they vary the unit?
**Records as:** "Builds evidence in one long sentence, then lands the point in a 3–5 word line (sample 2: *'…and the dashboard never loaded. Not once.'*). Uses one-sentence paragraphs to mark turns."
**Anti-pattern:** "Varied sentence structure." (Says nothing.)
---
## 3. Stance & POV
**Looking for:** where the writer stands relative to the reader and the material.
- **Person:** first-person observer ("I noticed") vs. protagonist ("I built") vs. corporate "we" vs. second-person "you".
- **Certainty:** confident assertion vs. hedged exploration. Do they say "this is" or "this might be"?
- **Relationship:** teacher (explaining down), peer (alongside), provocateur (pushing), guide (leading).
**How to find it:** look at the verbs around claims. Count "I/we/you". Note whether opinions arrive grounded (evidence first) or asserted (claim first). A writer can be confident *and* honest — note if they pair strong claims with explicit limitations.
**Records as:** "First-person observer, not protagonist — watches and reports rather than stars. Confident but evidence-first: the claim arrives after the proof, never before (sample 3). Peer stance, never teacherly."
**Anti-pattern:** "Authoritative yet approachable." (Marketing-speak, unfalsifiable.)
---
## 4. Opening instinct
**Looking for:** the move that earns the first line. Common real ones: a specific sensory scene; a concrete number or fact; a claim that divides; a thing that doesn't sit right. And the **anti-openings** — what they never open with (a definition? a market truism? a reaction to someone else?).
**How to find it:** read only the first 1–2 sentences of each sample. Is there a repeated *kind* of opening? That repetition is the instinct.
**Records as:** "Opens on a specific lived scene with a number (sample 1: *'47 minutes'*). Never opens by reacting to someone else's idea or with a market generality."
**Anti-pattern:** "Strong, attention-grabbing hooks."
---
## 5. Closing instinct
**Looking for:** how they land. A sharpened worldview? An unanswered question? A callback to the opening (Blundell's "close echoes open")? An understated one-liner? A refusal to wrap up neatly?
**How to find it:** read only the last paragraph of each sample. Note whether they end on argument, emotion, question, or quiet.
**Records as:** "Closes on a worldview the reader can't unsee, often with one understated line; sometimes echoes the opening scene. Never ends on a CTA or a preview list."
**Anti-pattern:** "Memorable conclusions."
---
## 6. Signature moves
**Looking for:** the 1–3 things that are *unmistakably this person* — the fingerprint. Could be a structural beat (always a wrong-turn before the answer), a recurring aside, a specific verbal tic (a signature closing phrase), a habit of naming a thing precisely, a kind of joke.
**How to find it:** the move that, if removed, would make a real reader say "that doesn't sound like them." Often the writer is half-aware of it. Quote a clear instance.
**Records as:** "Signature beat: a short understated closing line that resonates with their thesis (e.g. a recurring sign-off phrase). Always shows a wrong turn before the solution."
**Anti-pattern:** listing generic techniques everyone uses. A signature move is *distinctive*, not just present.
---
## 7. Humor & emotional register
**Looking for:** the emotional key the writing plays in, and what the reader is meant to *feel*.
- **Humor:** dry / warm / absent / self-deprecating / sharp. How often.
- **Target feeling:** invited into a new view? let in on something? challenged? reassured? Note the *primary* feeling and what they deliberately avoid (e.g. avoids making the reader feel corrected).
**How to find it:** notice your own reaction reading it. Find the line that produced it; quote that line.
**Records as:** "Dry wit, used sparingly (≤once per piece). Target feeling: 'I can't unsee this now' — recognition first, then a cognitive shift. Deliberately avoids a confrontational register; the reader is invited, never corrected."
**Anti-pattern:** "Engaging and relatable tone."
---
## Register vs. voice — the contrast read
After extracting voice from the *common* signal, look at what *differs* between samples written for different surfaces. Those differences are register, not voice:
| Surface | What flexes |
|---|---|
| Newsletter / essay | longest, most evidence, full arc, scene-led openings |
| LinkedIn | mid-length, one insight, professional opener |
| X / thread | shortest, one claim per unit, sharpest hook, no warm-up |
| Internal memo | plain, fast, low ornament, decision-first |
If you only have samples from one surface, **say so** — record register as "unknown for [other channels]" rather than guessing. Guessing register is a soft form of fabrication.
---
## Confidence tags
- **high** — trait is clear and consistent across all samples; quote it and move on.
- **medium** — present but in only one sample, or weakly; record it, tag medium, note which sample.
- **low / needs more samples** — you suspect it but can't cleanly quote it. Record the suspicion in "Gaps & how to strengthen," not in the voice body.
A partial-but-honest voice print is useful. A complete-but-invented one is worse than nothing — it teaches downstream skills to write a person who doesn't exist.