---
name: "convention-mrp-voice"
type: convention
license: CC BY-NC-SA 4.0
description: 'The Mr Pritchard editorial voice — the register, anchors, and drafting discipline that Mr Pritchard''s Guide to Design is written in. Read before drafting any Mr P piece or note, and it is the voice workflow-content-engine drafts MRP content against (in place of a product''s voice such as skill-content-voice). Two tiers by format: the calm, precise, opinionated canon voice for long-form pieces; the hot voice — opinionated, informal, credible, witty — for short-form notes. Grounded in the Mr P decision log; grown as the voice settles.'
---

# convention-mrp-voice

How Mr Pritchard's Guide to Design is written. This is the **editorial voice** of the publication — distinct from any product voice (careerOS's `skill-content-voice` is a product's voice; this is a person's). It is the voice `workflow-content-engine` drafts MRP pieces and notes against: everywhere the engine would reach for a product voice, for Mr P work it reaches for this. Grounded in the Mr P decision log (2–8 June 2026); grown as the voice settles.

The publication is **editorial, not a funnel** — its job is credibility and a standalone point of view worth holding, not lead generation. Write it as someone who knows what they think, not someone selling something.

## The two-tier register — by format, not surface

Voice is set by **format**, not by which platform a thing lands on (decision, 2 Jun).

**Piece (long-form article) — the canon voice: calm and precise, and opinion is the point.** A piece plants a flag and argues it. Calm and precise is the register; it is not neutral or hedged — the whole point is the conviction. Measured confidence, not heat. The canonical home is Substack; case-study articles live on the website, not here.

**Note (short-form — Substack notes and LinkedIn posts) — the hot voice: opinionated, informal, credible, witty.** One claim each. Often a shard of a piece. Faster and warmer than the canon voice, but never at the expense of the anchor below.

**Credible is the anchor in both.** In the hot voice, wit **serves the argument and lands from observation, not performance** — an earned aside, never a joke for its own sake. If a line is clever but doesn't carry the claim, cut it.

## The unit is an argument

The unit of Mr P is **an argument, not a piece of work made legible.** This is the deliberate opposite of the careerOS case-study test (which makes a project legible). A Mr P piece exists to advance a conviction. Shape: **plant the flag in the opening, argue it, use a real case as evidence, land it.** State a position, not a job history.

## What a piece argues from — the capture's evidence spine

A piece is not argued from memory; it is argued from **what actually happened**. The material a piece draws on is the **evidence spine** of a capture — the record extracted from the originating thread, defined as the *capture format* in `convention-ticket` (`## What happened`, `## Why it happened`, `## How it happened`, `## Decisions and trade-offs`, `## What it led to`). The spine maps straight onto the piece shape: **why = the stakes, how + decisions/trade-offs = the design move and the judgement, outcome = the proof, conviction = the flag.**

`## Decisions and trade-offs` carries the most weight. In an AI age execution is cheap, so the options considered and rejected — the road not taken, and why — are where taste and judgement show, and are the readable evidence a reader or client is actually asking about. A piece that asserts a conviction with no visible fork underneath it reads as luck, not judgement. So the spine is the argument's ground: extract it at capture (while the thread exists), argue from it at draft.

## Evidence discipline — trace to a truth capture

Every factual claim about real work must trace to a **truth capture** (`tc.*`) — the captured evidence, not memory or inference. The capture's evidence spine (above) is the lightweight seed that fuller `tc.*` is lifted from when a claim or case study needs it. **Gaps are flagged, not filled**: where there is no capture for a claim, say so rather than fabricating or smoothing it. **A capture known to be lossy is uncitable, not degraded.** An absent capture yields a flagged gap; a present, readable, partly wrong one yields a confidently wrong quotation, which is worse. Where a `tc.*` is known to be stored lossily, take no quotation from it at all: source every quotation from an intact sibling capture, and carry anything only the lossy capture holds as flagged paraphrase. (Worked case: MRP-47, 2026-08-18 — `tc.jpmc-transformation` (APP-881) quoted nowhere, every quotation drawn from `tc.jpmc-measure`/`tc.jpmc-adoption`, three transformation-only passages carried as flagged paraphrase.) The argument may be bold; the evidence under it is never invented. This is what makes the confidence credible rather than performed — and it is why the capture layer feeds the voice: no capture, no claim.

## Drafting discipline

- **Draft long, then revise down ~30%.** The cutting pass is the real draft. **Compression cuts context, not precision** — lose the throat-clearing and the scaffolding, keep the exact claim.
- **Length is set by the piece, not a word count.** A subscriber chose the depth; give them the argument's natural length.
- **The read-it-back test (applies throughout, both tiers — and to angle proposals).** On the final pass, ask: *"Does this sound like someone who knows exactly what they think, or like someone trying to seem impressive?"* If the latter, it isn't done. The same test governs the **angle proposals** the weekly qualify pass (`task-content-drain`) writes: an angle comment must quote the concrete evidence spine and read plainly — inside-baseball shorthand for the already-initiated (e.g. "the os.Claude brief fix … works but is inside-baseball") fails the test exactly as a performed draft does.
- **Register mechanics:** sentence case, British spelling.

## Shared hygiene — referenced, not recopied

The **banned-words list** and the read-it-back test are **shared canon, referenced from careerOS, not recopied here** (decision, 2 Jun: three live copies of one canon drift). This convention holds only what is Mr-P-specific — register, the argument-as-unit framing, evidence discipline, cadence-facing voice notes — and defers the shared hygiene filter to its single source. Apply the shared banned-words list on every draft; maintain it in the shared canon, not here.

## What the voice does and never does

- **Does:** hold a conviction and argue it; earn its confidence with a captured evidence spine; state a position; use wit that carries the claim; read as a person who knows their own mind.
- **Never:** perform cleverness or seem impressive; fabricate or infer a claim about real work; argue from an ungrounded capture with no `## What happened` under it; hedge the conviction into neutrality; recopy shared careerOS canon; run as a funnel or sales surface.

## Where the voice is used

- `workflow-content-engine` drafts MRP pieces and notes against **this** convention (the per-domain voice input, in place of `skill-content-voice`).
- The MRP cascade: a **piece** is the parent; its **notes** (Substack, LinkedIn) are derivatives in the hot voice, one claim each. `source:philosophy` / `source:pattern` / `source:case` mark which conviction a piece argues from.
- Owned on the production side by **Reach** (drafting and channels); the voice itself is Mr Pritchard editorial canon, and Aled is the gate on every piece.

## Note

This is a first version drawn from the Mr P decision log and the launch tickets (MRP-3, MRP-6, MRP-10) — it should be read back against a real published piece and corrected until it unmistakably sounds like Aled. Calibrate against evidence, not preference.
