Writing Antipatterns
Overview
Most writing advice is stated as if it were universal. Almost none of it is. "Lead with a concrete example" is excellent advice in an explainer and actively harmful in an API reference, where a worked example gets mistaken for the general rule. "Name the actor in every sentence" is right in a plan and wrong in a specification, where the subject is a system. The unit that decides whether a move is craft or a defect is not "good writing" — it is genre.
This skill owns the catalogue of how writing fails, tagged by genre: a short list of defects that
hold everywhere (no lookup needed), and a longer list whose correctness inverts depending on which
of eight genres the prose belongs to. It does not own the positive craft each genre practises —
that stays with the skill that owns the genre (technical-writing, explanatory-writing,
executive-communication, architecture, semantic-consulting-coach). This skill is the
negative space: what not to do, and why the same move is sometimes exactly what to do one genre
over.
The genre map
Diátaxis gives four genres, all documentation. It has nothing to say about a board paper, because a board paper documents nothing — its reader is choosing, not learning, doing, or looking up. The map below is Diátaxis plus the four genres the rest of the Meaningfy skill suite implies.
| Genre | Reader's job | Owning skill | Register signature |
|-------|--------------|---------------|--------------------|
| Tutorial | learn by doing | technical-writing | second person, literal steps, low surprise |
| How-to | complete a known task | technical-writing | terse, imperative, goal stated first |
| Reference | look something up | technical-writing | agentless declarative, tables, exhaustive |
| Explanation | build understanding | explanatory-writing | one metaphor, example beside claim, rhythm |
| Decision (persuasive) | choose and commit now | executive-communication, proposal-writing | answer first, MECE pyramid, hedging banned |
| Decision record | reconstruct the reasoning later | architecture (ADR) | context first, options visible, consequences named |
| Contract / specification | be held to it | proposal-writing (SoW), architecture (OpenAPI/LinkML/use cases) | enumerated, precise, no persuasion, no metaphor |
| Coaching dialogue | think before committing | semantic-consulting-coach | questions over statements, no structure until asked |
Two consequences worth acting on:
- Decision splits into two genres that invert each other. Persuasive decision writing leads with the answer and suppresses rejected options (the reader has ten seconds). A decision record does the opposite: context, considered options, outcome, consequences. An ADR that opens with the outcome has destroyed its only reason to exist.
- Coaching inverts the whole executive method. In
semantic-consulting-coach's Exploration mode, producing a governing thought, a pyramid, or a canvas before alignment is confirmed is explicitly a defection, not craft — the strongest technique in the suite becomes an antipattern one skill away.
Domain is not genre. A sales pitch, a PM status update, or a BI insight memo is a domain
flavour of one of the eight genres above, not a ninth genre — a sales pitch is Decision; a BI memo
is Explanation or Decision depending on its job. The one genuine gap is the internal working
plan (no current skill owns a team's own working plan; it defaults to Reference by default and
reads as inventory) — noted here, not solved here; a future plan-writing skill is the fix.
Model- and diagram-structure antipatterns (free strings where a controlled vocabulary belongs,
god-classes, unlabelled duplication in a C4 diagram) are a different axis — prose vs model
structure — owned by modelling-conventions and
architecture respectively, cited here, never duplicated.
Unguarded defects — apply in every genre, no lookup needed
U1. Genre fusion. Two or more genres bound into one artefact, one register. Cue: name the single reader and the single job; either answer needing "and also" means fused. Fix: split, then link. Licensed exception: paired artefacts (proposal + SoW, a White use case and its Blue realisation) — separate documents, single-sourced shared content, each with a named reader. Pairing is not fusion.
U2. The bland sentence. Stays true if you swap in a competitor's name. Cue: substitute the org name; if nothing breaks, cut it. Reference is allowed to be dry (precision without ornament) — that is not the same as bland (ornament without content).
U3. Non-parallel enumeration. List items switch grammatical form (a noun, a nominalisation, a participle, chained by semicolons). Cue: read the first word of every list item — same part of speech, or not.
U4. Prose compensating for missing structure. Meaning smuggled into a label or sentence because the artefact lacks an element to carry it — a comma-spliced sentence doing the work of three table rows, a paragraph doing the work of a decision register. Fix: create the missing element — one relationship per modality, one row per decision, one sentence per claim.
U5. Weasel words / unattributed authority. "Studies show," "some experts believe," vague
quantifiers ("significant," "many") — a claim asserted while dodging accountability for it by
attributing it to nobody in particular. Distinct from the guarded Hedging entry: hedging
qualifies the author's own claim; a weasel word launders a claim through an unnamed source.
Cue: for every claim shaped "X shows/suggests/believes Y," can you name X? If not, it's a weasel
word.
Fix: name the source and the figure, or drop the claim — the same discipline as
company-voice.md's "grounded; name the figure and its source."
U6. Terminology drift. The same referent named three different ways across one document (feature/tool/functionality; capitalisation drift). Cue: grep for near-synonyms pointing at the same referent; if a reader must infer they're the same thing, it's drift. Fix: one term, defined once — see the guarded Deferred definitions entry, which is about when a term is defined; this is about whether it stays the same term.
Guarded (genre-conditional) antipatterns
Thirteen entries, each stating where a move fails, where the same move is required, the guard
question that tells them apart, and the fix — full text in
references/inversion-matrix.md. Load only the genre you need:
| Genre | Reference file |
|-------|-----------------|
| Tutorial | references/tutorial.md |
| How-to | references/how-to.md |
| Reference | references/reference.md |
| Explanation | references/explanation.md |
| Decision (persuasive) | references/decision.md |
| Decision record | references/decision-record.md |
| Contract / specification | references/contract.md |
| Coaching dialogue | references/coaching.md |
Each per-genre file is a pointer table only (move, status, link to the canonical entry) — it
does not restate an entry's smell/fix/cue text. That restraint is deliberate: this catalogue would
otherwise commit its own Duplication & cardinality antipattern (inversion-matrix.md, entry
A1).
For the teaching version of this frame — why the rules invert, worked examples, and a
self-application check against the catalogue's own output — see
references/explainer.md.
Self-check
- Have you named the single genre this prose belongs to, before applying any rule from this catalogue?
- Does the fix you're about to apply come from the guarded entry's guard question, or are you pattern-matching from a different genre you're more fluent in?
- If you're adding a new antipattern: is it one row in
inversion-matrix.md, with pointer-table rows added to the affected per-genre files — not prose copied into each? - If you're tempted to add a new genre: does it already map onto one of the eight (see "Domain is not genre" above)?
Boundary & Related Skills
Owns: the genre map (the cross-cutting synthesis no other skill states), the unguarded defect list, and the canonical inversion matrix of genre-conditional antipatterns.
Does NOT own: the voice profile (company-voice.md),
the Diátaxis quadrant definitions or the craft moves that realise them
(explanatory-writing), doc placement/form
(technical-writing), the persuasive structure method
(executive-communication), or model/diagram-structure
antipatterns (modelling-conventions,
architecture).
Related: technical-writing, explanatory-writing, executive-communication,
proposal-writing, decision-package, architecture, semantic-consulting-coach.