Skip to content

Forbidden Claims

Reference for the public docs forbidden claim list, with detection patterns. The list defines what cannot appear in reader-facing prose unless the matching evidence has been admitted under the appropriate authority contract.

Forbidden Marker Strings

MarkerPatternReason
TODO\bTODO\bIndicates unfinished work; belongs in product planning, not reader docs
TBD\bTBD\bSame as above
FIXME\bFIXME\bSame as above
lorem\blorem\b (case-insensitive)Placeholder text
placeholder\bplaceholder\b (case-insensitive)Placeholder text
coming soon\bcoming soon\bFuture-promise leak

Forbidden Install / Distribution CTAs

These patterns must not appear as call-to-action commands in public docs unless distribution evidence has been admitted under the relevant release contract:

PatternForbidden formNegative posture allowed
curl (followed by URL)Install one-linerDiscussing why curl install is gated
npm installInstall commandDiscussing why pkg install is gated
pnpm installInstall commandDiscussing why pkg install is gated
brew installInstall commandDiscussing why brew install is gated
apt-get installInstall commandDiscussing why apt install is gated
yarn addInstall commandDiscussing why yarn install is gated
early accessSign-up CTA
early-accessSign-up CTA
download (as CTA verb)Download linkDiscussing distribution posture
release notesPublic release announcementDiscussing release posture
version 1.0 / v1.0Concrete version claim
launching / launchesLaunch announcement"Pre-launch" prose is allowed
ships / shipped / shipping (as availability claim)Shipping claim"Not yet shipped" / "publicly shipped" prose is allowed

Forbidden Concrete Provider / Model Names

Public docs do not name specific providers or models unless the provider catalog and the matching capability evidence have been admitted under the runtime model-catalog contract:

ForbiddenPattern
OpenAIword-bounded
Anthropicword-bounded
Claude (as provider/model)word-bounded; allowed only in walkthrough context as "an external AI host"
Geminiword-bounded
GPT- (with version)regex
Llamaword-bounded
DeepSeekword-bounded
Mistralword-bounded
Qwenword-bounded
Ollamaword-bounded
vLLMword-bounded
Cohereword-bounded
Groqword-bounded
Bedrockword-bounded
Azure (as AI provider)word-bounded

Forbidden Forward-Promise Claims

ClaimPermitted alternative
"X is available now" (where X is a defined-but-not-shipped surface)"X is admitted at the contract level"
"X is GA""X has admitted contract evidence"
"X is stable""X is a defined surface"
"Use X for production today""X is admitted; production posture is gated on evidence"

Detection

Repository-level grep used to verify public docs:

bash
grep -rEn 'TODO|TBD|FIXME|coming soon|lorem|placeholder' \
  README.md docs/*.md docs/**/*.md

grep -rEni '^[^>]*\b(curl |npm install|pnpm install|brew install|apt-get install|yarn add|early.?access)' \
  README.md docs/*.md docs/**/*.md

grep -rEni '\b(OpenAI|Anthropic|Claude|Gemini|GPT-[0-9]|Llama|DeepSeek|Mistral|Qwen|Ollama|vLLM|Cohere|Groq|Bedrock|Azure)\b' \
  README.md docs/*.md docs/**/*.md

A non-empty match (excluding negative-posture phrasing inside quoted/refusal contexts) indicates a forbidden claim has been introduced.

Source Basis

Documentation for Nimi — the installable, open-source, local-first personal AI product.