politeness-bloat — Politeness Token Removal

Severity: WARN (default) · Auto-fix: Yes · Category: 📝 Verbosity

What It Does

Detects unnecessary politeness words that consume tokens without affecting model output quality. LLMs respond identically to "Write a function" and "Please kindly write a function" — the polite version just costs more.

Default Detected Words

The following words/phrases are flagged by default:

Word / phrase	Tokens wasted
please	~1
kindly	~1
i would appreciate	~3
thank you	~2
be so kind as to	~5
if possible	~2

You can customize the list — see Configuration.

Severity: allow_politeness

The allow_politeness flag is not intuitive

• allow_politeness: false (default) → findings fire at WARN level
• allow_politeness: true → findings fire at INFO level (demoted)

Setting allow_politeness: true does not disable the rule — it just reduces severity so it won't fail a --fail-level warn pipeline. To disable entirely, set the rule to false.

Example

Input:

Please kindly help me write a Python function.
I would appreciate it if you could add error handling.
Thank you.

With allow_politeness: false (default):

[ WARN ] politeness-bloat (line 1)
  'please' adds ~1.5 tokens without semantic value. Remove it.

[ WARN ] politeness-bloat (line 1)
  'kindly' adds ~1.5 tokens without semantic value. Remove it.

[ WARN ] politeness-bloat (line 2)
  'i would appreciate' adds ~1.5 tokens without semantic value. Remove it.

[ WARN ] politeness-bloat (line 3)
  'thank you' adds ~1.5 tokens without semantic value. Remove it.

With allow_politeness: true:

[ INFO ] politeness-bloat (line 1)
  'please' adds ~1.5 tokens. Consider removing (~1.5 tokens).
  [This is allowed per configuration — removing improves token efficiency.]

After --fix:

Write a Python function.
Add error handling.

Cost Impact

At 10,000 calls/day and $0.005/1k tokens, removing 3 politeness tokens per prompt saves:

3 tokens × $0.005/1000 × 10,000 calls/day × 30 days = $4.50/month

Small per-prompt, and it compounds across all your prompts.

False Positives

"Please" in quoted strings or examples — the rule matches case-insensitively as a whole word, so please inside a quoted string like "Please enter your name:" will still trigger. If your prompt is about a UI element that uses "please", disable the rule or adjust your prompt to use a placeholder.

"If possible" as a genuine constraint — if you mean "do X if it's feasible given the constraints", consider replacing with a clearer conditional rather than suppressing the rule.

Configuration

rules:
  politeness_bloat:
    enabled: true
    allow_politeness: false    # false = WARN (default), true = INFO
    words:                     # override the full list
      - please
      - kindly
      - i would appreciate
      - thank you
      - be so kind as to
      - if possible
      - could you please
      - would you mind
    savings_per_hit: 1.5       # tokens saved per occurrence (used in output messages)

fix:
  politeness_bloat: true       # auto-remove on --fix

Add custom words

To add words to the default list, you must specify the full list:

rules:
  politeness_bloat:
    words:
      - please
      - kindly
      - i would appreciate
      - thank you
      - be so kind as to
      - if possible
      - with your assistance    # custom addition
      - at your earliest convenience  # custom addition

Disable

rules:
  politeness_bloat: false