Default Behavior Competition

The model always has a default behavior for any task — what it does without instructions. When instructions conflict or are too weak, the model reverts to this default completely. The default is an ever-present competitor.

Antipatterns

• Hedged instructions opposing the default. Writing "you might want to use ruff instead of black" when the model defaults to black will not override the default. Hedged modality produces zero behavioral change.
• Conflicting instructions expecting a compromise. Two instructions that disagree ("use tabs" vs "use spaces") do not produce a blend. The model reverts to its default (typically spaces), identical to no instruction at all.
• Abstract instructions against specific defaults. Writing "format code consistently" when the model already has a specific default formatter changes nothing. The instruction must name the exact tool and behavior to displace the default.

Pass / Fail

Pass

# Formatting

Use `ruff` for all formatting. Run `ruff format .` before committing.
*NEVER run `black` or manual formatting.*

Fail

# Formatting

Consider using a consistent code formatter.
You might want to format code before committing.

Fix

Work with the default or overwhelm it:

• If the desired behavior aligns with the model's default: lighter instructions suffice
• If the desired behavior opposes the default: maximum strength required — name exact constructs, use direct commands, place last in context. Any weakness leaves the default unchanged.
• Never rely on conflicting instructions to produce "average" behavior — conflict produces default behavior, identical to no instruction at all.

Limitations

This is an informational diagnostic. The model's default behavior for a given task cannot be directly measured — this rule flags instructions that are likely too weak to override defaults based on their specificity and modality.