select_strategy

basemode.detect.select_strategy

def select_strategy(
    model: str,
    override: str | None = None,
    *,
    allow_user_override: bool = True,
) -> StrategyChoice

Resolve the continuation strategy for a model and report what decided it. detect_strategy wraps this and returns the strategy instance instead.

from basemode import select_strategy

choice = select_strategy("moonshot/kimi-k3")
print(choice.name, choice.source)  # prefill registry

Returns

StrategyChoice(name, source), where source is one of:

Source	Meaning
explicit	Came from the override argument
user	Pinned on this machine by basemode bench --save
registry	The prompt_method verified for this model
heuristic	Derived from the model name

Resolution order

1. override, if set — raises ValueError if it isn't a known strategy.
2. A locally pinned strategy for this model, unless allow_user_override=False.
3. The prompt_method recorded for the model in the verified-models registry.
4. Model-name heuristics: Claude → prefill (or system when no_prefill applies), known completion models → completion, FIM families → fim, everything else → system.

Steps 2 and 3 are skipped when the strategy they name is prefill and the model carries the no_prefill quirk, so stale data degrades to the heuristic rather than to a failed request.

Pass allow_user_override=False in scripts that generate committed data, so one developer's local pin can't leak into a shared file.

See Strategies for the surrounding workflow, and Model Normalization for provider-prefix and alias handling.