Using Selector Shortcuts

The Selector class provides static methods that serve as shortcuts for commonly used span selections. These shortcuts are supported via TruApp, as well as TruChain and TruLlama when using LangChain and LlamaIndex frameworks, respectively.

Evaluating App Input

To evaluate the application input, use Selector.select_record_input():

from trulens.core import Metric, Selector

f_coherence = Metric(
    implementation=provider.coherence,
    name="Coherence",
    selectors={
        "text": Selector.select_record_input(),
    },
)

This is equivalent to the explicit selector:

from trulens.core import Metric, Selector
from trulens.otel.semconv.trace import SpanAttributes

f_coherence = Metric(
    implementation=provider.coherence,
    name="Coherence",
    selectors={
        "text": Selector(
            span_type=SpanAttributes.SpanType.RECORD_ROOT,
            span_attribute=SpanAttributes.RECORD_ROOT.INPUT,
        ),
    },
)

Evaluating App Output

To evaluate the application output, use Selector.select_record_output():

from trulens.core import Metric, Selector

f_coherence = Metric(
    implementation=provider.coherence,
    name="Coherence",
    selectors={
        "text": Selector.select_record_output(),
    },
)

This is equivalent to the explicit selector:

from trulens.core import Metric, Selector
from trulens.otel.semconv.trace import SpanAttributes

f_coherence = Metric(
    implementation=provider.coherence,
    name="Coherence",
    selectors={
        "text": Selector(
            span_type=SpanAttributes.SpanType.RECORD_ROOT,
            span_attribute=SpanAttributes.RECORD_ROOT.OUTPUT,
        ),
    },
)

Evaluating Retrieved Context

To evaluate retrieved context, use Selector.select_context():

from trulens.core import Metric, Selector

f_groundedness = Metric(
    implementation=provider.groundedness_measure_with_cot_reasons,
    name="Groundedness",
    selectors={
        "source": Selector.select_context(collect_list=True),
        "statement": Selector.select_record_output(),
    },
)

This is equivalent to the explicit selector:

from trulens.core import Metric, Selector
from trulens.otel.semconv.trace import SpanAttributes

f_groundedness = Metric(
    implementation=provider.groundedness_measure_with_cot_reasons,
    name="Groundedness",
    selectors={
        "source": Selector(
            span_type=SpanAttributes.SpanType.RETRIEVAL,
            span_attribute=SpanAttributes.RETRIEVAL.RETRIEVED_CONTEXTS,
            collect_list=True,
        ),
        "statement": Selector.select_record_output(),
    },
)

Using collect_list

The collect_list parameter controls how multiple matching spans are handled:

• collect_list=True: Concatenates all matching span attributes into a single value for evaluation
• collect_list=False: Evaluates each span attribute individually and aggregates results

This is particularly useful for context relevance where you want to evaluate each context chunk separately:

from trulens.core import Metric, Selector
import numpy as np

f_context_relevance = Metric(
    implementation=provider.context_relevance_with_cot_reasons,
    name="Context Relevance",
    selectors={
        "question": Selector.select_record_input(),
        "context": Selector.select_context(collect_list=False),
    },
    agg=np.mean,
)

Summary of Selector Shortcuts

Shortcut	Span Type	Span Attribute
Selector.select_record_input()	RECORD_ROOT	RECORD_ROOT.INPUT
Selector.select_record_output()	RECORD_ROOT	RECORD_ROOT.OUTPUT
Selector.select_context(collect_list=...)	RETRIEVAL	RETRIEVAL.RETRIEVED_CONTEXTS