Skip to content

:hedgehog: PostHog is the leading platform for building self-driving products. Our developer tools – AI observability, analytics, session replay, flags, experiments, error tracking, logs, and more – capture all the context agents need to diagnose problems, uncover opportunities, and ship fixes. Steer it all from Slack, web, desktop, or the MCP.

feat(aio): survey responses in the trace feedback mode (#124376)

Carlos Marchal3dfc586a7277
diff --git a/.github/new-events-schema-targets.txt b/.github/new-events-schema-targets.txt
index 039dc5118c29..58ea04a7c0e9 100644
--- a/.github/new-events-schema-targets.txt
+++ b/.github/new-events-schema-targets.txt
@@ -84,4 +84,5 @@ products/ai_observability/backend/api/test/test_evaluations.py
products/ai_observability/backend/api/test/test_personal_spend.py
products/revenue_analytics/backend/views/test/test_mrr_views.py
products/surveys/backend/api/test/test_survey.py
+products/surveys/backend/responses/test_linked.py
products/web_analytics/backend/test/test_heatmaps_api.py
diff --git a/frontend/snapshots.yml b/frontend/snapshots.yml
index 926b69289cc7..1f7db71eb495 100644
--- a/frontend/snapshots.yml
+++ b/frontend/snapshots.yml
@@ -9391,9 +9391,9 @@ snapshots:
scenes-app-ai-observability-trace-view-evals-chip--some-failed--light:
hash: v1.k794b7964.f6f244cf98f06733927cd0a5ee7d7904908c6489b4583e71d0e445f5b6c30397.SbII8dL5ykXeviCDXpciR9TWuevdn0XLgEzEU3PejDM
scenes-app-ai-observability-trace-view-feedback--empty--dark:
- hash: v1.k794b7964.7fcd6990cb4124d3c2102fdf9a61729a7147140dd202705dcc13d148340a6aae.K6Y6bMLOoabdYZYDhkpEgdaWMIKc8bxHYqARMBjcJsg
+ hash: v1.k794b7964.fca1924b8daff044a790370ddce702358924ee88525bf928737c743e3009c028.pHlq57K3zTFY0KCkf5WuAg3vg7A3ckNEK9pTqmXRzns
scenes-app-ai-observability-trace-view-feedback--empty--light:
- hash: v1.k794b7964.1a5e59c324cf81fcaa6ca778eb489f6c873ab83cef4eebe79b2404dc7ade7bff.QkZ2cNKAkwNU5zYhWlqYB4ev6HG9zI5oBp7nzYZ7wbg
+ hash: v1.k794b7964.1685723a7ce4615e61ab1eae93ec659c8ac6e9009ff45edf539b42a81e7222df.XnJCmw3orsFHNdhjmajchf11Ppw9R3gk3Iw485976Rw
scenes-app-ai-observability-trace-view-feedback--focused-on-a-step--dark:
hash: v1.k794b7964.ccf120a9d03e413fbd5e98441fe4c4ccc12223da1923981b43717bfdb3132227.xXmxciFfHeD25MoC3s_b7W-W_6iMvjW1RgYbWkIRknw
scenes-app-ai-observability-trace-view-feedback--focused-on-a-step--light:
@@ -9407,9 +9407,9 @@ snapshots:
scenes-app-ai-observability-trace-view-feedback--loading--light:
hash: v1.k794b7964.21d886cdffe610104020c41adba7daed122c2558b46c64f3db996310820a92a5.Kgd6p3D4p6GKOXi1u6wCVFfNxiinZZKyqXFS7o--x14
scenes-app-ai-observability-trace-view-feedback--narrow--dark:
- hash: v1.k794b7964.407140f0509c3f0f44cfccebeb90073405fb276b49f5e68bbe01e453de2e51b8.awZbgmLAyB0Lj58ifmDYcfThU0faXTmmrMsaZkmhQMw
+ hash: v1.k794b7964.650f8fefa313d380d34f5aa5da88f49ed9d36086eca1d77108761a2ece9b73d7.AEIig19ivaFqHzUC0cMFMvAEJXxLSe3jkjlbrQTuuEs
scenes-app-ai-observability-trace-view-feedback--narrow--light:
- hash: v1.k794b7964.c196f001ea7d126b256b1e773a12b4804486b5b9467a74c9cd90bc928f03953c.dkmJGOz-kuZFaNJat6eo9aN7nfSpyo_jHxIgNUgnYo4
+ hash: v1.k794b7964.47a1c1e10d4781aff843a4490d0b7290b3f6dd2e8b80220dac9069436c7d6634.h0AWs12xPjgkdKNY7ojcYaX3G_rzscD5lX5X4Pgz3PM
scenes-app-ai-observability-trace-view-feedback--scores-and-text--dark:
hash: v1.k794b7964.42534f604afe39beb23e4c09f89d84809bc98e4f2b59fedb79e5a31144f8c500.2i0dy5bXglrLz9xRUb8jss1TSydnGhnJeWRoGNn6uac
scenes-app-ai-observability-trace-view-feedback--scores-and-text--light:
@@ -9418,6 +9418,14 @@ snapshots:
hash: v1.k794b7964.b070900020433ca79a06ea06e9d662e53127c9eac7bb8f9db7d56b0965f20615.u7YGWkBWd6FhUE3YrpBUW6vQXiCoSMApTE9jWdTkvis
scenes-app-ai-observability-trace-view-feedback--scores-only--light:
hash: v1.k794b7964.96177e830cd7e7306e202523e1fc27a44436de5e7956ce78836c02e4c8be708f.BYDlsBquGF3TY13KOFeLtmIIfPaVhmoXhqqOYgA1FFg
+ scenes-app-ai-observability-trace-view-feedback--survey-read-failed--dark:
+ hash: v1.k794b7964.65a560fe236aa7a2d8a0f7cc3c62625bf337d9853fcc363124ac7ed28c864715.8O9AR-jMFC_e1LJdc3MuB-TZER2xM5LH8xplsZoKROA
+ scenes-app-ai-observability-trace-view-feedback--survey-read-failed--light:
+ hash: v1.k794b7964.13fbfd6b2e29b1d8f6b9273595e361038b99a6f15593ca3bcd592de39e26acf7.1LjqUgWMDG_SuF4yaFI_FTjt--pY0gU7h3hDWZmXoeI
+ scenes-app-ai-observability-trace-view-feedback--with-surveys--dark:
+ hash: v1.k794b7964.177690f62823852dfc3e7597448f748d5c779cd7e3c132bc5ec20a6fa040336d.kU7x2Svq6sPvpg0bs90Vb0obQ9A5GEouEZ-CRKdPmko
+ scenes-app-ai-observability-trace-view-feedback--with-surveys--light:
+ hash: v1.k794b7964.bf831c89fb9249d4e52a983f09857489fea2a096fca6d18d15f86453e2e23991.oYVFY4WtphJ0iSW3tMmGJa3BcWZReKFHkVzn89v1Ldk
scenes-app-ai-observability-trace-view-input-and-output--empty-object--dark:
hash: v1.k794b7964.6ed3f97001d8b595545a78f37627d3e09c188a3dcd782c8ab3389c65cbd4210f.2U9W7fXNr-OZRXfkmXy1ICFu5-dSKZUsMMWOtbf4PWw
scenes-app-ai-observability-trace-view-input-and-output--empty-object--light:
@@ -9850,6 +9858,10 @@ snapshots:
hash: v1.k794b7964.f48c22bbab5113e9ef6290dad04daae1285be7f8932d41fa8ace16ace664e502.oxJblySth_77826dzwFkmd_PGo7EfwpvdJx8GlFzrgw
scenes-app-ai-observability-trace-view-trace-summary-bar--with-git-branch--light:
hash: v1.k794b7964.f6e85ad3ddc7ab047875da5570247b6bab1038c36f5ec2bd3d6be709296f9595.Z2RX9IwREaeGSEESRc0k_qVZIf3cwVzLtss3gO9CGh0
+ scenes-app-ai-observability-trace-view-trace-summary-bar--with-thumbs-feedback--dark:
+ hash: v1.k794b7964.7362a5c721272d156b1af5133afcdd78e0672d8336d622d54deaf29226d0d699.1p5tXtQY1JGCciuOqhrEfLLBXzdiFwOJv__OVCg0Cvo
+ scenes-app-ai-observability-trace-view-trace-summary-bar--with-thumbs-feedback--light:
+ hash: v1.k794b7964.156eaf59c0759646ca88ad40c1dae1bcd335485e02b01ca58fbf6f36e0f0818a.BWpOPt4q8glW5PSG0TFKdVz-VfsdES6PHDhQ4y3L96g
scenes-app-ai-observability-trace-view-trace-timeline--agent-run--dark:
hash: v1.k794b7964.96ceaef4cd703d36753e7b6071612547f6854073f804a52b50d9c7c6fd6d5718.i6JAjAgngbls2Ro2Z9eLdkQwjEiXw1Zt2Fnz32p1wVs
scenes-app-ai-observability-trace-view-trace-timeline--agent-run--light:
@@ -9911,9 +9923,9 @@ snapshots:
scenes-app-ai-observability-trace-view-trace-view--error-trace--light:
hash: v1.k794b7964.bf3111e7a39fc9d9c9dd2acaeb3e0a6dd264005eaf7630e02f83e76f0b31c6c4.mILO7jkQuM6lNdlXFNhncF9TyaT5CuSfze_ookKOiz4
scenes-app-ai-observability-trace-view-trace-view--feedback-mode--dark:
- hash: v1.k794b7964.df4d60c2a4558c739f9658ba5ba5a85d1b6988925f88210eb8973c2f72a96c9d.i7rmESrElWpU9f6K3xEfiWUCq2LmGVcGzGXrNEhvtXY
+ hash: v1.k794b7964.889d686dada73683048a402f1c14ba362813f9021c78dc2127a890e85d62c3a2.4W3H-CludFIrDuvCeTaRSE_lpT5WO7ZMp_rFoq69TAA
scenes-app-ai-observability-trace-view-trace-view--feedback-mode--light:
- hash: v1.k794b7964.4008b7bc041e33d6457038242a83d34e06aa0064f52e08241bc4a40112c71cbd.3inRKhCjXf4qBLuXls4GrocWk1OxTlSJbTSCOtgyH7w
+ hash: v1.k794b7964.740b5ccc6508dc0ebe7feea3a0484c54c411fe458a7e302bee274a8d107cedd3.SQNSmOiyuMYYNVPGLa7UQlPWuHETHX0X5kje6g2oP2Y
scenes-app-ai-observability-trace-view-trace-view--generation-selected--dark:
hash: v1.k794b7964.d31bb379f48e59e75e81826b08e59e01c8729349348209e37312494cc2f520f3.SD-1JnMC8iAmFz0cQLskows96wdUBctK7pFjw0EHCvM
scenes-app-ai-observability-trace-view-trace-view--generation-selected--light:
diff --git a/products/ai_observability/backend/facade/api.py b/products/ai_observability/backend/facade/api.py
index 435b96cfe02b..cb9cdc72aef7 100644
--- a/products/ai_observability/backend/facade/api.py
+++ b/products/ai_observability/backend/facade/api.py
@@ -3,8 +3,17 @@ from datetime import date, datetime
from posthog.schema import HogQLPropertyFilter
+from posthog.exceptions_capture import capture_exception
from posthog.models import Team, User
+from products.surveys.backend.facade.api import (
+ LinkedSurveyAnswer,
+ LinkedSurveyResponse,
+ LinkedSurveyResponses,
+ SurveyRef,
+ get_survey_responses_linked_to,
+)
+
from ..logic.evaluations.evaluation_metas import load_evaluation_meta, load_evaluation_metas
from ..logic.evaluations.overview import evaluation_result, group_by_evaluation, overview_totals, pass_rate_overview
from ..logic.evaluations.overview_queries import load_bucket_counts, load_category_counts
@@ -37,6 +46,7 @@ from ..logic.traces.summary import (
)
from ..logic.traces.summary_generation import generate_summary
from ..logic.traces.summary_store import fail_abandoned_summary, request_summary, viewer_summary
+from ..logic.traces.survey_feedback import THUMBS_UP, answer_kind, count_thumbs, survey_window
from ..logic.traces.timeline import TimelineRow
from ..logic.traces.trace import Trace
from ..logic.traces.trace_queries import (
@@ -297,17 +307,49 @@ def _to_trace_cluster(cluster: TraceCluster) -> contracts.TraceCluster:
)
+NO_LINKED_SURVEYS = LinkedSurveyResponses(responses=(), unanswered=())
+
+
def get_trace_feedback(
- team: Team, user: User | None, trace_id: str, timestamp_hint: datetime | None
+ team: Team, user: User | None, trace_id: str, timestamp_hint: datetime | None, *, include_surveys: bool
) -> contracts.TraceFeedback:
- """End-user feedback on a trace. An unknown trace has none."""
- feedback = TraceFeedback(trace_id, load_trace_feedback_rows(team, user, trace_id, timestamp_hint))
+ """End-user feedback on a trace, from its feedback events and the survey responses linked to it. An unknown trace
+ has none. Survey responses are read only with `include_surveys`, and a failed survey read leaves them out."""
+ rows = load_trace_feedback_rows(team, user, trace_id, timestamp_hint)
+ feedback = TraceFeedback(trace_id, rows)
+ survey_responses_status: contracts.TraceSurveyResponsesStatus = "ok" if include_surveys else "not_permitted"
+ linked = NO_LINKED_SURVEYS
+ window = survey_window(rows, timestamp_hint)
+ if include_surveys and window is not None:
+ try:
+ linked = get_survey_responses_linked_to(
+ team, user, property_name="$ai_trace_id", property_value=trace_id, since=window.start, until=window.end
+ )
+ except Exception as error:
+ # Survey responses are one feedback source among several, so their failure must not hide the others.
+ capture_exception(error)
+ survey_responses_status = "failed"
+ survey_responses = linked.responses
+ thumbs = count_thumbs(survey_responses)
scores = [_to_feedback_score(score) for score in feedback.scores]
- responses = [_to_feedback_text(text) for text in feedback.texts]
+ responses: list[contracts.TraceFeedbackText | contracts.TraceSurveyResponse] = sorted(
+ [
+ *(_to_feedback_text(text) for text in feedback.texts),
+ *(_to_survey_response(response, trace_id) for response in survey_responses),
+ ],
+ key=lambda response: response.timestamp,
+ )
return contracts.TraceFeedback(
scores=scores,
responses=responses,
- summary=contracts.TraceFeedbackSummary(total=len(scores) + len(responses)),
+ unanswered=[
+ contracts.TraceUnansweredSurvey(survey=_to_survey(survey.survey), timestamp=survey.shown_at)
+ for survey in linked.unanswered
+ ],
+ survey_responses_status=survey_responses_status,
+ summary=contracts.TraceFeedbackSummary(
+ total=len(scores) + len(responses), thumbs_up=thumbs.up, thumbs_down=thumbs.down
+ ),
)
@@ -323,6 +365,45 @@ def _to_feedback_text(text: FeedbackText) -> contracts.TraceFeedbackText:
)
+def _to_survey(survey: SurveyRef) -> contracts.TraceSurvey:
+ return contracts.TraceSurvey(id=survey.id, name=survey.name)
+
+
+def _to_survey_response(response: LinkedSurveyResponse, trace_id: str) -> contracts.TraceSurveyResponse:
+ return contracts.TraceSurveyResponse(
+ kind="survey",
+ id=response.submission_id,
+ node_id=trace_id,
+ timestamp=response.submitted_at,
+ survey=_to_survey(response.survey),
+ answers=[_to_survey_answer(answer) for answer in response.answers],
+ complete=response.complete,
+ unavailable=response.unavailable,
+ )
+
+
+def _to_survey_answer(answer: LinkedSurveyAnswer) -> contracts.TraceSurveyAnswer:
+ question = answer.question_text
+ values = list(answer.value) if isinstance(answer.value, tuple) else [answer.value]
+ match answer_kind(answer):
+ case "thumbs":
+ return contracts.TraceThumbsAnswer(kind="thumbs", question=question, up=answer.value == THUMBS_UP)
+ case "rating":
+ rating = answer.rating
+ return contracts.TraceRatingAnswer(
+ kind="rating",
+ question=question,
+ value=", ".join(values),
+ scale=rating.scale if rating else None,
+ lower_label=rating.lower_bound_label if rating else None,
+ upper_label=rating.upper_bound_label if rating else None,
+ )
+ case "choice":
+ return contracts.TraceChoiceAnswer(kind="choice", question=question, choices=values)
+ case "text":
+ return contracts.TraceTextAnswer(kind="text", question=question, text=", ".join(values))
+
+
def _to_evaluation_summary(summary: EvaluationSummary) -> contracts.TraceEvaluationSummary:
return contracts.TraceEvaluationSummary(
passed=summary.passed,
diff --git a/products/ai_observability/backend/facade/contracts.py b/products/ai_observability/backend/facade/contracts.py
index c1645696c4e2..59c1461ffe2e 100644
--- a/products/ai_observability/backend/facade/contracts.py
+++ b/products/ai_observability/backend/facade/contracts.py
@@ -211,13 +211,84 @@ class TraceFeedbackText(Contract):
timestamp: datetime
+class TraceSurvey(Contract):
+ id: str
+ name: str | None = Field(description="Null when the survey no longer exists.")
+
+
+class TraceThumbsAnswer(Contract):
+ kind: Literal["thumbs"]
+ question: str
+ up: bool
+
+
+class TraceRatingAnswer(Contract):
+ kind: Literal["rating"]
+ question: str
+ value: str
+ scale: int | None
+ lower_label: str | None
+ upper_label: str | None
+
+
+class TraceChoiceAnswer(Contract):
+ kind: Literal["choice"]
+ question: str
+ choices: list[str]
+
+
+class TraceTextAnswer(Contract):
+ kind: Literal["text"]
+ question: str
+ text: str
+
+
+TraceSurveyAnswer = Annotated[
+ TraceThumbsAnswer | TraceRatingAnswer | TraceChoiceAnswer | TraceTextAnswer, Field(discriminator="kind")
+]
+
+
+class TraceSurveyResponse(Contract):
+ kind: Literal["survey"]
+ id: str
+ node_id: str = Field(
+ description="The tree node the response targets: the trace id, as a survey response covers the whole trace."
+ )
+ timestamp: datetime
+ survey: TraceSurvey
+ answers: list[TraceSurveyAnswer]
+ complete: bool = Field(description="False when the user left the survey before its end.")
+ unavailable: Literal["survey_not_found", "no_questions"] | None = Field(
+ description="Why the answers cannot be shown: the survey was deleted, or it has no questions."
+ )
+
+
+class TraceUnansweredSurvey(Contract):
+ survey: TraceSurvey
+ timestamp: datetime = Field(description="When the survey was first shown.")
+
+
+TraceSurveyResponsesStatus = Literal["ok", "failed", "not_permitted"]
+
+
+TraceFeedbackResponse = Annotated[TraceFeedbackText | TraceSurveyResponse, Field(discriminator="kind")]
+
+
class TraceFeedbackSummary(Contract):
total: int = Field(description="Scores plus responses.")
+ thumbs_up: int = Field(description="Thumbs up answers to 2-point survey rating questions.")
+ thumbs_down: int = Field(description="Thumbs down answers to 2-point survey rating questions.")
class TraceFeedback(Contract):
scores: list[TraceFeedbackScore] = Field(description="Oldest first.")
- responses: list[TraceFeedbackText] = Field(description="What users said, oldest first.")
+ responses: list[TraceFeedbackResponse] = Field(description="What users said, oldest first.")
+ unanswered: list[TraceUnansweredSurvey] = Field(
+ description="Surveys shown with the trace and not answered, oldest first."
+ )
+ survey_responses_status: TraceSurveyResponsesStatus = Field(
+ description="ok when survey responses are included, failed when the survey read failed, not_permitted when the token lacks the survey:read and query:read scopes."
+ )
summary: TraceFeedbackSummary
diff --git a/products/ai_observability/backend/logic/traces/survey_feedback.py b/products/ai_observability/backend/logic/traces/survey_feedback.py
new file mode 100644
index 000000000000..d37f7279ca46
--- /dev/null
+++ b/products/ai_observability/backend/logic/traces/survey_feedback.py
@@ -0,0 +1,54 @@
+from collections.abc import Sequence
+from datetime import UTC, datetime, time, timedelta
+from typing import Literal
+
+from posthog.dataclasses import frozen
+
+from products.surveys.backend.facade.api import LinkedSurveyAnswer, LinkedSurveyResponse
+
+from .trace_queries import FeedbackEventRow
+
+SurveyAnswerKind = Literal["thumbs", "rating", "choice", "text"]
+
+# Surveys stores a 2-point rating as "1" for thumbs up and "2" for thumbs down, whatever its display.
+THUMBS_UP = "1"
+THUMBS_DOWN = "2"
+
+# Users answer a survey about a trace some time after it ran, so the read looks past the trace's own day.
+SURVEY_DAYS_AFTER_TRACE = 30
+
+
+@frozen
+class SurveyWindow:
+ start: datetime
+ end: datetime
+
+
+@frozen
+class ThumbsCount:
+ up: int
+ down: int
+
+
+def survey_window(rows: Sequence[FeedbackEventRow], timestamp_hint: datetime | None) -> SurveyWindow | None:
+ """The trace's UTC day and the `SURVEY_DAYS_AFTER_TRACE` days after it. The trace's day is the earliest of the hint
+ and the feedback rows. None when neither says when the trace ran."""
+ known = [moment for moment in (timestamp_hint, *(row.timestamp for row in rows)) if moment is not None]
+ if not known:
+ return None
+ start = datetime.combine(min(known).astimezone(UTC).date(), time(), tzinfo=UTC)
+ return SurveyWindow(start=start, end=start + timedelta(days=1 + SURVEY_DAYS_AFTER_TRACE))
+
+
+def answer_kind(answer: LinkedSurveyAnswer) -> SurveyAnswerKind:
+ if answer.question_type == "rating":
+ is_thumbs = answer.rating is not None and answer.rating.scale == 2 and answer.value in (THUMBS_UP, THUMBS_DOWN)
+ return "thumbs" if is_thumbs else "rating"
+ if answer.question_type in ("single_choice", "multiple_choice"):
+ return "choice"
+ return "text"
+
+
+def count_thumbs(responses: Sequence[LinkedSurveyResponse]) -> ThumbsCount:
+ thumbs = [answer.value for response in responses for answer in response.answers if answer_kind(answer) == "thumbs"]
+ return ThumbsCount(up=thumbs.count(THUMBS_UP), down=thumbs.count(THUMBS_DOWN))
diff --git a/products/ai_observability/backend/presentation/views.py b/products/ai_observability/backend/presentation/views.py
index b71ae0058564..cf86aef4746f 100644
--- a/products/ai_observability/backend/presentation/views.py
+++ b/products/ai_observability/backend/presentation/views.py
@@ -14,11 +14,13 @@ from posthog.api.mixins import ValidatedRequest, validated_request
from posthog.api.routing import TeamAndOrgViewSetMixin
from posthog.clickhouse.query_tagging import Feature, tag_queries
from posthog.models.user import User
+from posthog.permissions import get_authenticator_scopes
from posthog.rate_limit import (
AIObservabilitySummarizationBurstThrottle,
AIObservabilitySummarizationDailyThrottle,
AIObservabilitySummarizationSustainedThrottle,
)
+from posthog.scopes import scopes_not_covered
from ..facade import api, contracts
from .query_concurrency import query_concurrency_slots
@@ -64,6 +66,15 @@ NODE_ID_PARAMETER = OpenApiParameter(
OpenApiParameter.PATH,
description="The node id, sent as the unpadded base64url encoding of its UTF-8 bytes. The root node's id is the trace id.",
)
+SURVEY_RESPONSE_SCOPES = ["survey:read", "query:read"]
+
+
+def _token_allows(request: ValidatedRequest, required_scopes: list[str]) -> bool:
+ """True for session auth, which carries no scopes."""
+ scopes = get_authenticator_scopes(getattr(request, "successful_authenticator", None))
+ return scopes is None or "*" in scopes or not scopes_not_covered(scopes, required_scopes)
+
+
THROTTLED_RESPONSE = OpenApiResponse(description="Too many queries are running for the project or organization.")
NOT_APPROVED_DETAIL = "AI data processing must be approved by your organization before using summaries."
NOT_A_USER_DETAIL = "Summaries are only available to signed-in users and their personal API keys."
@@ -199,12 +210,17 @@ class TraceViewSet(TeamAndOrgViewSetMixin, viewsets.ViewSet):
)
@action(detail=True, methods=["GET"], url_path="feedback")
def feedback(self, request: ValidatedRequest, pk: str, **kwargs: Any) -> Response:
- """End-user feedback on a trace: the scores and the comments its feedback events carry, each with the step it targets."""
+ """End-user feedback on a trace: scores and comments from its feedback events, survey responses linked to it, and surveys shown but not answered."""
try:
trace_id = decode_trace_id_segment(pk)
except MalformedTraceIdSegmentError:
return Response({"detail": "Malformed encoded trace id."}, status=status.HTTP_400_BAD_REQUEST)
- feedback = self._read(request, lambda user, hint: api.get_trace_feedback(self.team, user, trace_id, hint))
+ # Survey responses are survey data, so a token also needs the scopes the survey responses endpoint requires.
+ include_surveys = _token_allows(request, SURVEY_RESPONSE_SCOPES)
+ feedback = self._read(
+ request,
+ lambda user, hint: api.get_trace_feedback(self.team, user, trace_id, hint, include_surveys=include_surveys),
+ )
return Response(feedback.model_dump(mode="json", by_alias=True))
@validated_request(
diff --git a/products/ai_observability/backend/tests/logic/traces/test_survey_feedback.py b/products/ai_observability/backend/tests/logic/traces/test_survey_feedback.py
new file mode 100644
index 000000000000..c54a1843823a
--- /dev/null
+++ b/products/ai_observability/backend/tests/logic/traces/test_survey_feedback.py
@@ -0,0 +1,115 @@
+from datetime import UTC, datetime, timedelta
+
+from parameterized import parameterized
+
+from products.ai_observability.backend.logic.traces.survey_feedback import (
+ SurveyAnswerKind,
+ SurveyWindow,
+ ThumbsCount,
+ answer_kind,
+ count_thumbs,
+ survey_window,
+)
+from products.ai_observability.backend.logic.traces.trace_queries import FeedbackEventRow
+from products.surveys.backend.facade.api import LinkedSurveyAnswer, LinkedSurveyResponse, RatingScale, SurveyRef
+
+from .rows import T0, make_feedback_row
+
+THUMBS = RatingScale(scale=2, display="emoji", lower_bound_label=None, upper_bound_label=None)
+
+
+def make_answer(
+ question_type: str, value: str | tuple[str, ...], rating: RatingScale | None = None
+) -> LinkedSurveyAnswer:
+ return LinkedSurveyAnswer(
+ question_id="q-1",
+ question_index=0,
+ question_text="Was this answer useful?",
+ question_type=question_type,
+ rating=rating,
+ value=value,
+ )
+
+
+def make_response(*answers: LinkedSurveyAnswer, minute: int = 0) -> LinkedSurveyResponse:
+ return LinkedSurveyResponse(
+ survey=SurveyRef(id="survey-1", name="Answer quality"),
+ submission_id=f"sub-{minute}",
+ submitted_at=T0 + timedelta(minutes=minute),
+ complete=True,
+ answers=answers,
+ unavailable=None,
+ )
+
+
+ [
+ ("a 2-point emoji rating", "rating", THUMBS, "1", "thumbs"),
+ (
+ "a 2-point number rating",
+ "rating",
+ RatingScale(scale=2, display="number", lower_bound_label=None, upper_bound_label=None),
+ "2",
+ "thumbs",
+ ),
+ ("a 2-point rating with another value", "rating", THUMBS, "3", "rating"),
+ (
+ "a 5-point rating",
+ "rating",
+ RatingScale(scale=5, display="number", lower_bound_label="Bad", upper_bound_label="Great"),
+ "4",
+ "rating",
+ ),
+ ("a single choice", "single_choice", None, "Accuracy", "choice"),
+ ("a multiple choice", "multiple_choice", None, ("Accuracy", "Tone"), "choice"),
+ ("an open question", "open", None, "Too long", "text"),
+ ("a link question", "link", None, "https://example.com", "text"),
+ ]
+)
+def test_answer_kind(
+ _name: str,
+ question_type: str,
+ rating: RatingScale | None,
+ value: str | tuple[str, ...],
+ expected: SurveyAnswerKind,
+) -> None:
+ assert answer_kind(make_answer(question_type, value, rating)) == expected
+
+
+def test_counts_thumbs_up_and_down_across_responses() -> None:
+ responses = [
+ make_response(make_answer("rating", "1", THUMBS), make_answer("open", "Fine."), minute=1),
+ make_response(make_answer("rating", "2", THUMBS), minute=2),
+ make_response(make_answer("rating", "1", THUMBS), minute=3),
+ ]
+
+ assert count_thumbs(responses) == ThumbsCount(up=2, down=1)
+
+
+ [
+ (
+ "the hint's day, when feedback arrived days later",
+ [make_feedback_row(uuid="m1", metric_value_json="1", timestamp=T0 + timedelta(days=4))],
+ T0,
+ datetime(2026, 9, 1, tzinfo=UTC),
+ ),
+ (
+ "the earliest row's day, when the hint is later",
+ [
+ make_feedback_row(uuid="m1", metric_value_json="1", timestamp=T0 + timedelta(days=2)),
+ make_feedback_row(uuid="e0", event="$ai_span", timestamp=T0 + timedelta(hours=13)),
+ ],
+ T0 + timedelta(days=3),
+ datetime(2026, 9, 1, tzinfo=UTC),
+ ),
+ ("the hint's day without rows", [], T0 + timedelta(days=3), datetime(2026, 9, 4, tzinfo=UTC)),
+ ("nothing without rows or a hint", [], None, None),
+ ]
+)
+def test_survey_window_covers_the_trace_day_and_the_30_days_after_it(
+ _name: str, rows: list[FeedbackEventRow], timestamp_hint: datetime | None, start: datetime | None
+) -> None:
+ window = survey_window(rows, timestamp_hint)
+
+ assert window == (SurveyWindow(start=start, end=start + timedelta(days=31)) if start else None)
diff --git a/products/ai_observability/backend/tests/test_facade_api.py b/products/ai_observability/backend/tests/test_facade_api.py
index 862b7ac7264d..69f3e9566ead 100644
--- a/products/ai_observability/backend/tests/test_facade_api.py
+++ b/products/ai_observability/backend/tests/test_facade_api.py
@@ -3,7 +3,7 @@ from datetime import UTC, datetime, timedelta
import pytest
from posthog.test.base import BaseTest, ClickhouseTestMixin
-from unittest.mock import patch
+from unittest.mock import call, patch
from parameterized import parameterized
@@ -28,15 +28,24 @@ from products.ai_observability.backend.logic.traces.summary_store import (
summary_key,
write_summary,
)
-from products.ai_observability.backend.logic.traces.trace_queries import LoadedTrace
+from products.ai_observability.backend.logic.traces.trace_queries import FeedbackEventRow, LoadedTrace
from products.ai_observability.backend.models.evaluations import Evaluation
from products.event_definitions.backend.models.property_definition import PropertyDefinition
+from products.surveys.backend.facade.api import (
+ LinkedSurveyAnswer,
+ LinkedSurveyResponse,
+ LinkedSurveyResponses,
+ RatingScale,
+ SurveyRef,
+ UnansweredSurvey,
+)
from .logic.traces.rows import (
CLUSTERING_JOB_ID,
make_cluster,
make_cluster_run_event,
make_evaluation_result_event,
+ make_feedback_row,
make_row,
)
from .property_access import hide_event_property
@@ -72,6 +81,144 @@ def test_missing_trace_raises_not_found() -> None:
api.get_trace(team=None, user=None, trace_id="missing", timestamp_hint=None) # type: ignore[arg-type]
+NO_SURVEYS = LinkedSurveyResponses(responses=(), unanswered=())
+
+
+def _survey_answer(
+ question: str, question_type: str, value: str | tuple[str, ...], scale: int | None = None
+) -> LinkedSurveyAnswer:
+ rating = (
+ RatingScale(scale=scale, display="number", lower_bound_label="Bad", upper_bound_label="Great")
+ if scale
+ else None
+ )
+ return LinkedSurveyAnswer(
+ question_id=question,
+ question_index=0,
+ question_text=question,
+ question_type=question_type,
+ rating=rating,
+ value=value,
+ )
+
+
+def test_feedback_maps_scores_texts_and_survey_responses_oldest_first() -> None:
+ rows = (
+ make_feedback_row(uuid="gen-uuid", event="$ai_generation", generation_id="gen-a", timestamp=T0),
+ make_feedback_row(uuid="metric", metric_name="helpfulness", metric_value_json="0", parent_id="gen-a"),
+ make_feedback_row(
+ uuid="text",
+ event="$ai_feedback",
+ feedback_text_json='"Too long, but correct."',
+ timestamp=T0 + timedelta(minutes=2),
+ ),
+ )
+ survey = SurveyRef(id="survey-1", name="Answer quality")
+ linked = LinkedSurveyResponses(
+ responses=(
+ LinkedSurveyResponse(
+ survey=survey,
+ submission_id="sub-1",
+ submitted_at=T0 + timedelta(minutes=1),
+ complete=False,
+ answers=(
+ _survey_answer("Useful?", "rating", "1", scale=2),
+ _survey_answer("How good?", "rating", "4", scale=5),
+ _survey_answer("What could be better?", "multiple_choice", ("Accuracy", "Tone")),
+ _survey_answer("Anything else?", "open", "Shorter, please."),
+ ),
+ unavailable=None,
+ ),
+ ),
+ unanswered=(
+ UnansweredSurvey(survey=SurveyRef(id="survey-2", name="Follow-up"), shown_at=T0 + timedelta(minutes=3)),
+ ),
+ )
+ with (
+ patch.object(api, "load_trace_feedback_rows", return_value=rows),
+ patch.object(api, "get_survey_responses_linked_to", return_value=linked),
+ ):
+ feedback = api.get_trace_feedback(
+ team=None, # type: ignore[arg-type]
+ user=None,
+ trace_id="trace-1",
+ timestamp_hint=None,
+ include_surveys=True,
+ )
+
+ assert [(score.name, score.value, score.node_id) for score in feedback.scores] == [("helpfulness", "0", "gen-uuid")]
+ assert [(response.kind, response.id, response.node_id) for response in feedback.responses] == [
+ ("survey", "sub-1", "trace-1"),
+ ("text", "text", "trace-1"),
+ ]
+ survey_response = feedback.responses[0]
+ assert isinstance(survey_response, contracts.TraceSurveyResponse)
+ assert (survey_response.survey.name, survey_response.complete) == ("Answer quality", False)
+ assert [answer.model_dump() for answer in survey_response.answers] == [
+ {"kind": "thumbs", "question": "Useful?", "up": True},
+ {
+ "kind": "rating",
+ "question": "How good?",
+ "value": "4",
+ "scale": 5,
+ "lower_label": "Bad",
+ "upper_label": "Great",
+ },
+ {"kind": "choice", "question": "What could be better?", "choices": ["Accuracy", "Tone"]},
+ {"kind": "text", "question": "Anything else?", "text": "Shorter, please."},
+ ]
+ assert [(item.survey.id, item.timestamp) for item in feedback.unanswered] == [
+ ("survey-2", T0 + timedelta(minutes=3))
+ ]
+ assert feedback.summary == contracts.TraceFeedbackSummary(total=3, thumbs_up=1, thumbs_down=0)
+
+
+SURVEY_READ = call(
+ None,
+ None,
+ property_name="$ai_trace_id",
+ property_value="trace-1",
+ since=datetime(2026, 9, 1, tzinfo=UTC),
+ until=datetime(2026, 10, 2, tzinfo=UTC),
+)
+WITH_FEEDBACK = (make_feedback_row(uuid="metric", metric_value_json="1", timestamp=T0),)
+
+
+ [
+ ("a trace with feedback", WITH_FEEDBACK, True, None, [SURVEY_READ], "ok"),
+ ("a trace without feedback or a hint", (), True, None, [], "ok"),
+ ("a token without the survey scopes", WITH_FEEDBACK, False, None, [], "not_permitted"),
+ ("a failed survey read", WITH_FEEDBACK, True, RuntimeError("timeout"), [SURVEY_READ], "failed"),
+ ]
+)
+def test_feedback_reads_the_surveys_linked_to_the_trace_from_its_day(
+ _name: str,
+ rows: tuple[FeedbackEventRow, ...],
+ include_surveys: bool,
+ survey_error: Exception | None,
+ expected_reads: list[object],
+ expected_status: str,
+) -> None:
+ with (
+ patch.object(api, "load_trace_feedback_rows", return_value=rows),
+ patch.object(
+ api, "get_survey_responses_linked_to", return_value=NO_SURVEYS, side_effect=survey_error
+ ) as read_surveys,
+ patch.object(api, "capture_exception"),
+ ):
+ feedback = api.get_trace_feedback(
+ team=None, # type: ignore[arg-type]
+ user=None,
+ trace_id="trace-1",
+ timestamp_hint=None,
+ include_surveys=include_surveys,
+ )
+
+ assert read_surveys.call_args_list == expected_reads
+ assert (feedback.survey_responses_status, len(feedback.scores)) == (expected_status, len(rows))
+
+
class TestGetTraceEvaluationResults(ClickhouseTestMixin, BaseTest):
def _detector(self, team: Team | None = None, *, deleted: bool = False, stored_config: object = None) -> str:
evaluation = Evaluation.objects.create(
diff --git a/products/ai_observability/backend/tests/test_trace_views.py b/products/ai_observability/backend/tests/test_trace_views.py
index f067ecd8ee5f..ecd8c1fefb96 100644
--- a/products/ai_observability/backend/tests/test_trace_views.py
+++ b/products/ai_observability/backend/tests/test_trace_views.py
@@ -465,6 +465,25 @@ class TestTraceFeedbackView(ClickhouseTestMixin, APIBaseTest):
("text", "Too long, but correct.", "trace-1")
]
assert body["summary"]["total"] == 2
+ assert body["surveyResponsesStatus"] == "ok"
+
+ @parameterized.expand(
+ [
+ ("only the trace scope", ["llm_analytics:read"], "not_permitted"),
+ ("the survey response scopes too", ["llm_analytics:read", "survey:read", "query:read"], "ok"),
+ ]
+ )
+ def test_includes_survey_responses_only_for_a_token_with_the_survey_scopes(
+ self, _name: str, scopes: list[str], expected: str
+ ) -> None:
+ _create_feedback(self.team)
+ headers = {"authorization": f"Bearer {self.create_personal_api_key_with_scopes(scopes)}"}
+ self.client.logout()
+
+ response = self.client.get(self._url(_encode("trace-1")), headers=headers)
+
+ assert response.status_code == 200
+ assert (response.json()["surveyResponsesStatus"], response.json()["summary"]["total"]) == (expected, 2)
def test_an_unknown_trace_has_no_feedback(self) -> None:
response = self.client.get(self._url(_encode("trace-2")))
diff --git a/products/ai_observability/frontend/generated/api.schemas.ts b/products/ai_observability/frontend/generated/api.schemas.ts
index 46b95cf35287..465014a8e56b 100644
--- a/products/ai_observability/frontend/generated/api.schemas.ts
+++ b/products/ai_observability/frontend/generated/api.schemas.ts
@@ -1534,16 +1534,98 @@ export interface TraceFeedbackTextApi {
timestamp: string
}
+export interface TraceSurveyApi {
+ id: string
+ /** Null when the survey no longer exists. */
+ name: string | null
+}
+
+export interface TraceThumbsAnswerApi {
+ kind: 'thumbs'
+ question: string
+ up: boolean
+}
+
+export interface TraceRatingAnswerApi {
+ kind: 'rating'
+ question: string
+ value: string
+ scale: number | null
+ lowerLabel: string | null
+ upperLabel: string | null
+}
+
+export interface TraceChoiceAnswerApi {
+ kind: 'choice'
+ question: string
+ choices: string[]
+}
+
+export interface TraceTextAnswerApi {
+ kind: 'text'
+ question: string
+ text: string
+}
+
+/**
+ * Why the answers cannot be shown: the survey was deleted, or it has no questions.
+ */
+export type TraceSurveyResponseApiUnavailable =
+ | (typeof TraceSurveyResponseApiUnavailable)[keyof typeof TraceSurveyResponseApiUnavailable]
+ | null
+
+export const TraceSurveyResponseApiUnavailable = {
+ SurveyNotFound: 'survey_not_found',
+ NoQuestions: 'no_questions',
+} as const
+
+export interface TraceSurveyResponseApi {
+ kind: 'survey'
+ id: string
+ /** The tree node the response targets: the trace id, as a survey response covers the whole trace. */
+ nodeId: string
+ timestamp: string
+ survey: TraceSurveyApi
+ answers: (TraceThumbsAnswerApi | TraceRatingAnswerApi | TraceChoiceAnswerApi | TraceTextAnswerApi)[]
+ /** False when the user left the survey before its end. */
+ complete: boolean
+ /** Why the answers cannot be shown: the survey was deleted, or it has no questions. */
+ unavailable: TraceSurveyResponseApiUnavailable
+}
+
+export interface TraceUnansweredSurveyApi {
+ survey: TraceSurveyApi
+ /** When the survey was first shown. */
+ timestamp: string
+}
+
+export type SurveyResponsesStatusEnumApi =
+ (typeof SurveyResponsesStatusEnumApi)[keyof typeof SurveyResponsesStatusEnumApi]
+
+export const SurveyResponsesStatusEnumApi = {
+ Ok: 'ok',
+ Failed: 'failed',
+ NotPermitted: 'not_permitted',
+} as const
+
export interface TraceFeedbackSummaryApi {
/** Scores plus responses. */
total: number
+ /** Thumbs up answers to 2-point survey rating questions. */
+ thumbsUp: number
+ /** Thumbs down answers to 2-point survey rating questions. */
+ thumbsDown: number
}
export interface TraceFeedbackApi {
/** Oldest first. */
scores: TraceFeedbackScoreApi[]
/** What users said, oldest first. */
- responses: TraceFeedbackTextApi[]
+ responses: (TraceFeedbackTextApi | TraceSurveyResponseApi)[]
+ /** Surveys shown with the trace and not answered, oldest first. */
+ unanswered: TraceUnansweredSurveyApi[]
+ /** ok when survey responses are included, failed when the survey read failed, not_permitted when the token lacks the survey:read and query:read scopes. */
+ surveyResponsesStatus: SurveyResponsesStatusEnumApi
summary: TraceFeedbackSummaryApi
}
diff --git a/products/ai_observability/frontend/generated/api.ts b/products/ai_observability/frontend/generated/api.ts
index e0c41c991663..fa106e06b304 100644
--- a/products/ai_observability/frontend/generated/api.ts
+++ b/products/ai_observability/frontend/generated/api.ts
@@ -766,7 +766,7 @@ export const getAiObservabilityTracesFeedbackRetrieveUrl = (
}
/**
- * End-user feedback on a trace: the scores and the comments its feedback events carry, each with the step it targets.
+ * End-user feedback on a trace: scores and comments from its feedback events, survey responses linked to it, and surveys shown but not answered.
*/
export const aiObservabilityTracesFeedbackRetrieve = async (
projectId: string,
diff --git a/products/ai_observability/frontend/redesign/trace/adapter/testFixtures.ts b/products/ai_observability/frontend/redesign/trace/adapter/testFixtures.ts
index 157602a99015..5048caf01eea 100644
--- a/products/ai_observability/frontend/redesign/trace/adapter/testFixtures.ts
+++ b/products/ai_observability/frontend/redesign/trace/adapter/testFixtures.ts
@@ -13,6 +13,7 @@ import type {
TraceNodeIOKeysApi,
TraceNodeMetadataApi,
TraceNodeStatsApi,
+ TraceSurveyResponseApi,
} from '../../../generated/api.schemas'
export function makeEvent(overrides: Partial<LLMTraceEvent> & { id: string }): LLMTraceEvent {
@@ -183,10 +184,31 @@ export function makeFeedbackText(overrides: Partial<TraceFeedbackTextApi> = {}):
}
}
+export function makeSurveyResponse(overrides: Partial<TraceSurveyResponseApi> = {}): TraceSurveyResponseApi {
+ return {
+ kind: 'survey',
+ id: 'submission-1',
+ nodeId: 'trace-1',
+ timestamp: '2026-09-01T10:18:00Z',
+ survey: { id: 'survey-1', name: 'Answer quality' },
+ answers: [{ kind: 'thumbs', question: 'Was this answer useful?', up: true }],
+ complete: true,
+ unavailable: null,
+ ...overrides,
+ }
+}
+
export function makeFeedback(overrides: Partial<TraceFeedbackApi> = {}): TraceFeedbackApi {
const scores = overrides.scores ?? [makeFeedbackScore()]
const responses = overrides.responses ?? [makeFeedbackText()]
- return { scores, responses, summary: { total: scores.length + responses.length }, ...overrides }
+ return {
+ scores,
+ responses,
+ unanswered: [],
+ surveyResponsesStatus: 'ok',
+ summary: { total: scores.length + responses.length, thumbsUp: 0, thumbsDown: 0 },
+ ...overrides,
+ }
}
export function makeEvaluationResults(overrides: Partial<TraceEvaluationResultsApi> = {}): TraceEvaluationResultsApi {
diff --git a/products/ai_observability/frontend/redesign/trace/adapter/toFeedback.test.ts b/products/ai_observability/frontend/redesign/trace/adapter/toFeedback.test.ts
index 39af5593b0f6..346b68e09ba3 100644
--- a/products/ai_observability/frontend/redesign/trace/adapter/toFeedback.test.ts
+++ b/products/ai_observability/frontend/redesign/trace/adapter/toFeedback.test.ts
@@ -1,27 +1,37 @@
-import { makeFeedback, makeFeedbackScore, makeFeedbackText, makeTraceResource } from './testFixtures'
+import {
+ makeFeedback,
+ makeFeedbackScore,
+ makeFeedbackText,
+ makeSurveyResponse,
+ makeTraceResource,
+} from './testFixtures'
import { stepFeedbackOf, toFeedbackChip, toFeedbackMode, toStepFeedback } from './toFeedback'
describe('toFeedback', () => {
test.each([
- ['some', 3, { total: 3 }],
- ['none', 0, null],
- ])('toFeedbackChip with %s feedback', (_name, total, expected) => {
- expect(toFeedbackChip({ total })).toEqual(expected)
+ ['thumbs', { total: 4, thumbsUp: 1, thumbsDown: 1 }, { total: 4, thumbsUp: 1, thumbsDown: 1 }],
+ ['nothing', { total: 0, thumbsUp: 0, thumbsDown: 0 }, null],
+ ])('toFeedbackChip with %s', (_name, summary, expected) => {
+ expect(toFeedbackChip(summary)).toEqual(expected)
})
test.each(['trace-1', 'constructor', '__proto__'])(
- 'toStepFeedback puts each entry on its step and trace-level entries on the root %s',
+ 'toStepFeedback puts each entry on its step and trace-level entries and survey responses on the root %s',
(traceId) => {
const score = makeFeedbackScore({ nodeId: 'gen-1' })
const traceScore = makeFeedbackScore({ id: 'score-2', nodeId: traceId })
const feedback = makeFeedback({
scores: [score, traceScore],
- responses: [makeFeedbackText({ nodeId: 'gen-1' }), makeFeedbackText({ id: 'text-2', nodeId: traceId })],
+ responses: [
+ makeFeedbackText({ nodeId: 'gen-1' }),
+ makeFeedbackText({ id: 'text-2', nodeId: traceId }),
+ makeSurveyResponse({ nodeId: traceId }),
+ ],
})
expect({ ...toStepFeedback(feedback) }).toEqual({
- 'gen-1': { scores: [score], textCount: 1 },
- [traceId]: { scores: [traceScore], textCount: 1 },
+ 'gen-1': { scores: [score], responseCount: 1 },
+ [traceId]: { scores: [traceScore], responseCount: 2 },
})
}
)
@@ -33,20 +43,27 @@ describe('toFeedback', () => {
const withFeedback = toStepFeedback(makeFeedback({ scores: [traceScore], responses: [] }))
expect(stepFeedbackOf({}, traceId)).toBeNull()
- expect(stepFeedbackOf(withFeedback, traceId)).toEqual({ scores: [traceScore], textCount: 0 })
+ expect(stepFeedbackOf(withFeedback, traceId)).toEqual({ scores: [traceScore], responseCount: 0 })
}
)
- test('toFeedbackMode names each step and keeps the root and an unknown step as trace-level', () => {
+ test('toFeedbackMode names each step, keeps the root and an unknown step as trace-level and survey responses as they come', () => {
+ const survey = makeSurveyResponse()
+ const unanswered = [{ survey: { id: 'survey-2', name: 'Follow-up' }, timestamp: '2026-09-01T10:19:00Z' }]
const feedback = makeFeedback({
scores: [makeFeedbackScore({ nodeId: 'gen-1' }), makeFeedbackScore({ id: 'score-2', nodeId: 'gone' })],
- responses: [makeFeedbackText({ nodeId: 'trace-1' })],
+ responses: [makeFeedbackText({ nodeId: 'trace-1' }), survey],
+ unanswered,
+ surveyResponsesStatus: 'failed',
})
const mode = toFeedbackMode({ feedback, tree: makeTraceResource().tree, traceId: 'trace-1', focusNodeId: null })
expect(mode.scores.map((row) => row.step)).toEqual([{ id: 'gen-1', name: 'gen-1' }, null])
- expect(mode.responses.map((item) => item.step)).toEqual([null])
+ expect(mode.responses.map((item) => item.step)).toEqual([null, null])
+ expect(mode.responses[1]).toEqual({ ...survey, step: null, focused: false })
+ expect(mode.unanswered).toEqual(unanswered)
+ expect(mode.surveysFailed).toBe(true)
})
test.each([
diff --git a/products/ai_observability/frontend/redesign/trace/adapter/toFeedback.ts b/products/ai_observability/frontend/redesign/trace/adapter/toFeedback.ts
index cc7c8ef2e426..b44d43a5e051 100644
--- a/products/ai_observability/frontend/redesign/trace/adapter/toFeedback.ts
+++ b/products/ai_observability/frontend/redesign/trace/adapter/toFeedback.ts
@@ -3,19 +3,21 @@ import { FeedbackChipData, FeedbackModeState, FeedbackPlacement, StepFeedback }
import { nodeLabels } from './toTraceTree'
export function toFeedbackChip(summary: TraceFeedbackSummaryApi): FeedbackChipData | null {
- return summary.total > 0 ? { total: summary.total } : null
+ return summary.total > 0
+ ? { total: summary.total, thumbsUp: summary.thumbsUp, thumbsDown: summary.thumbsDown }
+ : null
}
/** Trace-level entries sit on the root, whose id is the trace id. */
export function toStepFeedback(feedback: TraceFeedbackApi): Record<string, StepFeedback> {
// The root's key is the trace id, which is free-form, so a prototype key such as `constructor` must not resolve.
const byStep: Record<string, StepFeedback> = Object.create(null)
- const entry = (nodeId: string): StepFeedback => (byStep[nodeId] ??= { scores: [], textCount: 0 })
+ const entry = (nodeId: string): StepFeedback => (byStep[nodeId] ??= { scores: [], responseCount: 0 })
for (const score of feedback.scores) {
entry(score.nodeId).scores.push(score)
}
for (const response of feedback.responses) {
- entry(response.nodeId).textCount += 1
+ entry(response.nodeId).responseCount += 1
}
return byStep
}
@@ -50,5 +52,7 @@ export function toFeedbackMode({
status: 'ready',
scores: feedback.scores.map((score) => ({ ...score, ...placement(score.nodeId) })),
responses: feedback.responses.map((response) => ({ ...response, ...placement(response.nodeId) })),
+ unanswered: feedback.unanswered,
+ surveysFailed: feedback.surveyResponsesStatus === 'failed',
}
}
diff --git a/products/ai_observability/frontend/redesign/trace/adapter/traceSceneWiring.test.tsx b/products/ai_observability/frontend/redesign/trace/adapter/traceSceneWiring.test.tsx
index 168f788e85f6..0e0b5d73d00d 100644
--- a/products/ai_observability/frontend/redesign/trace/adapter/traceSceneWiring.test.tsx
+++ b/products/ai_observability/frontend/redesign/trace/adapter/traceSceneWiring.test.tsx
@@ -21,6 +21,7 @@ import {
makeFeedbackScore,
makeFeedbackText,
makeNodeDetail,
+ makeSurveyResponse,
makeTrace,
makeTraceCluster,
makeTraceResource,
@@ -304,6 +305,27 @@ describe('TraceScene', () => {
fireEvent.click(await findByText('helpfulness 0'))
expect(await findByText('What users said')).toBeTruthy()
})
+
+ it('shows thumbs in the header chip and the survey answers in the feedback', async () => {
+ useMocks({
+ get: {
+ [FEEDBACK_URL]: () => [
+ 200,
+ makeFeedback({
+ scores: [],
+ responses: [makeSurveyResponse()],
+ summary: { total: 1, thumbsUp: 1, thumbsDown: 0 },
+ }),
+ ],
+ },
+ })
+ const { findByLabelText, findByText } = renderNewView()
+
+ fireEvent.click(await findByLabelText('1 thumbs up, 0 thumbs down, 1 feedback in total'))
+
+ expect(await findByText('Answer quality')).toBeTruthy()
+ expect(await findByText('Thumbs up')).toBeTruthy()
+ })
})
describe('behind the flag gate', () => {
diff --git a/products/ai_observability/frontend/redesign/trace/adapter/traceViewAdapterLogic.test.ts b/products/ai_observability/frontend/redesign/trace/adapter/traceViewAdapterLogic.test.ts
index 8b54310881c1..183152a1b379 100644
--- a/products/ai_observability/frontend/redesign/trace/adapter/traceViewAdapterLogic.test.ts
+++ b/products/ai_observability/frontend/redesign/trace/adapter/traceViewAdapterLogic.test.ts
@@ -344,13 +344,14 @@ describe('traceViewAdapterLogic', () => {
await expectLogic(logic, () => logic.actions.showFeedback('gen-1', 'step')).toMatchValues({
mode: 'feedback',
feedbackFocusNodeId: 'gen-1',
- feedbackChip: { total: 2 },
+ feedbackChip: { total: 2, thumbsUp: 0, thumbsDown: 0 },
feedbackCountByNodeId: { 'gen-1': 1, 'trace-1': 1 },
})
expect(capture).toHaveBeenCalledWith('llma trace view feedback viewed', {
entry: 'step',
scores: 1,
responses: 1,
+ unanswered: 0,
})
await expectLogic(logic, () => logic.actions.setMode('feedback')).toMatchValues({
diff --git a/products/ai_observability/frontend/redesign/trace/adapter/traceViewAdapterLogic.ts b/products/ai_observability/frontend/redesign/trace/adapter/traceViewAdapterLogic.ts
index 0a4b8c480a65..b02dc70412bd 100644
--- a/products/ai_observability/frontend/redesign/trace/adapter/traceViewAdapterLogic.ts
+++ b/products/ai_observability/frontend/redesign/trace/adapter/traceViewAdapterLogic.ts
@@ -602,7 +602,10 @@ export const traceViewAdapterLogic = kea<traceViewAdapterLogicType>([
(s) => [s.stepFeedback],
(stepFeedback: Record<string, StepFeedback>): Record<string, number> =>
Object.fromEntries(
- Object.entries(stepFeedback).map(([nodeId, step]) => [nodeId, step.scores.length + step.textCount])
+ Object.entries(stepFeedback).map(([nodeId, step]) => [
+ nodeId,
+ step.scores.length + step.responseCount,
+ ])
),
],
selectedStepFeedback: [
@@ -680,6 +683,7 @@ export const traceViewAdapterLogic = kea<traceViewAdapterLogicType>([
entry,
scores: values.traceFeedback?.scores.length ?? null,
responses: values.traceFeedback?.responses.length ?? null,
+ unanswered: values.traceFeedback?.unanswered.length ?? null,
})
}
const syncSummary = (previousMode: TraceMode): void => {
diff --git a/products/ai_observability/frontend/redesign/trace/components/FeedbackChip.tsx b/products/ai_observability/frontend/redesign/trace/components/FeedbackChip.tsx
index 906f5bc5e44a..5854709f8f94 100644
--- a/products/ai_observability/frontend/redesign/trace/components/FeedbackChip.tsx
+++ b/products/ai_observability/frontend/redesign/trace/components/FeedbackChip.tsx
@@ -1,4 +1,4 @@
-import { IconMessage } from '@posthog/icons'
+import { IconMessage, IconThumbsDown, IconThumbsUp } from '@posthog/icons'
import { LemonButton } from '@posthog/lemon-ui'
import { FeedbackChipData } from '../types'
@@ -9,16 +9,33 @@ export interface FeedbackChipProps {
}
export function FeedbackChip({ chip, onClick }: FeedbackChipProps): JSX.Element {
+ const hasThumbs = chip.thumbsUp + chip.thumbsDown > 0
return (
<LemonButton
size="xsmall"
type="secondary"
- icon={<IconMessage />}
+ icon={hasThumbs ? undefined : <IconMessage />}
onClick={onClick}
tooltip="Show the feedback on this trace"
+ aria-label={
+ hasThumbs
+ ? `${chip.thumbsUp} thumbs up, ${chip.thumbsDown} thumbs down, ${chip.total} feedback in total`
+ : undefined
+ }
data-attr="trace-view-feedback-chip"
>
- {`Feedback ${chip.total}`}
+ {hasThumbs ? (
+ <span className="flex items-center gap-1">
+ <IconThumbsUp />
+ <span>{chip.thumbsUp}</span>
+ <IconThumbsDown />
+ <span>{chip.thumbsDown}</span>
+ <span className="text-secondary">·</span>
+ <span>{`Feedback ${chip.total}`}</span>
+ </span>
+ ) : (
+ `Feedback ${chip.total}`
+ )}
</LemonButton>
)
}
diff --git a/products/ai_observability/frontend/redesign/trace/components/FeedbackEmptyState.tsx b/products/ai_observability/frontend/redesign/trace/components/FeedbackEmptyState.tsx
index 7b294afdd649..aaa7180c5774 100644
--- a/products/ai_observability/frontend/redesign/trace/components/FeedbackEmptyState.tsx
+++ b/products/ai_observability/frontend/redesign/trace/components/FeedbackEmptyState.tsx
@@ -1,16 +1,24 @@
import { Link } from '@posthog/lemon-ui'
+import { urls } from 'scenes/urls'
+
const FEEDBACK_DOCS_URL = 'https://posthog.com/docs/ai-observability/collect-user-feedback'
export function FeedbackEmptyState(): JSX.Element {
return (
<div className="flex flex-col gap-1">
<p className="m-0 text-secondary">
- No feedback on this trace yet. Feedback shows here when your code sends scores or comments for it.
+ No feedback on this trace yet. Feedback shows here when your code sends scores or comments for it, or
+ when users answer a survey that sends its trace id.
</p>
- <Link to={FEEDBACK_DOCS_URL} target="_blank" data-attr="trace-view-feedback-docs-link">
- Send feedback from your code
- </Link>
+ <div className="flex flex-wrap gap-x-4 gap-y-1">
+ <Link to={urls.surveyWizard()} data-attr="trace-view-feedback-surveys-link">
+ Create a survey
+ </Link>
+ <Link to={FEEDBACK_DOCS_URL} target="_blank" data-attr="trace-view-feedback-docs-link">
+ Send feedback from your code
+ </Link>
+ </div>
</div>
)
}
diff --git a/products/ai_observability/frontend/redesign/trace/components/FeedbackSurveyCard.tsx b/products/ai_observability/frontend/redesign/trace/components/FeedbackSurveyCard.tsx
new file mode 100644
index 000000000000..bb107edc72b2
--- /dev/null
+++ b/products/ai_observability/frontend/redesign/trace/components/FeedbackSurveyCard.tsx
@@ -0,0 +1,47 @@
+import { LemonTag } from '@posthog/lemon-ui'
+
+import { cn } from 'lib/utils/css-classes'
+
+import type { TraceSurveyResponseApi, TraceSurveyResponseApiUnavailable } from '../../../generated/api.schemas'
+import { FeedbackPlacement } from '../types'
+import { FeedbackSurveyName } from './FeedbackSurveyName'
+import { scrollIntoNearest } from './scrollIntoNearest'
+import { ShortTimestamp } from './ShortTimestamp'
+import { SurveyAnswerView } from './SurveyAnswerView'
+
+const UNAVAILABLE_MESSAGES: Record<NonNullable<TraceSurveyResponseApiUnavailable>, string> = {
+ survey_not_found: "This survey was deleted, so its answers can't be shown.",
+ no_questions: 'This survey has no questions.',
+}
+
+export interface FeedbackSurveyCardProps {
+ response: TraceSurveyResponseApi & FeedbackPlacement
+ /** Brings the card into view when it mounts. */
+ isScrollTarget: boolean
+}
+
+export function FeedbackSurveyCard({ response, isScrollTarget }: FeedbackSurveyCardProps): JSX.Element {
+ return (
+ <li
+ ref={isScrollTarget ? scrollIntoNearest : undefined}
+ className={cn('flex flex-col gap-2 px-3 py-2', response.focused && 'bg-accent-highlight-secondary')}
+ >
+ <div className="flex flex-wrap items-center gap-x-2 gap-y-1 text-xs">
+ <FeedbackSurveyName survey={response.survey} />
+ {response.complete ? null : (
+ <LemonTag type="muted" size="small">
+ Partial response
+ </LemonTag>
+ )}
+ <span className="ml-auto text-secondary">
+ <ShortTimestamp time={response.timestamp} />
+ </span>
+ </div>
+ {response.unavailable ? (
+ <p className="m-0 text-sm text-secondary">{UNAVAILABLE_MESSAGES[response.unavailable]}</p>
+ ) : (
+ response.answers.map((answer, index) => <SurveyAnswerView key={index} answer={answer} />)
+ )}
+ </li>
+ )
+}
diff --git a/products/ai_observability/frontend/redesign/trace/components/FeedbackSurveyName.tsx b/products/ai_observability/frontend/redesign/trace/components/FeedbackSurveyName.tsx
new file mode 100644
index 000000000000..c7ac11132770
--- /dev/null
+++ b/products/ai_observability/frontend/redesign/trace/components/FeedbackSurveyName.tsx
@@ -0,0 +1,20 @@
+import { Link } from '@posthog/lemon-ui'
+
+import { urls } from 'scenes/urls'
+
+import type { TraceSurveyApi } from '../../../generated/api.schemas'
+
+export interface FeedbackSurveyNameProps {
+ survey: TraceSurveyApi
+}
+
+export function FeedbackSurveyName({ survey }: FeedbackSurveyNameProps): JSX.Element {
+ if (survey.name === null) {
+ return <span className="font-semibold text-secondary">Deleted survey</span>
+ }
+ return (
+ <Link to={urls.survey(survey.id)} className="font-semibold" data-attr="trace-view-feedback-survey-link">
+ {survey.name}
+ </Link>
+ )
+}
diff --git a/products/ai_observability/frontend/redesign/trace/components/FeedbackTextCard.tsx b/products/ai_observability/frontend/redesign/trace/components/FeedbackTextCard.tsx
index 1c0ef06c0daf..77e2753053bf 100644
--- a/products/ai_observability/frontend/redesign/trace/components/FeedbackTextCard.tsx
+++ b/products/ai_observability/frontend/redesign/trace/components/FeedbackTextCard.tsx
@@ -1,12 +1,13 @@
import { cn } from 'lib/utils/css-classes'
-import { FeedbackResponseItem } from '../types'
+import type { TraceFeedbackTextApi } from '../../../generated/api.schemas'
+import { FeedbackPlacement } from '../types'
import { FeedbackStepLink } from './FeedbackStepLink'
import { scrollIntoNearest } from './scrollIntoNearest'
import { ShortTimestamp } from './ShortTimestamp'
export interface FeedbackTextCardProps {
- item: FeedbackResponseItem
+ item: TraceFeedbackTextApi & FeedbackPlacement
onSelectStep: (nodeId: string) => void
/** Brings the card into view when it mounts. */
isScrollTarget: boolean
diff --git a/products/ai_observability/frontend/redesign/trace/components/StepFeedbackChips.tsx b/products/ai_observability/frontend/redesign/trace/components/StepFeedbackChips.tsx
index 728d2662dbe7..08e3c9c6da01 100644
--- a/products/ai_observability/frontend/redesign/trace/components/StepFeedbackChips.tsx
+++ b/products/ai_observability/frontend/redesign/trace/components/StepFeedbackChips.tsx
@@ -38,7 +38,7 @@ export function StepFeedbackChips({ feedback, onOpen }: StepFeedbackChipsProps):
{`+${hidden}`}
</LemonButton>
) : null}
- {feedback.textCount > 0 ? (
+ {feedback.responseCount > 0 ? (
<LemonButton
size="xsmall"
type="secondary"
@@ -47,7 +47,7 @@ export function StepFeedbackChips({ feedback, onOpen }: StepFeedbackChipsProps):
tooltip="Show what users said"
data-attr="trace-view-feedback-step-chip"
>
- {`Feedback ${feedback.textCount}`}
+ {`Feedback ${feedback.responseCount}`}
</LemonButton>
) : null}
</>
diff --git a/products/ai_observability/frontend/redesign/trace/components/SurveyAnswerValue.tsx b/products/ai_observability/frontend/redesign/trace/components/SurveyAnswerValue.tsx
new file mode 100644
index 000000000000..f8ef14b26103
--- /dev/null
+++ b/products/ai_observability/frontend/redesign/trace/components/SurveyAnswerValue.tsx
@@ -0,0 +1,43 @@
+import { IconThumbsDown, IconThumbsUp } from '@posthog/icons'
+import { LemonTag } from '@posthog/lemon-ui'
+
+import { SurveyAnswer } from '../types'
+
+export interface SurveyAnswerValueProps {
+ answer: SurveyAnswer
+}
+
+export function SurveyAnswerValue({ answer }: SurveyAnswerValueProps): JSX.Element {
+ switch (answer.kind) {
+ case 'thumbs':
+ return (
+ <span className="flex items-center gap-1 text-sm">
+ {answer.up ? <IconThumbsUp /> : <IconThumbsDown />}
+ <span>{answer.up ? 'Thumbs up' : 'Thumbs down'}</span>
+ </span>
+ )
+ case 'rating':
+ return (
+ <span className="flex flex-wrap items-baseline gap-x-2 text-sm">
+ <span className="font-semibold">
+ {answer.scale ? `${answer.value} / ${answer.scale}` : answer.value}
+ </span>
+ {answer.lowerLabel && answer.upperLabel ? (
+ <span className="text-xs text-secondary">{`${answer.lowerLabel} to ${answer.upperLabel}`}</span>
+ ) : null}
+ </span>
+ )
+ case 'choice':
+ return (
+ <span className="flex flex-wrap gap-1">
+ {answer.choices.map((choice) => (
+ <LemonTag key={choice} weight="normal">
+ {choice}
+ </LemonTag>
+ ))}
+ </span>
+ )
+ case 'text':
+ return <p className="m-0 whitespace-pre-wrap break-words text-sm">{answer.text}</p>
+ }
+}
diff --git a/products/ai_observability/frontend/redesign/trace/components/SurveyAnswerView.tsx b/products/ai_observability/frontend/redesign/trace/components/SurveyAnswerView.tsx
new file mode 100644
index 000000000000..43d3d74711d3
--- /dev/null
+++ b/products/ai_observability/frontend/redesign/trace/components/SurveyAnswerView.tsx
@@ -0,0 +1,15 @@
+import { SurveyAnswer } from '../types'
+import { SurveyAnswerValue } from './SurveyAnswerValue'
+
+export interface SurveyAnswerViewProps {
+ answer: SurveyAnswer
+}
+
+export function SurveyAnswerView({ answer }: SurveyAnswerViewProps): JSX.Element {
+ return (
+ <div className="flex flex-col gap-0.5">
+ <span className="text-xs text-secondary">{answer.question}</span>
+ <SurveyAnswerValue answer={answer} />
+ </div>
+ )
+}
diff --git a/products/ai_observability/frontend/redesign/trace/components/TraceFeedbackView.stories.tsx b/products/ai_observability/frontend/redesign/trace/components/TraceFeedbackView.stories.tsx
index 50d98f16b178..9f2cee1a6cea 100644
--- a/products/ai_observability/frontend/redesign/trace/components/TraceFeedbackView.stories.tsx
+++ b/products/ai_observability/frontend/redesign/trace/components/TraceFeedbackView.stories.tsx
@@ -1,6 +1,6 @@
import type { Meta, StoryObj } from '@storybook/react'
-import { FIXTURE_FEEDBACK_MODE, withWidth } from '../storyFixtures'
+import { FIXTURE_FEEDBACK_MODE, FIXTURE_FEEDBACK_MODE_WITH_SURVEYS, withWidth } from '../storyFixtures'
import { TraceFeedbackView } from './TraceFeedbackView'
const unfocused = {
@@ -22,8 +22,12 @@ type Story = StoryObj<typeof TraceFeedbackView>
export const ScoresAndText: Story = { args: { state: unfocused } }
export const FocusedOnAStep: Story = { args: { state: FIXTURE_FEEDBACK_MODE } }
export const ScoresOnly: Story = { args: { state: { ...unfocused, responses: [] } } }
-export const Narrow: Story = { args: { state: FIXTURE_FEEDBACK_MODE }, decorators: [withWidth(520)] }
-export const Empty: Story = { args: { state: { status: 'ready', scores: [], responses: [] } } }
+export const WithSurveys: Story = { args: { state: FIXTURE_FEEDBACK_MODE_WITH_SURVEYS } }
+export const Narrow: Story = { args: { state: FIXTURE_FEEDBACK_MODE_WITH_SURVEYS }, decorators: [withWidth(520)] }
+export const Empty: Story = {
+ args: { state: { status: 'ready', scores: [], responses: [], unanswered: [], surveysFailed: false } },
+}
+export const SurveyReadFailed: Story = { args: { state: { ...unfocused, responses: [], surveysFailed: true } } }
export const Loading: Story = {
args: { state: { status: 'loading' } },
parameters: { testOptions: { waitForLoadersToDisappear: false } },
diff --git a/products/ai_observability/frontend/redesign/trace/components/TraceFeedbackView.tsx b/products/ai_observability/frontend/redesign/trace/components/TraceFeedbackView.tsx
index 8cda0168b233..1e5355fb85aa 100644
--- a/products/ai_observability/frontend/redesign/trace/components/TraceFeedbackView.tsx
+++ b/products/ai_observability/frontend/redesign/trace/components/TraceFeedbackView.tsx
@@ -1,10 +1,12 @@
-import { LemonSkeleton } from '@posthog/lemon-ui'
+import { LemonBanner, LemonSkeleton } from '@posthog/lemon-ui'
import { FeedbackModeState } from '../types'
import { ErrorCallout } from './ErrorCallout'
import { FeedbackEmptyState } from './FeedbackEmptyState'
import { FeedbackScoresTable } from './FeedbackScoresTable'
+import { FeedbackSurveyCard } from './FeedbackSurveyCard'
import { FeedbackTextCard } from './FeedbackTextCard'
+import { UnansweredSurveyRow } from './UnansweredSurveyRow'
export interface TraceFeedbackViewProps {
state: FeedbackModeState
@@ -27,11 +29,22 @@ export function TraceFeedbackView({ state, onSelectStep, onRetry }: TraceFeedbac
if (state.status === 'error') {
return <ErrorCallout message="Feedback couldn't load." onRetry={onRetry} />
}
- if (state.scores.length === 0 && state.responses.length === 0) {
- return <FeedbackEmptyState />
+ const surveysFailedBanner = state.surveysFailed ? (
+ <LemonBanner type="warning" action={onRetry ? { children: 'Try again', onClick: onRetry } : undefined}>
+ Survey responses couldn't load, so only the feedback your code sent shows here.
+ </LemonBanner>
+ ) : null
+ if (state.scores.length === 0 && state.responses.length === 0 && state.unanswered.length === 0) {
+ return (
+ <div className="flex flex-col gap-4">
+ {surveysFailedBanner}
+ <FeedbackEmptyState />
+ </div>
+ )
}
return (
<div className="flex flex-col gap-4" data-attr="trace-view-feedback">
+ {surveysFailedBanner}
{state.scores.length > 0 ? (
<section className="flex flex-col gap-2">
<h3 className="m-0 text-sm font-semibold">Scores</h3>
@@ -46,13 +59,31 @@ export function TraceFeedbackView({ state, onSelectStep, onRetry }: TraceFeedbac
<section className="flex flex-col gap-2">
<h3 className="m-0 text-sm font-semibold">What users said</h3>
<ul className="m-0 list-none divide-y divide-primary rounded border border-primary bg-surface-primary p-0">
- {state.responses.map((item) => (
- <FeedbackTextCard
- key={item.id}
- item={item}
- onSelectStep={onSelectStep}
- isScrollTarget={item.id === scrollTargetId}
- />
+ {state.responses.map((item) =>
+ item.kind === 'text' ? (
+ <FeedbackTextCard
+ key={`text-${item.id}`}
+ item={item}
+ onSelectStep={onSelectStep}
+ isScrollTarget={item.id === scrollTargetId}
+ />
+ ) : (
+ <FeedbackSurveyCard
+ key={`survey-${item.id}`}
+ response={item}
+ isScrollTarget={item.id === scrollTargetId}
+ />
+ )
+ )}
+ </ul>
+ </section>
+ ) : null}
+ {state.unanswered.length > 0 ? (
+ <section className="flex flex-col gap-2">
+ <h3 className="m-0 text-sm font-semibold">Shown, not answered</h3>
+ <ul className="m-0 list-none divide-y divide-primary rounded border border-primary p-0">
+ {state.unanswered.map((item) => (
+ <UnansweredSurveyRow key={item.survey.id} item={item} />
))}
</ul>
</section>
diff --git a/products/ai_observability/frontend/redesign/trace/components/TraceSummaryBar.stories.tsx b/products/ai_observability/frontend/redesign/trace/components/TraceSummaryBar.stories.tsx
index a4c1f409af74..a102b0dc46b3 100644
--- a/products/ai_observability/frontend/redesign/trace/components/TraceSummaryBar.stories.tsx
+++ b/products/ai_observability/frontend/redesign/trace/components/TraceSummaryBar.stories.tsx
@@ -28,7 +28,11 @@ export const LongSlugId: Story = { ...wide, args: { traceId: 'support-chat-sessi
export const WithClusters: Story = { ...wide, args: { clusters: FIXTURE_CLUSTERS_CHIP } }
export const Narrow: Story = { args: { clusters: FIXTURE_CLUSTERS_CHIP }, decorators: [withWidth(420)] }
export const WithFeedback: Story = {
- args: { feedback: { chip: { total: 6 }, onClick: () => {} } },
+ args: { feedback: { chip: { total: 6, thumbsUp: 0, thumbsDown: 0 }, onClick: () => {} } },
+ decorators: [withWidth(420)],
+}
+export const WithThumbsFeedback: Story = {
+ args: { feedback: { chip: { total: 6, thumbsUp: 3, thumbsDown: 1 }, onClick: () => {} } },
decorators: [withWidth(420)],
}
diff --git a/products/ai_observability/frontend/redesign/trace/components/TraceView.stories.tsx b/products/ai_observability/frontend/redesign/trace/components/TraceView.stories.tsx
index c9c54500b26d..5c22249acb0c 100644
--- a/products/ai_observability/frontend/redesign/trace/components/TraceView.stories.tsx
+++ b/products/ai_observability/frontend/redesign/trace/components/TraceView.stories.tsx
@@ -12,7 +12,7 @@ import { SampleTraceFixture } from '../sampleFixtures/sampleTraceFixture'
import {
FIXTURE_EVAL_OUTCOMES,
FIXTURE_EVALS_CHIP,
- FIXTURE_FEEDBACK_MODE,
+ FIXTURE_FEEDBACK_MODE_WITH_SURVEYS,
FIXTURE_SUMMARY,
FIXTURE_TRACE,
withWidth,
@@ -135,9 +135,11 @@ function Playground({
onReload: () => {},
onSelectStep: selectAndShowSpans,
}}
- feedbackMode={FIXTURE_FEEDBACK_MODE}
- feedbackCount={FIXTURE_FEEDBACK_MODE.scores.length + FIXTURE_FEEDBACK_MODE.responses.length}
- feedbackCountByNodeId={{ 'trace-1': 3, 'gen-answer': 3 }}
+ feedbackMode={FIXTURE_FEEDBACK_MODE_WITH_SURVEYS}
+ feedbackCount={
+ FIXTURE_FEEDBACK_MODE_WITH_SURVEYS.scores.length + FIXTURE_FEEDBACK_MODE_WITH_SURVEYS.responses.length
+ }
+ feedbackCountByNodeId={{ 'trace-1': 7, 'gen-answer': 2 }}
/>
)
}
diff --git a/products/ai_observability/frontend/redesign/trace/components/UnansweredSurveyRow.tsx b/products/ai_observability/frontend/redesign/trace/components/UnansweredSurveyRow.tsx
new file mode 100644
index 000000000000..5132a46a3b48
--- /dev/null
+++ b/products/ai_observability/frontend/redesign/trace/components/UnansweredSurveyRow.tsx
@@ -0,0 +1,19 @@
+import type { TraceUnansweredSurveyApi } from '../../../generated/api.schemas'
+import { FeedbackSurveyName } from './FeedbackSurveyName'
+import { ShortTimestamp } from './ShortTimestamp'
+
+export interface UnansweredSurveyRowProps {
+ item: TraceUnansweredSurveyApi
+}
+
+export function UnansweredSurveyRow({ item }: UnansweredSurveyRowProps): JSX.Element {
+ return (
+ <li className="flex flex-wrap items-center gap-x-2 gap-y-1 px-3 py-1.5 text-xs text-secondary">
+ <FeedbackSurveyName survey={item.survey} />
+ <span className="ml-auto flex items-center gap-1">
+ <span>Shown</span>
+ <ShortTimestamp time={item.timestamp} />
+ </span>
+ </li>
+ )
+}
diff --git a/products/ai_observability/frontend/redesign/trace/storyFixtures.tsx b/products/ai_observability/frontend/redesign/trace/storyFixtures.tsx
index 147d6c39149b..a53e88c8c20a 100644
--- a/products/ai_observability/frontend/redesign/trace/storyFixtures.tsx
+++ b/products/ai_observability/frontend/redesign/trace/storyFixtures.tsx
@@ -425,7 +425,7 @@ export const FIXTURE_STEP_FEEDBACK: StepFeedback = {
feedbackScore('score-latency', 'latency_ok', 'true', 19),
feedbackScore('score-unnamed', null, '{"stars":3}', 20),
],
- textCount: 1,
+ responseCount: 1,
}
const answerStep = { id: 'gen-answer', name: 'draft-answer' }
@@ -458,6 +458,86 @@ export const FIXTURE_FEEDBACK_MODE: Extract<FeedbackModeState, { status: 'ready'
focused: true,
},
],
+ unanswered: [],
+ surveysFailed: false,
+}
+
+const answerQuality = { id: '0192f2a4-5c1e-7000-8000-00000000a001', name: 'Answer quality' }
+
+export const FIXTURE_FEEDBACK_MODE_WITH_SURVEYS: Extract<FeedbackModeState, { status: 'ready' }> = {
+ status: 'ready',
+ scores: FIXTURE_FEEDBACK_MODE.scores.map((row) => ({ ...row, focused: false })),
+ responses: [
+ { ...FIXTURE_FEEDBACK_MODE.responses[0], focused: false },
+ {
+ kind: 'survey',
+ id: 'submission-full',
+ nodeId: 'trace-1',
+ timestamp: '2026-09-01T10:25:00Z',
+ survey: answerQuality,
+ answers: [
+ { kind: 'thumbs', question: 'Was this answer useful?', up: true },
+ {
+ kind: 'rating',
+ question: 'How good was it?',
+ value: '4',
+ scale: 5,
+ lowerLabel: 'Bad',
+ upperLabel: 'Great',
+ },
+ { kind: 'choice', question: 'What could be better?', choices: ['Accuracy', 'Tone'] },
+ { kind: 'text', question: 'Anything else?', text: 'Shorter answers, please.' },
+ ],
+ complete: true,
+ unavailable: null,
+ step: null,
+ focused: false,
+ },
+ {
+ kind: 'survey',
+ id: 'submission-partial',
+ nodeId: 'trace-1',
+ timestamp: '2026-09-01T10:26:00Z',
+ survey: answerQuality,
+ answers: [{ kind: 'thumbs', question: 'Was this answer useful?', up: false }],
+ complete: false,
+ unavailable: null,
+ step: null,
+ focused: false,
+ },
+ {
+ kind: 'survey',
+ id: 'submission-deleted',
+ nodeId: 'trace-1',
+ timestamp: '2026-09-01T10:27:00Z',
+ survey: { id: '0192f2a4-5c1e-7000-8000-00000000a002', name: null },
+ answers: [],
+ complete: true,
+ unavailable: 'survey_not_found',
+ step: null,
+ focused: false,
+ },
+ {
+ kind: 'survey',
+ id: 'submission-empty',
+ nodeId: 'trace-1',
+ timestamp: '2026-09-01T10:28:00Z',
+ survey: { id: '0192f2a4-5c1e-7000-8000-00000000a003', name: 'Draft survey' },
+ answers: [],
+ complete: true,
+ unavailable: 'no_questions',
+ step: null,
+ focused: false,
+ },
+ ],
+ unanswered: [
+ {
+ survey: { id: '0192f2a4-5c1e-7000-8000-00000000a004', name: 'Follow-up' },
+ timestamp: '2026-09-01T10:29:00Z',
+ },
+ { survey: { id: '0192f2a4-5c1e-7000-8000-00000000a005', name: null }, timestamp: '2026-09-01T10:30:00Z' },
+ ],
+ surveysFailed: false,
}
export const FIXTURE_SUMMARY: TraceSummaryContentApi = {
diff --git a/products/ai_observability/frontend/redesign/trace/types.ts b/products/ai_observability/frontend/redesign/trace/types.ts
index f297d2b2e635..9eacda449170 100644
--- a/products/ai_observability/frontend/redesign/trace/types.ts
+++ b/products/ai_observability/frontend/redesign/trace/types.ts
@@ -6,6 +6,8 @@ import type {
TraceNodeDetailApi,
TraceNodeEvaluationRollupOutcomeEnumApi,
TraceSummaryContentApi,
+ TraceSurveyResponseApi,
+ TraceUnansweredSurveyApi,
} from '../../generated/api.schemas'
export type TraceNodeKind = TraceNodeApi['kind']
@@ -166,18 +168,29 @@ export interface FeedbackPlacement {
export type FeedbackScoreRow = TraceFeedbackScoreApi & FeedbackPlacement
-export type FeedbackResponseItem = TraceFeedbackTextApi & FeedbackPlacement
+/** Survey responses link to the whole trace, so their `step` is always null. */
+export type FeedbackResponseItem = (TraceFeedbackTextApi | TraceSurveyResponseApi) & FeedbackPlacement
+
+export type SurveyAnswer = TraceSurveyResponseApi['answers'][number]
export type FeedbackModeState =
| { status: 'loading' }
| { status: 'error' }
- | { status: 'ready'; scores: FeedbackScoreRow[]; responses: FeedbackResponseItem[] }
+ | {
+ status: 'ready'
+ scores: FeedbackScoreRow[]
+ responses: FeedbackResponseItem[]
+ unanswered: TraceUnansweredSurveyApi[]
+ surveysFailed: boolean
+ }
export interface FeedbackChipData {
total: number
+ thumbsUp: number
+ thumbsDown: number
}
export interface StepFeedback {
scores: TraceFeedbackScoreApi[]
- textCount: number
+ responseCount: number
}
diff --git a/products/surveys/backend/facade/api.py b/products/surveys/backend/facade/api.py
index 3760051e18ab..1e713871e868 100644
--- a/products/surveys/backend/facade/api.py
+++ b/products/surveys/backend/facade/api.py
@@ -3,5 +3,27 @@ from products.surveys.backend.desktop_feedback import (
read_desktop_feedback_media,
submit_desktop_feedback,
)
+from products.surveys.backend.responses.linked import (
+ LinkedResponseUnavailable,
+ LinkedSurveyAnswer,
+ LinkedSurveyResponse,
+ LinkedSurveyResponses,
+ RatingScale,
+ SurveyRef,
+ UnansweredSurvey,
+ get_survey_responses_linked_to,
+)
-__all__ = ["DesktopFeedbackUnavailable", "read_desktop_feedback_media", "submit_desktop_feedback"]
+__all__ = [
+ "DesktopFeedbackUnavailable",
+ "LinkedResponseUnavailable",
+ "LinkedSurveyAnswer",
+ "LinkedSurveyResponse",
+ "LinkedSurveyResponses",
+ "RatingScale",
+ "SurveyRef",
+ "UnansweredSurvey",
+ "get_survey_responses_linked_to",
+ "read_desktop_feedback_media",
+ "submit_desktop_feedback",
+]
diff --git a/products/surveys/backend/responses/linked.py b/products/surveys/backend/responses/linked.py
new file mode 100644
index 000000000000..d23ad1b99135
--- /dev/null
+++ b/products/surveys/backend/responses/linked.py
@@ -0,0 +1,356 @@
+"""Survey responses linked to another product's record through an event property, such as `$ai_trace_id`.
+
+Responses merge per submission like the responses API in `fetch_rows`: the same response events, grouped by
+`SUBMISSION_GROUPING_KEY`, each question keeping its latest non-empty answer, archived responses left out. This read
+also returns every picked choice of a multiple choice question, maps translated choices back to the base language,
+matches survey ids however their case is written, and says whether each submission was completed.
+"""
+
+import json
+from collections.abc import Sequence
+from datetime import datetime
+from typing import Any, Literal, cast
+from uuid import UUID
+
+from django.db.models import QuerySet
+
+from posthog.hogql import ast
+from posthog.hogql.parser import parse_expr, parse_select
+from posthog.hogql.query import execute_hogql_query
+
+from posthog.clickhouse.client.connection import Workload
+from posthog.clickhouse.query_tagging import Product, tag_queries, tags_context
+from posthog.dataclasses import frozen
+from posthog.models import Team, User
+
+from products.access_control.backend.facade.user_access_control import UserAccessControl
+from products.surveys.backend.models import Survey
+from products.surveys.backend.responses.fetch_rows import (
+ RESPONSE_EVENT_FILTER,
+ SUBMISSION_GROUPING_KEY,
+ resolve_question_metadata,
+)
+from products.surveys.backend.util import get_archived_response_uuids
+
+LINKED_SURVEYS_LIMIT = 10
+# Per survey, newest first.
+LINKED_SUBMISSIONS_LIMIT = 100
+_CHOICE_TYPES = ("single_choice", "multiple_choice")
+
+
+@frozen
+class SurveyRef:
+ id: str
+ # None when no survey of the team has this id.
+ name: str | None
+
+
+@frozen
+class RatingScale:
+ scale: int | None
+ display: str | None
+ lower_bound_label: str | None
+ upper_bound_label: str | None
+
+
+@frozen
+class LinkedSurveyAnswer:
+ question_id: str
+ question_index: int
+ question_text: str
+ question_type: str
+ # Set for rating questions only.
+ rating: RatingScale | None
+ # Every picked choice for a multiple choice question, the answer as text otherwise.
+ value: str | tuple[str, ...]
+
+
+LinkedResponseUnavailable = Literal["survey_not_found", "no_questions"]
+
+
+@frozen
+class LinkedSurveyResponse:
+ survey: SurveyRef
+ submission_id: str
+ submitted_at: datetime
+ complete: bool
+ answers: tuple[LinkedSurveyAnswer, ...]
+ # Why the answers cannot be read, when they cannot.
+ unavailable: LinkedResponseUnavailable | None
+
+
+@frozen
+class UnansweredSurvey:
+ survey: SurveyRef
+ shown_at: datetime
+
+
+@frozen
+class LinkedSurveyResponses:
+ responses: tuple[LinkedSurveyResponse, ...]
+ unanswered: tuple[UnansweredSurvey, ...]
+
+
+@frozen
+class _SurveyActivity:
+ survey_id: str
+ raw_ids: tuple[str, ...]
+ responses: int
+ first_shown_at: datetime | None
+
+
+@frozen
+class _Link:
+ property_name: str
+ property_value: str
+ since: datetime
+ until: datetime | None
+
+ def condition(self) -> ast.Expr:
+ exprs: list[ast.Expr] = [
+ ast.CompareOperation(
+ op=ast.CompareOperationOp.Eq,
+ left=ast.Field(chain=["properties", self.property_name]),
+ right=ast.Constant(value=self.property_value),
+ ),
+ ast.CompareOperation(
+ op=ast.CompareOperationOp.GtEq,
+ left=ast.Field(chain=["timestamp"]),
+ right=ast.Constant(value=self.since),
+ ),
+ ]
+ if self.until is not None:
+ exprs.append(
+ ast.CompareOperation(
+ op=ast.CompareOperationOp.LtEq,
+ left=ast.Field(chain=["timestamp"]),
+ right=ast.Constant(value=self.until),
+ )
+ )
+ return ast.And(exprs=exprs)
+
+
+def _canonical_uuid(value: str) -> str | None:
+ try:
+ return str(UUID(value))
+ except ValueError:
+ return None
+
+
+def _execute(team: Team, user: User | None, name: str, query: ast.SelectQuery) -> list[Sequence[Any]]:
+ with tags_context(product=Product.SURVEYS):
+ tag_queries(name=name)
+ response = execute_hogql_query(query=query, team=team, user=user, query_type=name, workload=Workload.ONLINE)
+ return list(response.results or [])
+
+
+def _read_activity(team: Team, user: User | None, link: _Link) -> list[_SurveyActivity]:
+ query = cast(
+ ast.SelectQuery,
+ parse_select(
+ f"""
+ SELECT
+ properties.`$survey_id` AS raw_survey_id,
+ countIf({RESPONSE_EVENT_FILTER}) AS responses,
+ countIf(event = 'survey shown') AS shown,
+ minIf(timestamp, event = 'survey shown') AS first_shown_at
+ FROM events
+ WHERE (event = 'survey shown' OR {RESPONSE_EVENT_FILTER}) AND {{link}}
+ GROUP BY raw_survey_id
+ ORDER BY min(timestamp)
+ LIMIT {{limit}}
+ """,
+ {"link": link.condition(), "limit": ast.Constant(value=LINKED_SURVEYS_LIMIT)},
+ ),
+ )
+ # Survey ids are UUIDs that some SDKs write in upper case, so each survey gathers every spelling of its id.
+ grouped: dict[str, _SurveyActivity] = {}
+ for raw_id, responses, shown, first_shown_at in _execute(team, user, "survey_linked_activity", query):
+ if not raw_id:
+ continue
+ survey_id = _canonical_uuid(raw_id) or raw_id
+ seen = grouped.get(survey_id)
+ shown_at = first_shown_at if shown else None
+ if seen is None:
+ grouped[survey_id] = _SurveyActivity(
+ survey_id=survey_id, raw_ids=(raw_id,), responses=responses, first_shown_at=shown_at
+ )
+ else:
+ grouped[survey_id] = _SurveyActivity(
+ survey_id=survey_id,
+ raw_ids=(*seen.raw_ids, raw_id),
+ responses=seen.responses + responses,
+ first_shown_at=min(filter(None, (seen.first_shown_at, shown_at)), default=None),
+ )
+ return list(grouped.values())
+
+
+@frozen
+class _SurveyAccess:
+ readable: dict[str, Survey]
+ # Surveys the team has but the viewer may not read.
+ hidden: set[str]
+
+
+def _survey_access(team: Team, user: User | None, survey_ids: Sequence[str]) -> _SurveyAccess:
+ uuids = [survey_id for survey_id in survey_ids if _canonical_uuid(survey_id)]
+ surveys: QuerySet[Survey] = Survey.objects.filter(team_id=team.pk, id__in=uuids)
+ existing = {str(survey_id) for survey_id in surveys.values_list("id", flat=True)}
+ if user is not None:
+ surveys = UserAccessControl(user=user, team=team).filter_queryset_by_access_level(surveys, resource="survey")
+ readable = {str(survey.id): survey for survey in surveys}
+ return _SurveyAccess(readable=readable, hidden=existing - readable.keys())
+
+
+def _submissions_query(
+ activity: _SurveyActivity, link: _Link, questions: list[dict[str, Any]], archived: set[str]
+) -> ast.SelectQuery:
+ placeholders: dict[str, ast.Expr] = {
+ "link": link.condition(),
+ "survey_ids": ast.Tuple(exprs=[ast.Constant(value=raw_id) for raw_id in activity.raw_ids]),
+ "grouping_key": parse_expr(SUBMISSION_GROUPING_KEY),
+ "limit": ast.Constant(value=LINKED_SUBMISSIONS_LIMIT),
+ }
+ inner_answers: list[str] = []
+ outer_answers: list[str] = []
+ for question in questions:
+ index = question["index"]
+ is_multiple_choice = question["type"] == "multiple_choice"
+ placeholders[f"q_idx_{index}"] = ast.Constant(value=index)
+ placeholders[f"q_id_{index}"] = ast.Constant(value=question["id"])
+ placeholders[f"q_mc_{index}"] = ast.Constant(value=is_multiple_choice)
+ inner_answers.append(f"getSurveyResponse({{q_idx_{index}}}, {{q_id_{index}}}, {{q_mc_{index}}}) AS raw_{index}")
+ present = f"notEmpty(raw_{index})" if is_multiple_choice else f"isNotNull(raw_{index})"
+ outer_answers.append(f"argMaxIf(raw_{index}, timestamp, {present}) AS answer_{index}")
+ having = ""
+ if archived:
+ # Archiving records the submission's latest uuid, which is what argMax(uuid) returns.
+ having = "HAVING uuid NOT IN {archived}"
+ placeholders["archived"] = ast.Tuple(exprs=[ast.Constant(value=uuid) for uuid in archived])
+ return cast(
+ ast.SelectQuery,
+ parse_select(
+ f"""
+ SELECT
+ submission_key,
+ argMax(uuid, timestamp) AS uuid,
+ max(timestamp) AS submitted_at,
+ countIf(event = 'survey sent' AND completed NOT IN ('false', '0')) > 0 AS complete
+ {"".join(f", {column}" for column in outer_answers)}
+ FROM (
+ SELECT
+ uuid,
+ event,
+ timestamp,
+ {{grouping_key}} AS submission_key,
+ coalesce(toString(properties.`$survey_completed`), '') AS completed
+ {"".join(f", {column}" for column in inner_answers)}
+ FROM events
+ WHERE {RESPONSE_EVENT_FILTER} AND properties.`$survey_id` IN {{survey_ids}} AND {{link}}
+ )
+ GROUP BY submission_key
+ {having}
+ ORDER BY submitted_at DESC
+ LIMIT {{limit}}
+ """,
+ placeholders,
+ ),
+ )
+
+
+def _rating_scale(question: dict[str, Any]) -> RatingScale:
+ scale = question.get("scale")
+ return RatingScale(
+ scale=scale if isinstance(scale, int) else None,
+ display=question.get("display"),
+ lower_bound_label=question.get("lowerBoundLabel"),
+ upper_bound_label=question.get("upperBoundLabel"),
+ )
+
+
+def _answer_value(raw: Any, question: dict[str, Any]) -> str | tuple[str, ...] | None:
+ # Answers to a translated survey arrive in the respondent's language, so choices map back to the base language.
+ choice_map: dict[str, str] = question["choice_map"] if question["type"] in _CHOICE_TYPES else {}
+ if question["type"] == "multiple_choice":
+ choices = tuple(str(choice) for choice in (json.loads(item) for item in raw or []) if choice not in (None, ""))
+ return tuple(choice_map.get(choice, choice) for choice in choices) or None
+ if raw in (None, ""):
+ return None
+ return choice_map.get(str(raw), str(raw))
+
+
+def _answers(values: Sequence[Any], questions: list[dict[str, Any]], survey: Survey) -> tuple[LinkedSurveyAnswer, ...]:
+ raw_questions = survey.questions or []
+ answers: list[LinkedSurveyAnswer] = []
+ for question, raw in zip(questions, values, strict=True):
+ value = _answer_value(raw, question)
+ if value is None:
+ continue
+ answers.append(
+ LinkedSurveyAnswer(
+ question_id=question["id"],
+ question_index=question["index"],
+ question_text=question["text"],
+ question_type=question["type"],
+ rating=_rating_scale(raw_questions[question["index"]]) if question["type"] == "rating" else None,
+ value=value,
+ )
+ )
+ return tuple(answers)
+
+
+def _read_responses(
+ team: Team, user: User | None, link: _Link, activity: _SurveyActivity, survey: Survey | None
+) -> list[LinkedSurveyResponse]:
+ questions = resolve_question_metadata(survey) if survey is not None else []
+ unavailable: LinkedResponseUnavailable | None = (
+ "survey_not_found" if survey is None else "no_questions" if not questions else None
+ )
+ archived = get_archived_response_uuids(survey.id, team.pk) if survey is not None else set()
+ ref = SurveyRef(id=activity.survey_id, name=survey.name if survey is not None else None)
+ rows = _execute(team, user, "survey_linked_responses", _submissions_query(activity, link, questions, archived))
+ return [
+ LinkedSurveyResponse(
+ survey=ref,
+ submission_id=str(submission_key),
+ submitted_at=submitted_at,
+ complete=bool(complete),
+ answers=_answers(values, questions, survey) if survey is not None else (),
+ unavailable=unavailable,
+ )
+ for submission_key, _uuid, submitted_at, complete, *values in rows
+ ]
+
+
+def get_survey_responses_linked_to(
+ team: Team,
+ user: User | None,
+ *,
+ property_name: str,
+ property_value: str,
+ since: datetime,
+ until: datetime | None = None,
+) -> LinkedSurveyResponses:
+ """Responses whose events carry `property_name = property_value`, merged per submission as the responses API
+ merges them, with each question's text and type, oldest first. Also the surveys shown with no response.
+
+ A response to a survey the team no longer has is kept, marked unavailable. Surveys the user cannot read are left
+ out. Reads at most `LINKED_SURVEYS_LIMIT` surveys and `LINKED_SUBMISSIONS_LIMIT` submissions per survey."""
+ link = _Link(property_name=property_name, property_value=property_value, since=since, until=until)
+ activities = _read_activity(team, user, link)
+ access = _survey_access(team, user, [activity.survey_id for activity in activities])
+ responses: list[LinkedSurveyResponse] = []
+ unanswered: list[UnansweredSurvey] = []
+ for activity in activities:
+ if activity.survey_id in access.hidden:
+ continue
+ survey = access.readable.get(activity.survey_id)
+ if activity.responses:
+ responses.extend(_read_responses(team, user, link, activity, survey))
+ elif activity.first_shown_at is not None:
+ ref = SurveyRef(id=activity.survey_id, name=survey.name if survey is not None else None)
+ unanswered.append(UnansweredSurvey(survey=ref, shown_at=activity.first_shown_at))
+ return LinkedSurveyResponses(
+ responses=tuple(sorted(responses, key=lambda response: response.submitted_at)),
+ unanswered=tuple(sorted(unanswered, key=lambda survey: survey.shown_at)),
+ )
diff --git a/products/surveys/backend/responses/test_linked.py b/products/surveys/backend/responses/test_linked.py
new file mode 100644
index 000000000000..ee0c7cf9994a
--- /dev/null
+++ b/products/surveys/backend/responses/test_linked.py
@@ -0,0 +1,325 @@
+from datetime import UTC, datetime, timedelta
+from uuid import uuid4
+
+from posthog.test.base import APIBaseTest, ClickhouseTestMixin
+
+from parameterized import parameterized
+
+from posthog.constants import AvailableFeature
+from posthog.models.event.util import bulk_create_events
+
+from products.access_control.backend.models.access_control import AccessControl
+from products.surveys.backend.facade.api import LinkedSurveyResponses, RatingScale, get_survey_responses_linked_to
+from products.surveys.backend.models import Survey, SurveyResponseArchive
+
+T0 = datetime(2026, 9, 1, 10, 0, tzinfo=UTC)
+QUESTIONS: list[dict[str, object]] = [
+ {"id": "q-thumbs", "type": "rating", "question": "Was this answer useful?", "scale": 2, "display": "emoji"},
+ {
+ "id": "q-score",
+ "type": "rating",
+ "question": "How good was it?",
+ "scale": 5,
+ "display": "number",
+ "lowerBoundLabel": "Bad",
+ "upperBoundLabel": "Great",
+ },
+ {"id": "q-topic", "type": "single_choice", "question": "What was it about?", "choices": ["Billing", "Setup"]},
+ {
+ "id": "q-issues",
+ "type": "multiple_choice",
+ "question": "What could be better?",
+ "choices": ["Accuracy", "Tone", "Length"],
+ },
+ {"id": "q-comment", "type": "open", "question": "Anything else?"},
+]
+
+
+class TestLinkedSurveyResponses(ClickhouseTestMixin, APIBaseTest):
+ def setUp(self) -> None:
+ super().setUp()
+ self.survey = Survey.objects.create(team=self.team, name="Answer quality", type="api", questions=QUESTIONS)
+
+ def _event(
+ self,
+ event: str,
+ *,
+ survey_id: str | None = None,
+ minute: int = 1,
+ trace_id: str = "trace-1",
+ uuid: str | None = None,
+ properties: dict[str, object] | None = None,
+ ) -> dict[str, object]:
+ return {
+ "event": event,
+ "event_uuid": uuid or str(uuid4()),
+ "distinct_id": "user-1",
+ "team": self.team,
+ "timestamp": T0 + timedelta(minutes=minute),
+ "properties": {
+ "$survey_id": survey_id or str(self.survey.id),
+ "$ai_trace_id": trace_id,
+ **(properties or {}),
+ },
+ }
+
+ def _read(self, user_given: bool = True) -> LinkedSurveyResponses:
+ return get_survey_responses_linked_to(
+ self.team,
+ self.user if user_given else None,
+ property_name="$ai_trace_id",
+ property_value="trace-1",
+ since=T0,
+ )
+
+ def _answers(self, linked: LinkedSurveyResponses) -> list[list[tuple[str, object]]]:
+ return [[(answer.question_id, answer.value) for answer in response.answers] for response in linked.responses]
+
+ def test_merges_a_submission_split_across_events(self) -> None:
+ bulk_create_events(
+ [
+ self._event(
+ "survey sent",
+ minute=1,
+ properties={
+ "$survey_submission_id": "sub-1",
+ "$survey_response_q-score": "4",
+ "$survey_completed": False,
+ },
+ ),
+ self._event(
+ "survey dismissed",
+ minute=2,
+ properties={
+ "$survey_submission_id": "sub-1",
+ "$survey_response_q-comment": "Too long, but correct.",
+ "$survey_partially_completed": True,
+ },
+ ),
+ ]
+ )
+
+ linked = self._read()
+
+ assert self._answers(linked) == [[("q-score", "4"), ("q-comment", "Too long, but correct.")]]
+ assert [(response.submission_id, response.complete) for response in linked.responses] == [("sub-1", False)]
+
+ @parameterized.expand(
+ [
+ ("one survey sent without a completion flag", [("survey sent", {})], True),
+ (
+ "an incomplete send, then a completed send",
+ [("survey sent", {"$survey_completed": False}), ("survey sent", {"$survey_completed": True})],
+ True,
+ ),
+ (
+ "an incomplete send, then a partial dismissal",
+ [
+ ("survey sent", {"$survey_completed": False}),
+ ("survey dismissed", {"$survey_partially_completed": True}),
+ ],
+ False,
+ ),
+ ]
+ )
+ def test_complete(self, _name: str, events: list[tuple[str, dict[str, object]]], expected: bool) -> None:
+ bulk_create_events(
+ [
+ self._event(
+ event,
+ minute=index + 1,
+ properties={"$survey_submission_id": "sub-1", "$survey_response_q-comment": "Fine.", **properties},
+ )
+ for index, (event, properties) in enumerate(events)
+ ]
+ )
+
+ assert [response.complete for response in self._read().responses] == [expected]
+
+ def test_keeps_repeat_submissions_apart(self) -> None:
+ bulk_create_events(
+ [
+ self._event(
+ "survey sent",
+ minute=1,
+ properties={"$survey_submission_id": "sub-1", "$survey_response_q-score": "2"},
+ ),
+ self._event(
+ "survey sent",
+ minute=2,
+ properties={"$survey_submission_id": "sub-2", "$survey_response_q-score": "5"},
+ ),
+ ]
+ )
+
+ assert self._answers(self._read()) == [[("q-score", "2")], [("q-score", "5")]]
+
+ def test_events_without_a_submission_id_are_each_a_response(self) -> None:
+ first, second = str(uuid4()), str(uuid4())
+ bulk_create_events(
+ [
+ self._event("survey sent", minute=1, uuid=first, properties={"$survey_response_q-score": "3"}),
+ self._event("survey sent", minute=2, uuid=second, properties={"$survey_response_q-score": "4"}),
+ ]
+ )
+
+ assert [response.submission_id for response in self._read().responses] == [first, second]
+
+ def test_reads_only_events_linked_to_the_value_since_the_start(self) -> None:
+ bulk_create_events(
+ [
+ self._event("survey sent", minute=1, properties={"$survey_response_q-comment": "Linked."}),
+ self._event(
+ "survey sent", minute=2, trace_id="trace-2", properties={"$survey_response_q-comment": "Other."}
+ ),
+ self._event("survey sent", minute=-5, properties={"$survey_response_q-comment": "Too early."}),
+ ]
+ )
+
+ assert self._answers(self._read()) == [[("q-comment", "Linked.")]]
+
+ def test_matches_a_survey_id_written_in_upper_case(self) -> None:
+ bulk_create_events(
+ [
+ self._event(
+ "survey sent",
+ survey_id=str(self.survey.id).upper(),
+ properties={"$survey_response_q-comment": "Upper."},
+ )
+ ]
+ )
+
+ linked = self._read()
+
+ assert [(response.survey.id, response.survey.name) for response in linked.responses] == [
+ (str(self.survey.id), "Answer quality")
+ ]
+
+ def test_reads_every_kind_of_answer_with_its_question(self) -> None:
+ bulk_create_events(
+ [
+ self._event(
+ "survey sent",
+ properties={
+ "$survey_response_q-thumbs": "1",
+ "$survey_response_q-score": "4",
+ "$survey_response_q-topic": "Billing",
+ "$survey_response_q-issues": ["Accuracy", "Tone"],
+ "$survey_response_q-comment": "Too long, but correct.",
+ },
+ )
+ ]
+ )
+
+ [response] = self._read().responses
+
+ assert [
+ (answer.question_text, answer.question_type, answer.value, answer.rating) for answer in response.answers
+ ] == [
+ (
+ "Was this answer useful?",
+ "rating",
+ "1",
+ RatingScale(scale=2, display="emoji", lower_bound_label=None, upper_bound_label=None),
+ ),
+ (
+ "How good was it?",
+ "rating",
+ "4",
+ RatingScale(scale=5, display="number", lower_bound_label="Bad", upper_bound_label="Great"),
+ ),
+ ("What was it about?", "single_choice", "Billing", None),
+ ("What could be better?", "multiple_choice", ("Accuracy", "Tone"), None),
+ ("Anything else?", "open", "Too long, but correct.", None),
+ ]
+
+ def test_maps_translated_choices_back_to_the_base_language(self) -> None:
+ translated = Survey.objects.create(
+ team=self.team,
+ name="Answer quality, translated",
+ type="api",
+ questions=[
+ {**QUESTIONS[2], "translations": {"fr": {"choices": ["Facturation", "Installation"]}}},
+ {**QUESTIONS[3], "translations": {"fr": {"choices": ["Exactitude", "Ton", "Longueur"]}}},
+ ],
+ )
+ bulk_create_events(
+ [
+ self._event(
+ "survey sent",
+ survey_id=str(translated.id),
+ properties={"$survey_response_q-topic": "Facturation", "$survey_response_q-issues": ["Ton"]},
+ )
+ ]
+ )
+
+ assert self._answers(self._read()) == [[("q-topic", "Billing"), ("q-issues", ("Tone",))]]
+
+ def test_reports_a_response_to_a_missing_survey(self) -> None:
+ missing = str(uuid4())
+ bulk_create_events([self._event("survey sent", survey_id=missing, properties={"$survey_response": "Hi."})])
+
+ [response] = self._read().responses
+
+ assert (response.survey.id, response.survey.name, response.answers, response.unavailable) == (
+ missing,
+ None,
+ (),
+ "survey_not_found",
+ )
+
+ def test_reports_a_survey_without_questions(self) -> None:
+ empty = Survey.objects.create(team=self.team, name="Empty", type="api", questions=[])
+ bulk_create_events([self._event("survey sent", survey_id=str(empty.id))])
+
+ assert [(response.survey.name, response.unavailable) for response in self._read().responses] == [
+ ("Empty", "no_questions")
+ ]
+
+ def test_a_shown_survey_without_a_response_is_unanswered(self) -> None:
+ shown_only = Survey.objects.create(team=self.team, name="Follow-up", type="api", questions=QUESTIONS)
+ bulk_create_events(
+ [
+ self._event("survey shown", minute=1),
+ self._event("survey sent", minute=2, properties={"$survey_response_q-comment": "Fine."}),
+ self._event("survey shown", survey_id=str(shown_only.id), minute=3),
+ self._event("survey shown", survey_id=str(shown_only.id), minute=4),
+ ]
+ )
+
+ linked = self._read()
+
+ assert [(survey.survey.name, survey.shown_at) for survey in linked.unanswered] == [
+ ("Follow-up", T0 + timedelta(minutes=3))
+ ]
+ assert [response.survey.name for response in linked.responses] == ["Answer quality"]
+
+ def test_omits_surveys_the_viewer_cannot_access(self) -> None:
+ self.organization.available_product_features = [
+ {"key": AvailableFeature.ACCESS_CONTROL, "name": AvailableFeature.ACCESS_CONTROL}
+ ]
+ self.organization.save()
+ AccessControl.objects.create(
+ team=self.team, resource="survey", access_level="none", organization_member=self.organization_membership
+ )
+ bulk_create_events(
+ [
+ self._event("survey sent", minute=1, properties={"$survey_response_q-comment": "Hidden."}),
+ self._event("survey shown", minute=2),
+ ]
+ )
+
+ assert self._read() == LinkedSurveyResponses(responses=(), unanswered=())
+ assert len(self._read(user_given=False).responses) == 1
+
+ def test_excludes_archived_responses(self) -> None:
+ archived, kept = str(uuid4()), str(uuid4())
+ bulk_create_events(
+ [
+ self._event("survey sent", minute=1, uuid=archived, properties={"$survey_response_q-comment": "Old."}),
+ self._event("survey sent", minute=2, uuid=kept, properties={"$survey_response_q-comment": "New."}),
+ ]
+ )
+ SurveyResponseArchive.objects.create(team=self.team, survey=self.survey, response_uuid=archived)
+
+ assert self._answers(self._read()) == [[("q-comment", "New.")]]
diff --git a/services/mcp/src/api/generated.ts b/services/mcp/src/api/generated.ts
index 2763003a5a2f..8c2024ff6721 100644
--- a/services/mcp/src/api/generated.ts
+++ b/services/mcp/src/api/generated.ts
@@ -117845,6 +117845,15 @@ export namespace Schemas {
offset: number;
}
+ export type SurveyResponsesStatusEnum = typeof SurveyResponsesStatusEnum[keyof typeof SurveyResponsesStatusEnum];
+
+
+ export const SurveyResponsesStatusEnum = {
+ Ok: 'ok',
+ Failed: 'failed',
+ NotPermitted: 'not_permitted',
+ } as const;
+
export type SurveyResultsWidgetCatalogEntryOpenApiWidgetType = typeof SurveyResultsWidgetCatalogEntryOpenApiWidgetType[keyof typeof SurveyResultsWidgetCatalogEntryOpenApiWidgetType];
@@ -121635,6 +121644,12 @@ export namespace Schemas {
git: TraceGit | null;
}
+ export interface TraceChoiceAnswer {
+ kind: 'choice';
+ question: string;
+ choices: string[];
+ }
+
export interface TraceCluster {
/** The clustering run that put the trace in this cluster. The cluster page is addressed by this id and the cluster id. */
runId: string;
@@ -121734,16 +121749,82 @@ export namespace Schemas {
timestamp: string;
}
+ export interface TraceSurvey {
+ id: string;
+ /** Null when the survey no longer exists. */
+ name: string | null;
+ }
+
+ export interface TraceThumbsAnswer {
+ kind: 'thumbs';
+ question: string;
+ up: boolean;
+ }
+
+ export interface TraceRatingAnswer {
+ kind: 'rating';
+ question: string;
+ value: string;
+ scale: number | null;
+ lowerLabel: string | null;
+ upperLabel: string | null;
+ }
+
+ export interface TraceTextAnswer {
+ kind: 'text';
+ question: string;
+ text: string;
+ }
+
+ /**
+ * Why the answers cannot be shown: the survey was deleted, or it has no questions.
+ */
+ export type TraceSurveyResponseUnavailable = typeof TraceSurveyResponseUnavailable[keyof typeof TraceSurveyResponseUnavailable] | null;
+
+
+ export const TraceSurveyResponseUnavailable = {
+ SurveyNotFound: 'survey_not_found',
+ NoQuestions: 'no_questions',
+ } as const;
+
+ export interface TraceSurveyResponse {
+ kind: 'survey';
+ id: string;
+ /** The tree node the response targets: the trace id, as a survey response covers the whole trace. */
+ nodeId: string;
+ timestamp: string;
+ survey: TraceSurvey;
+ answers: (TraceThumbsAnswer | TraceRatingAnswer | TraceChoiceAnswer | TraceTextAnswer)[];
+ /** False when the user left the survey before its end. */
+ complete: boolean;
+ /** Why the answers cannot be shown: the survey was deleted, or it has no questions. */
+ unavailable: TraceSurveyResponseUnavailable;
+ }
+
+ export interface TraceUnansweredSurvey {
+ survey: TraceSurvey;
+ /** When the survey was first shown. */
+ timestamp: string;
+ }
+
export interface TraceFeedbackSummary {
/** Scores plus responses. */
total: number;
+ /** Thumbs up answers to 2-point survey rating questions. */
+ thumbsUp: number;
+ /** Thumbs down answers to 2-point survey rating questions. */
+ thumbsDown: number;
}
export interface TraceFeedback {
/** Oldest first. */
scores: TraceFeedbackScore[];
/** What users said, oldest first. */
- responses: TraceFeedbackText[];
+ responses: (TraceFeedbackText | TraceSurveyResponse)[];
+ /** Surveys shown with the trace and not answered, oldest first. */
+ unanswered: TraceUnansweredSurvey[];
+ /** ok when survey responses are included, failed when the survey read failed, not_permitted when the token lacks the survey:read and query:read scopes. */
+ surveyResponsesStatus: SurveyResponsesStatusEnum;
summary: TraceFeedbackSummary;
}
diff --git a/tach.toml b/tach.toml
index d42becf3010c..ab8f165ead23 100644
--- a/tach.toml
+++ b/tach.toml
@@ -589,6 +589,8 @@ depends_on = [
"products.workflows", "products.access_control",
# Evaluations and taggers compile Hog through products.cdp.backend.validation and filters.
"products.cdp",
+ # Trace feedback reads trace-linked survey responses through the surveys facade.
+ "products.surveys",
]
layer = "modules"