: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.
Python ab-testingai-analyticsanalyticscdpdata-warehouseexperimentsfeature-flagsjavascriptproduct-analyticspythonreactsession-replaysurveystypescriptweb-analytics
from dataclasses import dataclass
from functools import cached_property
from typing import Any
from products.signals.backend.emission.registry import SignalEmitterOutput, SignalSourceTableConfig, get_signal_config
from products.warehouse_sources.backend.facade.types import ExternalDataSourceType
# Sentinel for error tracking signals (not a real ExternalDataSourceType)
ERROR_TRACKING = "error_tracking"
# Maps ExternalDataSourceType to the schema name used in the signal registry
_SOURCE_SCHEMA_MAP: dict[ExternalDataSourceType, str] = {
ExternalDataSourceType.ZENDESK: "tickets",
ExternalDataSourceType.GITHUB: "issues",
ExternalDataSourceType.LINEAR: "issues",
}
# Maps ExternalDataSourceType to (title_field, body_field) in the emitter's record format
_RECORD_FIELD_MAP: dict[ExternalDataSourceType, tuple[str, str]] = {
ExternalDataSourceType.ZENDESK: ("subject", "description"),
ExternalDataSourceType.GITHUB: ("title", "body"),
ExternalDataSourceType.LINEAR: ("title", "description"),
}
# Defaults for extra fields that Linear's _build_extra
_LINEAR_EXTRA_DEFAULTS: dict[str, Any] = {
"url": "",
"identifier": "",
"number": 0,
"priority": 0,
"priority_label": "",
"created_at": "",
"updated_at": "",
}
# Preamble text for error tracking signal descriptions, matching cymbal's format
_ERROR_TRACKING_PREAMBLES: dict[str, str] = {
"issue_created": "New error tracking issue created - this particular exception was observed for the first time",
"issue_reopened": "Previously resolved error tracking issue has reappeared",
"issue_spiking": "This error tracking issue is experiencing a spike in occurrences",
}
_ERROR_TRACKING_WEIGHTS: dict[str, float] = {
"issue_created": 0.4,
"issue_reopened": 0.7,
"issue_spiking": 0.7,
}
def _render_error_tracking_description(title: str, body: str, source_type: str) -> str:
"""Render an error tracking signal description matching cymbal's format from signals.rs."""
preamble = _ERROR_TRACKING_PREAMBLES.get(source_type, "Error tracking issue")
return f"{preamble}:\n{title}\n\n```\n{body}\n```"
def _noop_emitter(_team_id: int, _record: dict[str, Any]) -> SignalEmitterOutput | None:
return None
def _noop_fetcher(_team: Any, _config: SignalSourceTableConfig, _context: dict[str, Any]) -> list[dict[str, Any]]:
# Error tracking eval cases never go through the fetcher path — content is rendered directly
# via _render_error_tracking_description. The fetcher only exists to satisfy the registry contract.
return []
@dataclass
class EvalSignalSpec:
"""Specification for a single synthetic signal."""
source: ExternalDataSourceType | str # str for ERROR_TRACKING
title: str
body: str
source_type_override: str = "issue_created" # only used for error_tracking
def _to_record(self) -> dict[str, Any]:
title_field, body_field = _RECORD_FIELD_MAP[self.source] # type: ignore[index]
record: dict[str, Any] = {
"id": 1,
title_field: self.title,
body_field: self.body,
}
if self.source == ExternalDataSourceType.LINEAR:
record.update(_LINEAR_EXTRA_DEFAULTS)
return record
@cached_property
def config(self) -> SignalSourceTableConfig:
if self.source == ERROR_TRACKING:
return SignalSourceTableConfig(
source_product="error_tracking",
source_type=self.source_type_override,
emitter=_noop_emitter,
record_fetcher=_noop_fetcher,
partition_field="",
fields=(),
)
schema_name = _SOURCE_SCHEMA_MAP[self.source] # type: ignore[index]
config = get_signal_config(self.source.value, schema_name) # type: ignore[union-attr]
if config is None:
raise ValueError(f"No signal config registered for {self.source.value}/{schema_name}") # type: ignore[union-attr]
return config
@cached_property
def content(self) -> SignalEmitterOutput:
if self.source == ERROR_TRACKING:
return SignalEmitterOutput(
source_product="error_tracking",
source_type=self.source_type_override,
source_id="eval-0",
description=_render_error_tracking_description(self.title, self.body, self.source_type_override),
weight=_ERROR_TRACKING_WEIGHTS.get(self.source_type_override, 0.4),
extra={},
)
record = self._to_record()
content = self.config.emitter(0, record)
if content is None:
raise ValueError(f"Content empty for {self.source.value}") # type: ignore[union-attr]
return content
@dataclass
class EvalGroupSpec:
"""Specification for a single synthetic group."""
scenario: str
signals: list[EvalSignalSpec]
actionable: bool = True
safe: bool = True