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.

fix(signals): research new signals after wrong-repo dismissals (#124663)

Alex Lebedev77fb3e911b6d
diff --git a/docs/internal/signals-pr-lifecycle.md b/docs/internal/signals-pr-lifecycle.md
index e101bae8004a..b4cf3743d9f1 100644
--- a/docs/internal/signals-pr-lifecycle.md
+++ b/docs/internal/signals-pr-lifecycle.md
@@ -475,9 +475,15 @@ The gates apply to automatic implementation only.
Pressing Implement in the inbox starts a run whatever the links say.
Nothing re-evaluates a held-back report when its dependency's pull request opens, so it starts on the report's next pipeline evaluation or by hand.
-## Recurrence after a fixed verdict
+## Recurrence after a fixed verdict or wrong repository
A report dismissed as `already_fixed`, `fixed_outside_posthog`, or `pr_merged` can create a new report when the issue returns.
+A report dismissed as `wrong_repo` also creates a fresh report when a new matching signal arrives, even without a corrected repository.
+The original stays dismissed, and the new report uses the normal research promotion threshold.
+Its repository selection runs afresh with the project's saved correction feedback, without reusing the rejected selection.
+If selection picks the repository rejected by the parent's latest dismissal, the new report waits for repository input before research. This also covers projects with only one eligible repository and does not block unrelated reports from using it.
+The parent's corrected repository remains selection feedback rather than a selection copied onto the successor, so normal eligibility and source-visibility checks still apply.
+Research does not treat a wrong-repository dismissal as evidence of a previous fix.
The pipeline records the new report's parent with a typed `recurrence_of` report link.
A report created because a follow-up check failed on a resolved report also gets a typed `follow_up_of` link to that report, carrying the verdict as the link's reason.
Generic `related_to` links do not control signal assignment.
diff --git a/products/signals/ARCHITECTURE.md b/products/signals/ARCHITECTURE.md
index a59c80e46a44..17dee79c28b3 100644
--- a/products/signals/ARCHITECTURE.md
+++ b/products/signals/ARCHITECTURE.md
@@ -409,7 +409,8 @@ potential → candidate → in_progress → ready
# RESOLVED is terminal and never re-promotes: a recurrence spawns a fresh report, linked back to the
# resolved one via a recurrence_of report link (assign_and_emit_signal_activity). A report dismissed with a
# reason that claims the issue is fixed (already_fixed, fixed_outside_posthog, pr_merged) makes the
-# same claim, so it forks the same way; every other dismissal reason absorbs later signals silently
+# same claim, so it forks the same way. A wrong_repo dismissal also forks so repository selection
+# can run again; every other dismissal reason absorbs later signals silently
# (recurrence.py). The pipeline records each successor with a typed recurrence_of link. Generic related_to
# links do not control signal assignment. Matching follows this successor chain, including
# preference-dismissed reports that must keep absorbing signals. Repeated feedback does not break
@@ -1105,6 +1106,11 @@ A `wrong_assignee` dismissal from an older client or an agent also takes the cal
A bulk dismissal processes only the reports that name the caller, and it skips the per-report correction notes, because the dismissal already sends each scout one note for the batch.
**Wrong-repo dismissals.** The `wrong_repo` reason code carries an optional `corrected_repository` (`owner/repo`, refused with any other reason).
+The next matching signal starts a fresh report linked back by `recurrence_of`, while the original stays dismissed.
+Further matching signals follow that link to the same successor, which uses the normal weight threshold for research promotion.
+Repository selection runs afresh with the project's correction feedback; the successor does not reuse the dismissed report's selection or treat its findings as evidence of a previous fix.
+If selection returns the repository rejected by the parent's latest dismissal, the successor requests repository input before research, including when that repository is the only eligible candidate. This check applies only to the linked successor, so unrelated reports can still use that repository.
+The successor does not copy `corrected_repository` into its selection; the correction informs normal selection, which checks current eligibility and source visibility.
The dismissal artefact denormalizes `selected_repository` (the report's latest `repo_selection` at dismissal time) next to the correction, so a selection mistake is one self-contained record.
When the corrected repository is currently connected to the team, the state action also appends it as a user-attributed `repo_selection` artefact, latest-wins.
Restoring the report does not itself re-research; the appended selection means the report's next research run, once new signals re-promote it, targets the corrected repository instead of repeating the rejected pick.
diff --git a/products/signals/backend/artefact_schemas.py b/products/signals/backend/artefact_schemas.py
index 8ab3a3f98b53..d27521d70372 100644
--- a/products/signals/backend/artefact_schemas.py
+++ b/products/signals/backend/artefact_schemas.py
@@ -544,7 +544,7 @@ DISMISSAL_REASON_WRONG_REPO = "wrong_repo"
# Reason codes that claim the issue is fixed, rather than stating a preference about the report.
# A later matching signal contradicts the claim, so the grouping stage treats these dismissals the
# way it treats a resolved report: the recurrence gets a fresh report (see `recurrence.py`). The
-# `wontfix_*` codes and the rest stay sinks, because they say "I do not want this".
+# `wrong_repo` can also start a fresh report, but does not claim a previous fix.
FIXED_DISMISSAL_REASONS = frozenset({"already_fixed", "fixed_outside_posthog", "pr_merged"})
# Bounds shared by the state API and this schema, so the generic artefact endpoint cannot store a
# dismissal the state API would reject. Readers scan these rows in bulk (`repo_corrections`), so an
diff --git a/products/signals/backend/recurrence.py b/products/signals/backend/recurrence.py
index c4229368a65a..ffd04f123ba9 100644
--- a/products/signals/backend/recurrence.py
+++ b/products/signals/backend/recurrence.py
@@ -1,45 +1,74 @@
-"""Recurrence of an issue a report already claimed was fixed.
+"""Route new evidence after a fixed verdict or a rejected repository choice.
A resolved report and a report dismissed as fixed make the same factual claim: the issue is gone.
-A later matching signal is direct evidence that the claim was wrong, so both must surface the
-recurrence instead of absorbing it. The grouping stage reads this module to tell the two kinds of
-dismissal apart.
+A wrong-repo dismissal leaves the issue unresolved. Both need a fresh report for later matching
+signals, but only fixed verdicts supply prior-fix context to research.
"""
from datetime import datetime
from pydantic import ValidationError
-from products.signals.backend.artefact_schemas import FIXED_DISMISSAL_REASONS, Dismissal
+from products.signals.backend.artefact_schemas import DISMISSAL_REASON_WRONG_REPO, FIXED_DISMISSAL_REASONS, Dismissal
from products.signals.backend.enums import ReportLinkKind
from products.signals.backend.models import SignalReport, SignalReportArtefact
-from products.signals.backend.typed_report_links import incoming_links
+from products.signals.backend.repo_corrections import sanitized_repository
+from products.signals.backend.typed_report_links import incoming_links, outgoing_links
-def fixed_dismissal_at(report: SignalReport) -> datetime | None:
- """When the report was dismissed with a claim that the issue is fixed, or None.
-
- Only the latest dismissal counts: dismissals stack, so an earlier "already fixed" that a
- reviewer has since overruled with a `wontfix_*` must not keep forking reports.
- """
- latest = (
- SignalReportArtefact.objects.filter(
- report_id=report.id,
- team_id=report.team_id,
+def _latest_dismissal(*, team_id: int, report_id: str) -> SignalReportArtefact | None:
+ return (
+ SignalReportArtefact.objects.using("default")
+ .filter(
+ report_id=report_id,
+ team_id=team_id,
type=SignalReportArtefact.ArtefactType.DISMISSAL,
)
.order_by("-created_at")
- .values_list("content", "created_at")
+ .only("content", "created_at")
.first()
)
+
+
+def _dismissal_at(report: SignalReport, reasons: frozenset[str]) -> datetime | None:
+ latest = _latest_dismissal(team_id=report.team_id, report_id=str(report.id))
if latest is None:
return None
- content, created_at = latest
try:
- reason = Dismissal.model_validate_json(content).reason
+ reason = Dismissal.model_validate_json(latest.content).reason
except ValidationError:
return None
- return created_at if reason in FIXED_DISMISSAL_REASONS else None
+ return latest.created_at if reason in reasons else None
+
+
+def fixed_dismissal_at(report: SignalReport) -> datetime | None:
+ """When the report was dismissed with a claim that the issue is fixed, or None.
+
+ Only the latest dismissal counts: dismissals stack, so an earlier "already fixed" that a
+ reviewer has since overruled with a `wontfix_*` must not keep forking reports.
+ """
+ return _dismissal_at(report, FIXED_DISMISSAL_REASONS)
+
+
+def recurrence_dismissal_at(report: SignalReport) -> datetime | None:
+ return _dismissal_at(report, FIXED_DISMISSAL_REASONS | {DISMISSAL_REASON_WRONG_REPO})
+
+
+def recurrence_rejects_repository(*, team_id: int, report_id: str, repository: str) -> bool:
+ for edge in outgoing_links(team_id=team_id, report_id=report_id, kinds=(ReportLinkKind.RECURRENCE_OF,)):
+ latest = _latest_dismissal(team_id=team_id, report_id=edge.target_id)
+ if latest is None:
+ continue
+ try:
+ dismissal = Dismissal.model_validate_json(latest.content)
+ except ValidationError:
+ continue
+ if (
+ dismissal.reason == DISMISSAL_REASON_WRONG_REPO
+ and sanitized_repository(dismissal.selected_repository) == repository.strip().lower()
+ ):
+ return True
+ return False
def _next_recurrence_report(report: SignalReport, *, lock: bool) -> SignalReport | None:
diff --git a/products/signals/backend/temporal/agentic/select_repository.py b/products/signals/backend/temporal/agentic/select_repository.py
index 92adbe2bc0ef..2e5967bd31c8 100644
--- a/products/signals/backend/temporal/agentic/select_repository.py
+++ b/products/signals/backend/temporal/agentic/select_repository.py
@@ -13,6 +13,7 @@ from posthog.temporal.common.heartbeat import Heartbeater
from posthog.temporal.common.scoped import scoped_temporal
from posthog.temporal.common.utils import aretry_on_db_connection_drop, close_db_connections
+from products.signals.backend.recurrence import recurrence_rejects_repository
from products.signals.backend.report_generation.select_repo import (
RepoSelectionResult,
persisted_repo_selection,
@@ -162,6 +163,18 @@ async def select_repository_activity(input: SelectRepositoryInput) -> RepoSelect
spend_signal_id=input.signal_id,
sandbox_environment_id=sandbox_env_id,
)
+ # The shared selector takes a sole eligible repository without reading corrections.
+ # A recurrence must honor its parent's rejection before spending on research there.
+ if result.repository is not None and await database_sync_to_async(
+ recurrence_rejects_repository, thread_sensitive=False
+ )(team_id=input.team_id, report_id=input.report_id, repository=result.repository):
+ result = RepoSelectionResult(
+ repository=None,
+ reason=(
+ "This repository was rejected for the previous report. "
+ "Choose a repository to continue research."
+ ),
+ )
logger.info(
"signals repo selection completed",
report_id=input.report_id,
diff --git a/products/signals/backend/temporal/grouping.py b/products/signals/backend/temporal/grouping.py
index 559f4fc8740f..e99801201b3f 100644
--- a/products/signals/backend/temporal/grouping.py
+++ b/products/signals/backend/temporal/grouping.py
@@ -38,7 +38,7 @@ from products.signals.backend.enums import ReportLinkKind, ReportLinkWritePath,
from products.signals.backend.models import SignalEmissionRecord, SignalReport, SignalReportArtefact
from products.signals.backend.quota import capture_signal_report_quota_paused, self_driving_quota_gate
from products.signals.backend.receivers import _is_safety_suppressed
-from products.signals.backend.recurrence import fixed_dismissal_at
+from products.signals.backend.recurrence import recurrence_dismissal_at
from products.signals.backend.report_merge import signal_target_report
from products.signals.backend.signal_metadata import EMBEDDING_MODEL
from products.signals.backend.spend import begin_report_spend, complete_signal_pipeline, signal_spend_summaries
@@ -913,20 +913,18 @@ async def assign_and_emit_signal_activity(input: AssignAndEmitSignalInput) -> As
promotion_suppressed=False,
next_research_bucket=None,
)
- # A report dismissed as fixed claims the issue is gone, exactly as a resolved report
- # does, so this signal contradicts it and must not be absorbed. The other dismissal
- # codes state a preference about the report, and a sink is the right answer for
- # them (see recurrence.py).
+ # Wrong-repo feedback rejects the repository choice rather than the issue, so new
+ # evidence needs fresh research, just as it does after a fixed verdict.
report = signal_target_report(report, lock=True)
- dismissed_as_fixed_at = (
- fixed_dismissal_at(report) if report.status == SignalReport.Status.SUPPRESSED else None
+ dismissed_for_recurrence_at = (
+ recurrence_dismissal_at(report) if report.status == SignalReport.Status.SUPPRESSED else None
)
# Resolved reports are terminal — never reopen them. When a signal would have grouped
# into an already-resolved report, the issue it fixed has recurred (or a related one
# has), so we start a fresh report and link it to the resolved report via a
# `recurrence_of` report link. The research agent is later handed that
# resolved report as context (see report.py).
- if report.effective_status == SignalReport.Status.RESOLVED or dismissed_as_fixed_at is not None:
+ if report.effective_status == SignalReport.Status.RESOLVED or dismissed_for_recurrence_at is not None:
parent_report = report
inherit_content = not _is_safety_suppressed(str(parent_report.id), input.team_id)
report = SignalReport.objects.create(
@@ -979,8 +977,7 @@ async def assign_and_emit_signal_activity(input: AssignAndEmitSignalInput) -> As
)
# Promotion rules by status:
- # - SUPPRESSED: never promoted. A report dismissed as fixed never receives new signals
- # either (a recurrence spawns a fresh report above).
+ # - SUPPRESSED: never promoted. Fixed and wrong-repo dismissals spawn a fresh report above.
# - RESOLVED: terminal — never receives new signals (a recurrence spawns a fresh report above).
# - POTENTIAL: promote once total_weight >= WEIGHT_THRESHOLD and signal_count >= signals_at_run
# (snooze gate, defaults to 0). Uncapped — a report's first research always runs.
diff --git a/products/signals/backend/test/test_agentic_report_activity.py b/products/signals/backend/test/test_agentic_report_activity.py
index 0399cefcf541..baa19d99e3e2 100644
--- a/products/signals/backend/test/test_agentic_report_activity.py
+++ b/products/signals/backend/test/test_agentic_report_activity.py
@@ -300,6 +300,7 @@ def _build_signals() -> list[SignalData]:
[
(SignalReport.Status.RESOLVED, None, True),
(SignalReport.Status.SUPPRESSED, "already_fixed", True),
+ (SignalReport.Status.SUPPRESSED, "wrong_repo", False),
# Archived because nobody wants to see it, not because it was fixed: there is no prior fix
# to reason about, so the research agent is told nothing.
(SignalReport.Status.SUPPRESSED, "wontfix_intentional", False),
diff --git a/products/signals/backend/test/test_assign_and_emit_signal.py b/products/signals/backend/test/test_assign_and_emit_signal.py
index 0f4b8b801d8b..3aced3b7ddc1 100644
--- a/products/signals/backend/test/test_assign_and_emit_signal.py
+++ b/products/signals/backend/test/test_assign_and_emit_signal.py
@@ -3,7 +3,7 @@ import random
from datetime import timedelta
import pytest
-from unittest.mock import AsyncMock, patch
+from unittest.mock import AsyncMock, MagicMock, patch
from django.utils import timezone
@@ -13,6 +13,7 @@ from asgiref.sync import sync_to_async
from posthog.models import Organization, Team
from posthog.sync import database_sync_to_async
+from products.signals.backend.agent_runtime import DEFAULT_RUNTIME
from products.signals.backend.artefact_attribution import ArtefactAttribution
from products.signals.backend.artefact_schemas import Dismissal, RelatedTo, ReportLink
from products.signals.backend.daily_limit import DailyReportLimitGate
@@ -21,6 +22,10 @@ from products.signals.backend.models import SignalEmissionRecord, SignalReport,
from products.signals.backend.quota import SelfDrivingQuotaGate
from products.signals.backend.report_merge import MERGE_DISMISSAL_REASON
from products.signals.backend.spend import complete_signal_pipeline, finish_report_spend, record_llm_spend
+from products.signals.backend.temporal.agentic.select_repository import (
+ SelectRepositoryInput,
+ select_repository_activity,
+)
from products.signals.backend.temporal.grouping import (
WEIGHT_THRESHOLD,
AssignAndEmitSignalInput,
@@ -541,6 +546,7 @@ async def _suppressed_report(team, reason: str) -> SignalReport:
("already_fixed", True),
("fixed_outside_posthog", True),
("pr_merged", True),
+ ("wrong_repo", True),
("wontfix_intentional", False),
("wontfix_irrelevant", False),
("analysis_wrong", False),
@@ -549,10 +555,7 @@ async def _suppressed_report(team, reason: str) -> SignalReport:
],
)
@pytest.mark.parametrize("unsafe_parent", [False, True])
-async def test_suppressed_report_forks_only_for_a_fixed_dismissal(ateam, reason, forks, unsafe_parent):
- # A dismissal that claims the issue is fixed is contradicted by a new signal, so the signal
- # starts a fresh report, exactly as a resolved report does. Every other reason states a
- # preference about the report, so a sink stays the right answer.
+async def test_suppressed_report_forks_for_fixed_or_wrong_repo_dismissals(ateam, reason, forks, unsafe_parent):
parent = await _suppressed_report(ateam, reason)
if unsafe_parent:
await database_sync_to_async(SignalReportArtefact.objects.create)(
@@ -590,10 +593,76 @@ async def test_suppressed_report_forks_only_for_a_fixed_dismissal(ateam, reason,
@pytest.mark.asyncio
@pytest.mark.django_db
-async def test_latest_dismissal_decides_whether_a_suppressed_report_forks(ateam):
- # Dismissals stack, so an "already fixed" a reviewer has since overruled with a won't-fix must
- # stop forking reports.
- parent = await _suppressed_report(ateam, "already_fixed")
+ "reason,corrected,eligible,match_parent,overridden,expected",
+ [
+ ("wrong_repo", None, "example/website", True, False, None),
+ ("wrong_repo", "example/checkout", "example/website", True, False, None),
+ ("wrong_repo", "example/checkout", "example/checkout", True, False, "example/checkout"),
+ ("already_fixed", None, "example/website", True, False, "example/website"),
+ ("wrong_repo", None, "example/website", False, False, "example/website"),
+ ("wrong_repo", None, "example/website", True, True, "example/website"),
+ ],
+)
+async def test_recurrence_selection_does_not_reuse_a_rejected_single_repository(
+ ateam: Team,
+ reason: str,
+ corrected: str | None,
+ eligible: str,
+ match_parent: bool,
+ overridden: bool,
+ expected: str | None,
+) -> None:
+ parent = await database_sync_to_async(SignalReport.objects.create)(
+ team=ateam, status=SignalReport.Status.SUPPRESSED
+ )
+ await database_sync_to_async(SignalReportArtefact.append_dismissal)(
+ team_id=ateam.id,
+ report_id=str(parent.id),
+ content=Dismissal(reason=reason, selected_repository="Example/Website", corrected_repository=corrected),
+ attribution=ArtefactAttribution.system(),
+ )
+ assigned = await assign_and_emit_signal_activity(
+ _build_input(
+ ateam.id, _existing_match(str(parent.id)) if match_parent else _new_match(), weight=WEIGHT_THRESHOLD
+ )
+ )
+ assert assigned.report_id != str(parent.id)
+ if overridden:
+ await _dismiss(parent, "analysis_wrong")
+ github = MagicMock()
+ github.list_all_cached_repositories.return_value = [{"full_name": eligible, "private": True}]
+ selector = "products.tasks.backend.logic.repo_selection.agent"
+ activity = "products.signals.backend.temporal.agentic.select_repository"
+ with (
+ patch(f"{selector}.resolve_team_github_integration", return_value=github),
+ patch(f"{selector}._list_candidate_repos", return_value=[eligible]),
+ patch(f"{selector}._list_eligible_full_names", return_value={eligible}),
+ patch(f"{selector}.GitHubRepositoryFullCache") as cache,
+ patch(f"{selector}.MultiTurnSession.start", new_callable=AsyncMock) as start,
+ patch(
+ "products.signals.backend.report_generation.select_repo.resolve_agent_runtime", return_value=DEFAULT_RUNTIME
+ ),
+ patch(f"{activity}._resolve_sandbox_user_id", return_value=1),
+ patch(f"{activity}.get_or_create_signals_sandbox_env", return_value=None),
+ patch(f"{activity}.Heartbeater"),
+ ):
+ cache.return_value.sync_full_cache = AsyncMock()
+ result = await select_repository_activity(
+ SelectRepositoryInput(team_id=ateam.id, report_id=assigned.report_id, signals=[])
+ )
+
+ assert result.repository == expected
+ if expected is None:
+ assert "Choose a repository" in result.reason
+ start.assert_not_awaited()
+
+
[email protected]("reason", ["already_fixed", "wrong_repo"])
+async def test_latest_dismissal_decides_whether_a_suppressed_report_forks(ateam, reason):
+ parent = await _suppressed_report(ateam, reason)
await _dismiss(parent, "wontfix_intentional")
result = await assign_and_emit_signal_activity(
@@ -606,18 +675,19 @@ async def test_latest_dismissal_decides_whether_a_suppressed_report_forks(ateam)
@pytest.mark.asyncio
@pytest.mark.django_db
[email protected]("reason", ["already_fixed", "wrong_repo"])
@pytest.mark.parametrize("repeat_feedback", [False, True])
@pytest.mark.parametrize("restore_parent", [False, True])
-async def test_second_recurrence_signal_joins_the_open_fork(ateam, repeat_feedback, restore_parent):
+async def test_second_recurrence_signal_joins_the_open_fork(ateam, reason, repeat_feedback, restore_parent):
# A parent that absorbed hundreds of signals must produce one live recurrence report, not one
# per signal: the second signal accumulates on the first signal's fork.
- parent = await _suppressed_report(ateam, "already_fixed")
+ parent = await _suppressed_report(ateam, reason)
first = await assign_and_emit_signal_activity(
_build_input(ateam.id, _existing_match(str(parent.id)), weight=WEIGHT_THRESHOLD * 0.5)
)
if repeat_feedback:
- await _dismiss(parent, "already_fixed")
+ await _dismiss(parent, reason)
if restore_parent:
parent.status = SignalReport.Status.READY
await database_sync_to_async(parent.save)(update_fields=["status"])
diff --git a/products/signals/backend/views.py b/products/signals/backend/views.py
index 7d7fda73a063..c76649e85e4a 100644
--- a/products/signals/backend/views.py
+++ b/products/signals/backend/views.py
@@ -749,11 +749,13 @@ _DISMISSAL_REASON_HELP_TEXT = (
"codes claim the issue is gone, so a later signal about the same issue starts a fresh report linked "
"to this one. Fixed reason codes require state='suppressed' or state='resolved', not 'potential'. "
"The dismissal codes (report_unclear, analysis_wrong, wrong_assignee, "
- "wrong_repo, wontfix_*) go with state='suppressed' and absorb later signals silently. "
+ "wontfix_*) go with state='suppressed' and absorb later signals silently. "
"'wrong_assignee' is accepted from older clients, but a report sent to the wrong person still holds "
"work: remove the person from the report's reviewers with a removal_reason instead of suppressing it. "
"'wrong_repo' when the agent picked the wrong repository for this report, ideally with "
- "corrected_repository naming the right one. "
+ "corrected_repository naming the right one. The report stays suppressed, but a later matching "
+ "signal starts a fresh linked report with a new repository selection informed by the project's "
+ "correction feedback, even when no corrected_repository was supplied. "
"Use 'verified_manually' with state='resolved' to stop Monitoring after manual verification. "
"It requires Monitoring to be enabled and a report with a Monitoring period, and records no automatic "
"verification result or evidence amendment. "
diff --git a/products/signals/frontend/generated/api.schemas.ts b/products/signals/frontend/generated/api.schemas.ts
index 91ae8a226e95..fba97f18fe14 100644
--- a/products/signals/frontend/generated/api.schemas.ts
+++ b/products/signals/frontend/generated/api.schemas.ts
@@ -5064,7 +5064,7 @@ export interface SignalReportStateRequestApi {
* * `resolved` - resolved
* * `monitoring` - monitoring */
state: SignalReportStateEnumApi
- /** Optional canonical reason code recorded with the transition. Must be one of: already_fixed, report_unclear, analysis_wrong, wrong_assignee, wrong_repo, wontfix_intentional, wontfix_irrelevant, fixed_outside_posthog, verified_manually, no_action_needed, pr_merged, other — these match the inbox UI so the rationale renders as a labelled chip rather than a raw code. When the work this report asked for is done, the honest transition is state='resolved' with 'fixed_outside_posthog' (the fix landed without a pull request), 'pr_merged' (a pull request with the fix was merged but did not resolve the report on its own), or 'already_fixed' (it was fixed before the report was filed). Use 'no_action_needed' with state='resolved' when the report was useful but needs no implementation. A report that failed in processing resolves too, so a fix that landed is recorded as a fix rather than as a dismissal. These three codes claim the issue is gone, so a later signal about the same issue starts a fresh report linked to this one. Fixed reason codes require state='suppressed' or state='resolved', not 'potential'. The dismissal codes (report_unclear, analysis_wrong, wrong_assignee, wrong_repo, wontfix_*) go with state='suppressed' and absorb later signals silently. 'wrong_assignee' is accepted from older clients, but a report sent to the wrong person still holds work: remove the person from the report's reviewers with a removal_reason instead of suppressing it. 'wrong_repo' when the agent picked the wrong repository for this report, ideally with corrected_repository naming the right one. Use 'verified_manually' with state='resolved' to stop Monitoring after manual verification. It requires Monitoring to be enabled and a report with a Monitoring period, and records no automatic verification result or evidence amendment. Use 'other' together with a dismissal_note for anything that doesn't fit a code.
+ /** Optional canonical reason code recorded with the transition. Must be one of: already_fixed, report_unclear, analysis_wrong, wrong_assignee, wrong_repo, wontfix_intentional, wontfix_irrelevant, fixed_outside_posthog, verified_manually, no_action_needed, pr_merged, other — these match the inbox UI so the rationale renders as a labelled chip rather than a raw code. When the work this report asked for is done, the honest transition is state='resolved' with 'fixed_outside_posthog' (the fix landed without a pull request), 'pr_merged' (a pull request with the fix was merged but did not resolve the report on its own), or 'already_fixed' (it was fixed before the report was filed). Use 'no_action_needed' with state='resolved' when the report was useful but needs no implementation. A report that failed in processing resolves too, so a fix that landed is recorded as a fix rather than as a dismissal. These three codes claim the issue is gone, so a later signal about the same issue starts a fresh report linked to this one. Fixed reason codes require state='suppressed' or state='resolved', not 'potential'. The dismissal codes (report_unclear, analysis_wrong, wrong_assignee, wontfix_*) go with state='suppressed' and absorb later signals silently. 'wrong_assignee' is accepted from older clients, but a report sent to the wrong person still holds work: remove the person from the report's reviewers with a removal_reason instead of suppressing it. 'wrong_repo' when the agent picked the wrong repository for this report, ideally with corrected_repository naming the right one. The report stays suppressed, but a later matching signal starts a fresh linked report with a new repository selection informed by the project's correction feedback, even when no corrected_repository was supplied. Use 'verified_manually' with state='resolved' to stop Monitoring after manual verification. It requires Monitoring to be enabled and a report with a Monitoring period, and records no automatic verification result or evidence amendment. Use 'other' together with a dismissal_note for anything that doesn't fit a code.
*
* * `already_fixed` - Already fixed
* * `report_unclear` - Report is unclear to me
@@ -5701,7 +5701,7 @@ export interface SignalReportBulkStateRequestApi {
* * `resolved` - resolved
* * `monitoring` - monitoring */
state: SignalReportStateEnumApi
- /** Optional canonical reason code recorded with the transition. Must be one of: already_fixed, report_unclear, analysis_wrong, wrong_assignee, wrong_repo, wontfix_intentional, wontfix_irrelevant, fixed_outside_posthog, verified_manually, no_action_needed, pr_merged, other — these match the inbox UI so the rationale renders as a labelled chip rather than a raw code. When the work this report asked for is done, the honest transition is state='resolved' with 'fixed_outside_posthog' (the fix landed without a pull request), 'pr_merged' (a pull request with the fix was merged but did not resolve the report on its own), or 'already_fixed' (it was fixed before the report was filed). Use 'no_action_needed' with state='resolved' when the report was useful but needs no implementation. A report that failed in processing resolves too, so a fix that landed is recorded as a fix rather than as a dismissal. These three codes claim the issue is gone, so a later signal about the same issue starts a fresh report linked to this one. Fixed reason codes require state='suppressed' or state='resolved', not 'potential'. The dismissal codes (report_unclear, analysis_wrong, wrong_assignee, wrong_repo, wontfix_*) go with state='suppressed' and absorb later signals silently. 'wrong_assignee' is accepted from older clients, but a report sent to the wrong person still holds work: remove the person from the report's reviewers with a removal_reason instead of suppressing it. 'wrong_repo' when the agent picked the wrong repository for this report, ideally with corrected_repository naming the right one. Use 'verified_manually' with state='resolved' to stop Monitoring after manual verification. It requires Monitoring to be enabled and a report with a Monitoring period, and records no automatic verification result or evidence amendment. Use 'other' together with a dismissal_note for anything that doesn't fit a code.
+ /** Optional canonical reason code recorded with the transition. Must be one of: already_fixed, report_unclear, analysis_wrong, wrong_assignee, wrong_repo, wontfix_intentional, wontfix_irrelevant, fixed_outside_posthog, verified_manually, no_action_needed, pr_merged, other — these match the inbox UI so the rationale renders as a labelled chip rather than a raw code. When the work this report asked for is done, the honest transition is state='resolved' with 'fixed_outside_posthog' (the fix landed without a pull request), 'pr_merged' (a pull request with the fix was merged but did not resolve the report on its own), or 'already_fixed' (it was fixed before the report was filed). Use 'no_action_needed' with state='resolved' when the report was useful but needs no implementation. A report that failed in processing resolves too, so a fix that landed is recorded as a fix rather than as a dismissal. These three codes claim the issue is gone, so a later signal about the same issue starts a fresh report linked to this one. Fixed reason codes require state='suppressed' or state='resolved', not 'potential'. The dismissal codes (report_unclear, analysis_wrong, wrong_assignee, wontfix_*) go with state='suppressed' and absorb later signals silently. 'wrong_assignee' is accepted from older clients, but a report sent to the wrong person still holds work: remove the person from the report's reviewers with a removal_reason instead of suppressing it. 'wrong_repo' when the agent picked the wrong repository for this report, ideally with corrected_repository naming the right one. The report stays suppressed, but a later matching signal starts a fresh linked report with a new repository selection informed by the project's correction feedback, even when no corrected_repository was supplied. Use 'verified_manually' with state='resolved' to stop Monitoring after manual verification. It requires Monitoring to be enabled and a report with a Monitoring period, and records no automatic verification result or evidence amendment. Use 'other' together with a dismissal_note for anything that doesn't fit a code.
*
* * `already_fixed` - Already fixed
* * `report_unclear` - Report is unclear to me
diff --git a/products/signals/frontend/generated/api.zod.ts b/products/signals/frontend/generated/api.zod.ts
index ccae314f64aa..f219dea61069 100644
--- a/products/signals/frontend/generated/api.zod.ts
+++ b/products/signals/frontend/generated/api.zod.ts
@@ -1089,7 +1089,7 @@ export const SignalsReportsStateCreateBody = /* @__PURE__ */ zod.object({
)
.optional()
.describe(
- "Optional canonical reason code recorded with the transition. Must be one of: already_fixed, report_unclear, analysis_wrong, wrong_assignee, wrong_repo, wontfix_intentional, wontfix_irrelevant, fixed_outside_posthog, verified_manually, no_action_needed, pr_merged, other — these match the inbox UI so the rationale renders as a labelled chip rather than a raw code. When the work this report asked for is done, the honest transition is state='resolved' with 'fixed_outside_posthog' (the fix landed without a pull request), 'pr_merged' (a pull request with the fix was merged but did not resolve the report on its own), or 'already_fixed' (it was fixed before the report was filed). Use 'no_action_needed' with state='resolved' when the report was useful but needs no implementation. A report that failed in processing resolves too, so a fix that landed is recorded as a fix rather than as a dismissal. These three codes claim the issue is gone, so a later signal about the same issue starts a fresh report linked to this one. Fixed reason codes require state='suppressed' or state='resolved', not 'potential'. The dismissal codes (report_unclear, analysis_wrong, wrong_assignee, wrong_repo, wontfix_\*) go with state='suppressed' and absorb later signals silently. 'wrong_assignee' is accepted from older clients, but a report sent to the wrong person still holds work: remove the person from the report's reviewers with a removal_reason instead of suppressing it. 'wrong_repo' when the agent picked the wrong repository for this report, ideally with corrected_repository naming the right one. Use 'verified_manually' with state='resolved' to stop Monitoring after manual verification. It requires Monitoring to be enabled and a report with a Monitoring period, and records no automatic verification result or evidence amendment. Use 'other' together with a dismissal_note for anything that doesn't fit a code.\n\n\* `already_fixed` - Already fixed\n\* `report_unclear` - Report is unclear to me\n\* `analysis_wrong` - Agent's analysis is wrong\n\* `wrong_assignee` - Wrong assignee\n\* `wrong_repo` - Agent picked the wrong repository\n\* `wontfix_intentional` - Won't fix - intentional behavior\n\* `wontfix_irrelevant` - Not relevant to this project\n\* `fixed_outside_posthog` - Fixed outside PostHog\n\* `verified_manually` - Verified manually\n\* `no_action_needed` - No action needed\n\* `pr_merged` - PR was merged\n\* `other` - Something else…"
+ "Optional canonical reason code recorded with the transition. Must be one of: already_fixed, report_unclear, analysis_wrong, wrong_assignee, wrong_repo, wontfix_intentional, wontfix_irrelevant, fixed_outside_posthog, verified_manually, no_action_needed, pr_merged, other — these match the inbox UI so the rationale renders as a labelled chip rather than a raw code. When the work this report asked for is done, the honest transition is state='resolved' with 'fixed_outside_posthog' (the fix landed without a pull request), 'pr_merged' (a pull request with the fix was merged but did not resolve the report on its own), or 'already_fixed' (it was fixed before the report was filed). Use 'no_action_needed' with state='resolved' when the report was useful but needs no implementation. A report that failed in processing resolves too, so a fix that landed is recorded as a fix rather than as a dismissal. These three codes claim the issue is gone, so a later signal about the same issue starts a fresh report linked to this one. Fixed reason codes require state='suppressed' or state='resolved', not 'potential'. The dismissal codes (report_unclear, analysis_wrong, wrong_assignee, wontfix_\*) go with state='suppressed' and absorb later signals silently. 'wrong_assignee' is accepted from older clients, but a report sent to the wrong person still holds work: remove the person from the report's reviewers with a removal_reason instead of suppressing it. 'wrong_repo' when the agent picked the wrong repository for this report, ideally with corrected_repository naming the right one. The report stays suppressed, but a later matching signal starts a fresh linked report with a new repository selection informed by the project's correction feedback, even when no corrected_repository was supplied. Use 'verified_manually' with state='resolved' to stop Monitoring after manual verification. It requires Monitoring to be enabled and a report with a Monitoring period, and records no automatic verification result or evidence amendment. Use 'other' together with a dismissal_note for anything that doesn't fit a code.\n\n\* `already_fixed` - Already fixed\n\* `report_unclear` - Report is unclear to me\n\* `analysis_wrong` - Agent's analysis is wrong\n\* `wrong_assignee` - Wrong assignee\n\* `wrong_repo` - Agent picked the wrong repository\n\* `wontfix_intentional` - Won't fix - intentional behavior\n\* `wontfix_irrelevant` - Not relevant to this project\n\* `fixed_outside_posthog` - Fixed outside PostHog\n\* `verified_manually` - Verified manually\n\* `no_action_needed` - No action needed\n\* `pr_merged` - PR was merged\n\* `other` - Something else…"
),
dismissal_note: zod
.string()
@@ -1365,7 +1365,7 @@ export const SignalsReportsBulkStateCreateBody = /* @__PURE__ */ zod.object({
)
.optional()
.describe(
- "Optional canonical reason code recorded with the transition. Must be one of: already_fixed, report_unclear, analysis_wrong, wrong_assignee, wrong_repo, wontfix_intentional, wontfix_irrelevant, fixed_outside_posthog, verified_manually, no_action_needed, pr_merged, other — these match the inbox UI so the rationale renders as a labelled chip rather than a raw code. When the work this report asked for is done, the honest transition is state='resolved' with 'fixed_outside_posthog' (the fix landed without a pull request), 'pr_merged' (a pull request with the fix was merged but did not resolve the report on its own), or 'already_fixed' (it was fixed before the report was filed). Use 'no_action_needed' with state='resolved' when the report was useful but needs no implementation. A report that failed in processing resolves too, so a fix that landed is recorded as a fix rather than as a dismissal. These three codes claim the issue is gone, so a later signal about the same issue starts a fresh report linked to this one. Fixed reason codes require state='suppressed' or state='resolved', not 'potential'. The dismissal codes (report_unclear, analysis_wrong, wrong_assignee, wrong_repo, wontfix_\*) go with state='suppressed' and absorb later signals silently. 'wrong_assignee' is accepted from older clients, but a report sent to the wrong person still holds work: remove the person from the report's reviewers with a removal_reason instead of suppressing it. 'wrong_repo' when the agent picked the wrong repository for this report, ideally with corrected_repository naming the right one. Use 'verified_manually' with state='resolved' to stop Monitoring after manual verification. It requires Monitoring to be enabled and a report with a Monitoring period, and records no automatic verification result or evidence amendment. Use 'other' together with a dismissal_note for anything that doesn't fit a code.\n\n\* `already_fixed` - Already fixed\n\* `report_unclear` - Report is unclear to me\n\* `analysis_wrong` - Agent's analysis is wrong\n\* `wrong_assignee` - Wrong assignee\n\* `wrong_repo` - Agent picked the wrong repository\n\* `wontfix_intentional` - Won't fix - intentional behavior\n\* `wontfix_irrelevant` - Not relevant to this project\n\* `fixed_outside_posthog` - Fixed outside PostHog\n\* `verified_manually` - Verified manually\n\* `no_action_needed` - No action needed\n\* `pr_merged` - PR was merged\n\* `other` - Something else…"
+ "Optional canonical reason code recorded with the transition. Must be one of: already_fixed, report_unclear, analysis_wrong, wrong_assignee, wrong_repo, wontfix_intentional, wontfix_irrelevant, fixed_outside_posthog, verified_manually, no_action_needed, pr_merged, other — these match the inbox UI so the rationale renders as a labelled chip rather than a raw code. When the work this report asked for is done, the honest transition is state='resolved' with 'fixed_outside_posthog' (the fix landed without a pull request), 'pr_merged' (a pull request with the fix was merged but did not resolve the report on its own), or 'already_fixed' (it was fixed before the report was filed). Use 'no_action_needed' with state='resolved' when the report was useful but needs no implementation. A report that failed in processing resolves too, so a fix that landed is recorded as a fix rather than as a dismissal. These three codes claim the issue is gone, so a later signal about the same issue starts a fresh report linked to this one. Fixed reason codes require state='suppressed' or state='resolved', not 'potential'. The dismissal codes (report_unclear, analysis_wrong, wrong_assignee, wontfix_\*) go with state='suppressed' and absorb later signals silently. 'wrong_assignee' is accepted from older clients, but a report sent to the wrong person still holds work: remove the person from the report's reviewers with a removal_reason instead of suppressing it. 'wrong_repo' when the agent picked the wrong repository for this report, ideally with corrected_repository naming the right one. The report stays suppressed, but a later matching signal starts a fresh linked report with a new repository selection informed by the project's correction feedback, even when no corrected_repository was supplied. Use 'verified_manually' with state='resolved' to stop Monitoring after manual verification. It requires Monitoring to be enabled and a report with a Monitoring period, and records no automatic verification result or evidence amendment. Use 'other' together with a dismissal_note for anything that doesn't fit a code.\n\n\* `already_fixed` - Already fixed\n\* `report_unclear` - Report is unclear to me\n\* `analysis_wrong` - Agent's analysis is wrong\n\* `wrong_assignee` - Wrong assignee\n\* `wrong_repo` - Agent picked the wrong repository\n\* `wontfix_intentional` - Won't fix - intentional behavior\n\* `wontfix_irrelevant` - Not relevant to this project\n\* `fixed_outside_posthog` - Fixed outside PostHog\n\* `verified_manually` - Verified manually\n\* `no_action_needed` - No action needed\n\* `pr_merged` - PR was merged\n\* `other` - Something else…"
),
dismissal_note: zod
.string()
diff --git a/services/mcp/src/api/generated.ts b/services/mcp/src/api/generated.ts
index 3e12a0474052..2e31c0ec41b6 100644
--- a/services/mcp/src/api/generated.ts
+++ b/services/mcp/src/api/generated.ts
@@ -109904,7 +109904,7 @@ export namespace Schemas {
* * `resolved` - resolved
* * `monitoring` - monitoring */
state: SignalReportStateEnum;
- /** Optional canonical reason code recorded with the transition. Must be one of: already_fixed, report_unclear, analysis_wrong, wrong_assignee, wrong_repo, wontfix_intentional, wontfix_irrelevant, fixed_outside_posthog, verified_manually, no_action_needed, pr_merged, other — these match the inbox UI so the rationale renders as a labelled chip rather than a raw code. When the work this report asked for is done, the honest transition is state='resolved' with 'fixed_outside_posthog' (the fix landed without a pull request), 'pr_merged' (a pull request with the fix was merged but did not resolve the report on its own), or 'already_fixed' (it was fixed before the report was filed). Use 'no_action_needed' with state='resolved' when the report was useful but needs no implementation. A report that failed in processing resolves too, so a fix that landed is recorded as a fix rather than as a dismissal. These three codes claim the issue is gone, so a later signal about the same issue starts a fresh report linked to this one. Fixed reason codes require state='suppressed' or state='resolved', not 'potential'. The dismissal codes (report_unclear, analysis_wrong, wrong_assignee, wrong_repo, wontfix_*) go with state='suppressed' and absorb later signals silently. 'wrong_assignee' is accepted from older clients, but a report sent to the wrong person still holds work: remove the person from the report's reviewers with a removal_reason instead of suppressing it. 'wrong_repo' when the agent picked the wrong repository for this report, ideally with corrected_repository naming the right one. Use 'verified_manually' with state='resolved' to stop Monitoring after manual verification. It requires Monitoring to be enabled and a report with a Monitoring period, and records no automatic verification result or evidence amendment. Use 'other' together with a dismissal_note for anything that doesn't fit a code.
+ /** Optional canonical reason code recorded with the transition. Must be one of: already_fixed, report_unclear, analysis_wrong, wrong_assignee, wrong_repo, wontfix_intentional, wontfix_irrelevant, fixed_outside_posthog, verified_manually, no_action_needed, pr_merged, other — these match the inbox UI so the rationale renders as a labelled chip rather than a raw code. When the work this report asked for is done, the honest transition is state='resolved' with 'fixed_outside_posthog' (the fix landed without a pull request), 'pr_merged' (a pull request with the fix was merged but did not resolve the report on its own), or 'already_fixed' (it was fixed before the report was filed). Use 'no_action_needed' with state='resolved' when the report was useful but needs no implementation. A report that failed in processing resolves too, so a fix that landed is recorded as a fix rather than as a dismissal. These three codes claim the issue is gone, so a later signal about the same issue starts a fresh report linked to this one. Fixed reason codes require state='suppressed' or state='resolved', not 'potential'. The dismissal codes (report_unclear, analysis_wrong, wrong_assignee, wontfix_*) go with state='suppressed' and absorb later signals silently. 'wrong_assignee' is accepted from older clients, but a report sent to the wrong person still holds work: remove the person from the report's reviewers with a removal_reason instead of suppressing it. 'wrong_repo' when the agent picked the wrong repository for this report, ideally with corrected_repository naming the right one. The report stays suppressed, but a later matching signal starts a fresh linked report with a new repository selection informed by the project's correction feedback, even when no corrected_repository was supplied. Use 'verified_manually' with state='resolved' to stop Monitoring after manual verification. It requires Monitoring to be enabled and a report with a Monitoring period, and records no automatic verification result or evidence amendment. Use 'other' together with a dismissal_note for anything that doesn't fit a code.
*
* * `already_fixed` - Already fixed
* * `report_unclear` - Report is unclear to me
@@ -110272,7 +110272,7 @@ export namespace Schemas {
* * `resolved` - resolved
* * `monitoring` - monitoring */
state: SignalReportStateEnum;
- /** Optional canonical reason code recorded with the transition. Must be one of: already_fixed, report_unclear, analysis_wrong, wrong_assignee, wrong_repo, wontfix_intentional, wontfix_irrelevant, fixed_outside_posthog, verified_manually, no_action_needed, pr_merged, other — these match the inbox UI so the rationale renders as a labelled chip rather than a raw code. When the work this report asked for is done, the honest transition is state='resolved' with 'fixed_outside_posthog' (the fix landed without a pull request), 'pr_merged' (a pull request with the fix was merged but did not resolve the report on its own), or 'already_fixed' (it was fixed before the report was filed). Use 'no_action_needed' with state='resolved' when the report was useful but needs no implementation. A report that failed in processing resolves too, so a fix that landed is recorded as a fix rather than as a dismissal. These three codes claim the issue is gone, so a later signal about the same issue starts a fresh report linked to this one. Fixed reason codes require state='suppressed' or state='resolved', not 'potential'. The dismissal codes (report_unclear, analysis_wrong, wrong_assignee, wrong_repo, wontfix_*) go with state='suppressed' and absorb later signals silently. 'wrong_assignee' is accepted from older clients, but a report sent to the wrong person still holds work: remove the person from the report's reviewers with a removal_reason instead of suppressing it. 'wrong_repo' when the agent picked the wrong repository for this report, ideally with corrected_repository naming the right one. Use 'verified_manually' with state='resolved' to stop Monitoring after manual verification. It requires Monitoring to be enabled and a report with a Monitoring period, and records no automatic verification result or evidence amendment. Use 'other' together with a dismissal_note for anything that doesn't fit a code.
+ /** Optional canonical reason code recorded with the transition. Must be one of: already_fixed, report_unclear, analysis_wrong, wrong_assignee, wrong_repo, wontfix_intentional, wontfix_irrelevant, fixed_outside_posthog, verified_manually, no_action_needed, pr_merged, other — these match the inbox UI so the rationale renders as a labelled chip rather than a raw code. When the work this report asked for is done, the honest transition is state='resolved' with 'fixed_outside_posthog' (the fix landed without a pull request), 'pr_merged' (a pull request with the fix was merged but did not resolve the report on its own), or 'already_fixed' (it was fixed before the report was filed). Use 'no_action_needed' with state='resolved' when the report was useful but needs no implementation. A report that failed in processing resolves too, so a fix that landed is recorded as a fix rather than as a dismissal. These three codes claim the issue is gone, so a later signal about the same issue starts a fresh report linked to this one. Fixed reason codes require state='suppressed' or state='resolved', not 'potential'. The dismissal codes (report_unclear, analysis_wrong, wrong_assignee, wontfix_*) go with state='suppressed' and absorb later signals silently. 'wrong_assignee' is accepted from older clients, but a report sent to the wrong person still holds work: remove the person from the report's reviewers with a removal_reason instead of suppressing it. 'wrong_repo' when the agent picked the wrong repository for this report, ideally with corrected_repository naming the right one. The report stays suppressed, but a later matching signal starts a fresh linked report with a new repository selection informed by the project's correction feedback, even when no corrected_repository was supplied. Use 'verified_manually' with state='resolved' to stop Monitoring after manual verification. It requires Monitoring to be enabled and a report with a Monitoring period, and records no automatic verification result or evidence amendment. Use 'other' together with a dismissal_note for anything that doesn't fit a code.
*
* * `already_fixed` - Already fixed
* * `report_unclear` - Report is unclear to me
diff --git a/services/mcp/src/generated/signals/api.ts b/services/mcp/src/generated/signals/api.ts
index d8e85788e19a..a4075be8f36e 100644
--- a/services/mcp/src/generated/signals/api.ts
+++ b/services/mcp/src/generated/signals/api.ts
@@ -926,7 +926,7 @@ export const SignalsReportsStateCreateBody = () => zod.object({
)
.optional()
.describe(
- "Optional canonical reason code recorded with the transition. Must be one of: already_fixed, report_unclear, analysis_wrong, wrong_assignee, wrong_repo, wontfix_intentional, wontfix_irrelevant, fixed_outside_posthog, verified_manually, no_action_needed, pr_merged, other — these match the inbox UI so the rationale renders as a labelled chip rather than a raw code. When the work this report asked for is done, the honest transition is state='resolved' with 'fixed_outside_posthog' (the fix landed without a pull request), 'pr_merged' (a pull request with the fix was merged but did not resolve the report on its own), or 'already_fixed' (it was fixed before the report was filed). Use 'no_action_needed' with state='resolved' when the report was useful but needs no implementation. A report that failed in processing resolves too, so a fix that landed is recorded as a fix rather than as a dismissal. These three codes claim the issue is gone, so a later signal about the same issue starts a fresh report linked to this one. Fixed reason codes require state='suppressed' or state='resolved', not 'potential'. The dismissal codes (report_unclear, analysis_wrong, wrong_assignee, wrong_repo, wontfix_\*) go with state='suppressed' and absorb later signals silently. 'wrong_assignee' is accepted from older clients, but a report sent to the wrong person still holds work: remove the person from the report's reviewers with a removal_reason instead of suppressing it. 'wrong_repo' when the agent picked the wrong repository for this report, ideally with corrected_repository naming the right one. Use 'verified_manually' with state='resolved' to stop Monitoring after manual verification. It requires Monitoring to be enabled and a report with a Monitoring period, and records no automatic verification result or evidence amendment. Use 'other' together with a dismissal_note for anything that doesn't fit a code.\n\n\* `already_fixed` - Already fixed\n\* `report_unclear` - Report is unclear to me\n\* `analysis_wrong` - Agent's analysis is wrong\n\* `wrong_assignee` - Wrong assignee\n\* `wrong_repo` - Agent picked the wrong repository\n\* `wontfix_intentional` - Won't fix - intentional behavior\n\* `wontfix_irrelevant` - Not relevant to this project\n\* `fixed_outside_posthog` - Fixed outside PostHog\n\* `verified_manually` - Verified manually\n\* `no_action_needed` - No action needed\n\* `pr_merged` - PR was merged\n\* `other` - Something else…"
+ "Optional canonical reason code recorded with the transition. Must be one of: already_fixed, report_unclear, analysis_wrong, wrong_assignee, wrong_repo, wontfix_intentional, wontfix_irrelevant, fixed_outside_posthog, verified_manually, no_action_needed, pr_merged, other — these match the inbox UI so the rationale renders as a labelled chip rather than a raw code. When the work this report asked for is done, the honest transition is state='resolved' with 'fixed_outside_posthog' (the fix landed without a pull request), 'pr_merged' (a pull request with the fix was merged but did not resolve the report on its own), or 'already_fixed' (it was fixed before the report was filed). Use 'no_action_needed' with state='resolved' when the report was useful but needs no implementation. A report that failed in processing resolves too, so a fix that landed is recorded as a fix rather than as a dismissal. These three codes claim the issue is gone, so a later signal about the same issue starts a fresh report linked to this one. Fixed reason codes require state='suppressed' or state='resolved', not 'potential'. The dismissal codes (report_unclear, analysis_wrong, wrong_assignee, wontfix_\*) go with state='suppressed' and absorb later signals silently. 'wrong_assignee' is accepted from older clients, but a report sent to the wrong person still holds work: remove the person from the report's reviewers with a removal_reason instead of suppressing it. 'wrong_repo' when the agent picked the wrong repository for this report, ideally with corrected_repository naming the right one. The report stays suppressed, but a later matching signal starts a fresh linked report with a new repository selection informed by the project's correction feedback, even when no corrected_repository was supplied. Use 'verified_manually' with state='resolved' to stop Monitoring after manual verification. It requires Monitoring to be enabled and a report with a Monitoring period, and records no automatic verification result or evidence amendment. Use 'other' together with a dismissal_note for anything that doesn't fit a code.\n\n\* `already_fixed` - Already fixed\n\* `report_unclear` - Report is unclear to me\n\* `analysis_wrong` - Agent's analysis is wrong\n\* `wrong_assignee` - Wrong assignee\n\* `wrong_repo` - Agent picked the wrong repository\n\* `wontfix_intentional` - Won't fix - intentional behavior\n\* `wontfix_irrelevant` - Not relevant to this project\n\* `fixed_outside_posthog` - Fixed outside PostHog\n\* `verified_manually` - Verified manually\n\* `no_action_needed` - No action needed\n\* `pr_merged` - PR was merged\n\* `other` - Something else…"
),
dismissal_note: zod
.string()
@@ -1364,7 +1364,7 @@ export const SignalsReportsBulkStateCreateBody = () => zod.object({
)
.optional()
.describe(
- "Optional canonical reason code recorded with the transition. Must be one of: already_fixed, report_unclear, analysis_wrong, wrong_assignee, wrong_repo, wontfix_intentional, wontfix_irrelevant, fixed_outside_posthog, verified_manually, no_action_needed, pr_merged, other — these match the inbox UI so the rationale renders as a labelled chip rather than a raw code. When the work this report asked for is done, the honest transition is state='resolved' with 'fixed_outside_posthog' (the fix landed without a pull request), 'pr_merged' (a pull request with the fix was merged but did not resolve the report on its own), or 'already_fixed' (it was fixed before the report was filed). Use 'no_action_needed' with state='resolved' when the report was useful but needs no implementation. A report that failed in processing resolves too, so a fix that landed is recorded as a fix rather than as a dismissal. These three codes claim the issue is gone, so a later signal about the same issue starts a fresh report linked to this one. Fixed reason codes require state='suppressed' or state='resolved', not 'potential'. The dismissal codes (report_unclear, analysis_wrong, wrong_assignee, wrong_repo, wontfix_\*) go with state='suppressed' and absorb later signals silently. 'wrong_assignee' is accepted from older clients, but a report sent to the wrong person still holds work: remove the person from the report's reviewers with a removal_reason instead of suppressing it. 'wrong_repo' when the agent picked the wrong repository for this report, ideally with corrected_repository naming the right one. Use 'verified_manually' with state='resolved' to stop Monitoring after manual verification. It requires Monitoring to be enabled and a report with a Monitoring period, and records no automatic verification result or evidence amendment. Use 'other' together with a dismissal_note for anything that doesn't fit a code.\n\n\* `already_fixed` - Already fixed\n\* `report_unclear` - Report is unclear to me\n\* `analysis_wrong` - Agent's analysis is wrong\n\* `wrong_assignee` - Wrong assignee\n\* `wrong_repo` - Agent picked the wrong repository\n\* `wontfix_intentional` - Won't fix - intentional behavior\n\* `wontfix_irrelevant` - Not relevant to this project\n\* `fixed_outside_posthog` - Fixed outside PostHog\n\* `verified_manually` - Verified manually\n\* `no_action_needed` - No action needed\n\* `pr_merged` - PR was merged\n\* `other` - Something else…"
+ "Optional canonical reason code recorded with the transition. Must be one of: already_fixed, report_unclear, analysis_wrong, wrong_assignee, wrong_repo, wontfix_intentional, wontfix_irrelevant, fixed_outside_posthog, verified_manually, no_action_needed, pr_merged, other — these match the inbox UI so the rationale renders as a labelled chip rather than a raw code. When the work this report asked for is done, the honest transition is state='resolved' with 'fixed_outside_posthog' (the fix landed without a pull request), 'pr_merged' (a pull request with the fix was merged but did not resolve the report on its own), or 'already_fixed' (it was fixed before the report was filed). Use 'no_action_needed' with state='resolved' when the report was useful but needs no implementation. A report that failed in processing resolves too, so a fix that landed is recorded as a fix rather than as a dismissal. These three codes claim the issue is gone, so a later signal about the same issue starts a fresh report linked to this one. Fixed reason codes require state='suppressed' or state='resolved', not 'potential'. The dismissal codes (report_unclear, analysis_wrong, wrong_assignee, wontfix_\*) go with state='suppressed' and absorb later signals silently. 'wrong_assignee' is accepted from older clients, but a report sent to the wrong person still holds work: remove the person from the report's reviewers with a removal_reason instead of suppressing it. 'wrong_repo' when the agent picked the wrong repository for this report, ideally with corrected_repository naming the right one. The report stays suppressed, but a later matching signal starts a fresh linked report with a new repository selection informed by the project's correction feedback, even when no corrected_repository was supplied. Use 'verified_manually' with state='resolved' to stop Monitoring after manual verification. It requires Monitoring to be enabled and a report with a Monitoring period, and records no automatic verification result or evidence amendment. Use 'other' together with a dismissal_note for anything that doesn't fit a code.\n\n\* `already_fixed` - Already fixed\n\* `report_unclear` - Report is unclear to me\n\* `analysis_wrong` - Agent's analysis is wrong\n\* `wrong_assignee` - Wrong assignee\n\* `wrong_repo` - Agent picked the wrong repository\n\* `wontfix_intentional` - Won't fix - intentional behavior\n\* `wontfix_irrelevant` - Not relevant to this project\n\* `fixed_outside_posthog` - Fixed outside PostHog\n\* `verified_manually` - Verified manually\n\* `no_action_needed` - No action needed\n\* `pr_merged` - PR was merged\n\* `other` - Something else…"
),
dismissal_note: zod
.string()
diff --git a/services/mcp/tests/unit/__snapshots__/tool-schemas/inbox-reports-bulk-set-state.json b/services/mcp/tests/unit/__snapshots__/tool-schemas/inbox-reports-bulk-set-state.json
index bbbfc69db598..a2479ff65e52 100644
--- a/services/mcp/tests/unit/__snapshots__/tool-schemas/inbox-reports-bulk-set-state.json
+++ b/services/mcp/tests/unit/__snapshots__/tool-schemas/inbox-reports-bulk-set-state.json
@@ -13,7 +13,7 @@
"type": "string"
},
"dismissal_reason": {
- "description": "Optional canonical reason code recorded with the transition. Must be one of: already_fixed, report_unclear, analysis_wrong, wrong_assignee, wrong_repo, wontfix_intentional, wontfix_irrelevant, fixed_outside_posthog, verified_manually, no_action_needed, pr_merged, other — these match the inbox UI so the rationale renders as a labelled chip rather than a raw code. When the work this report asked for is done, the honest transition is state='resolved' with 'fixed_outside_posthog' (the fix landed without a pull request), 'pr_merged' (a pull request with the fix was merged but did not resolve the report on its own), or 'already_fixed' (it was fixed before the report was filed). Use 'no_action_needed' with state='resolved' when the report was useful but needs no implementation. A report that failed in processing resolves too, so a fix that landed is recorded as a fix rather than as a dismissal. These three codes claim the issue is gone, so a later signal about the same issue starts a fresh report linked to this one. Fixed reason codes require state='suppressed' or state='resolved', not 'potential'. The dismissal codes (report_unclear, analysis_wrong, wrong_assignee, wrong_repo, wontfix_*) go with state='suppressed' and absorb later signals silently. 'wrong_assignee' is accepted from older clients, but a report sent to the wrong person still holds work: remove the person from the report's reviewers with a removal_reason instead of suppressing it. 'wrong_repo' when the agent picked the wrong repository for this report, ideally with corrected_repository naming the right one. Use 'verified_manually' with state='resolved' to stop Monitoring after manual verification. It requires Monitoring to be enabled and a report with a Monitoring period, and records no automatic verification result or evidence amendment. Use 'other' together with a dismissal_note for anything that doesn't fit a code.\n\n* `already_fixed` - Already fixed\n* `report_unclear` - Report is unclear to me\n* `analysis_wrong` - Agent's analysis is wrong\n* `wrong_assignee` - Wrong assignee\n* `wrong_repo` - Agent picked the wrong repository\n* `wontfix_intentional` - Won't fix - intentional behavior\n* `wontfix_irrelevant` - Not relevant to this project\n* `fixed_outside_posthog` - Fixed outside PostHog\n* `verified_manually` - Verified manually\n* `no_action_needed` - No action needed\n* `pr_merged` - PR was merged\n* `other` - Something else…",
+ "description": "Optional canonical reason code recorded with the transition. Must be one of: already_fixed, report_unclear, analysis_wrong, wrong_assignee, wrong_repo, wontfix_intentional, wontfix_irrelevant, fixed_outside_posthog, verified_manually, no_action_needed, pr_merged, other — these match the inbox UI so the rationale renders as a labelled chip rather than a raw code. When the work this report asked for is done, the honest transition is state='resolved' with 'fixed_outside_posthog' (the fix landed without a pull request), 'pr_merged' (a pull request with the fix was merged but did not resolve the report on its own), or 'already_fixed' (it was fixed before the report was filed). Use 'no_action_needed' with state='resolved' when the report was useful but needs no implementation. A report that failed in processing resolves too, so a fix that landed is recorded as a fix rather than as a dismissal. These three codes claim the issue is gone, so a later signal about the same issue starts a fresh report linked to this one. Fixed reason codes require state='suppressed' or state='resolved', not 'potential'. The dismissal codes (report_unclear, analysis_wrong, wrong_assignee, wontfix_*) go with state='suppressed' and absorb later signals silently. 'wrong_assignee' is accepted from older clients, but a report sent to the wrong person still holds work: remove the person from the report's reviewers with a removal_reason instead of suppressing it. 'wrong_repo' when the agent picked the wrong repository for this report, ideally with corrected_repository naming the right one. The report stays suppressed, but a later matching signal starts a fresh linked report with a new repository selection informed by the project's correction feedback, even when no corrected_repository was supplied. Use 'verified_manually' with state='resolved' to stop Monitoring after manual verification. It requires Monitoring to be enabled and a report with a Monitoring period, and records no automatic verification result or evidence amendment. Use 'other' together with a dismissal_note for anything that doesn't fit a code.\n\n* `already_fixed` - Already fixed\n* `report_unclear` - Report is unclear to me\n* `analysis_wrong` - Agent's analysis is wrong\n* `wrong_assignee` - Wrong assignee\n* `wrong_repo` - Agent picked the wrong repository\n* `wontfix_intentional` - Won't fix - intentional behavior\n* `wontfix_irrelevant` - Not relevant to this project\n* `fixed_outside_posthog` - Fixed outside PostHog\n* `verified_manually` - Verified manually\n* `no_action_needed` - No action needed\n* `pr_merged` - PR was merged\n* `other` - Something else…",
"enum": [
"already_fixed",
"report_unclear",
diff --git a/services/mcp/tests/unit/__snapshots__/tool-schemas/inbox-reports-set-state.json b/services/mcp/tests/unit/__snapshots__/tool-schemas/inbox-reports-set-state.json
index 0db05c43aca1..9ce4d36f7355 100644
--- a/services/mcp/tests/unit/__snapshots__/tool-schemas/inbox-reports-set-state.json
+++ b/services/mcp/tests/unit/__snapshots__/tool-schemas/inbox-reports-set-state.json
@@ -13,7 +13,7 @@
"type": "string"
},
"dismissal_reason": {
- "description": "Optional canonical reason code recorded with the transition. Must be one of: already_fixed, report_unclear, analysis_wrong, wrong_assignee, wrong_repo, wontfix_intentional, wontfix_irrelevant, fixed_outside_posthog, verified_manually, no_action_needed, pr_merged, other — these match the inbox UI so the rationale renders as a labelled chip rather than a raw code. When the work this report asked for is done, the honest transition is state='resolved' with 'fixed_outside_posthog' (the fix landed without a pull request), 'pr_merged' (a pull request with the fix was merged but did not resolve the report on its own), or 'already_fixed' (it was fixed before the report was filed). Use 'no_action_needed' with state='resolved' when the report was useful but needs no implementation. A report that failed in processing resolves too, so a fix that landed is recorded as a fix rather than as a dismissal. These three codes claim the issue is gone, so a later signal about the same issue starts a fresh report linked to this one. Fixed reason codes require state='suppressed' or state='resolved', not 'potential'. The dismissal codes (report_unclear, analysis_wrong, wrong_assignee, wrong_repo, wontfix_*) go with state='suppressed' and absorb later signals silently. 'wrong_assignee' is accepted from older clients, but a report sent to the wrong person still holds work: remove the person from the report's reviewers with a removal_reason instead of suppressing it. 'wrong_repo' when the agent picked the wrong repository for this report, ideally with corrected_repository naming the right one. Use 'verified_manually' with state='resolved' to stop Monitoring after manual verification. It requires Monitoring to be enabled and a report with a Monitoring period, and records no automatic verification result or evidence amendment. Use 'other' together with a dismissal_note for anything that doesn't fit a code.\n\n* `already_fixed` - Already fixed\n* `report_unclear` - Report is unclear to me\n* `analysis_wrong` - Agent's analysis is wrong\n* `wrong_assignee` - Wrong assignee\n* `wrong_repo` - Agent picked the wrong repository\n* `wontfix_intentional` - Won't fix - intentional behavior\n* `wontfix_irrelevant` - Not relevant to this project\n* `fixed_outside_posthog` - Fixed outside PostHog\n* `verified_manually` - Verified manually\n* `no_action_needed` - No action needed\n* `pr_merged` - PR was merged\n* `other` - Something else…",
+ "description": "Optional canonical reason code recorded with the transition. Must be one of: already_fixed, report_unclear, analysis_wrong, wrong_assignee, wrong_repo, wontfix_intentional, wontfix_irrelevant, fixed_outside_posthog, verified_manually, no_action_needed, pr_merged, other — these match the inbox UI so the rationale renders as a labelled chip rather than a raw code. When the work this report asked for is done, the honest transition is state='resolved' with 'fixed_outside_posthog' (the fix landed without a pull request), 'pr_merged' (a pull request with the fix was merged but did not resolve the report on its own), or 'already_fixed' (it was fixed before the report was filed). Use 'no_action_needed' with state='resolved' when the report was useful but needs no implementation. A report that failed in processing resolves too, so a fix that landed is recorded as a fix rather than as a dismissal. These three codes claim the issue is gone, so a later signal about the same issue starts a fresh report linked to this one. Fixed reason codes require state='suppressed' or state='resolved', not 'potential'. The dismissal codes (report_unclear, analysis_wrong, wrong_assignee, wontfix_*) go with state='suppressed' and absorb later signals silently. 'wrong_assignee' is accepted from older clients, but a report sent to the wrong person still holds work: remove the person from the report's reviewers with a removal_reason instead of suppressing it. 'wrong_repo' when the agent picked the wrong repository for this report, ideally with corrected_repository naming the right one. The report stays suppressed, but a later matching signal starts a fresh linked report with a new repository selection informed by the project's correction feedback, even when no corrected_repository was supplied. Use 'verified_manually' with state='resolved' to stop Monitoring after manual verification. It requires Monitoring to be enabled and a report with a Monitoring period, and records no automatic verification result or evidence amendment. Use 'other' together with a dismissal_note for anything that doesn't fit a code.\n\n* `already_fixed` - Already fixed\n* `report_unclear` - Report is unclear to me\n* `analysis_wrong` - Agent's analysis is wrong\n* `wrong_assignee` - Wrong assignee\n* `wrong_repo` - Agent picked the wrong repository\n* `wontfix_intentional` - Won't fix - intentional behavior\n* `wontfix_irrelevant` - Not relevant to this project\n* `fixed_outside_posthog` - Fixed outside PostHog\n* `verified_manually` - Verified manually\n* `no_action_needed` - No action needed\n* `pr_merged` - PR was merged\n* `other` - Something else…",
"enum": [
"already_fixed",
"report_unclear",