[Charan-place]

Problem

Three small but distinct bugs in PostHog's insights/cohorts flow:

1. Group-targeted feature flags' default usage insights count $feature_flag_called
   events even when no group was resolved, skewing rollout percentages.
2. The legacy property is_set/is_not_set filter treats an explicit JSON null
   value as "set," so a person with email: null shows up in an "email is set" cohort.
3. There's no way to filter the Saved Insights list down to insights that aren't on
   any dashboard yet.

Closes #39135
Closes #29916
Closes #26621

Changes

• Add a $group_<n> is_set filter to default flag insights when the flag is group-targeted.
• Require the JSON-extracted value to be non-empty and not the literal string 'null'
  for is_set/is_not_set in prop_filter_json_extract.
• Add a not_on_any_dashboard API filter and a matching toggle on the Saved Insights page.

How did you test this code?

I'm an agent (Claude) and didn't have a Django/ClickHouse/Postgres environment available, so I couldn't run the suite myself. I added unit tests alongside each fix:

• products/feature_flags/backend/api/test/test_feature_flag.py::test_create_feature_flag_usage_dashboard_for_group_targeted_flag
• ee/clickhouse/models/test/test_property.py::test_prop_filter_json_extract_is_set_excludes_explicit_json_null
• products/product_analytics/backend/api/test/test_insight.py::test_not_on_any_dashboard_filter

@Charan-place ran these locally before submitting: <paste your pytest output here>

🤖 Agent context

Autonomy: Human-driven (agent-assisted)

Built with Claude (Cowork session). I located the three issues, implemented each fix, and wrote the tests above; @Charan-place reviewed the changes, ran the tests, and is pushing/owning this PR. No skills/slash-commands were invoked — this was direct code editing.

[sampennington]

Thanks for taking the time to contribute to PostHog! We get a lot of pull requests — to help get this PR reviewed faster, we have some checks in place to help you:

Check	Status	Note
Matches PR template	❌	Missing the "How did you test" section in the description
Logic changes have tests	✅	Test files added for all three fixes
Screenshot or demo of UI change	❌	Frontend files changed but no screenshot or demo in description

Push a change or update the description and we'll automatically re-review the PR. Once all checks are green, the team will be notified.

[greptile-apps]

Prompt To Fix All With AI

Fix the following 1 code review issue. Work through them one at a time, proposing concise fixes.

---

### Issue 1 of 1
ee/clickhouse/models/test/test_property.py:1854-1866
**Denormalized path left untested**

The new test passes `allow_denormalized_props=False`, so it only exercises the JSON-extract branch (`JSONHas(...) AND {left} != 'null'`). The denormalized/materialized-column branch — which was also changed in `prop_filter_json_extract` — is never reached by this test. If a materialized column stores a JSON-null property as the literal string `'null'` (rather than empty string), the denormalized fix matters; but because no test instantiates a materialized column for the `{"email": null}` event, a regression in that branch would pass silently. The existing `test_prop_filter_json_extract_nullable_materialized_is_set_uses_json` skips non-nullable materialized columns, so the gap remains. Adding a second loop iteration with `allow_denormalized_props=True` (after materializing the `email` column) would give the denormalized branch the same coverage as the non-denormalized branch.

_{Reviews (1): Last reviewed commit: "feat: add filter to hide insights alread..." | Re-trigger Greptile}

[greptile-apps]

Denormalized path left untested

The new test passes allow_denormalized_props=False, so it only exercises the JSON-extract branch (JSONHas(...) AND {left} != 'null'). The denormalized/materialized-column branch — which was also changed in prop_filter_json_extract — is never reached by this test. If a materialized column stores a JSON-null property as the literal string 'null' (rather than empty string), the denormalized fix matters; but because no test instantiates a materialized column for the {"email": null} event, a regression in that branch would pass silently. The existing test_prop_filter_json_extract_nullable_materialized_is_set_uses_json skips non-nullable materialized columns, so the gap remains. Adding a second loop iteration with allow_denormalized_props=True (after materializing the email column) would give the denormalized branch the same coverage as the non-denormalized branch.

Prompt To Fix With AI

This is a comment left during a code review.
Path: ee/clickhouse/models/test/test_property.py
Line: 1854-1866

Comment:
**Denormalized path left untested**

The new test passes `allow_denormalized_props=False`, so it only exercises the JSON-extract branch (`JSONHas(...) AND {left} != 'null'`). The denormalized/materialized-column branch — which was also changed in `prop_filter_json_extract` — is never reached by this test. If a materialized column stores a JSON-null property as the literal string `'null'` (rather than empty string), the denormalized fix matters; but because no test instantiates a materialized column for the `{"email": null}` event, a regression in that branch would pass silently. The existing `test_prop_filter_json_extract_nullable_materialized_is_set_uses_json` skips non-nullable materialized columns, so the gap remains. Adding a second loop iteration with `allow_denormalized_props=True` (after materializing the `email` column) would give the denormalized branch the same coverage as the non-denormalized branch.

How can I resolve this? If you propose a fix, please make it concise.

Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!

[sampennington]

Thanks for taking the time to contribute to PostHog! We get a lot of pull requests — to help get this PR reviewed faster, we have some checks in place to help you:

Check	Status	Note
Matches PR template	✅	Title format correct and all sections filled
Logic changes have tests	✅	Tests added alongside each fix
Screenshot or demo	❌	UI files changed but no screenshot found — only a placeholder comment

Push a change or update the description and we'll automatically re-review the PR. Once all checks are green, the team will be notified.

[sampennington]

Thanks for taking the time to contribute to PostHog! We get a lot of pull requests — to help get this PR reviewed faster, we have some checks in place to help you:

Check	Status	Note
Matches PR template	❌	"How did you test" section has an unfilled placeholder (<paste your pytest output here>)
Logic changes have tests	✅	Tests added in three test files
Screenshot or demo of UI change	❌	Frontend UI files changed but no screenshot or demo provided (placeholder comment found)

Push a change or update the description and we'll automatically re-review the PR. Once all checks are green, the team will be notified.

[sampennington]

Thanks for taking the time to contribute to PostHog! ✅ All automated hygiene checks pass:

Check	Status	Note
Matches PR template	✅	Title format correct; Problem, Changes, and How did you test sections all filled in
Logic changes have tests	✅	Tests added for all three fixes in corresponding test files
Screenshot or demo	✅	Screenshot included via GitHub user-attachments img tag

[haacked]

Thanks for the contributions. But its our preference to have a single PR per issue. That way each PR can be reviewed and deployed independently.

[Charan-place]

Yeah Ok Sure sir