Multistate indicator selection / table validation fails across most states and sections

[saqibmanan]

Found while verifying #31's CI run. Real, currently-live, and confirmed independent of two other fixes merged the same day.

What's failing

TestMultiStateIndicatorsMapView, TestMultiStateIndicatorsChartView, TestMultiStateIndicatorsTableView, and TestSectionCoverageByState in tests/test_analytics.py — across most states and most sections (Hazard, Exposure, Vulnerability, Government Response), across Map, Chart, and Table views.

Two failure shapes:

Map/Chart — the indicator click itself fails:

AssertionError: ❌ 2/4 indicators FAILED
    • Elderly population: Failed to select indicator
    • Children population: Failed to select indicator

Table — indicator selection succeeds, data validation fails:

AssertionError: ❌ 2/2 indicators FAILED (Table View)
    • Total Monthly Rainfall: Table validation failed
    • Mean intensity of inundation (removing 0 intensity pixels): Table validation failed

Table-view failures are expensive: the shard with the bulk of them ran 2h28m, largely from 61 pytest-rerunfailures reruns — each failure gets retried and fails again.

Confirmed independent of #28 and #29

Both PRs were merged into main (2026-09-08) before the CI run that produced this. This failure pattern reproduced identically in two separate full-regression runs on #31 — one before merging #28/#29 into the branch, one after. So it isn't the disaster-hub navigation issue #28 fixed, and it isn't the self-healing false-positive issue #29 fixed.

Leading theory, not yet checked

There's an existing documented requirement (second-brain memory project_analytics_district_selection) that Chart and Table views need district + revenue circle selected right after select_view(), or they show no data — the dropdowns don't even render until a view is active. Worth checking first whether TestMultiStateIndicatorsChartView/TableView actually do that selection correctly before assuming the indicator locators themselves are stale.

Separately: this PR (#31) did find and fix 6 stale GovtResponseLocators leaf locators — but those were in select_govt_response_option(), a different, previously-dead-code method never called by any test before #31. TestMultiStateIndicatorsMapView etc. use select_indicator_by_text() instead, which has not been audited for the same kind of drift.

Not done

Deliberately not investigated further in this session — flagged and deferred rather than bolted onto an unrelated PR. This card exists so it doesn't get lost.

[saqibmanan]

to be taken in next sprint.