Get In Focus

Curated 4-source-group × 5-category grid for the homepage.

Reads the pre-materialized InFocusOut payload from in_focus_grid (single-row table, refreshed by the scheduler every ~5 min via compute_and_store_in_focus_grid). One indexed row fetch + Pydantic JSON validation = ~5ms backend latency.

See compute_and_store_in_focus_grid for the writer side and the _IN_FOCUS_SQL / _PULSE_SQL doc-blocks for the curation algorithm:

Two-pass fill:

1. Priority-5 candidates (_IN_FOCUS_SQL). For every (group, category) cell, the top-N markets (by score = |Δprob| × cf) whose parent event carries that category. Same logic for all four strips.

2. Pulse extension (_PULSE_SQL, 'others' + 'manifold' strips). Pulse rows fill empty cells in those two strips with the next-best market from the same source(s) whose parent event sits OUTSIDE the priority 5 (Health, Economics, Climate, etc.). Drained via an empty-first policy — each pulse market goes to the cell with the fewest current candidates — so a strip with 15 pulse markets and 5 cells distributes ~3 per cell rather than piling 10 into cell 0.

   Pulse is NOT applied to Polymarket or Kalshi. Their priority-5 coverage is dense enough that a missing cell is signal ("nothing in Elections moved >2pp on Polymarket today") rather than a layout bug. Manifold is pulse-extended because, despite high market volume overall, its category mix skews away from Culture/World — the strip would otherwise leave those cells empty.

Both queries respect the _IN_FOCUS_BLOCKED_QUESTION_PATTERNS title kill-switch — see that constant for the list and how to extend it.