L2 — Player vs Bowler-Type × Phase

How a batter performs against each of 6 bowler types (pace, leg-spin, off-spin, left-arm orthodox, left-arm wrist-spin, left-arm pace) within each of the 3 phases (powerplay, middle, death).

Period: IPL 2026. Sample-size floor: ≥20 balls faced. Suppressed cells render the URI with the parent L1 link instead of metrics.

Conditioning dimensions

bowler_type — pre-classified from each ball's recorded bowling style on the live ball-by-ball feed (per-delivery, not the player roster). Falls back to "unknown" for unparseable styles; unknown is surfaced as a separate bucket, never silently merged.
phase — powerplay (overs 1-6), middle (7-15), death (16-20). Classified from each ball's over-position on the live feed.

Sample-size floor + rationale

≥20 balls faced per cell.

Smallest cell in the 6 × 3 grid; 20 balls faced is the doctrine §3.1 batting floor relaxed for the more granular conditioning.

Suppression rule

URIs stable across versions per the extension doc §5.2. Below floor: page renders the suppression message, MCP + REST APIs return `suppressed: true` + `metrics: null`. Above floor: positive claim with metrics.

Update cadence

Recomputed after every captured fixture (auto-capture cron) + daily full rebuild as a safety net.

Known limitations

Phase A period is single-season IPL 2026. Sample sizes roughly double when Phase B widens the period to include IPL 2025 historical (CricketMind integration).
Bowler "unknown" bucket captures unparseable styles — surfaced separately so the share is visible.
§10.5 distribution check: ~16.5% of cells clear the ≥20 balls floor in the Phase A corpus. Remaining cells render as suppressed URIs (URI stability preserved per §5.2).

Change log

2026-05-13 · Phase 1 launch. Period: IPL 2026 only. Floor: ≥20 balls faced.

Consumed by

UC1: /players/{slug}/vs/{bowler-type}/phase/{phase}/