Mpc Results

Read back what MPC actually did. Board Round 2 #26, tech design M4-15.

Without this the data accumulates in mpc_pitches and nobody can look at it: rows, and no answer.

Read in this order:

1. rank_distribution — is the matcher better than chance, and does anyone ever click past rank 2? This settles the 3-vs-5 question without splitting traffic.
2. by_arm.avg_cards — the VALIDITY check. If the challenger arm cannot fill its card count, a null result is the harness's fault.
3. guardrails — bounces and opt-outs. ruzora.com carries every outbound channel, so this stops the experiment on its own.
4. coverage — PRD §6.1's independent kill rule: below 50% the assignment logic is broken however good the reply rate looks.

Read-only. No auth, matching the ticks above: it exposes counts and rates, never a candidate name, a company, or an email address.