Council Round-3 citation source-diversity gate

In the Round-3 verdict parser (after each judge's verdict JSON is returned, around the citation-validation step), in addition to enforcing the existing 'minimum 3 citations' rule, also enforce that the cited deep-report sources span ≥2 distinct report types from the 9 deep_investigation report families (market_reality, competitor, icp, buyer_sim, moat, pricing, distribution, risk, 90_day_plan). If a judge's citations all reference the same report type, mark that judge's R3 verdict as ok=false with reason='single_source_grounding' and exclude it from the final tally — drop to 2-judge majority.

Judges that anchor their entire R3 verdict on (e.g.) only 'risk' will be excluded. Predicted effect: ~10-20% of R3 verdicts that survive the citation-count check fail source-diversity. Net: STRONG_BUILD/KILL escalations from R3 will be drawn from a more grounded set of judges, reducing the rate at which a single deep-report dominates a council outcome.

Sample 20 historical R3 verdicts (re-parse from moves.output_json). If <5% would have failed source-diversity, the gate is a no-op and shouldn't ship. If >40% would have failed, the gate kills too many R3s and the council degenerates to 2-judge mode habitually, which itself is a load-bearing change — reject and revisit.

Rejected by filter_score. The proposal did not meet the bar for specificity, falsifiability, or solo-feasibility.