COMPO LAB · LAB-008 · VERSION 1.0

Candidate
Pipeline.

Проходит ли bounded candidate population через явный deterministic funnel без Cartesian explosion?

STATUS: SUPPORTED · 1000 GENERATED · 117 AFTER DIVERSITY PRUNING

Research question

Проверить pipeline semantic binding → local measurements → hard filter → soft scoring → deduplication → diversity pruning с seed и lineage для каждого кандидата.

Actual funnel

1000 generated → 950 semantic-bound → 815 hard-valid → 815 soft-scored → 586 deduplicated → 117 diversity-pruned.

Lineage completeness: 100%. Peak candidate budget: 1000. Same seed повторяем; другой seed меняет digest.

{"status":"SUPPORTED","repeatable_same_seed":true,"different_seed_changes_digest":true,"lineage_complete":true,"peak_candidates":1000}

Interpretation

Эксперимент подтверждает воспроизводимость и порядок стадий на synthetic pipeline. Это не утверждение production thresholds: размеры funnel и pruning ratio остаются исследовательскими параметрами.

Canonical impact

CANONICAL_CHANGE_REQUIRED = NO. Canonical modules не изменялись.

Artifacts

LAB_SPEC.jsonassertions.jsonoutput/summary.jsonoutput/raw_results.csvoutput/raw_results.jsonrun_lab_008.ps1Related: Compo 7.06
VISITS / ...