Remove probe sequential species experiment; keep probe pick cap

Tune probe selection budget and optional sequential species pass

Consolidate probe configs and eval artifacts on main

Polish Gradio UI layout/tooltips and remove dead helper code

Speed up retrieval exact/alias fasttext path and rerun n30 eval

Add n30 caption-evident set and per-group display blend overrides

Switch Stage3 to explicit-only no-why selection, drop bear probe, and set k=1 defaults

Consolidate pending pipeline, structural, and analysis updates

Remove dead retrieval/display helpers and simplify debug paths

Remove Stage3 bootstrap rerank and add experiment memory log

Simplify Stage3 chunking to interleave-only and add eval diagnostics

Add eval audit tools, caption-evident set, and logging

Record non-fatal pipeline issues in eval JSONL outputs

Fix constraints: body_type, species, gender are multi-select

Add tag categorization pipeline for e621 checklist

Improve topless/bottomless definitions to prevent confusion

Redesign structural inference as group-based system with wiki data

Rewrite structural inference prompt for better Llama 3.1 8B performance

Add structural tag inference (Stage 3s) and compact eval output

Default min_why to strong_implied; add retrieval gap analysis script

Normalize GT annotations: expand implications, exclude non-evaluable tags

Add tag implication expansion (fox→canine→canid→mammal)

Add --min-why threshold to filter Stage 3 selections by confidence level

Add diagnostic eval metrics, why-distribution tracking, and generic character filter

Add alias-based character tag filtering for Stage 3