Govtrack

ask_pipeworx, ask_pipeworx_beta, and ask_pipeworx_grounded are near-duplicate routers — the beta currently behaves identically to stable, creating a genuine misselection trap. The six polymarket tools (arbitrage, edges, bet_research, fill_risk, kalshi_spread, edge_tracker) also have fuzzy boundaries, with several overlapping on 'find a trade opportunity'.

Most names follow a consistent snake_case verb-first convention (resolve_entity, validate_claim, compare_entities, search_within) and the polymarket_*/pipeworx_* family prefixes are regular. But there are deviations: ask_pipeworx breaks the underscore pattern, bet_research drops the polymarket_* prefix despite belonging to that family, and ai_visibility_check / generate_llms_txt deviate from verb-first naming.

At 31 tools this exceeds the 25+ threshold, and much of the surface is meta-tooling (discover_tools, suggest_questions, pipeworx_feedback, pipeworx_trending), memory utilities (remember/recall/forget), and subscription management rather than core data access. The 5,621-tool routing breadth justifies size, but the count is bloated by utilities peripheral to the stated purpose.

The primary research lifecycle is well covered: resolve_entity → ask_pipeworx/deep_research → entity_profile → compare_entities → recent_changes → validate_claim/ask_pipeworx_grounded → search_within, with stable citations throughout. Prediction-market workflows are likewise complete (research, edges, arbitrage, fill-risk, cross-venue spread, persistence). No obvious dead-ends for the core purpose — the gaps are off-domain additions (llms.txt generation, npm scanning) that add bloat rather than leave holes.