[dinethlive]

Brings all nine examples onto Lumin's current tool surface. Every route was run live against the real server (Sonnet 5.5) and returned 200 with toolsFailed=0.

Why

A first live pass of all 12 routes found two failures:

• horary-desk /api/ask: Lumin now rejects a place of judgement without judgement_datetime. The prompt said "Moment of judgment: now" and documented the old birth-style fields, so the call failed. The route now fixes the moment server-side and every horary tool takes the judgement_* fields.
• health-risk-analyzer: output was cut off at 24,000 tokens. The budget is now 128,000 (streamed, so unused room costs nothing).

The server also renamed twelve _advanced tools (old names work for one more release) and split bundles into step tools. This PR moves every app off the old names and onto step tools where an app reads one block.

Per app

App	Change
horary-desk	judgement_* fields everywhere, including get_horary_serial sessions. get_horary_number_intuition replaces the renamed validator. Lost questions use the four lost-article step tools. New cards: the number methods side by side, and both ruling-planet definitions.
health-risk-analyzer	get_csl_condition, get_surgery_dasha_periods, get_chronic_recovery, get_malignancy_signatures replace bundles. get_paradigm_panel replaces the tallying get_multi_system_verdict.
kundli-match	get_couple_csl_factors + get_couple_dasha_sync replace get_compatibility_advanced, with no overall score (the old one was an unlabelled average). Adds Punarphoo and KSK's discord rule. Ashta Koota now scores the male chart as the groom, as the tool defines it.
weather-windows	The channel scores are raw signed numbers, not 0 to 100; the old app rescaled them and showed invented levels and an outdoor rating. It now shows the engine's bands and dispositions. Season banner reads get_weather_csl_fourth; each window shows Sapta Nadi.
muhurta-scheduler	find_auspicious_triangulation replaces get_muhurta_advanced, read to the last page.
today-panel	New ruling-planets card from get_ruling_planets_at_judgement.
products-matcher, wellness-matcher	get_nakshatra_pada and get_shadbala_index; the 0 to 100 strength is labelled as Lumin's index, not the BPHS Shadbala.
career-fit	No tool changes; the prompt's description of run_pre_verdict_audit now matches what the tool computes.

Shared: the run log named every tool unknown (the result block carries only tool_use_id); it now names each one. scripts/tool-names.json is regenerated from a live tools/list (240 to 315). Root docs drop the renamed names.

Checks

• bun run check:tools: no unknown names (315 known)
• bun run typecheck: all ten packages exit 0
• bun run build: all nine apps exit 0
• Live: all 12 routes 200, toolsFailed=0. today and elect re-run on fresh production servers after the last edits.

Known open items

• health-risk-analyzer is heavy: about 45 tool calls and roughly 950k input tokens per reading. Prompt caching would cut that.
• horary-desk: a lost question whose gate answers (rather than withholds) pages through about 10 calls and has not been run live.
• muhurta-scheduler README still has typed counts, which CONTRIBUTING forbids.
• today-panel (predates this PR): on a date other than today, the Choghadiya and hora "current" period is still today's.

🤖 Generated with Claude Code