{"server":"io.github.SamYatesSmith/tru8","count":1,"changes":[{"tool":"tru8_check","kind":"changed","observed_at":"2026-10-12T12:05:26.702Z","fields":["description"],"description_before":"Evidence research for a factual claim or article URL.\n\n    Submit a claim as text or paste an article URL. URLs are auto-detected\n    and trigger article mode: the pipeline extracts claims from the page\n    and auto-selects up to 5 for evidence research.\n\n    Tiers (in fallback order):\n    - lookup    (~£0.02, instant) — cached prior analysis\n    - consensus (~£0.03, instant) — cross-user aggregate landscape (k≥3 checks)\n    - quick     (~£0.07, ~15s) — web search + heuristic classification\n    - full      (~£0.15, ~60-90s) — web + specialist APIs, LLM classification, coverage recovery\n\n    Charges based on tier actually executed, not tier requested.\n    Identical calls within ten minutes return the same check, so a\n    transport retry never runs — or charges — twice.\n    Set max_tier to control maximum spend per call.\n\n    Output structure:\n    - claims[].claimMap.elements[] — verifiable sub-claims with state\n      (supported/disputed/unresolved)\n    - claims[].claimMap.elements[].evidenceRefs[] — evidence mapped to elements\n      with relationship (supports/challenges/context) and reasoning\n    - claims[].evidence[] — sources classified by tier (primary/reporting/\n      commentary) and type (data/official/news/analysis/opinion/academic)\n    - claims[].claimMap.orientation — mechanical summary from element states\n    - _meta — execution metadata: executedTier, chargedPence, limitations\n\n    Per-argument guidance lives on the parameters themselves (see the\n    Annotated/Field declarations above), which is where a client can actually\n    read it — a docstring \"Args:\" block never reaches the tool's inputSchema.\n    ","description_after":"Evidence research for a factual claim or article URL.\n\n    Submit a claim as text or paste an article URL. URLs are auto-detected\n    and trigger article mode: the pipeline extracts claims from the page\n    and auto-selects up to 5 for evidence research.\n\n    Tiers (in fallback order):\n    - lookup    (~£0.02, instant) — cached prior analysis\n    - consensus (~£0.03, instant) — cross-user aggregate landscape (k≥3 checks)\n    - quick     (~£0.07, ~15s) — web search + heuristic classification\n    - full      (~£0.15, ~60-90s) — web + specialist APIs, LLM classification, coverage recovery\n\n    Charges based on tier actually executed, not tier requested.\n    Identical calls within ten minutes return the same check, so a\n    transport retry never runs — or charges — twice.\n    Set max_tier to control maximum spend per call.\n\n    Output structure:\n    - claims[].claimMap.elements[] — verifiable sub-claims with state\n      (supported/disputed/unresolved)\n    - claims[].claimMap.elements[].evidenceRefs[] — evidence mapped to elements\n      with relationship (supports/challenges/context) and reasoning\n    - claims[].evidence[] — sources classified by tier (primary/reporting/\n      commentary) and type (data/official/news/analysis/opinion/academic)\n    - claims[].claimMap.orientation — mechanical summary from element states\n    - _meta — execution metadata: executedTier, chargedPence, limitations\n\n    A run that has not finished after several minutes returns\n    {checkId, status, chargedPence, message} instead of the result: fetch it\n    later with tru8_get_result (free). A run that FAILS raises an error and is\n    not charged.\n\n    Per-argument guidance lives on the parameters themselves (see the\n    Annotated/Field declarations above), which is where a client can actually\n    read it — a docstring \"Args:\" block never reaches the tool's inputSchema.\n    ","description_similarity":0.88,"before_hash":"3c91e9af7b43f5cc797e66cd6d2dcf1f","after_hash":"143c0bce6c835cd2468a4ce0dc703948"}],"next_before":null,"next_actions":[]}