cytado

Call to check progress of a background harvest started by check_corpus with provision:true. Free. Returns status (queued | running | done | failed | unknown), a ready flag and current coverage. Poll every 30-60 seconds; when ready is true, call ground_citation or get_chapter_sources again — the corpus now includes the harvested sources. Pass the SAME topic/outline/langs you used in check_corpus, or coverage will be computed for a different topic string.

check_harvest_status

FREE preview by default: call BEFORE ground_citation or get_chapter_sources to learn whether the corpus already covers a topic (coverage counts) and whether a full run would trigger a paid harvest. With provision:true it actually starts a background harvest job when coverage is insufficient and returns a jobId — poll it with check_harvest_status (jobs take 5-30 min). Billing: preview (provision omitted or false) is always free; provision:true charges per topic ONLY if a harvest actually starts (sufficient coverage = nothing charged, no job).

check_corpus

Call when writing a chapter or section of an academic text and you need several sources with page-tagged text excerpts to cite from. Returns up to `limit` sources, each with metadata (title, authors, year, DOI/URL) and excerpts marked with printed page numbers (=== STRONA N ===), plus a `paginated` flag (can this source support a page-precise footnote), a `tier` flag (wprost = directly on topic, tlo = background) and `wsparcie` — does the returned excerpt ITSELF support the chapter (wprost/posrednie/brak, null = not assessed); cite freely from wprost, read-and-qualify posrednie, and do not hang chapter claims on brak. Billing: charged per NEW document returned (already-charged and background documents are free); zero sources costs nothing. Use check_corpus first if you want to know coverage before calling.

get_chapter_sources

Call when you have factual claims that need real academic citations with printed page numbers. For each claim, returns a verified source (title, authors, year, DOI/URL, page number, short quote) or found:false — this tool never invents sources. Resolves from an existing corpus in seconds. WORKFLOW — call this BEFORE writing the final text: (1) plan the claims your text will make, (2) ground them with this tool, (3) write the text ONCE, with footnote markers (¹, ², …) already at the grounded sentences and the full references underneath (authors, title, journal/publisher, year, page — e.g. “A. Kowalska, Tytuł pracy, „Czasopismo” 2023, s. 45.”). Do NOT write a draft without citations and then rewrite it — for long texts (chapters, multi-paragraph sections) ground claims section by section and write each section once, with its footnotes in place. Never show the user grounded text without visible citations. If claims come back found:false, the corpus does not cover them yet: call check_corpus with provision:true for the underlying topic, wait for the harvest job (5-30 min, poll with check_harvest_status), then call this tool again. Each source carries `wsparcie` — does the returned snippet ITSELF back the claim: "wprost" (snippet states it — cite as-is), "posrednie" (same phenomenon, context only — read before citing and qualify), "brak" (snippet does not state it — treat like found:false for this exact sentence and pick another source or rephrase), null (not assessed). Billing: only resolved claims are charged (per citation); unresolved ones are free. The same document for the same topic is never charged twice.

ground_citation