GLO Law

Case-name metadata search over the local CourtListener mirror. Punctuation-insensitive trigram match on the case name ("Brown v board of education" matches "Brown v. Board of Education"). Returns cluster rows: id, case_name, case_name_short, date_filed, precedential_status, citation_count, slug. Metadata only: no opinion text, and not concept search (it finds a case by its name, not by what it is about). Case law in this mirror ends 2026-09-30 (exact scan of all 10,174,547 case records): a thin or empty result for recent authority means the collection ends there, not that no such case exists. Every response carries a "note" stating that response’s source and coverage limits.

Keyword full-text search (BM25) over opinion body text in the local mirror. Keyword, not concept: matches the exact words given (case-insensitive) with no stemming ("petition" does not match "petitions"); all words must appear; "double quotes" mark an exact phrase. A case that states the concept in other words does not appear; semantic_search covers that gap. Coverage (exact scan 2026-07-18): the best extracted text per record is searched; 8,751,062 records carry 500+ characters, 1,993,069 carry 1-499, and 1,798 have no text and cannot match. Returns opinion_id, cluster_id, rank (BM25), case metadata and a matched-text snippet with [hl]...[/hl] around hits. Case law in this mirror ends 2026-09-30 (exact scan of all 10,174,547 case records): a thin or empty result for recent authority means the collection ends there, not that no such case exists. Responses carry body_as_of, body_pending (state, exact count, read_at) and body_note, which say when the stored opinion text is older than the case records while a refresh is written; a count that could not be read makes an empty result inconclusive. A cold index may answer 503 briefly; the server retries once. Every response carries a "note" stating that response’s source and coverage limits.

Semantic (vector) search over embedded opinion chunks: finds cases by concept or plain-language description when there is no fixed term of art. Returns one row per case: opinion_id, cluster_id, metadata, matched_chunk, distance, similarity, matched_text (the passage the case matched on) and matched_text_status; same_case_other_opinions counts other opinions of the same case, and duplicate_records names a second record holding identical text. Limits: ranking covers only a partial embedded subset of the 10,903,917 opinion records (coverage_note gives it), so relevant authority that is not embedded does not appear; similarity is a relevance signal only, not treatment, validity or "good law"; matched_text is re-derived from the opinion text and present only when that reconstruction agrees with the stored chunk count (about 90% of opinions), otherwise null with matched_text_status giving the reason. A matched_text marks where the match happened, not a holding. Case law in this mirror ends 2026-09-30 (exact scan of all 10,174,547 case records): a thin or empty result for recent authority means the collection ends there, not that no such case exists. Degraded state: the route depends on an embedding service that may be down; it then returns a 503 error carrying coverage_note, which means concept search is unavailable, not that nothing matched (fts is a separate route). Every response carries a "note" and a coverage_note. The sources tool returns this tool’s full coverage statement.

Given one opinion id, returns other opinions whose wording overlaps it, ranked by keyword match (BM25 more-like-this). Keyword overlap only: not a finding that the neighbors address the same issue, reach the same result or remain good law, and not complete. The seed opinion and its own case are excluded. Case law in this mirror ends 2026-09-30 (exact scan of all 10,174,547 case records): a thin or empty result for recent authority means the collection ends there, not that no such case exists. Responses carry body_as_of, body_pending (state, exact count, read_at) and body_note, which say when the stored opinion text is older than the case records while a refresh is written; a count that could not be read makes an empty result inconclusive. A cold index may answer 503 briefly; the server retries once. Every response carries a "note" stating that response’s source and coverage limits.

Resolves a reporter citation "<volume> <reporter> <page>" (e.g. 347 U.S. 483) to the matching case cluster(s) in the local mirror, optionally with member opinions and parallel citations. A 404 means the citation is not in this mirror, not that the case does not exist. A resolved citation shows the case is in the mirror; it is not a good-law, treatment or validity signal. One citation can sit on several records: all are returned up to 25, and capped=true means more records carry it than are shown. Every response carries a "note" stating that response’s source and coverage limits. A miss carries one too.

Full case (cluster) detail by cluster id: cluster metadata, member opinions and parallel citations. Metadata only; the opinion tool returns full text. Every response carries a "note" stating that response’s source and coverage limits.

Full text of one opinion by opinion id, plus metadata. The text is the best stored text extracted from the mirror (HTML stripped; body_source names the extraction source and fallback_used flags a lower-precedence one), bounded to 200k characters with truncated and text_available flags. Not live CourtListener, not legal advice, and no treatment, "good law" or citation-validity status. Responses carry body_as_of, body_pending (state, exact count, read_at) and body_note, which say when the stored opinion text is older than the case records while a refresh is written; a count that could not be read makes an empty result inconclusive. body_pending_note, when present, says whether this text is the older build or not yet written. Every response carries a "note" stating that response’s source and coverage limits.

Given one opinion id, lists the cases it cites (direction=authorities) or the cases that cite it (direction=cited_by). Textual reference only: an edge means one opinion’s text names the other; it is not treatment, agreement, validity or "good law", and not a Shepard’s or KeyCite result. At most 1000 references are examined; those resolving to a loaded case are shown (most-cited first), the rest are counted in out_of_corpus_count, and examined_capped=true marks a floor. depth is how often the citing text names the case, not importance. References are per opinion: same_case_opinions gives each opinion of a record its own count and most_references_on names one carrying more (Miranda v. Arizona: record 107252’s combined text 49; lead opinion 9423233, 57,482). also_carried_by names other records sharing the reporter citation, each with its own count; counts are never pooled. The reference table misses citations in the stored text: 22.7% of references on a preregistered sample had no edge (95% interval 18.9-27.0%, n=405), concentrated in some cases, so a thin cited_by list is not evidence a case is rarely cited. with_passages=true (cited_by only) adds citing_passage, the citing court’s own sentence copied verbatim, found only where the citing opinion repeats the reporter citation (first occurrence; at most 10 opinions a request), with passage_note and passage_error. Case law in this mirror ends 2026-09-30 (exact scan of all 10,174,547 case records): a thin or empty result for recent authority means the collection ends there, not that no such case exists. Every response carries a "note" and a coverage_note. The sources tool returns this tool’s full coverage statement.

Given one opinion id, returns the one-line parentheticals later courts wrote about that case, verbatim. A parenthetical is the later court’s own characterisation: not a holding of the described case, not a summary written by this service, and not treatment, validity or "good law". Each item names the opinion that wrote it (case name, date, citation_count). Near-duplicates ship pre-grouped by CourtListener/FLP (group_id); one row per group is returned (highest score, ties by id; that pick is this service’s, and score has no documented meaning), and group_size counts the lines it stands for. Coverage is partial and skewed toward frequently cited opinions (6,561,861 parentheticals): an empty result means none was extracted for this opinion. At most 2000 are examined a request (examined_capped=true marks a floor); unattributable_in_window counts lines whose author is absent from the mirror. same_case_opinions and most_parentheticals_on name a sibling opinion carrying more (Miranda v. Arizona: 107252 has 1; lead opinion 9423233, 1,524). Case law in this mirror ends 2026-09-30 (exact scan of all 10,174,547 case records): a thin or empty result for recent authority means the collection ends there, not that no such case exists. Every response carries a "note" and a coverage_note. The sources tool returns this tool’s full coverage statement.

Later courts, in their own words, reporting that one case was overruled, abrogated, superseded, reversed or vacated. Takes a cluster id (the case-record id), not an opinion id. Every row carries the citing court’s own sentence, cut from its stored text; the label is a filter over those sentences, not a verdict, and 68.7% of findings are expressly partial ("overruled on other grounds"). Not a validity verdict and not Shepardizing: an automated, unverified reading with no positive signal, so zero rows is not evidence that a case remains good law. Rows include known false positives; a read guard (version 2) withholds ten known false shapes and counts them in declined_by_read_guard with read_guard_note. Precision, hand-read after version 2: 0.9862 (95% interval 0.9511-0.9962, 145 of 150 kept rows); 1.0000 (0.9124-1.0000, 40 of 40) for cases cited 10,000+ times. Recall for this collection is unknown; the only figure, 0.6477 (0.5780-0.7116, n=193), is an upper bound on the Supreme Court stratum. The overruling opinion itself is usually not among the rows (Roe carries no row from Dobbs). Five labels only; "followed" and "distinguished" are not loaded. Counts are for this record alone; also_carried_by names sibling records. unrelayable_in_window counts rows dropped because no sentence could be cut; capped=true marks a slice; order is citing-opinion date. Case law in this mirror ends 2026-09-30 (exact scan of all 10,174,547 case records): a thin or empty result for recent authority means the collection ends there, not that no such case exists. Every response carries a "note" and a coverage_note. The sources tool returns this tool’s full coverage statement.

Checks the case citations in a document the user supplies (a draft filing, brief or motion) against this collection and court-reported negative history, showing which cited cases to read before filing. Nothing is ever cleared: there is no pass mark or "verified" state. Per citation: reported_negative_history (a later opinion’s sentence matched as reporting it overruled, abrogated, superseded, reversed or vacated, in records[].reported_by; unverified, with known false positives), no_row_found (nothing found, not a clearance), not_in_collection (echoed as written, never corrected or remapped). Short forms (Id., supra, bare names) are counted in short_form_counts, not checked; statutes, rules and secondary sources are not checked. quotes[]: quoted words near a citation compared with the stored text, status found (pin cite and support not checked), not_found_in_stored_text (with stored_passage and agrees_for_words; not a finding of invention), no_text_stored or not_checked; block quotations are not checked; the join is a nearest-neighbour guess (attribution_method, attribution_distance). protected_info: a pattern scan for identifiers court rules bar from public filings (Fed. R. Civ. P. 5.2(a), Kan. S. Ct. R. 24(b)), values masked, with category, rule_refs and offsets; a finding is not a ruling, zero findings is not a clearance, and not_detectable lists what it cannot see. Bounds: 200 citations, 25 detailed flags (flagged_without_detail), citations_truncated, multiple_records, zero_result_detail. The document is not stored or logged. Negative-history limits are those of the treatment tool. Case law in this mirror ends 2026-09-30 (exact scan of all 10,174,547 case records): a thin or empty result for recent authority means the collection ends there, not that no such case exists. Every response carries a "headline", a "note" and a coverage_note. The sources tool returns this tool’s full coverage statement.

Exact lookup of one statute section or one regulation section by citation, in a point-in-time snapshot of published law. Returns the text verbatim with jurisdiction, corpus, citation, heading, path, source_url, as_of, act_status, last_amended_year and repaired. Enacted text only: not case law or annotations, no treatment or "good law" judgment, and nothing in a result means the text is current; as_of on each row is its snapshot date. Default source: 2,123,652 rows in four snapshots, never blended (state statutes from open-us-law 2026.08, as_of 2026-08-14; state statutes from Justia 2019-2021 where the newer snapshot lacks the section, with GA, NC, PR and VI Justia-only; the United States Code; and 9,888 Kansas Administrative Regulations from the 2022 bound edition). source=<code> reaches one regulation source (cfr for federal, or a state code), each with its own note, coverage_note, defect floor, held units and citation form, never mixed with another. Regulations outside the loaded sources are absent from this mirror, which says nothing about whether a rule exists. A miss is not an absence: structured_lookup_supported=false means that lookup form cannot match for the jurisdiction, and cite_normalized and cite_forms_tried report the forms tried. Row flags: ends_at_colon (text stopping where a table or list is announced; its absence does not show the text is whole), page_overlap_repeats_removed (raw=true returns the stored text unchanged), last_amended_year_older_than_text. A held unit answers kind "held" with its reasons and no text. Every response, including 400 and 404, carries "note" and "coverage_note". The sources tool returns this tool’s full coverage statement and, per source code, its coverage, citation forms, served and held counts and read dates.

Keyword (BM25) search over the body text of statute sections and, in the default source, Kansas administrative regulations, in the same snapshot as the statute tool; source=<code> searches one regulation source alone. Without jurisdiction=, results mix states: on 6,852 benchmark housing-law questions run unconfined, the top-ranked section came from a state other than the asker’s in 97.7% of them. Each row carries its jurisdiction, corpus and as_of; statute and regulation rows can appear together. Rank is textual match strength only. Keyword, not concept, with no stemming: very common words are dropped, duplicates collapse, at most 40 words run, matched any-of, so a top hit may share one word with the query; terms_searched and terms_dropped report what ran, and "double quotes" require a phrase. Only the body is searched, not the heading. Scores drift in their last digits between restarts, so close rows can swap places. Returns metadata, a snippet with [hl]...[/hl] marks and a rank; the statute tool returns the full section. A section not returned is a miss, not an absence. ends_at_colon is flagged as in the statute tool. A missing keyword index returns a visible 503, not zero matches. Every response, including 400 and 503, carries "note" and "coverage_note". The sources tool returns this tool’s full coverage statement and, per source code, its coverage and counts.

Lists the opinions whose text cites one U.S. Code section. Federal only: no state statute-citation links exist, and a state citation returns a visible error. An entry means the opinion’s text names the section; it does not say how the court treated it and carries no treatment, validity or "good law" judgment. Zero results is the majority outcome: only 25,093 of 60,444 parseable federal sections (41.5%) have any extracted citing opinion, so an empty list is not evidence that no case cites the statute. Order is most recently filed first and nothing else. The scan covered 10,745,929 opinion records and predates the 2026-09-30 refresh. Every figure is a lower bound with named extraction gaps (no section marker, subsections dropped, only the first section of a list, letter-suffixed sections, ranges). A trailing subsection given is stripped and reported in subsection_dropped. Resolution rates: occurrence-level 0.8870, distinct-pair 0.1697; resolution is not correctness. Precision on a 120-span blind sample: 0.9845 (90% Wilson 0.9526-0.9950); the second extractor alone 0.8667 (0.7785-0.9232); recall unmeasured. parentheticals[] carries any court-written parenthetical verbatim (95.8% of citations have none). section_in_snapshot=false with snapshot_detail marks a section absent from the U.S. Code snapshot; the citation is shown as written, never remapped. Every response, including 400 and 503, carries "note" and "coverage_note". The sources tool returns this tool’s full coverage statement.

Internal policies of a state department of corrections, or of the Federal Bureau of Prisons (state="bop", federal), as the department posts them, by number or by keyword search (query; no stemming). Not law: not statutes, regulations or court rules (a Maine policy the department marks as adopted as a rule answers apa_marked true; for Kentucky GLO makes no claim about legal effect either way). A department can change its policies without notice, and policies it does not post are not here. An answer gives the number, title, the dates the file prints, the version replaced where the file names one, the date GLO read the file, and the department’s coverage_note; a revoked policy is listed with the successor its notice names; a number the department lists but does not post answers not posted; a held number answers its status and reason, never text. Departments loaded: one code each in the state enum, among them "ks-go" (Kansas facility General Orders); many are partial, with served and held counts per department. "hi" (Hawaii) serves text for only 40 of 204 posted files, machine-read with no human read; every other row of its list answers its reason, never text. A state absent from the enum is not loaded, which says nothing about whether its department has such a policy. The sources tool returns, per state code, the department’s source, read date, served and held counts, held reasons and number forms. Give number or query, not both.

Counts a deadline from a date the user gives, under a rule whose text is in GLO’s mirror, and returns every counting step beside that text. Counted: K.S.A. 60-206 (Kansas civil procedure; a period in days the user supplies from the rule or order that sets it, forward or backward, with three days added after service by mail or by leaving it with the clerk) and K.A.R. 44-15-101b (Kansas Department of Corrections grievances: 15 days from discovery excluding Saturdays, Sundays and holidays, never later than one year after the event), Fed. R. Civ. P. 6 and Fed. R. App. P. 26 (federal civil and appellate time, a period in days the user supplies, with the three days 6(d) and 26(c) add) and 28 U.S.C. 2107(a) (thirty days after entry, counted by FRAP 26(a)). Where the text can be read more than one way, each reading is counted with its own date and earliest is the date safe under every reading. Not counted: periods in hours. The Kansas Supreme Court’s holiday orders are held for 2025 to 2027 only, counted for K.S.A. 60-206 as one reading beside a reading without them; other orders, days declared by order, and states’ holidays other than Kansas’s list are not in the mirror. Returns status, readings (id, label, date, steps), earliest, latest, warnings (each applies to the date given), the rule row (citation, verbatim text, corpus, as_of, source_url, ambiguities, exceptions, last_verified), detail, query and a note. Pure computation from the stored rule text, with no database query. Counting is information, not legal advice. The sources tool returns this tool’s full coverage statement.

Returns GLO’s coverage statements as text, so each tool’s and each source’s limits can be read on request. With a tool name alone: that tool’s full statement of coverage, measured limits and field meanings ("server" gives the full server overview). With a source or state code: for statute and statute_search, that regulation source’s coverage, citation forms, served and held counts and read dates (e.g. "cfr", "fl", "de"); for corrections_policy, that department’s (e.g. "ks", "bop"). Returns tool, source, text, argument_descriptions, sources_available and a note. Read-only and served from the tool layer itself, with no database query. Each data tool’s own response note and coverage_note remain the record for that response.