kausalflow/lightkone / components/search-query/versions/1/symbolon.toml · raw

[contract]
id = "lightkone/search-query"
version = 1
title = "Search a paper library with lightkone's query language"
description = """
Filters and orders a list of papers by a query string, as lightkone's search
box does. Returns the ids of the matching papers, in display order.

Query terms are separated by whitespace. A term is one of:

  field:value or field:"quoted value"   a filter (field names are ASCII
                                         letters, digits and _, any case)
  "quoted phrase"                        one word that may contain spaces
  anything else                          one word

Filters (all must hold, except year:, where only the last year: term that
counts applies):
  tag:T       some tag of the paper equals T after folding
  author:A    A is a substring of the folded authors joined with " ; "
  year:Y      year equals Y; year:A-B: A <= year <= B (missing year never matches).
              The value is split at "-": A is the first part, B the second,
              later parts are ignored. When A is not a number, or is 0, the
              term does not count (an earlier year: term stays in force).
              When B is missing, empty, 0 or not a number, B is A. So
              year:2017- means 2017 and year:-2017 does not count. A number
              here is ASCII digits with an optional fraction; other
              spellings are unspecified.
  in:F        the paper is in folder F or a folder inside it ("F/..."),
              compared after folding; trailing slashes on F are ignored
  has:X / no:X  for X in pdf, abstract, arxiv, notes. has:pdf accepts either
              pdf or file; no:pdf looks at pdf only. Unknown X is ignored.
  is:starred, is:recent (has been opened), is:trash. Unknown values are ignored.
Any other field:value term is an ordinary word, including the colon. So is a
field with nothing after the colon (author: is the word author:), and an
unknown field with a quoted value is one word, quotes included: o:"a b".

Trashed papers appear only with is:trash, and then only trashed papers appear.

Words: every word must be a substring of the paper's folded title (5 points),
folded author text (4 points) or folded rest (1 point): abstract, venue,
tags joined by spaces, key, doi, arxiv and notes, the non-empty ones joined
by single spaces. A paper's score is the sum over words; a paper where any
word scores 0 is left out.

Folding: Unicode NFKD, remove U+0300..U+036F, then lower case. Values of
has:, no: and is: are only lower-cased. Empty strings, false, and missing
fields are all "absent".

Order. sort=added: newest added first, except that a query with is:recent
orders by opened, and otherwise a query with words orders by relevance.
relevance: higher score first, then newest added. year: higher year first,
missing as 0. title: ascending by the Unicode Collation Algorithm
with the CLDR root collation at its default (tertiary) strength, which is
what JavaScript's a.localeCompare(b) does with no locale or options: letters
first, then accents, then case, lower case first ("apple" < "Apple" <
"Äpple" < "banana"); punctuation and spaces are not ignored and sort before
digits, digits before letters. opened: latest opened first, never-opened
last, then newest added. Papers that compare equal keep their input order.

Full-text search is out of scope.
"""
regenerability = "disposable"
license = "Proprietary (kausalflow)"
authors = ["kausalflow/lightkone"]

[entrypoint]
function = "run"

[requires]
network = []
credentials = []
state = []
permissions = []

[spec]
seed = 0
max_examples = 150
timeout_seconds = 300