Fields and typos

A shopper's words are looked for in the fields their type lists, in priority order, and a long word is found even with a typo. Here is how the home store searches its products:

curl -s -H "Authorization: Bearer $key" $api/search-settings \
  | jq -c '.types[] | select(.type == "product") | .fields.source, [.fields.fields[].field], .typos'

"set"
["title","brand","keywords","attributes.color","attributes.material","attributes.upholstery","attributes.style","attributes.model_number","attributes.gtin","description"]
{"source":"built_in","enabled":true,"one_typo":5,"two_typos":9}

A word found in an earlier field ranks a hit higher, so the title comes first. Each setting's source says where it comes from: set in types.json, derived from the catalog, or built_in, OrbSearch's default.

Reference: search · GET /api/v1/search-settings

TyposCopy Anchor Link

A word of 5 letters or more finds what holds it with one typo, and from 9 letters with two. A typo is a letter added, left out or changed, or two neighbours swapped; one at the first letter counts as two. "Teppcih" swaps two letters of "Teppich" (rug), and the trace's rank step names the typo:

curl -s 'http://127.0.0.1:7800/v1/search?query=Teppcih&debug=1' \
  | jq -r '.trace.steps[] | select(.step == "rank") | .sections[] | select(.type == "product")
      | .hits[0] | .entity, (.why[] | select(.because.kind == "text") | .sentence)'

product:TEPPICH-MARE
Found by every word of the query, with 1 typo (“teppcih” as “Teppich”), in title, description and words.

words holds what the index adds, such as the titles of a product's categories. A shorter word finds only as written or by its start, so "Lnud" does not find the Lund sofa.

Reference: typos · rank

Identifiers exempt from typosCopy Anchor Link

A part number one character off names another product, so the words of an attribute in no_typos are found only as written or by their start. Where a type leaves no_typos out, every index run derives it from the catalog, with a reason for each identifier the type searches:

curl -s -H "Authorization: Bearer $key" $api/search-settings \
  | jq -c '.types[] | select(.type == "product") | .exempt | .source, (.open[] | {attribute, reason})'

"derived"
{"attribute":"model_number","reason":{"by":"few","values":0}}
{"attribute":"gtin","reason":{"by":"few","values":0}}

An identifier is exempt as unique when 90 % of at least 10 values belong to one product alone. Values products share, such as cross-reference numbers, stay open as shared, and too few to tell as few: the home store's catalog holds no model numbers or GTINs.

An exemption covers the attribute's own words; the same word in a title still finds with a typo.

Reference: no_typos

A word the search does not look forCopy Anchor Link

When a product holds the shopper's word in a field the search skips, POST $api/preview/find says so, as the playground's Why isn't it here? does. The Fjord bed's finish is "Geölt" (oiled), but the product type does not search its finish:

curl -s -H "Authorization: Bearer $key" -H 'Content-Type: application/json' \
  -d '{"request": {"query": "geölt"}, "entity": "product:BETT-FJORD"}' $api/preview/find \
  | jq -c '.verdict.held[], .next[0]'

{"word":"geolt","field":"attributes.finish","holds":"Geölt","why":"not_searched"}
{"step":"open_search_settings","type":"product"}

held names the word as the engine reads it, the field that holds it and why: not_searched, or exempt_from_typos for an exempt attribute that holds it a typo away. The first next step opens the type's search settings.

Reference: held · POST /api/v1/preview/find

Change how a type is searchedCopy Anchor Link

Changes go into a draft, which you publish when it is right. Here the product type's words forgive a typo from 4 letters on.

draft=$(curl -s -X POST -H "Authorization: Bearer $key" -H 'Content-Type: application/x-ndjson' \
  --data-binary @tests/fixtures/home/catalog.jsonl "$api/imports?name=catalog.jsonl" | jq -r .import.id)
version=$(curl -s -H "Authorization: Bearer $key" $api/imports/$draft/search-settings | jq -r .version)
jq -n --arg version "$version" '{version: $version, typos: {enabled: true, one_typo: 4, two_typos: 9}}' \
  | curl -s -X PATCH -H "Authorization: Bearer $key" -H 'Content-Type: application/json' -d @- \
    $api/imports/$draft/search-settings/product \
  | jq -c '.settings.types[] | select(.type == "product") | .typos'

{"source":"set","enabled":true,"one_typo":4,"two_typos":9}

curl -s -H "Authorization: Bearer $key" $api/imports/$draft/publication | jq -r .course.sentence
curl -s -X DELETE -H "Authorization: Bearer $key" $api/imports/$draft | jq -r .state

Goes live within moments: types.json changed only settings the engine applies without reindexing.
discarded

How it goes liveCopy Anchor Link

The order of the fields and the typos are settings the engine takes at once. The rest changes what the index holds for every product:

The shop answers with what is live while a rebuild runs (Releases). A preview searches the live index, so a draft's typo settings show only once it is live.

NextCopy Anchor Link

• Synonyms: words that find each other, such as "couch" and "sofa".
• How search understands a query: how the words left as text are matched.
• Releases: draft, preview, publish and roll back.