These rules from the Relativity/dtSearch documentation can't be fully verified from term text alone. Use them as a checklist while drafting. Every item below cites the source page in the official Relativity documentation.
Query design
- Use several specific STRs instead of one broad query with many OR’d concepts. Broad queries are hard to review, hard to validate against a Dictionary, and often over-inclusive. [STR overview]
- Minimize the number of saved-search field conditions layered on top of the STR. Every condition adds a place where the STR’s intent can be quietly narrowed. [Create STR]
- Consider smaller, targeted indexes (by custodian, date range, or file type) rather than one large index. Smaller indexes validate faster and behave more predictably. [dtSearch overview]
- Missing highlights in a hit document do not mean the search failed. Highlighting is a display artifact and can lag or be omitted for complex proximity results. [Running a dtSearch]
- Ignore isolated highlights inside proximity-hit documents — the proximity operator groups matches internally, but the highlighter shows every individual token. [Syntax options]
Validation workflow
- Use the Dictionary to validate a term against actual indexed tokens. Dictionary is capped at ~2,000 unique expansions per query — broad wildcards and stemming can silently exceed this. [Dictionary search]
- Prefer Dictionary validation over broad wildcards. If a wildcard expands to more than the Dictionary can show, you don’t truly know what it will match. [Dictionary search]
- Validate wildcard, stemming, fuzzy, and phonic expansions in the Dictionary before running the STR against production data. [Syntax options]
- Enable auto-recognition for
date(), mail(), and creditcard() pattern types before you rely on them, but only when necessary — auto-recognition slows indexing and increases index size. [Syntax options]
- Record: which index/searchable set was used, the alphabet file version, the noise word list, the STR terms as run, and the run timestamp — for defensibility. [Create STR]
Index-dependent behavior
- The Fuzziness Level control is not supported in STR searches. Use
word% notation instead if you need approximate matching. [Syntax options: fuzzy]
- Accent sensitivity depends on how the index was built. This validator cannot verify accent handling from term text alone. [Alphabet file]
- Underscore does not create a word break by default.
foo_bar is indexed as one token, not two. [Alphabet file]
- Reserved symbols —
? * % ~ # = : & ( ) — are stripped by default. To search them as literals, add them to the alphabet file and fully rebuild the index. [Searching for symbols]
- Searches are case-insensitive. dtSearch normalizes indexed text to lowercase; uppercase in the query has no effect except inside
##"..." regex, where uppercase will fail to match. [Regex caveats]
Documentation URLs point to the current RelativityOne pages. If you’re on Server, replace /RelativityOne/ with /Server2026/ (or your version) in the URL.