# NIKKE ASS — query syntax reference ASS searches ~430k dialogue lines (main story, events, sudden stories, side stories, bond episodes and advise scripts) across the EN, KO and JA scripts. EN carries the full ~429k lines; KO and JA carry ~388k each. The filter builder writes the syntax below for you, so power users can type it directly. The search runs against **one language at a time** — the EN/KO/JA toggle, or `lang:` in the query (which wins; anything else falls back to the toggle). Speaker filters resolve from English data names and speaker keys in every language. ## Free-text terms rapi lines containing the word "rapi" "counted the days" exact phrase, adjacency preserved comm* prefix wildcard (commander, commands…) c?me ? matches one character (came, come…) **English terms match whole words.** `rapi` matches the word *rapi*, not *rapid*; `work` does not match *word*; `or` (quoted: `"or"`) matches only the word itself, not *order* or *work*. Punctuation splits words, so `rapi` also matches the possessive *rapi's*, but never a substring like *rapid*. This is deliberate — substring matching used to flood results with lines merely containing the same letters. **KO/JA terms match substrings.** Korean and Japanese are not whitespace-segmented, so those scripts use character-bigram matching that behaves exactly like substring search. - Plain English terms are case-insensitive whole words. - Quoted phrases keep exact substring semantics (spaces, punctuation, and adjacency); a single-word quote behaves as a whole word. - Wildcards run against the term dictionary: `*` is any run of characters, `?` is exactly one. `comm*` uses the index prefix; `*thing` scans it. - Broad queries are budgeted: patterns or term soups that would expand to a large share of the corpus (`*a*`, `a*`, `the OR you OR …`) are rejected with `query too broad` — add a more specific word or filter. More than 24 terms or field filters is rejected as `query too complex`. Common words are still searchable on their own; they stream the release-ordered index. - The Filters panel's **Text match** selector maps to: *Contains* (whole word), *Starts with* (prefix), *Ends with* (suffix), *Exact* (whole-line equality). It applies to plain terms only — quoted phrases and wildcards use their own semantics. - Hyphenated and apostrophised terms (`BA-01`, `don't`) match as written; quote only phrases and operator words (`"or"`, `"not"`). ## Boolean operators space AND (default) OR either side NOT / - / ! exclude — NOT ark, -ark, !ark all work ( ) grouping speaker:chime to:crown (source:overzone OR source:redash) -ark kind:event NOT commander -kind:bond marian Operators and field names are case-insensitive (`or`, `Or`, `OR`; `Speaker:` and `speaker:` both work). To search for the literal word, quote it: `"or"`, `"not"`. ## Fields | Field | Aliases | Matches | | ------------------ | ------------ | ---------------------------------------------- | | `speaker:` | `from:` | who is speaking | | `to:` | — | who is being addressed | | `portrait:` | `skin:` | a specific speaker variant | | `kind:` | — | story category | | `chapter:` | — | main story chapter (or event code) | | `source:` | — | one specific chapter/event | | `year:` | `released:` | release year of the source | | `lang:` | — | which translation to search (global) | | `sort:` | — | result order (global) | ### speaker: / from: Matches the speaking character. Characters include all their variants and aliases: speaker:anis Anis + Anis: Sparkling Summer + Anis: Star … speaker:marian Marian, Modernia, and every variant speaker:commander the player (Self / lines) A few names alias into others: `modernia` (and the misspelling `modenia`) finds Marian, and `lilith` finds Liliweiss. Values are English data names or speaker keys (`crown_n`). `from:` is the same field under the name the filter builder uses — handy next to `to:`. ### to: Matches lines where the character is the **addressee**. ASS infers addressees at ingest time from: - **Scene pairing** — scenes with exactly two participants: every line one of them speaks is addressed to the other. - **Vocatives** — lines that name someone: `Chime, watch out!`, `Hey Chime!`, `… right, Chime?` Narration, monologue and choice lines are never matched as addressees. Inference always runs on the English script, so results are the same in every language. `to:` is a heuristic — great for "find where X talks to Y", not an official script annotation. to:crown speaker:chime to:crown to:commander ### portrait: / skin: Matches one specific variant rather than the whole character: portrait:crown_n the Crown variant with key crown_n portrait:330 every variant sharing resource 330 ### kind: kind:main main story (also scene, story) kind:event archive events (also events, archive) kind:sudden sudden stories kind:side side stories (also sidestory, sidestories) kind:bond bond episodes (also bonds) kind:advise advise / counsel scripts (also advice, counsel, counseling) kind:relic Lost Relics collectible lore (also relics, memorial, memorials) Advise scripts are searchable but are not one of the builder's Content checkboxes — type `kind:advise` to reach them. Lost Relics have their own Content checkbox and open in the relic browser rather than the transcript. ### chapter: Numbers target main story chapters; text targets event codes: chapter:12 main story chapter 12 chapter:[10 TO 20] chapters 10–20 (inclusive) chapter:>=40 chapter 40 and up chapter:overzone the Overzone event (same as source:overzone) Ranges use `[ ]` for inclusive and `{ }` for exclusive bounds: `chapter:{10 TO 20]`. ### source: A single chapter or event: source:overzone event code source:22 main story chapter 22 (by number) source:#139 internal id (shown when you filter from the reader) ### year: / released: Filters by the release year of the chapter or event the line belongs to: year:2023 lines from sources released in 2023 year:[2019 TO 2021] inclusive year range year:>=2024 everything from 2024 on rapi year:<2021 early Rapi lines only Caveat: sources without a known release date are ordered just after the newest dated source, so the most recent years include undated material. ### lang: (global) Overrides the language toggle for this search. Only one language is searched per query: lang:ja 指揮官 search the JA script for 指揮官 lang:ko rapi search the KO script for "rapi" ### sort: (global) sort:new newest release first (also desc, latest) sort:old oldest first (the default) ## Combining fields Fields are AND'd unless you group with `OR`: speaker:chime to:crown Chime speaking to Crown (speaker:anis OR speaker:neon) kind:event speaker:marian "BA-01" -kind:bond the BA-01 line, not bond episodes ## Filter builder The **Filters** panel builds these tokens for you: its From/To/Portrait/ Chapter/Release-year and Exclude inputs map to `from:`/`to:`/`portrait:`/ `source:`/`year:` and `-` terms, the Content checkboxes map to `kind:`, and Sort maps to `sort:`. The query preview under the builder shows the text it will produce. **Convert to query** turns the current panel state into a text query you can edit by hand. **Copy link** shares the same filters as a URL, so a filter-only search (`/?kind=bond`, `?speaker=marian&kind=scene`) reproduces without any text. Two builder controls are global settings rather than query tokens: the language toggle and **Text match**. Text match applies to plain terms only; quoted phrases and wildcards use their own semantics. ## Gated content Scenes from sources the game has not released yet are excluded from search (snippets are spoilers) and locked in the reader until the acknowledgment phrase is entered. Gated sources are badged in the sidebar; there is no query syntax to unlock them — the gate lifts on its own at release.