domain tier

`domain-transcript-shard`

Local, deterministic structural decomposition of speaker-turn transcripts (court transcripts, depositions, interviews, meeting minutes): line-number parsing, speaker-turn grouping, citation-phrase detection, and pattern-based statement classification, emitting an opaque-ID shard with every real name and citation spelling replaced -- the AI never sees the raw prose. Implements the document-structural-decomposition gap named in ADR 0035 (semantic mediation). Pure logic, zero I/O, zero AI calls -- it produces input for a disclosure-minimized AI question, it never asks one itself.

Local, deterministic structural decomposition of speaker-turn transcripts (court transcripts, depositions, interviews, meeting minutes): line-number parsing, speaker-turn grouping, citation-phrase detection, and pattern-based statement classification, emitting an opaque-ID shard with every real name and citation spelling replaced -- the AI never sees the raw prose. Implements the document-structural-decomposition gap named in ADR 0035 (semantic mediation). Pure logic, zero I/O, zero AI calls -- it produces input for a disclosure-minimized AI question, it never asks one itself.

Tier	domain
Role	unclassified (baselined)
Path	`crates/domain/transcript-shard`
Edition	2021
Targets	transcript-shard, domain_transcript_shard
Public items	12 across 0 modules
Tests	11

What it is for

Local, deterministic structural decomposition of speaker-turn transcripts: numbered lines -> speaker turns -> pattern-classified statement types -> an opaque-ID shard.

This is the "document structural decomposition" layer named as a real gap in docs/architecture/adr/0035-privacy-preserving-semantic- mediation.md's inventory ("nothing in the inventory parses an arbitrary document into structure in this direction"). Every step here is string parsing and pattern matching -- nothing in this crate calls an AI, makes a network request, or shells out. It produces the minimal disclosure shard a caller would hand to an AI for one narrow question; it never asks that question itself.

Applies to any speaker-turn transcript -- court transcripts, depositions, interviews, meeting minutes -- not only legal proceedings. StatementType's ProceduralCourtesy/ProceduralRuling variants are formal-proceeding-flavored, but the parsing/detection machinery itself is generic (Gate 4.5 AC-generic; Gate 1.5 consensus, docs/userstories/sprint-3.48-transcript-shard.md).

Capabilities

`crate root`

Local, deterministic structural decomposition of speaker-turn

Item
`fn split_speaker_prefix(text : & str) -> Option <(String, String)>`
`fn detect_repeated_title_case_phrases(texts : & & str) -> Vec <String>`

`RawLine`

Local, deterministic structural decomposition of speaker-turn

Item
`pub struct RawLine`
`fn parse_raw_lines(text : & str) -> Vec <RawLine>`

`Shard`

Local, deterministic structural decomposition of speaker-turn

Item
`pub struct Shard`
`fn build_shard(turns : & Turn, task : & str, allowed_outputs : Vec <String>) -> Shard`

`StatementRef`

Local, deterministic structural decomposition of speaker-turn

Item
`pub struct StatementRef`

`StatementType`

Local, deterministic structural decomposition of speaker-turn

Item
`pub enum StatementType`
`fn classify_statement(text : & str) -> StatementType`

`Turn`

Local, deterministic structural decomposition of speaker-turn

Item
`pub struct Turn`
`fn group_turns(lines : & RawLine) -> Vec <Turn>`
`fn detect_citation_phrases(turns : & Turn) -> Vec <String>`

How to use it

No examples/ target and no doctest in this crate's rustdoc. The tests listed under Verification are the closest executable usage.

Module structure

No public modules: the crate root is its whole surface.

Public surface

`crate root`

Item	What it is
`pub struct RawLine`	One physical line of the source text: an optional leading line number (the transcript's own numbering) and the text after it
`fn parse_raw_lines(text : & str) -> Vec <RawLine>`	Parse raw transcript text into numbered lines
`fn split_speaker_prefix(text : & str) -> Option <(String, String)>`	If `text` starts with an all-uppercase word sequence followed by a colon (`"LORD PANNICK:"`, `"THE PRESIDENT:"`), split it into `(speaker, rest)`
`pub struct Turn`	One speaker's contiguous turn: who spoke, the transcript line range it spans, and the full text (continuation lines merged with a space)
`fn group_turns(lines : & RawLine) -> Vec <Turn>`	Group numbered lines into speaker turns
`pub enum StatementType`	A turn's structural type, decided by pattern alone -- no AI
`fn classify_statement(text : & str) -> StatementType`	Classify one turn's text
`fn detect_repeated_title_case_phrases(texts : & & str) -> Vec <String>`	Find candidate document/case-citation phrases across any set of texts: runs of 2+ consecutive Title-Case words, repeated at least twice across all of them
`fn detect_citation_phrases(turns : & Turn) -> Vec <String>`	Same detection, over a transcript's turns specifically -- thin wrapper kept for callers already working with `Turn`s.
`pub struct StatementRef`	One statement in the emitted shard: an opaque speaker ID, the opaque document IDs it references, and its structural type
`pub struct Shard`	The minimal-disclosure shard: opaque parties, opaque documents, typed statements referencing them, and the fixed question being asked.
`fn build_shard(turns : & Turn, task : & str, allowed_outputs : Vec <String>) -> Shard`	Build a shard from turns: detect citation phrases, assign each unique speaker and each citation an opaque ID in first-appearance order, and classify every turn

No pub use re-exports: every item above is declared in this crate.

Boundary

Reaches into foundation.

Shares tier domain with 41 other crates: domain-agreements, domain-ai-report, domain-billing, domain-catalog, domain-classify, domain-comments, domain-competitive-intel, domain-contact, … (41 total).

_What this crate deliberately does NOT own is a judgment. No committed registry records one for it, so none is stated here._

Where it sits

Tier (ontology)	`domain`
Architectural role (taxonomy)	`unclassified (baselined)`
Location	`crates/domain/transcript-shard`
Vocabulary in force (lexicon)	`current`

Tier flow. Which tiers this crate's own edges cross.

flowchart LR
  n_domain["domain"] --> n_foundation["foundation"]

Dependencies

Runtime, in this workspace.

Crate	Tier	Optional	Only on
`foundation-text-signals`	foundation	no	always

Runtime, from outside the workspace.

Crate	Requirement	Features	Optional	Only on
`clap`	`^4`	`derive`	no	always
`serde`	`^1`	`derive`, `derive`	no	always
`serde_json`	`^1`	—	no	always

Development, in this workspace.

Crate	Tier	Optional	Only on
`tools-cli-conformance`	tools	no	always

Build. None.

Depended on by. 4 workspace crates.

Signal flow — what reaches this crate, and what it reaches.

flowchart LR
  n_domain_prose_shard["domain-prose-shard"] -->|uses| SELF
  n_tools_doc_ask["tools-doc-ask"] -->|uses| SELF
  n_tools_doc_shape["tools-doc-shape"] -->|uses| SELF
  n_tools_knowitall["tools-knowitall"] -->|uses| SELF
  SELF["domain-transcript-shard"]
  SELF -->|development| n_tools_cli_conformance["tools-cli-conformance"]
  SELF -->|runtime| n_foundation_text_signals["foundation-text-signals"]
  classDef self fill:#1f883d,stroke:#1f883d,color:#fff;
  class SELF self;

Feature flags

No Cargo features are defined: every capability is unconditional, so no consumer can receive a half-wired crate.

Targets

Kind	Name	Source
bin	`transcript-shard`	`src/main.rs`
lib	`domain_transcript_shard`	`src/lib.rs`

Error model

No public error type was detected: no public item declares a type named *Error, and no public signature returns one.

Operational characteristics

Property	Evidence
async public surface	none detected
async runtime	none detected
database access	none detected
network I/O	none detected
unsafe code	none detected
environment variables	none detected

No unsafe block, unsafe fn, unsafe impl or unsafe trait was found by the parser anywhere in this crate's source.

Configuration

No environment variable is read with a literal name anywhere in this crate. A variable whose key is computed at run time cannot be listed here, and is not claimed to be absent.

4 workspace crates depend on this one: domain-prose-shard, tools-doc-ask, tools-doc-shape, tools-knowitall.

Verification

Kind	Count
Unit tests	11
Integration tests	0
Examples	0
Doctests	0

Evidence by module. How often each public module is named by something executable.

Module	Tests	Examples	Consumers
`crate root`	12	0	13

What the tests establish, by name:

build_shard_assigns_stable_opaque_ids_and_never_leaks_the_real_citation_spelling — src/lib.rs
classify_statement_matches_every_turn_in_the_real_fixture — src/lib.rs
detect_citation_phrases_finds_de_keyser_and_nothing_else — src/lib.rs
detect_repeated_title_case_phrases_keeps_a_two_word_article_led_proper_name_intact — src/lib.rs
detect_repeated_title_case_phrases_merges_a_sentence_initial_occurrence_with_a_mid_sentence_one — src/lib.rs
group_turns_merges_continuations_and_tracks_the_real_line_range — src/lib.rs
opaque_ids_restart_per_call_not_persisted_across_documents — src/lib.rs
parse_raw_lines_reads_leading_numbers_and_keeps_continuation_lines_unnumbered — src/lib.rs
real_ocr_noise_still_produces_the_same_shard_shape_as_the_clean_fixture — src/lib.rs
split_speaker_prefix_recognizes_multi_word_and_rejects_prose — src/lib.rs
cli_definition_conforms — src/main.rs

Documentation coverage

Measure	Documented	Total
Public items with rustdoc	12	12
Public modules with a `//!` block	0	0

pie showData
    title Public items with rustdoc
    "Documented" : 12
    "No rustdoc detected" : 0

Metrics

Metric	Value
Rust source files	2
Source lines	632
Code lines	458
Public API items	12
Public modules	0
Tests	11
Examples	0
Cargo features	0
Direct runtime dependencies	4
Workspace reverse dependencies	4

pie showData
    title Public API by kind
    "enum" : 1
    "function" : 7
    "struct" : 4

pie showData
    title Rust source composition
    "Code" : 458
    "Blank or comment" : 174

Generation

Rendered by tools-corpus corpus readme from repository evidence alone, renderer schema 2, lexicon current. No model, network service or database was consulted. Regenerate with tools-corpus corpus readme --write; verify with --check.

All domain · Manual