Soranoha

用語集 / Glossary

Terminology used throughout Soranoha for corpus artifacts, release records, and encoding concepts.

For internal names of parser and rendering behavior, see parser and rendering invariants (docs/parser-invariants.md).

admission

The decision to include a work in a release. A work is admitted when its rights standing has been established against a recorded basis and that basis still holds at the moment of publication.

Admission is independent of validation: a work can pass validation without being admitted, and publication does not guarantee encoding quality. Each release manifest records the admission evidence relied upon.

assessment

The evaluation that produces an admission basis: what is claimed about a work's rights standing, on what evidence, effective from when, and in which jurisdiction (Japan).

Assessment records facts with premises, not permanent verdicts. A fact records what was established, from which evidence, as of which date. If evidence becomes unreachable, the fact becomes unavailable rather than remaining asserted, so a work can drop out of a release without any new claim having been registered against it. Assessment is not legal advice, and Soranoha is not a rights clearinghouse. See reliance for the basis used for most works.

reliance

The default admission basis: Aozora Bunko's published copyright-expired classification for a specific edition, recorded as an attributed upstream assertion alongside the raw page bytes. A recorded exception or conflicting reviewed finding prevents reliance.

Reliance is re-checked against the live source at release time. Retained evidence cannot substitute for a current check.

fidelity

The proportion of source markup preserved in the resulting encoding. It measures markup preservation: rights standing comes from assessment, and correctness from validation.

Two evidence reports accompany each build:

Fidelity measures presence, not interpretive correctness. Ambiguities are recorded directly in the published TEI as interpretation-problem notes.

work, edition, document

Three distinct concepts maintained throughout:

Cite the identifier for the work, the identifier plus source_content_hash for the edition, and the artifact id for exact bytes. See work identifiers and citing Soranoha.

artifact

One published file, identified by its content: snh:1:<type>:<sha256>. The types published per work are tei, plaintext, markdown, and tei-validation.

Artifact IDs name bytes directly and can be verified by computing the SHA-256 hash locally.

manifest

The signed record of one release. It lists every admitted work with its identifier, its source_content_hash, and the ID and byte length of each of its four artifacts; the rights grant; admission evidence; the release catalog; and the parent manifest digest.

The manifest is authoritative. Site pages, indexes, and reading views are generated downstream from the manifest. https://soranoha.org/releases/latest serves the current manifest.

chain

The append-only sequence of manifests, each referencing its predecessor back to genesis.

Earlier releases remain immutable. A withdrawal removes a work from current distribution; it cannot alter historical manifests.

Signing roles are separated between publication and governance keys, with governance keys kept offline. See the protocol specification (docs/design/snh-protocol-v1.md).

release

One published state of the corpus: a manifest, its signature, and the artifacts it names. Identified by the manifest's SHA-256 hash (the release head).

Citing a release head identifies the exact corpus state referenced.

catalog

The bibliography of every work in a release, including identifiers, titles, readings, contributors, first publication details, orthographic style, NDC classification, Aozora Bunko card URL, and source edition. Published as https://soranoha.org/catalog.json.

The catalog is part of the signed release. It contains bibliographic facts rather than presentation-layer artifacts like download filenames or formatted citations.

withdrawal

Removing a work from current distribution via a signed governance event specifying a reason code and public statement.

Withdrawal halts distribution from /works/<identifier>/ and the active catalog. It does not delete historical records: prior signed manifests continue to name the work, and its artifacts remain addressable by hash. Withdrawal records are served at https://soranoha.org/withdrawn/<identifier>.json.

For rights inquiries or withdrawal requests, see rights.

body-v1

The reading policy defining canonical reading text, used by the plaintext projection, the site reading view, and external annotation layers.

It traverses the TEI body and resolves alternatives deterministically: selecting apparatus lemmas, choosing corrected or regularized editorial readings over source originals, keeping base text rather than ruby annotations, and omitting editorial notes or figure descriptions. Unresolved gaiji are excluded from annotatable spans rather than replaced.

The name is versioned so that an annotation can say which policy produced the offsets it uses. That is what keeps the annotation checkable if the policy ever changes, rather than silently reinterpreted against a different reading.

gaiji (外字)

A character the Aozora Bunko source file could not encode, written as a prose description, usually with a JIS X 0213 code point: ※[#「特のへん+廴+聿」、第3水準1-87-71].

Published TEI includes both the resolved character and the original source marker verbatim, allowing resolutions to be verified or challenged. See the worked example.

ruby (ルビ)

A reading printed alongside or above base characters (furigana). Written as 蓮池《はすいけ》 in source text; represented in TEI as <ruby> containing <rb> (base) and <rt> (reading).

Ruby readings appear in the TEI encoding and are omitted from plaintext projections, which retain base text only.

source offset

A byte range into the UTF-8 decoded Aozora Bunko source text, recorded for almost every element in the published TEI.

Offsets let a claim about the corpus be stated against the source rather than against a particular tool's output. The text they index is named once per document: primary-text-hash in <sourceDesc> is the sha256 of the Aozora Bunko text member as distributed, while the offsets index that member's UTF-8 decoding. Applying them to the wrong bytes therefore makes the mismatch visible.

snh

The publication protocol: the format of manifests, governance events, admission evidence, and artifact IDs, along with verifier rules. Also the prefix of the extension vocabulary in published TEI (https://w3id.org/soranoha/ns/tei).

Used to verify the corpus independently or build conforming tools. See the protocol specification (docs/design/snh-protocol-v1.md) and the TEI extension vocabulary.