Guidelines for the corpora repository

What this repository is

A corpus about corpora, its tooling, and its reading onto itself. It records positions, not the behaviour of any running system.

It must satisfy its own definitions. This corpus is a corpus, so every rule stated here binds top-corpus/. A rule tightened past the point where the documents stating it still qualify is a defect, not a stricter standard.

This file is structure rather than content, so the rules about standing do not reach it. It is reachable by reference from outside any corpus, and must stay readable by someone who has read no corpus at all.

Layout

• standard-concept/ is a second corpus, of commonly reused concepts that are not about corpora as such. The top corpus reads onto it, as it does onto any corpus.
• top-corpus/ is the corpus. top-corpus/CORPUS.md declares it and is its finding aid; top-corpus/structure/ and top-corpus/content/ hold its two theories; top-corpus/interpretation/ holds this corpus read onto itself and onto top-corpus/tools/. Those directory names are fixed for every corpus.
• top-corpus/tools/ holds checks on well-formedness, soundness and validity, whether performed by a program or by a reader.
• top-corpus/drafts/ holds unsettled material bound for one of the above.
• top-corpus/prompts/ holds prompts that prime a reader for a task. A prompt that produces a finding about material is a check and lives in top-corpus/tools/.

These directories are listed because the declaration is part of the design. What each is for does not depend on how much is in it, and a count of what a directory holds does not belong anywhere: it is wrong the moment anything lands, and listing the directory answers it.

An expectation is different and does belong. Whether a directory will ever fill is not recoverable from the filesystem, and none yet and none ever call for opposite responses. Such a declaration is authoritative in a corpus's {CORPUS.md}; a directory's own README may be consistent with it and is not the authority.

Marks

Sections of a corpus carry marks: how much to believe a section, and what a reader needs in order to read it. The marks are not interpretable without their definitions.

Read top-corpus/logic/marks.md. It is short, it says what a section is, and it points at anything it does not define itself.

The same instruction serves any corpus, since the definitions live in one file that other corpora cite rather than copy. A corpus in another repository points at that file by path and gives this instruction unchanged.

Reading one document cheaply

These documents are addressed by heading rather than read end to end. To see what one holds without opening it, take its heading map first:

grep -nE '^#{1,6} ' top-corpus/structure/some_document.md

When the question is broad, read the whole corpus instead. Do not sample: a partial read leaves you unable to tell whether the corpus is silent on something or you have simply not read the part that answers it.

Provisional terms

These are in use and not settled. Renaming them is expected, and listing them here keeps the set finite and findable.

Term	Provisional because
alignment	Used and defined nowhere here.
misalignment	Same, and it names a category as well as an instance.

alignment appears in top-corpus/interpretation/README.md only. misalignment appears there and in top-corpus/structure/completeness_precondition.md.

Retired, and not to be reintroduced. working corpus, because it names where attention is rather than a structural position. It is not provisional: it has been replaced, and appears only in the passage of top-corpus/structure/corpus-relationships.md that retires it.

defining corpus is a relation, not a name. A is the defining corpus for B is settled and in use. It is not a synonym for top corpus: many corpora may be defining corpora, and there is exactly one top corpus. Same passage.

top corpus is the settled term for the corpus nothing is above. It is an arbitrarily designated name, borrowed from a technical sense defined elsewhere, and a reader needs nothing of that sense to use it.

Terminology needed only by one definition is declared local to that site rather than renamed to avoid a collision elsewhere. See top-corpus/structure/local-terminology.md. The authoritative list of confined terms is kept there and not repeated here, in top-corpus/structure/corpus-relationships.md, "Where the name is borrowed from".

Status

Created 2026-08-30 as a skeleton. This guide states the intended shape so that triage has a target to sort against. It is not a record of what is here.