Glossary
A one-page reference for terminology used across the scriptorium docs. Each entry is two to four sentences and links to the relevant concept page or knowledge note for the longer treatment. This page exists so the rest of the docs can link to a single definition; cross-references should point here rather than duplicating definitions inline.
Accuracy guard
Section titled “Accuracy guard”A guard of kind accuracy: it exists because overriding it
would make the skill’s output wrong, not merely riskier. Examples are
no-parametric-citations (no citations from model memory),
preserve-quantities-verbatim and no-false-verification-claims.
Accuracy guards always default to refuse, and neither a per-run
answer nor meta.overrides can override them; scriptorium validate
rejects an override that names one.
See: Author authority.
Argumentative flow
Section titled “Argumentative flow”The logical and structural coherence of a section’s prose —
whether each paragraph’s topic position connects to prior context,
whether the stress position carries new information, whether the
argument arc is visible to the reader. The argumentative-flow
skill operates on this property of a section under a
preservation contract.
See: skills reference — argumentative-flow.
Core claim
Section titled “Core claim”A load-bearing assertion the manuscript is arguing — what a
reader of the abstract should walk away believing. Declared in
MANUSCRIPT_STATE.yaml under core_claims. Critique skills test
whether the prose actually supports each declared claim;
transformation skills must preserve declared claims verbatim
(none added, none dropped).
See: Schema reference, Start here — what scriptorium operates on.
Declared work / declared-work scope
Section titled “Declared work / declared-work scope”What scriptorium operates on: prose the author has already
written, or scaffolding the author has committed to in
MANUSCRIPT_STATE.yaml (claims, weaknesses, terminology,
audience, target venue). By default scriptorium does not produce
prose from blankness; the proposer role (Hayes 2012) stays with the
author. The boundary is enforced by scope guards, which
the author can override.
See: Declared-work scope,
knowledge/conventions/declared-work-scope.
Design memo
Section titled “Design memo”A standalone architectural decision record kept under
docs/design/ (and rendered into the site by the docs
preprocessor). Memos document why a design choice was made — what
alternatives were considered, what evidence drove the decision,
what would prompt revisiting. They are the longer-form companion
to the knowledge layer.
EQUATOR Network
Section titled “EQUATOR Network”The international network that maintains reporting-guideline
checklists for health research — CONSORT (RCTs; 2025 version),
SPIRIT (trial protocols; 2025 version), STROBE (observational),
PRISMA (systematic reviews), ARRIVE (animal
research), TRIPOD / TRIPOD+AI (prediction models), STARD
(diagnostic accuracy), CARE (case reports), COREQ (qualitative),
CHEERS (economic evaluations), and their AI extensions
(CONSORT-AI, SPIRIT-AI). reporting-guideline-fit infers which
checklist applies to a manuscript;
reporting-guideline-compliance runs the checklist against the
prose.
See: https://www.equator-network.org/.
A refusal or check a skill declares under guards: in its
manifest.yaml, with an id, a kind (policy, scope or
accuracy), a one-line summary, an optional source and a
default (ask or refuse). When a guard fires, the skill states
the concern and its source in one line, at every guidance level.
Policy and scope guards can be overridden;
accuracy guards cannot.
See: Author authority,
knowledge/conventions/author-authority.
Guidance level
Section titled “Guidance level”The author’s dial over how much detail and how many findings
scriptorium surfaces per skill invocation. Schema enum at
meta.guidance_level: terse (small handful, big-picture only),
standard (focused list — the default), full (every finding
above threshold, with rationale). Set the level to match the
bandwidth you have for revisions, not the level that feels
rigorous.
See: Guidance level,
knowledge/conventions/guidance-level.
Knowledge note
Section titled “Knowledge note”A standalone evidence-grounded document under knowledge/ that
documents what the literature says about some aspect of scholarly
writing or peer review. Each skill’s manifest.yaml lists the
knowledge notes its design decisions ground in. The full
structure of a knowledge note is documented in the
knowledge layer README.
See: Knowledge layer.
MANUSCRIPT_STATE.yaml
Section titled “MANUSCRIPT_STATE.yaml”The shared editorial-state file every scriptorium project
carries. Lives at the root of the manuscript repository. Records
what the manuscript is arguing (core_claims), the limitations
already acknowledged (known_weaknesses),
terminology.preferred / forbidden / synonyms, style
(tone, voice, audience), constraints, the target_venue, the
document_phase, meta.guidance_level and meta.overrides.
Every skill reads
it; only init writes it. It is a local file; nothing in
scriptorium uploads or syncs it.
See: Schema reference, the worked example.
Override
Section titled “Override”The author’s decision to let a skill proceed past a policy or scope
guard. It is given either as a per-run answer, which wins
in either direction, or as a standing entry under meta.overrides
in MANUSCRIPT_STATE.yaml naming the guard and the skills (or
"*"). A skill that proceeds past a guard lists it on an
“Overrides applied” line in its output.
See: Author authority, Schema reference.
Preservation contract
Section titled “Preservation contract”The explicit list of properties a transformation skill
(argumentative-flow, compression) is required to preserve
verbatim when it modifies prose: every citation key, every
numerical value / p-value / confidence interval / unit, the
declared terminology.preferred / forbidden / synonyms
lists, declared core_claims, and the author’s hedging strength
and stance markers. Any shift in hedging is reported as a
per-edit note rather than absorbed silently. The contract is
what scriptorium guarantees; sentence-level “voice preservation”
is explicitly not.
See: Landing — voice preservation, the honest version,
semantic-preservation.
Reviewer simulation lens
Section titled “Reviewer simulation lens”One of four attentional perspectives reviewer-simulation applies
to a manuscript: methodological skeptic, domain expert,
translational/clinical, statistical. Each lens emits the same
six-section structure (Major Critiques, Minor Critiques, Fatal
Concerns, Enthusiasm Drivers, Suggested Revisions, Acceptance
Risk). The term is lens, not persona. A lens is an attentional
focus on what a reader of that disposition would notice; a persona
implies impersonating a specific reviewer, which the skill does
not do.
See: skills reference — reviewer-simulation,
Case study for example output.
A single-responsibility unit of scriptorium functionality.
Implemented as a directory under skills/ containing
manifest.yaml (machine-readable metadata), SKILL.md (the
prompt for Claude Code), prompt.md (the platform-neutral
prompt), README.md (the human-readable contract), and
examples/. Skills do not auto-invoke; the author chooses which
one to run and when.
See: Skills reference for the catalog.
Skill category
Section titled “Skill category”One of six categorical labels in a skill’s manifest.yaml:
- critique — assesses prose; does not modify the manuscript. Emits structured findings.
- validation — checks against an external standard (reporting guideline, authorship policy); does not modify.
- normalization — enforces declared style and terminology; suggests edits, never auto-applies.
- transformation — modifies prose under a preservation contract.
- meta — orientation or explanation; no manuscript modification.
- utility / onboarding — bootstrap; modifies only
MANUSCRIPT_STATE.yaml(currently onlyinit).
The category determines what the skill can produce and the guarantees that come with that production.
See: Skills reference — categorisation axes.
Workflow stage
Section titled “Workflow stage”Where in the writing process scriptorium fits. The
document_phase.current field in MANUSCRIPT_STATE.yaml records
the stage (outline / draft / review / revision /
submission / post-submission / accepted). Scriptorium
occupies the translator and evaluator roles in Hayes’ 2012
cognitive-process model — most useful at draft through
submission. The author owns the proposer role; by default,
outline and ideation are out of scope.
See: Workflow stage.
Related
Section titled “Related”- Start here — the conceptual map.
- Declared-work scope — what scriptorium operates on.
- Guidance level — the author’s dial.
- Author authority — which guards the author can override.
- Workflow stage — where in the writing process scriptorium fits.
- Skills reference — the catalog.
- Schema reference —
every field in
MANUSCRIPT_STATE.yaml.