Self-hosted, Self-Improving, Federated Everything App — dev management, agent orchestration, persistent mind, digital twin, second brain, and creative pipeline from a single dashboard over Tailscale to port the user into the OS
JavaScript ai-agentsdashboarddeveloper-toolsdigital-twinquantified-selfsecond-brainself-hostedtailscale
/**
* Editorial check taxonomy (#2842 split of checkInfra.js) — the declarative
* vocabulary every check entry is validated against: scopes, kinds, severities,
* the staleness `sources` allow-list, and the UI field types.
*/
export const CHECK_SCOPES = Object.freeze(['series', 'issue', 'scene', 'noun']);
// A check's `scope` may be a single scope string OR an array of scopes (#1628).
// A dual-scope check is one rule meaningful at more than one granularity — e.g.
// `relationships.reciprocity` reasons series-wide but could also flag a newly
// added relationship per issue — declared `scope: ['series', 'issue']` instead of
// being split into two near-duplicate checks. These helpers normalize either form
// so every consumer (the load-time guard, the catalog grouping, the dry-run plan)
// fans a check across exactly its declared scopes.
//
// `normalizeCheckScopes` returns the declared scopes deduped and in canonical
// CHECK_SCOPES order; an unknown/empty input returns []. `primaryCheckScope`
// returns the first declared scope (canonical order) for the consumers that still
// bucket/sort by a single value — for a single-scope check it is just that scope.
export function normalizeCheckScopes(scope) {
const raw = Array.isArray(scope) ? scope : (scope == null ? [] : [scope]);
return CHECK_SCOPES.filter((s) => raw.includes(s));
}
export const primaryCheckScope = (scope) => normalizeCheckScopes(scope)[0] || null;
export const CHECK_KINDS = Object.freeze(['deterministic', 'llm']);
export const CHECK_SEVERITIES = Object.freeze(['high', 'medium', 'low']);
export const SEVERITIES = CHECK_SEVERITIES;
// The inputs a check can read, declared per-check via `sources` (#1387). The
// staleness runner (server/services/pipeline/editorial/checkRunner.js) fingerprints
// EXACTLY a check's declared sources, so a finding only goes stale when content
// the check actually analyzed drifts — editing the style guide no longer marks a
// naming finding stale, and editing the ticking clock no longer marks every
// canon-only finding stale. Every token here must have a matching resolver in the
// runner's `SOURCE_RESOLVERS` (a load-time guard there fails fast if they drift).
// - 'manuscript' — the stitched manuscript corpus (implies needsManuscript)
// - 'canon' — the universe/series canon (characters, relationships, objects)
// - 'continuityBible' — the series CONTINUITY-BIBLE facts ledger (#1305): the
// extracted ground-truth facts the timeline/canon-contradiction
// check (#1581) reconciles the prose against — `[{ category,
// subject, statement, issueNumber }]` across the bible's
// categories (physical, age, dates/elapsed time, location,
// possession, world rules, who-knows-what). The runner fetches
// it via `getFactsLedger` (gated on this source) and injects
// `ctx.continuityBible` (the facts array).
// - 'series.styleGuide' — the series style guide (tense/POV/rating/reading level)
// - 'series.arc.tickingClock' — the series arc's ticking clock
// - 'reverseOutline' — the cached reverse-outline scene segmentation (#1286);
// scenes carry components/povCharacter/charactersPresent.
// The runner fetches it (gated on this source) and injects
// `ctx.reverseOutline` (the scenes array).
// - 'reverseOutline.plotlines' — the cached reverse-outline's PLOTLINE list (#1286):
// `[{ id, label, kind, color }]` plus the per-scene
// `plotlineId`/`secondaryPlotlineId` tags. The runner fetches
// the outline (gated on this source) and injects
// `ctx.reverseOutlinePlotlines`. The plot-structure check (#1310)
// reconciles dropped subplots against these tagged plotlines —
// which start and then fizzle without a resolution scene.
// - 'series.arc.readerMap' — the series arc's authored reader-map (#1299): the
// writer-logged hooks (questions planted) and payoffs
// (their resolutions). The Chekhov check reconciles its
// detected setups/payoffs against these authored ones.
// - 'editorialArcs' — the detected per-character arc directions from the series
// editorial analysis aggregate (#1295). The runner fetches it
// (gated on this source) and injects `ctx.editorialArcs`
// (`[{ name, arcDirection, issueCount, isProtagonist }]`). This
// is the coarse, DETECTED arc signal — distinct from the
// AUTHORED `series.characterArcs` model below.
// - 'series.characterArcs' — the AUTHORED per-character story arcs (#1293):
// `series.characterArcs[]` (`{ characterId, characterName,
// want, need, startState, endState, transitions[] }`). The
// arc.transitions check reconciles detected change moments
// against these authored transitions + flat-arc warnings.
// - 'series.characterArcs.evolution'
// — the OPTIONAL five-stage evolution lens nested in those
// authored arcs (`series.characterArcs[].evolution`, #6440):
// the staged causal chain tested belief → external pressure →
// choice → cost paid → final behavioral proof, plus the DECLARED
// outcome (full-change / tragic-refusal / flat-testing /
// partial-open). A NARROWER token than 'series.characterArcs' on
// purpose: character.secondary-arc and arc.climax-agency read the
// lens but not the want/need/start/end model, so pointing them at
// the whole arcs token would stale their findings on an edit they
// never saw (the same split 'comicScript.layout' makes against
// 'comicScript.pacing'). Fingerprinted as the RENDERED lens block,
// so a deleted transition beat that turns a stage anchor stale
// stales the finding too.
// - 'storyboard.shots' — the per-issue storyboard shot lists
// (`stages.storyboards.scenes[].shots[]`) the
// visual-continuity check (#1315) reasons over:
// each shot carries `shotType` / `screenDirection`
// / `continuityFromShotId` (server/lib/shotGrammar.js).
// Served off the already-loaded `ctx.issues` (no
// extra I/O); the runner injects `ctx.storyboardScenes`
// (a flat list of `{ issueNumber, scene }` for every
// issue that has storyboard scenes).
// - 'comicScript' — every issue's AUTHORITATIVE comic content, keyed by
// issue number: the edited comic-pages split
// (`stages.comicPages.pages[]`) when present, else the
// generated `stages.comicScript.output`. The
// lettering-density check (#1313) counts per-panel/per-page
// word + balloon load over it (caption/dialogue/SFX only).
// Both comic-source checks read the same parsed pages via
// `comicLetteringIssues(ctx.issues)` (`[{ number, pages }]`);
// the runner fingerprints ONLY the lettering fields for this
// token, so a visual-description edit doesn't stale a
// lettering finding.
// - 'comicScript.pacing' — the same parsed comic pages, for the page-turn-beats
// LLM check (#1314), which reads each panel's visual
// `description` (+ caption/dialogue/SFX text) for its prompt
// digest. Distinct token from 'comicScript' because that
// broader read means a description edit must stale a page-turn
// finding while the lettering token stays put (and vice-versa).
// - 'comicScript.layout' — LAYOUT ONLY (per-page panel COUNT) for the panel-rhythm
// check (#1314), which reads nothing but counts. Separate from
// 'comicScript.pacing' so a text-only edit (reword a caption /
// description without adding/removing a panel) does NOT stale a
// rhythm finding — the splash/crowding/grid verdict can't have
// changed. All three comic tokens share `ctx.issues` (no extra I/O).
export const EDITORIAL_SOURCES = Object.freeze([
'manuscript',
'canon',
'continuityBible',
'series.styleGuide',
'series.arc.tickingClock',
'series.arc.readerMap',
// The authored foreshadowing ledger (#2172): the arc-overview-emitted
// plant → reinforce → payoff seeds the Chekhov check reconciles its detected
// setups/payoffs against. Lives on the already-loaded series record (no extra
// I/O); fingerprinting it stales Chekhov findings when the author edits the ledger.
'series.arc.foreshadowing',
'series.arc.themes',
// The author-supplied real-world fact reference the opt-in research.fact-accuracy
// check reconciles the prose against (#1588). Lives on the already-loaded series
// record (no extra I/O); fingerprinting it stales fact findings when the author
// edits the reference.
'series.factReference',
'reverseOutline',
'reverseOutline.plotlines',
'editorialArcs',
'series.characterArcs',
'series.characterArcs.evolution',
'storyboard.shots',
'comicScript',
'comicScript.pacing',
'comicScript.layout',
// Each issue's PROSE-stage text SPECIFICALLY (#1589) — NOT the default manuscript
// precedence (comicScript ▸ teleplay ▸ prose), which for a hybrid issue returns
// the comic script. The comic↔prose-sync check reads this to compare prose against
// the comic; fingerprinting it separately means a prose edit stales a sync finding
// without a comic-only edit doing so (and vice-versa, via comicScript.pacing).
'prose',
]);
// Default per-run finding cap for user-defined checks (#1346) — mirrors the
// built-in LLM checks' `maxFindings` default so a long manuscript can't flood
// the review. Defined up here so the custom-check prompt builder and config
// schema (both below) share one source.
export const CUSTOM_CHECK_MAX_FINDINGS_DEFAULT = 12;
// The serializable config-field types a check can declare for its UI form.
// `configSchema` (a Zod schema) stays the validation authority on the server;
// `configFields` is the wire-safe *render* descriptor the Editorial Checks UI
// reads to build the per-check config form (the Zod schema can't cross the wire).
// Keep this in lockstep with the controls EditorialCheckCard's ConfigField
// renders — only advertise a type the UI can actually draw (no 'select' until
// the <select> control + an `options` contract land).
export const CHECK_FIELD_TYPES = Object.freeze(['number', 'boolean', 'text']);