feat(domain): creative module + sourceright/citeweft handoff

SKILL.md

@@ -211,6 +211,8 @@ the material:
   [governance.md](references/governance.md).
 - Legal-regulatory prose (contracts, advice, affidavits, notices): read
   [legal.md](references/legal.md) with governance.
+- Creative or narrative prose: read
+  [creative.md](references/creative.md).
 - Claims with contradictions or reasoning failures: read
   [reasoning-failures.md](references/reasoning-failures.md).
 - FOI-O final manuscript editing after semantic and citation review: read
@@ -219,7 +221,7 @@ the material:
 
 Load more than one content reference only when the material genuinely crosses
 domains. Reasoning guidance supplements a content reference; it does not replace
-technical, academic, clinical, legal, or governance rules.
+technical, academic, clinical, legal, creative, or governance rules.
 
 For low-density or clearly human-authored prose, make only the smallest
 defensible edits.
@@ -234,6 +236,7 @@ Read these files for the full pattern catalog, examples, and remediation guidanc
 - [Clinical and patient-safety prose](references/clinical.md)
 - [Policy, governance, and compliance prose](references/governance.md)
 - [Legal and regulatory prose](references/legal.md)
+- [Creative and narrative prose](references/creative.md)
 - [Reasoning failures and self-contradictions](references/reasoning-failures.md)
 - [FOI-O final editorial workflow and evidence receipt](references/foio-editorial.md)
 

SKILL_PROFESSIONAL.md

@@ -18,6 +18,7 @@ Use this variant for technical, policy, academic, and client-facing prose. Keep
 - [Clinical module](references/clinical.md) — health research and patient-safety prose
 - [Governance module](references/governance.md) — policy and compliance
 - [Legal module](references/legal.md) — contracts, advice, and regulatory filings
+- [Creative module](references/creative.md) — fiction and narrative prose
 - [Reasoning module](references/reasoning-failures.md) — reasoning failures and contradictions
 - [FOI-O editorial workflow](references/foio-editorial.md) — final manuscript pass and receipt
 
@@ -29,6 +30,7 @@ Use this variant for technical, policy, academic, and client-facing prose. Keep
    * Health research or clinical-safety prose -> Core + Clinical, plus Academic or Governance as required
    * Policy, risk, or compliance writing -> Core + Governance
    * Legal-regulatory prose -> Core + Governance + Legal
+   * Creative or narrative prose -> Core + Creative
     * Reasoning failures or self-contradictions -> Core + Reasoning
     * FOI-O final manuscript editing -> Core + Academic + Governance + FOI-O
    * Otherwise -> Core only
@@ -85,4 +87,4 @@ Vary sentence rhythm with short and long lines. Use specific details instead of
 
 For severity tiers and false-positive guidance, read [Core patterns](references/core-patterns.md) (sections **SEVERITY CLASSIFICATION** and **DETECTION GUIDANCE**).
 
-Available reference files: core-patterns.md, technical.md, academic.md, clinical.md, governance.md, legal.md, reasoning-failures.md, foio-editorial.md.
+Available reference files: core-patterns.md, technical.md, academic.md, clinical.md, governance.md, legal.md, creative.md, reasoning-failures.md, foio-editorial.md.

docs/citation-manager-boundary.md

@@ -2,25 +2,56 @@
 
 ## Status
 
-Superseded on 2026-06-10 by sourceright extraction (track `modernization_20260610`).
+Active handoff contract for Authentext (updated 2026-08-12 on track
+`domain-and-citation_20260811`). Supersedes the 2026-06-10 sourceright-only
+extraction note from track `modernization_20260610`.
 
 ## Decision
 
-Citation and reference management no longer lives in this repository. The former experimental tree (`experiments/citation_ref_manager/`), the `humanizer-cite` MCP tool, the `scripts/research/citation-normalize.js` helper, and the `src/references.json` / `src/research_references.md` bibliography files have been removed.
+Citation and reference management does not live in this repository. Authentext
+**detects** bibliography or reference-list work, **refuses** to generate,
+complete, reformat, or "fix" those lists, and **hands off** to the citation
+authority when one is present.
 
-**Permanent home:** the standalone **sourceright** project for citation normalization, reference verification, and bibliography maintenance.
+**Citation authority (optional integration):**
 
-## What stays in Humanizer
+- [edithatogo/sourceright](https://github.com/edithatogo/sourceright) —
+  reference triage and verification infrastructure (CSL canonicalisation,
+  provider evidence, review queues, bibliography exports).
+- [edithatogo/citeweft](https://github.com/edithatogo/citeweft) — neutral,
+  auditable scholarly document extraction core (span-preserving reference and
+  citation-callout extraction). CiteWeft does not assert bibliographic truth.
 
-Humanizer-next remains a writing-skill repository. Academic **content patterns** that detect fake or malformed AI citations (for example patterns A9/A10 in the academic module) are part of the humanizer skill contract and are **not** citation-manager tooling.
+Discovery is optional: a sibling clone, `AUTHENTXT_SOURCERIGHT_PATH` /
+`AUTHENTXT_CITEWEFT_PATH`, or a configured local skill directory. When those
+tools are absent, Authentext still refuses bibliography edits and points at the
+GitHub homes above. See `scripts/lib/citation-handoff.js`.
+
+Do not install sourceright or citeweft into this repository. Do not rebuild a
+citation database here.
+
+## What stays in Authentext
+
+Authentext remains a writing-skill repository. Academic **content patterns**
+that detect fake or malformed AI citations (for example patterns A9/A10 in the
+academic module) are part of the skill contract and are **not** citation-manager
+tooling. Clinical and legal modules likewise refuse reference-list edits and
+keep citations unverified by construction.
 
 ## Rationale
 
-- The maintained surface is `SKILL.md`, `SKILL_PROFESSIONAL.md`, and the `src/` fragments that compile them.
-- Citation-manager code did not feed the compiled skill outputs and duplicated work now owned by sourceright.
-- Removing it narrows the repo contract and avoids implying that Humanizer installs or maintains a reference database.
+- The maintained surface is `SKILL.md`, `SKILL_PROFESSIONAL.md`, and the `src/`
+  fragments that compile them.
+- Citation-manager code did not feed the compiled skill outputs and duplicated
+  work now owned by sourceright / citeweft.
+- Refusing locally either way avoids implying that Authentext installs or
+  maintains a reference database.
 
 ## History
 
-- **2026-03-14:** Citation manager moved from `src/` to `experiments/citation_ref_manager/` (see prior version of this document).
-- **2026-06-10:** Full extraction to sourceright; all citation-manager paths deleted from humanizer-next.
+- **2026-03-14:** Citation manager moved from `src/` to
+  `experiments/citation_ref_manager/` (see prior version of this document).
+- **2026-06-10:** Full extraction to sourceright; all citation-manager paths
+  deleted from humanizer-next.
+- **2026-08-12:** Authentext duty restated as detect / refuse / hand off;
+  citeweft named beside sourceright.

references/academic.md

@@ -253,7 +253,7 @@ Keep "ANOVA" only if the source already names it.
 
 ## CITATION AND REFERENCING
 
-Reference lists are **out of scope**. Do not generate, complete, reformat, or "fix" a bibliography. Do not invent missing references, DOIs, PMIDs, page ranges, or author lists. Leave the reference list untouched and report the gap.
+Reference lists are **out of scope**. Do not generate, complete, reformat, or "fix" a bibliography. Do not invent missing references, DOIs, PMIDs, page ranges, or author lists. Leave the reference list untouched and report the gap. Hand bibliography work to sourceright / citeweft when present (sibling clone, AUTHENTXT_SOURCERIGHT_PATH / AUTHENTXT_CITEWEFT_PATH, or configured skill); when absent, refuse locally and point to https://github.com/edithatogo/sourceright and https://github.com/edithatogo/citeweft.
 
 In-text citation tokens that already exist in the source are protected spans. You may flag a vague attribution (Pattern A1) or a padded span (Pattern A10). You may not rewrite the reference list to match.
 

references/clinical.md

@@ -151,7 +151,7 @@ Keep the effect size, interval, and limitation already in the source. Do not upg
 
 - Invent a dose, interval, or contraindication
 - Swap `serious` and `severe`
-- Generate, complete, or reformat a reference list
+- Generate, complete, or reformat a reference list (hand off to sourceright / citeweft when present; refuse locally either way)
 - Claim the pass makes the document clinically safe
 
 ---

references/creative.md

@@ -0,0 +1,187 @@
+# Module: Creative
+
+## Description
+
+This module applies to creative and narrative prose: fiction, narrative
+non-fiction, profiles, speeches, and personal essays. It repairs editorial
+defects with verified diffs. It does not optimise against detector scores and
+does not claim undetectability.
+
+Prefer subtractive and reordering edits over generative ones. Diagnose
+**structural** tells before surface style. Surface vocabulary and punctuation
+tells are low-weight here.
+
+**When to Apply:**
+
+- Fiction and short stories
+- Narrative non-fiction and personal essays
+- Profiles and speeches with a narrative arc
+- Creative-narrative document profiles
+
+**When NOT to Apply:**
+
+- Clinical, legal, or regulatory instruments
+- Submitted academic manuscripts that are not creative work
+- Technical reference documentation
+
+Authentext does not invent biographical facts for non-fiction. For fiction,
+invented detail is the work only when the author is writing new fiction; do not
+invent plot, setting, or character specifics the author has not supplied when
+the task is editorial repair of an existing draft.
+
+---
+
+## STRUCTURAL TELLS FIRST
+
+About nine in ten of the useful human-versus-AI fiction signal is narrative
+structure, not surface style. Diagnose these before scrubbing diction:
+
+1. **Thematic over-explanation** — the text names its theme instead of letting
+   scenes carry it.
+2. **Single-track plots** — every scene advances one plot line; no digression,
+   countercurrent, or wasted motion.
+3. **Tidy resolution** — conflicts close cleanly; moral ambiguity is flattened.
+4. **Low moral ambiguity** — characters are sorted into neat virtues and vices.
+5. **Specificity deficit** — places, objects, and bodies stay generic
+   ("a small town", "her eyes", "the old house").
+
+Report structural findings with passage anchors. Prefer cuts and reordering
+over new invented scenes.
+
+---
+
+## SPECIFICITY: ASK, DO NOT INVENT
+
+When a specificity gap blocks a good edit, **ask the author**. Do not invent
+the missing detail to make the prose sound human.
+
+- Non-fiction / memoir / profile: inventing a street name, employer, diagnosis,
+  or quotation is a Critical failure.
+- Fiction draft under editorial repair: do not invent plot beats, names, or
+  sensory specifics the author did not write. Ask, or leave the generic span
+  and report it.
+- Fiction drafting when the user explicitly asks you to write new fiction is
+  outside this repair module's invent-nothing rule for story content; still
+  do not invent real-world citations or living persons' private facts.
+
+---
+
+## SURFACE TELLS (LOW WEIGHT)
+
+Core patterns still apply, but weight them lightly for creative prose.
+Intentional fragments, repetition, tonal shifts, and unusual punctuation can
+be voice. Prefer leaving a deliberate quirk over flattening it into "clean"
+prose.
+
+---
+
+## MARKET POLICY AT EXPORT
+
+When the user is preparing a submission or export, surface relevant market
+policy if it changes what they must disclose. Do not invent a market's rule.
+
+Documented examples (check the named market before treating these as current):
+
+- **Clarkesworld** bans AI-assisted submissions.
+- **Authors Guild** certification is de minimis for some disclosure contexts.
+
+Authentext does not file the disclosure, certify compliance, or claim the
+pass makes a manuscript market-ready.
+
+---
+
+## CITATION AND REFERENCE LISTS
+
+Creative work that includes a bibliography or works-cited list still follows
+the Authentext refuse path: do not generate, complete, reformat, or "fix" a
+reference list. Hand that work to sourceright / citeweft when present; when
+absent, refuse locally and point to those projects
+(`https://github.com/edithatogo/sourceright`,
+`https://github.com/edithatogo/citeweft`).
+
+---
+
+## CREATIVE PATTERNS
+
+### Pattern R1: Thematic Billboarding
+
+**Problem:** AI announces the theme in thesis language.
+
+**Severity:** High
+
+**Before:**
+
+> This story explores the profound nature of grief and the human condition.
+
+**After:**
+
+> Cut the billboard. Keep the scene that already shows the grief. Do not invent
+> a new elegiac paragraph to replace the summary.
+
+### Pattern R2: Tidy Moral Closure
+
+**Problem:** AI resolves every tension into a neat lesson.
+
+**Severity:** Medium
+
+**Action:** Prefer leaving unresolved tension the source already has. Do not
+add a moral coda the author did not write.
+
+### Pattern R3: Invented Specificity
+
+**Problem:** AI fills a generic noun with a plausible proper noun or sensory
+detail that is not in the source.
+
+**Severity:** Critical for non-fiction; Critical for repair-of-draft fiction
+when the detail is new invention.
+
+**Action:** Ask the author, or leave the generic span and report the gap.
+
+---
+
+## SEVERITY CLASSIFICATION
+
+### Critical
+
+- Invented biographical or factual specifics in non-fiction
+- Invented plot, names, or sensory detail during editorial repair of a draft
+- Bibliography generation or "fixing"
+
+### High
+
+- Pattern R1: Thematic billboarding that replaces scene
+- Structural single-track flattening that drops an existing countercurrent
+
+### Medium
+
+- Pattern R2: Tidy moral closure
+- Over-smoothing of intentional fragments
+
+### Low
+
+- Surface vocabulary tells without a structural cluster
+
+---
+
+## BEST PRACTICES
+
+### Do
+
+- Diagnose structure before diction
+- Ask the author when specificity is missing
+- Keep intentional roughness that reads as voice
+- Surface named market AI policies at export time when relevant
+
+### Don't
+
+- Invent the missing detail to beat a detector
+- Flatten moral ambiguity the source already holds
+- Generate or repair reference lists
+- Treat Clarkesworld or Authors Guild notes as universal law without checking
+  the named market
+
+---
+
+_Module Version: 3.2.0_
+_Last Updated: 2026-08-12_
+_Applies to: Fiction, narrative non-fiction, profiles, speeches, personal essays_

references/governance.md

@@ -24,7 +24,7 @@ This module applies to governance writing: policies, risk assessments, complianc
 
 **When NOT to Apply:**
 
-- Creative writing
+- Creative writing (load [creative.md](creative.md) when that module is present)
 - Marketing materials
 - Informal communications
 

references/legal.md

@@ -143,7 +143,7 @@ Prefer the smaller edit. If either risk is present, leave the span and report it
 
 - Report omission-versus-invention for the edit set
 - Keep quotations and pinpoints exactly
-- Point citation verification at sourceright or citeweft when those tools are in scope; otherwise refuse bibliography work locally
+- Point citation verification at sourceright or citeweft (https://github.com/edithatogo/sourceright, https://github.com/edithatogo/citeweft) when those tools are in scope; otherwise refuse bibliography work locally
 
 ### Don't
 

scripts/check-sync-clean.js

@@ -16,6 +16,7 @@ const TARGET_FILES = [
   'references/clinical.md',
   'references/governance.md',
   'references/legal.md',
+  'references/creative.md',
   'references/reasoning-failures.md',
 ];