moai-domain-humanize
Post-editing specialist that removes "AI tells" from generated text and rewrites it to read as human-authored, while preserving meaning. This is the editing counterpart to text generation: it does not write new content, it refines how existing content is said. Covers Korean, English, Japanese, and Chinese, across two genre surfaces: prose (columns, reports, blog posts, formal documents) and marketing copy (headlines, CTAs, landing pages, brand storytelling, slide titles). Each language module carries a prose catalogue and a copy-layer catalogue; the shared machinery below (severity model, dual grading, mode-specific guardrails) applies uniformly.
Quick Reference
Operating Principles (4)
- Meaning preservation is the top rule. Facts, numbers, statistics, named entities, quotations, citations, and the author's stance/certainty stay intact. Any meaning drift forces a rollback. In copy mode, "meaning" is defined by the fact anchors plus the core promise/benefit — see the copy-mode guard below.
- Evidence-based edits only. Every change must trace to a detected tell on a specific span. Stylistic "improvements" unconnected to a catalogued tell are themselves an over-editing signal and are forbidden.
- Genre and register preservation. Humanize within the source register — academic stays academic, casual stays casual. Never push formal text into slang or vice versa. Copy and slide genres apply their own structural rules (noun-phrase title boundaries, appeal-vs-informational voice) defined in each module's copy layer.
- Over-editing prevention. In prose mode, flag at >30% change (WARN) and halt at >50% change (forced stop / human review) — above 50% you are regenerating, not humanizing. In copy mode, the change-rate guard is REPLACED by the fact-anchor preservation guard (see Over-Editing Guardrails below).
Genre Mode Selection (Prose vs Copy)
Two operating genres select which guardrail and grading table apply. Default from the text's genre; an explicit user instruction overrides.
| Mode |
Genres |
Over-editing guard |
Grading table |
| Prose mode (default) |
column, report, blog, formal/official document |
Change-rate guard (WARN >30%, HALT >50%) |
Prose-mode grades |
| Copy mode |
marketing copy, headline, CTA, landing page, brand story, slides |
Fact-anchor preservation guard |
Copy-mode grades |
Processing Mode Selection (Fast / Strict)
- Fast mode (default, up to ~5,000 chars): a single pass — detect, rewrite, self-verify against the meaning-preservation checklist.
- Strict mode (long or high-stakes text, or when requested): separate stages — detect → surgical rewrite → content-fidelity audit (facts/figures/stance unchanged) → naturalness review. Re-run a second pass when the result lands at Grade C.
Output Contract
Return two things:
- The humanized text.
- A short change report: categories hit (with counts), the final quality grade (A/B/C/D), and — in prose mode — the estimated percent changed. When a guardrail fires, state it explicitly (prose mode: WARN at >30%, HALT at >50%; copy mode: any fact-anchor loss).
Common Severity Model (shared by all 4 languages)
Each tell carries one severity tier. Detectors gate by occurrence count and overlap, because a single tell rarely proves AI authorship — confidence comes from clustering.
| Tier |
Name |
Rule |
| S1 |
Decisive |
A single occurrence strongly confirms AI authorship → remove on first occurrence. |
| S2 |
Strong |
Acceptable at 1–2 instances → remove at 3 or more. |
| S3 |
Weak |
Problematic only when overlapping other tells → downgrade-only contributor. |
Common Quality Grades (shared by all 4 languages — dual tables)
Graded after the rewrite. The genre mode selects the table: prose mode grades on residual tells plus change rate; copy mode grades on residual S1 plus fact-anchor integrity, with NO change-rate band.
Prose-Mode Grade Table
Residual S1/S2 counts plus improvement % (= proportion of detected tells removed without introducing new ones).
| Grade |
Criteria |
Action |
| A |
0 residual S1, ≤2 residual S2, ≥70% improvement |
Pass — reads as human-authored |
| B |
0 residual S1, ≤4 residual S2, ≥50% improvement |
Pass — minor polish remains |
| C |
1–2 residual S1, OR <50% improvement, OR over-edit WARN (>30%) |
Trigger a second pass |
| D |
≥3 residual S1, OR over-edit HALT (>50%), OR meaning drift detected |
Request human review; do not auto-ship |
Copy-Mode Grade Table
Residual S1 (including the module's copy-layer S1 tells), fact-anchor integrity, and self-verification — no change-rate band, because a legitimate headline rewrite routinely changes most of its characters while preserving every anchor.
| Grade |
Criteria |
Action |
| A |
0 residual S1, 0 fact-anchor loss, self-verification passed |
Pass — ships as human copy |
| B |
0 residual S1, ≤1 conservative fact-anchor concern |
Pass with an explicit note |
| C |
1 residual S1, OR self-verification partially failed |
Trigger a second pass |
| D |
2+ residual S1, OR 2+ fact-anchor losses |
Request human review; do not auto-ship |
Hard rule (both modes): any residual S1 caps the grade at C; any meaning-distortion flag forces D. S3 tells affect the grade only when ≥3 of them overlap and reinforce an S1/S2 finding.
Over-Editing Guardrails (shared)
Prose mode — change-rate guard. Change rate = the proportion of the text altered; target band ~5–30%.
- >30% changed → WARN. Surface a caution and cap at Grade C until each edit is justified by a detected tell. Note: padding-removal legitimately shrinks text, so a length drop alone is not a violation — flag when meaning-bearing spans are altered.
- >50% changed → HALT. Stop and require human confirmation; revert to the last safe state.
- Conservative judgment near the thresholds. This skill carries no quantitative measurement layer, so the change rate is an LLM estimate, not a reproducible metric. Treat a borderline estimate as OVER the threshold: near ~30%, issue the WARN; near ~50%, HALT. Bias toward caution so an over-edit never slips through on an optimistic estimate. (Known limitation: without a computed metric, before/after improvement percentages are estimates as well — report them as such.)
Copy mode — fact-anchor preservation guard (REPLACES the change-rate guard). In copy mode, meaning invariance is anchored differently: numbers, dates, prices, proper nouns, and legal notation are preserved 100% character-intact, AND the core promise/benefit of the copy keeps its meaning — while expression and sentence structure MAY be rewritten freely. The change-rate guard does not apply, because copy humanization legitimately rewrites most of a headline; the guard that replaces it is absolute on anchors:
- Any altered number, date, price, proper noun, or legal notation → rollback of that edit.
- Core promise/benefit drift → rollback. The rewritten copy must promise the same thing to the same audience.
- No invented specifics. Replacing vague copy with concrete claims is only allowed when the concrete facts exist in the source or are supplied by the author.
Meaning-Preservation Checklist (shared, all must hold)
- Anchor facts first — fix the claims, numbers, names, dates, and certainty level before editing.
- Edit at sentence/phrase level, not whole-document regeneration.
- Add no new facts — never invent specifics to replace vagueness; simplify instead, or flag for the author.
- Drop no load-bearing facts — removing an inflated wrapper must keep the substantive claim inside.
- Preserve genuine certainty/hedging and technical terminology verbatim.
- Final diff check — compare facts, tone, certainty, and examples against the original; revert any edit that drifts.
Invariant Ledger and Delta Audit
Two techniques harden the meaning-preservation machinery above: the Invariant Ledger makes the boundary explicit before editing, and the Delta Audit makes the survival check systematic after editing. They thread into the workflow (steps 2 and 6) rather than replacing any step, and they reinforce — never replace — the severity model, grades, guardrails, and the meaning-preservation checklist.
Invariant Ledger (pre-edit boundary)
Before editing, record an Invariant Ledger — the explicit list of what MUST survive the humanization pass unchanged. This is the written, checkable form of "Anchor facts first" (checklist item 1). Capture every item across the four categories:
- facts (with evidence boundaries)
- identifiers (commands, paths, URLs, status values, error codes, product names)
- conditions / numbers / dates / versions / units / comparisons
- exceptions / limitations / risks / uncertainty / approvals / rollback / next-actions
Fidelity rule. Never silently add, remove, narrow, broaden, strengthen, or weaken a ledger item. The wording is free to change; the commitment the text makes is not.
Mark each item supplied or inferred. A supplied item is something the source text actually asserts — it is hard-anchored, and any drift on it triggers a rollback (see Delta Audit). An inferred item is an adjacent benefit or guarantee the source never stated but a reader might assume — it is recorded for reviewer awareness only, and dropping it during humanization is NOT a rollback trigger, because the original never promised it. When an item is left unmarked, treat it as supplied: the fail-safe direction is preservation.
Depth by processing mode.
- Fast mode: a lightweight inline anchor list that still covers every one of the four categories above. Fast is shorter in FORM, never narrower in CATEGORY COVERAGE — even a short text gets a line for each category that applies.
- Strict mode: an explicit written boundary document with each item enumerated and marked supplied/inferred.
Delta Audit (post-edit verification)
After the edit pass completes, run a Delta Audit — compare the output against the Invariant Ledger before grading. This is the systematic form of "Final diff check" (checklist item 6), across three axes:
- Claim & intent parity — every claim the source makes, the output still makes, at the same strength and with the same intent.
- Survival check — every ledger identifier, number, condition, limitation, and risk is still present and unchanged.
- Audience / tone / purpose fit — the output still addresses the same reader, register, and goal.
Also flag any ambiguity the edit newly introduced: an unresolved actor, unclear ownership or handoff, or a softened destructive-effect or approval/rollback caveat.
Rollback on a supplied-item violation. When the audit finds any supplied ledger item added, removed, narrowed, broadened, strengthened, or weakened, roll back that edit — the same meaning-drift rollback the Operating Principles already require. Removal of an item marked inferred is reported in the audit output but does NOT trigger a rollback.
Feeds grading. A ledger violation is a meaning-distortion flag, and a meaning-distortion flag forces Grade D in both modes per the existing hard rule (see Common Quality Grades) — the Delta Audit is the mechanism that detects it.
Language Routing
Each target language has its own tell catalogue (categories, before/after examples in the target language, per-category severity). Load the module that matches the text being edited:
| Language |
Module |
Source basis |
| Korean (한국어) |
modules/korean.md |
Original catalogue — prose (10 categories A–J) + copy layer (A-20…A-25, L-1…L-8, M-1…M-3) |
| English |
modules/english.md |
Web-researched catalogue — prose (EN-A…EN-J) + copy layer (ENC-1…ENC-9) |
| Japanese (日本語) |
modules/japanese.md |
Web-researched catalogue — prose (JA-01…JA-09) + copy layer (JA-10…JA-14) |
| Chinese (中文) |
modules/chinese.md |
Web-researched catalogue — prose (CN-A…CN-K) + copy layer (CN-L…CN-Q) |
The Korean module is an original catalogue; the English, Japanese, and Chinese modules are independently web-researched catalogues built on the same architecture. Each module's copy layer is language-native — copy tells do NOT transfer mechanically between languages (English headlines are natively terse; Japanese 体言止め is prestigious craft gated by frequency, not presence; Chinese 对偶/排比 is judged content-first, not by count) — so never apply one language's copy rules to another. The common severity model and quality grades above apply uniformly to every module — the modules add only the language-specific tell categories, severities, and example rewrites.
For mixed-language text, detect the dominant language and route to its module; apply each module independently to its spans when the text is genuinely multilingual.
Genre-Module Routing
Two genre modules stack ON TOP of the language routing above — they never replace the language module:
| Surface / invocation |
Additional module |
| Display-surface copy — landing page, slide/card deck, design-tool result copy |
modules/design-copy.md (genre structure rules + per-language native measures), loaded in addition to the matching language module |
| Post-generation QA-gate review — copy produced by another tool or workflow, reviewed before application |
modules/copy-review.md (review-only mode: detect and propose, never auto-apply; six-stage pipeline + per-language formula dictionaries) |
When both conditions hold (a QA-gate review of display-surface copy), load both genre modules alongside the language module. The Language Routing table above remains the language axis and is unchanged by this extension.
Implementation Guide
Workflow (per text)
- Identify language, genre mode, and processing mode. Pick the module by dominant language; pick prose vs copy mode by genre (see Genre Mode Selection); pick Fast vs Strict by length / stakes.
- Anchor facts — build the Invariant Ledger. Record the four-category ledger (facts; identifiers; conditions/numbers/dates/versions/units/comparisons; exceptions/limitations/risks/uncertainty/approvals/rollback/next-actions) that must survive unchanged, marking each item supplied or inferred (see Invariant Ledger and Delta Audit above). Fast mode uses a lightweight inline anchor list covering every category; Strict mode uses an explicit written boundary document. In copy mode these anchors are the guard itself.
- Detect tells. Scan against the module's catalogue — the prose categories in prose mode, plus the module's Copy Layer categories in copy mode. Record each hit with its category ID, span, and severity. Count occurrences (S2/S3 gate on repetition).
- Rewrite surgically. Edit only flagged spans. Replace each tell with a natural rendering in the same register. Do not touch unflagged text.
- Apply the mode's guardrail. Prose mode: estimate the change rate and apply the change-rate guard (WARN >30%, HALT >50%, conservative near the thresholds). Copy mode: verify every fact anchor and the core promise/benefit instead.
- Self-verify (Fast) or audit + review (Strict) — run the Delta Audit. Compare the output against the Invariant Ledger for claim & intent parity, identifier/number/condition/limitation/risk survival, and audience/tone/purpose fit (see Invariant Ledger and Delta Audit above). A supplied-item violation rolls back that edit and raises a meaning-distortion flag (forcing Grade D); an inferred-item removal is reported but not rolled back. Re-run the meaning-preservation checklist. In Strict mode, run the content-fidelity audit and naturalness review as separate stages.
- Grade. Count residual S1/S2 and improvement %; assign A/B/C/D. Second pass on C; human review on D.
- Emit the humanized text + change report.
Detection note (shared across languages)
Automated AI-text detectors are unreliable across these four languages (notably weak on CJK polite registers, where they false-positive on correct formal writing). This skill is a pattern-based editing tool, not a detection oracle: rely on the catalogued tell categories and the clustering-based severity gates, not on a detector's verdict.
Common pitfalls
- Re-injecting AI-ness. Rewriting AI text with a fresh full regeneration tends to add new tells. Favor surgical edits to flagged spans over wholesale rewriting.
- Fabricating specifics. When a module calls for concrete detail to replace vague filler and no real specifics exist, simplify or flag for the author — never invent.
- Style drift. "Cleaning up" beyond the flagged tells violates principle 2 and inflates the change rate toward the HALT threshold.
- Mixing registers mid-document. Keep one consistent register (and, for Japanese/Chinese, one consistent politeness/sentence-ending style) across the whole output.
Works Well With
sync-auditor: independent skeptical review. Use it to score whether the humanized output preserved meaning against the original and met the target grade.
Category-catalogue structure inspired by the im-not-ai (Humanize KR) project.
Version: 1.3.0
1---
2name: moai-domain-humanize
3description: >
4 AI text humanization and 윤문 (post-editing) specialist that detects and removes
5 AI tells while preserving meaning, facts, and figures. Covers Korean, English,
6 Japanese, and Chinese with a shared severity model (S1/S2/S3), quality grades
7 (A/B/C/D), and 30%/50% over-editing guardrails. Use to make AI-generated text
8 read as human-authored without changing what it says (de-ai, naturalness pass).
9
10when_to_use: >
11 Use for AI-text humanization and post-editing (윤문): detecting and
12 removing AI tells across Korean, English, Japanese, and Chinese,
13 applying the S1/S2/S3 severity model and quality grades while preserving
14 meaning, facts, and figures.
15
16license: Apache-2.0
17compatibility: Designed for Claude Code
18allowed-tools: Read, Write, Edit, Grep, Glob
19user-invocable: false
20metadata:
21 source: https://skillmd.com/skills/modu-ai/moai-domain-humanize
22 version: "1.3.0"
23 category: "domain"
24 status: "active"
25 updated: "2026-07-24"
26 tags: "humanize, ai-tell, 윤문, post-edit, naturalness, multilingual, copy"
27
28# MoAI Extension: Progressive Disclosure
29progressive_disclosure:
30 enabled: true
31 level1_tokens: 100
32 level2_tokens: 5000
33---
34
35# moai-domain-humanize
36
37Post-editing specialist that removes "AI tells" from generated text and rewrites it to read as human-authored, while preserving meaning. This is the **editing** counterpart to text generation: it does not write new content, it refines how existing content is said. Covers Korean, English, Japanese, and Chinese, across two genre surfaces: **prose** (columns, reports, blog posts, formal documents) and **marketing copy** (headlines, CTAs, landing pages, brand storytelling, slide titles). Each language module carries a prose catalogue and a copy-layer catalogue; the shared machinery below (severity model, dual grading, mode-specific guardrails) applies uniformly.
38
39---
40
41## Quick Reference
42
43### Operating Principles (4)
44
451. **Meaning preservation is the top rule.** Facts, numbers, statistics, named entities, quotations, citations, and the author's stance/certainty stay intact. Any meaning drift forces a rollback. In copy mode, "meaning" is defined by the fact anchors plus the core promise/benefit — see the copy-mode guard below.
462. **Evidence-based edits only.** Every change must trace to a detected tell on a specific span. Stylistic "improvements" unconnected to a catalogued tell are themselves an over-editing signal and are forbidden.
473. **Genre and register preservation.** Humanize *within* the source register — academic stays academic, casual stays casual. Never push formal text into slang or vice versa. Copy and slide genres apply their own structural rules (noun-phrase title boundaries, appeal-vs-informational voice) defined in each module's copy layer.
484. **Over-editing prevention.** In prose mode, flag at >30% change (WARN) and halt at >50% change (forced stop / human review) — above 50% you are regenerating, not humanizing. In copy mode, the change-rate guard is REPLACED by the fact-anchor preservation guard (see Over-Editing Guardrails below).
49
50### Genre Mode Selection (Prose vs Copy)
51
52Two operating genres select which guardrail and grading table apply. Default from the text's genre; an explicit user instruction overrides.
53
54| Mode | Genres | Over-editing guard | Grading table |
55|------|--------|--------------------|---------------|
56| **Prose mode** (default) | column, report, blog, formal/official document | Change-rate guard (WARN >30%, HALT >50%) | Prose-mode grades |
57| **Copy mode** | marketing copy, headline, CTA, landing page, brand story, slides | Fact-anchor preservation guard | Copy-mode grades |
58
59### Processing Mode Selection (Fast / Strict)
60
61- **Fast mode** (default, up to ~5,000 chars): a single pass — detect, rewrite, self-verify against the meaning-preservation checklist.
62- **Strict mode** (long or high-stakes text, or when requested): separate stages — detect → surgical rewrite → content-fidelity audit (facts/figures/stance unchanged) → naturalness review. Re-run a second pass when the result lands at Grade C.
63
64### Output Contract
65
66Return two things:
67
681. **The humanized text.**
692. **A short change report**: categories hit (with counts), the final quality grade (A/B/C/D), and — in prose mode — the estimated percent changed. When a guardrail fires, state it explicitly (prose mode: WARN at >30%, HALT at >50%; copy mode: any fact-anchor loss).
70
71---
72
73## Common Severity Model (shared by all 4 languages)
74
75Each tell carries one severity tier. Detectors gate by occurrence count and overlap, because a single tell rarely proves AI authorship — confidence comes from clustering.
76
77| Tier | Name | Rule |
78|------|------|------|
79| **S1** | Decisive | A single occurrence strongly confirms AI authorship → remove on first occurrence. |
80| **S2** | Strong | Acceptable at 1–2 instances → remove at 3 or more. |
81| **S3** | Weak | Problematic only when overlapping other tells → downgrade-only contributor. |
82
83## Common Quality Grades (shared by all 4 languages — dual tables)
84
85Graded **after** the rewrite. The genre mode selects the table: prose mode grades on residual tells plus change rate; copy mode grades on residual S1 plus fact-anchor integrity, with NO change-rate band.
86
87### Prose-Mode Grade Table
88
89Residual S1/S2 counts plus improvement % (= proportion of detected tells removed without introducing new ones).
90
91| Grade | Criteria | Action |
92|-------|----------|--------|
93| **A** | 0 residual S1, ≤2 residual S2, ≥70% improvement | Pass — reads as human-authored |
94| **B** | 0 residual S1, ≤4 residual S2, ≥50% improvement | Pass — minor polish remains |
95| **C** | 1–2 residual S1, OR <50% improvement, OR over-edit WARN (>30%) | Trigger a second pass |
96| **D** | ≥3 residual S1, OR over-edit HALT (>50%), OR meaning drift detected | Request human review; do not auto-ship |
97
98### Copy-Mode Grade Table
99
100Residual S1 (including the module's copy-layer S1 tells), fact-anchor integrity, and self-verification — no change-rate band, because a legitimate headline rewrite routinely changes most of its characters while preserving every anchor.
101
102| Grade | Criteria | Action |
103|-------|----------|--------|
104| **A** | 0 residual S1, 0 fact-anchor loss, self-verification passed | Pass — ships as human copy |
105| **B** | 0 residual S1, ≤1 conservative fact-anchor concern | Pass with an explicit note |
106| **C** | 1 residual S1, OR self-verification partially failed | Trigger a second pass |
107| **D** | 2+ residual S1, OR 2+ fact-anchor losses | Request human review; do not auto-ship |
108
109Hard rule (both modes): any residual S1 caps the grade at C; any meaning-distortion flag forces D. S3 tells affect the grade only when ≥3 of them overlap and reinforce an S1/S2 finding.
110
111### Over-Editing Guardrails (shared)
112
113**Prose mode — change-rate guard.** Change rate = the proportion of the text altered; target band ~5–30%.
114
115- **>30% changed → WARN.** Surface a caution and cap at Grade C until each edit is justified by a detected tell. Note: padding-removal legitimately shrinks text, so a length drop alone is not a violation — flag when meaning-bearing spans are altered.
116- **>50% changed → HALT.** Stop and require human confirmation; revert to the last safe state.
117- **Conservative judgment near the thresholds.** This skill carries no quantitative measurement layer, so the change rate is an LLM estimate, not a reproducible metric. Treat a borderline estimate as OVER the threshold: near ~30%, issue the WARN; near ~50%, HALT. Bias toward caution so an over-edit never slips through on an optimistic estimate. (Known limitation: without a computed metric, before/after improvement percentages are estimates as well — report them as such.)
118
119**Copy mode — fact-anchor preservation guard (REPLACES the change-rate guard).** In copy mode, meaning invariance is anchored differently: numbers, dates, prices, proper nouns, and legal notation are preserved 100% character-intact, AND the core promise/benefit of the copy keeps its meaning — while expression and sentence structure MAY be rewritten freely. The change-rate guard does not apply, because copy humanization legitimately rewrites most of a headline; the guard that replaces it is absolute on anchors:
120
121- **Any altered number, date, price, proper noun, or legal notation → rollback** of that edit.
122- **Core promise/benefit drift → rollback.** The rewritten copy must promise the same thing to the same audience.
123- **No invented specifics.** Replacing vague copy with concrete claims is only allowed when the concrete facts exist in the source or are supplied by the author.
124
125### Meaning-Preservation Checklist (shared, all must hold)
126
1271. Anchor facts first — fix the claims, numbers, names, dates, and certainty level before editing.
1282. Edit at sentence/phrase level, not whole-document regeneration.
1293. Add no new facts — never invent specifics to replace vagueness; simplify instead, or flag for the author.
1304. Drop no load-bearing facts — removing an inflated wrapper must keep the substantive claim inside.
1315. Preserve genuine certainty/hedging and technical terminology verbatim.
1326. Final diff check — compare facts, tone, certainty, and examples against the original; revert any edit that drifts.
133
134---
135
136## Invariant Ledger and Delta Audit
137
138Two techniques harden the meaning-preservation machinery above: the **Invariant Ledger** makes the boundary explicit *before* editing, and the **Delta Audit** makes the survival check systematic *after* editing. They thread into the workflow (steps 2 and 6) rather than replacing any step, and they reinforce — never replace — the severity model, grades, guardrails, and the meaning-preservation checklist.
139
140### Invariant Ledger (pre-edit boundary)
141
142Before editing, record an Invariant Ledger — the explicit list of what MUST survive the humanization pass unchanged. This is the written, checkable form of "Anchor facts first" (checklist item 1). Capture every item across the four categories:
143
1441. **facts** (with evidence boundaries)
1452. **identifiers** (commands, paths, URLs, status values, error codes, product names)
1463. **conditions / numbers / dates / versions / units / comparisons**
1474. **exceptions / limitations / risks / uncertainty / approvals / rollback / next-actions**
148
149**Fidelity rule.** Never silently add, remove, narrow, broaden, strengthen, or weaken a ledger item. The wording is free to change; the commitment the text makes is not.
150
151**Mark each item supplied or inferred.** A **supplied** item is something the source text actually asserts — it is hard-anchored, and any drift on it triggers a rollback (see Delta Audit). An **inferred** item is an adjacent benefit or guarantee the source never stated but a reader might assume — it is recorded for reviewer awareness only, and dropping it during humanization is NOT a rollback trigger, because the original never promised it. When an item is left unmarked, treat it as **supplied**: the fail-safe direction is preservation.
152
153**Depth by processing mode.**
154- **Fast mode**: a lightweight inline anchor list that still covers every one of the four categories above. Fast is shorter in FORM, never narrower in CATEGORY COVERAGE — even a short text gets a line for each category that applies.
155- **Strict mode**: an explicit written boundary document with each item enumerated and marked supplied/inferred.
156
157### Delta Audit (post-edit verification)
158
159After the edit pass completes, run a Delta Audit — compare the output against the Invariant Ledger before grading. This is the systematic form of "Final diff check" (checklist item 6), across three axes:
160
1611. **Claim & intent parity** — every claim the source makes, the output still makes, at the same strength and with the same intent.
1622. **Survival check** — every ledger identifier, number, condition, limitation, and risk is still present and unchanged.
1633. **Audience / tone / purpose fit** — the output still addresses the same reader, register, and goal.
164
165Also flag any ambiguity the edit newly introduced: an unresolved actor, unclear ownership or handoff, or a softened destructive-effect or approval/rollback caveat.
166
167**Rollback on a supplied-item violation.** When the audit finds any **supplied** ledger item added, removed, narrowed, broadened, strengthened, or weakened, roll back that edit — the same meaning-drift rollback the Operating Principles already require. Removal of an item marked **inferred** is reported in the audit output but does NOT trigger a rollback.
168
169**Feeds grading.** A ledger violation is a meaning-distortion flag, and a meaning-distortion flag forces **Grade D** in both modes per the existing hard rule (see Common Quality Grades) — the Delta Audit is the mechanism that detects it.
170
171---
172
173## Language Routing
174
175Each target language has its own tell catalogue (categories, before/after examples in the target language, per-category severity). Load the module that matches the text being edited:
176
177| Language | Module | Source basis |
178|----------|--------|--------------|
179| Korean (한국어) | `modules/korean.md` | Original catalogue — prose (10 categories A–J) + copy layer (A-20…A-25, L-1…L-8, M-1…M-3) |
180| English | `modules/english.md` | Web-researched catalogue — prose (EN-A…EN-J) + copy layer (ENC-1…ENC-9) |
181| Japanese (日本語) | `modules/japanese.md` | Web-researched catalogue — prose (JA-01…JA-09) + copy layer (JA-10…JA-14) |
182| Chinese (中文) | `modules/chinese.md` | Web-researched catalogue — prose (CN-A…CN-K) + copy layer (CN-L…CN-Q) |
183
184The Korean module is an original catalogue; the English, Japanese, and Chinese modules are independently web-researched catalogues built on the same architecture. Each module's copy layer is language-native — copy tells do NOT transfer mechanically between languages (English headlines are natively terse; Japanese 体言止め is prestigious craft gated by frequency, not presence; Chinese 对偶/排比 is judged content-first, not by count) — so never apply one language's copy rules to another. The common severity model and quality grades above apply uniformly to every module — the modules add only the language-specific tell categories, severities, and example rewrites.
185
186For mixed-language text, detect the dominant language and route to its module; apply each module independently to its spans when the text is genuinely multilingual.
187
188### Genre-Module Routing
189
190Two genre modules stack ON TOP of the language routing above — they never replace the language module:
191
192| Surface / invocation | Additional module |
193|----------------------|-------------------|
194| Display-surface copy — landing page, slide/card deck, design-tool result copy | `modules/design-copy.md` (genre structure rules + per-language native measures), loaded in addition to the matching language module |
195| Post-generation QA-gate review — copy produced by another tool or workflow, reviewed before application | `modules/copy-review.md` (review-only mode: detect and propose, never auto-apply; six-stage pipeline + per-language formula dictionaries) |
196
197When both conditions hold (a QA-gate review of display-surface copy), load both genre modules alongside the language module. The Language Routing table above remains the language axis and is unchanged by this extension.
198
199---
200
201## Implementation Guide
202
203### Workflow (per text)
204
2051. **Identify language, genre mode, and processing mode.** Pick the module by dominant language; pick prose vs copy mode by genre (see Genre Mode Selection); pick Fast vs Strict by length / stakes.
2062. **Anchor facts — build the Invariant Ledger.** Record the four-category ledger (facts; identifiers; conditions/numbers/dates/versions/units/comparisons; exceptions/limitations/risks/uncertainty/approvals/rollback/next-actions) that must survive unchanged, marking each item supplied or inferred (see Invariant Ledger and Delta Audit above). Fast mode uses a lightweight inline anchor list covering every category; Strict mode uses an explicit written boundary document. In copy mode these anchors are the guard itself.
2073. **Detect tells.** Scan against the module's catalogue — the prose categories in prose mode, plus the module's Copy Layer categories in copy mode. Record each hit with its category ID, span, and severity. Count occurrences (S2/S3 gate on repetition).
2084. **Rewrite surgically.** Edit only flagged spans. Replace each tell with a natural rendering in the same register. Do not touch unflagged text.
2095. **Apply the mode's guardrail.** Prose mode: estimate the change rate and apply the change-rate guard (WARN >30%, HALT >50%, conservative near the thresholds). Copy mode: verify every fact anchor and the core promise/benefit instead.
2106. **Self-verify (Fast) or audit + review (Strict) — run the Delta Audit.** Compare the output against the Invariant Ledger for claim & intent parity, identifier/number/condition/limitation/risk survival, and audience/tone/purpose fit (see Invariant Ledger and Delta Audit above). A supplied-item violation rolls back that edit and raises a meaning-distortion flag (forcing Grade D); an inferred-item removal is reported but not rolled back. Re-run the meaning-preservation checklist. In Strict mode, run the content-fidelity audit and naturalness review as separate stages.
2117. **Grade.** Count residual S1/S2 and improvement %; assign A/B/C/D. Second pass on C; human review on D.
2128. **Emit** the humanized text + change report.
213
214### Detection note (shared across languages)
215
216Automated AI-text detectors are unreliable across these four languages (notably weak on CJK polite registers, where they false-positive on correct formal writing). This skill is a **pattern-based editing tool**, not a detection oracle: rely on the catalogued tell categories and the clustering-based severity gates, not on a detector's verdict.
217
218### Common pitfalls
219
220- **Re-injecting AI-ness.** Rewriting AI text with a fresh full regeneration tends to add new tells. Favor surgical edits to flagged spans over wholesale rewriting.
221- **Fabricating specifics.** When a module calls for concrete detail to replace vague filler and no real specifics exist, simplify or flag for the author — never invent.
222- **Style drift.** "Cleaning up" beyond the flagged tells violates principle 2 and inflates the change rate toward the HALT threshold.
223- **Mixing registers mid-document.** Keep one consistent register (and, for Japanese/Chinese, one consistent politeness/sentence-ending style) across the whole output.
224
225---
226
227## Works Well With
228
229- `sync-auditor`: independent skeptical review. Use it to score whether the humanized output preserved meaning against the original and met the target grade.
230
231---
232
233Category-catalogue structure inspired by the im-not-ai (Humanize KR) project.
234
235Version: 1.3.0