Update to APODICTIC Development Editor

{
  "description": "Argument-shaped nonfiction development engine for op-ed, policy brief, testimony, academic argument, open letter, white paper, advocacy argument, legal brief, regulatory comment, expert affidavit, and writer-confirmed lens/exploratory essays. Provides Dialectical Clarity audit, Argument Red Team, Argument Persuasion, Argument Evidence Deep-Dive, Adversarial Evidence Review, Field Reconnaissance, and Citation Verifier as companion audits. Also covers nonfiction argument coaching and nonfiction pre-draft argument spine work. Invoked when intake resolves constraint=nonfiction + persuasive-argument form.\n",
  "included_files": [],
  "name": "nonfiction-argument-engine",
  "skill_md_contents": "---\nname: nonfiction-argument-engine\ndescription: >\n  Argument-shaped nonfiction development engine for op-ed, policy brief, testimony,\n  academic argument, open letter, white paper, advocacy argument, legal brief,\n  regulatory comment, expert affidavit, and writer-confirmed lens/exploratory essays. Provides Dialectical Clarity audit,\n  Argument Red Team, Argument Persuasion, Argument Evidence Deep-Dive, Adversarial\n  Evidence Review, Field Reconnaissance, and Citation Verifier as companion audits.\n  Also covers nonfiction argument coaching and nonfiction pre-draft argument spine work.\n  Invoked when intake resolves constraint=nonfiction + persuasive-argument form.\nversion: 2.13.3\n---\n\n# Nonfiction Argument Engine\n\n*A named, first-class skill within the APODICTIC Development Editor. The argument path\nhas always existed — this skill makes it explicit, discoverable, and bounded.*\n\n---\n\n## Canonical Source\n\nThis skill is the named orchestrator for argument-shaped nonfiction. It does not duplicate\nthe core-editor's spine; it specializes it for argument runs. The core-editor owns the\neditorial framework; this skill owns the argument-specific routing, audit activation, and\nstate management.\n\n**Upstream:** Invoked by `core-editor/SKILL.md §Delegation Rules — Nonfiction Argument Engine`\nwhen intake resolves `constraint=nonfiction + persuasive-argument form`.\n\n---\n\n## The Firewall\n\n*The no-content-invention firewall applies in argument mode exactly as in fiction mode.\nArgument editing = diagnosing inferential structure, not authoring claims.*\n\n<!-- Load `../core-editor/references/firewall.md` for the canonical Firewall definition.\n     This skill references the same file; there is exactly one canonical source. -->\n\nSee `core-editor/references/firewall.md` for the full definition. In argument context:\nthe Firewall means the engine diagnoses warrant gaps, objection architecture, and claim\nintegrity — it does NOT author new claims, invent evidence, or fill in the argument for\nthe writer. It also draws the line the warrant/premise split depends on: the engine returns\na **warrant verdict** (whether the reasoning warrants the conclusion — the inference axis,\nLocal Relevance + Local Sufficiency) and, separately, **premise-plausibility flags** (whether\na load-bearing premise is contestable — the acceptability axis, Local Acceptability). The flags\nsurface a premise for the writer to earn; they never rule it true or false, and they never\nchange the warrant verdict by themselves.\n\n---\n\n## Delegation Contract (core-editor → engine)\n\n**Trigger:** intake router resolves `constraint=nonfiction` AND form is in the persuasive-\nargument family per `core-editor/references/intake-router-runtime.md §6 Table A`:\n  - op-ed, policy brief, testimony (oral/written), academic article, open letter\n  - white paper, advocacy argument, legal brief, regulatory comment, expert affidavit\n  - grant proposal, book review, recommendation memo\n\n**NOT argument-shaped:** memoir, narrative nonfiction, creative nonfiction, personal essay\nwithout a discernible persuasive claim ladder. Those route via their own §4a rows.\n\n**Run-shape fields passed from core-editor:**\n  - `form` — the specific argument form (op-ed, policy brief, etc.)\n  - `register` — `asserted` (default) or writer-confirmed `generative`\n  - `register_confirmation` — `DEFAULT-ASSERTED`, `WRITER-CONFIRMED`, or `FORCED-ASSERTED`\n  - `audience_burden` — HIGH / MEDIUM / LOW per §Step 1 of Dialectical Clarity\n  - `stakes` — moral / epistemic / practical / mixed\n  - `high_stakes_flag` — boolean (triggers Hard Prerequisite tier for Field Recon)\n  - `constraint` — `nonfiction` (always set on this delegation path)\n\n**Artifact set returned to core-editor:** unchanged from the fiction path:\n  - Editorial letter (Argument-DE class)\n  - Findings Ledger (`apodictic.finding.v1` blocks with argument-cluster severities)\n  - `Argument_State.md` (Pass-10-Class rolling artifact; schema per argument-state-schema)\n  - Marked-up manuscript (Annotation Manifest + Annotated copy)\n\nThe synthesis spine is UNCHANGED: root causes → triage → adversarial self-check →\neditorial letter → mechanical validation. The argument-specific path only adds the\nargument-cluster audit activation and the Argument_State artifact.\n\n---\n\n## Scope Boundary\n\n**Owned (argument-shaped nonfiction):**\n- Dialectical Clarity (primary diagnostic audit)\n- Argument Red Team (adversarial pressure testing)\n- Argument Persuasion (audience-fit testing)\n- Argument Evidence Deep-Dive (evidence-chain integrity)\n- Adversarial Evidence Review (hostile-expert survivability)\n- Field Reconnaissance (literature-counterevidence surfacing)\n- Citation Verifier (citation provenance)\n- Argument_State.md (rolling argument spine state)\n- Nonfiction argument coaching (post-diagnosis revision guidance)\n- Nonfiction pre-draft argument spine (idea-stage claim ladder seeding)\n\n**Adjacent but NOT owned (flag, do not own):**\n- Narrative nonfiction / reported features → route via `genre/narrative-nonfiction.md`\n- Memoir / creative nonfiction → route via `genre/memoir-creative-nonfiction.md`\n- Fiction with argument themes (philosophical fiction, didactic fiction) → core-editor\n  handles via Pass 9 + Dialectical Clarity; this engine does NOT activate on fiction\n\n---\n\n## Owned References\n\nThis skill loads the following reference files:\n\n| Reference | Role |\n|-----------|------|\n| `core-editor/references/firewall.md` | Canonical Firewall definition (single source) |\n| `core-editor/references/argument-audits-routing.md` | Argument-cluster audit routing rows (§4a/§4b fragment) |\n| `core-editor/references/argument-audits-propagation.md` | Argument-cluster §4e signal-propagation rows (fragment) |\n| `core-editor/references/nonfiction-intake-routing.md` | Nonfiction triage routing fragment (§6 Table A + triage logic) |\n| `core-editor/references/synthesis-argument.md` | Argument-DE decision-layer schema (§8 Argument-DE class fragment) |\n| `core-editor/references/intake-router-runtime.md` | Full intake routing (for §6 Table A / form identification) |\n| `core-editor/references/pass-dependencies.md` | Pass resolver + audit tier definitions + §4e table |\n| `core-editor/references/run-synthesis.md` | Synthesis spine (shared with fiction path) |\n| `argument-audits/references/craft/rhetorical-stance-triage.md` | Argument-move stance taxonomy and pre-lock calibration protocol |\n\n---\n\n## Argument Engine Workflow\n\n### 1. Argument Intake\n\nLoad `core-editor/references/intake-router-runtime.md §6`. Run nonfiction triage per\nthe fragment `core-editor/references/nonfiction-intake-routing.md`:\n\n- Identify form, audience, stakes, constraint=high-stakes\n- Propose `generative` only for a lens/exploratory form and obtain writer confirmation;\n  otherwise record the asserted default. A high-stakes signal forces asserted.\n- Seed `Argument_State.md` with the initial `apodictic:argument_spine` block\n- Identify the Hard Prerequisite tier requirement (high-stakes signal → Field Recon first)\n\n### 2. Audit Activation (Pre-Pass)\n\nLoad `core-editor/references/argument-audits-routing.md` to resolve which argument-cluster\naudits activate at which tier per the run-shape:\n\n- Hard Prerequisite tier (high-stakes): Field Reconnaissance MUST complete before passes\n- Pre-DE Prerequisite tier (high-stakes): Citation Verifier runs before the DE begins\n- Auto-recommend before synthesis: Dialectical Clarity (always for argument-shaped runs)\n- Additional audits per §4a/§4b trigger logic in the routing fragment\n\n### 3. Pass Execution (Shared Spine)\n\nThe argument engine uses the SAME pass set as the fiction path (Passes 0, 1, 2, 5, 8\nfor the baseline floor; full pass set for expanded runs). Argument-specific passes are\nhandled by the audit cluster, not by new passes. The Findings Ledger structure is unchanged.\n\n### 4. Synthesis (Argument-DE Class)\n\nLoad `core-editor/references/synthesis-argument.md` for the Argument-DE decision-layer\nschema. The synthesis spine is unchanged; only the decision-layer section names differ\n(Protected Elements → Strengths / Protected Elements; Author Decisions → Editorial-Dispute\nTerritory). See `core-editor/references/run-synthesis.md §Step 8 Argument-DE class` for\nthe full schema.\n\nSignal propagation for argument-cluster audits follows `core-editor/references/argument-audits-propagation.md`\n(the §4e fragment for the argument cluster).\n\n### 5. Argument_State.md Update\n\nAfter synthesis, update `Argument_State.md` at the project root per the schema in\n`core-editor/references/run-synthesis.md §Argument_State`. This is the argument path's\nPass-10-Class rolling artifact — it persists across runs and is diffed on re-runs.\n\n### 6. Optional: author decisions and export\n\nOffer these only when the writer asks, or when the letter's next steps call for them:\n\n- **`apodictic-adjudicate`** — the author approves, rejects, or revises the claims and relationships in an existing reconstruction claim graph, one decision at a time, with resumable state beside the manuscript. It records decisions; it never drafts. Semantic screening of rejections is not built yet, so a nonempty rejection set blocks new approvals (the command explains this).\n- **AIF export** — a one-way, loss-aware projection of `Argument_State.md` into the AIF-Core graph format (Toulmin-style I-nodes, RA-nodes carrying the warrant, CA-nodes for typed objections) for argument-mapping tools: plugin-bundled `scripts/validate.sh argument-aif-export <project>/Argument_State.md --state-schema 0.2.0 --out <project>/Argument_AIF.json`, then `argument-aif-check` on the output. Every construct it cannot map is listed as a named loss, never invented. The export is never read back to set findings or severities.\n\n---\n\n## QA Guardrails\n\nAll core-editor QA guardrails apply (see `core-editor/SKILL.md §QA Guardrails`). Argument-specific additions:\n\n1. **Warrant verification before severity assignment.** Every WR-class flag must name the specific\n   inferential gap — \"the audience must supply X\" — not a generic \"needs more evidence.\"\n2. **Burden calibration.** Audience burden (HIGH/MEDIUM/LOW per §Step 1 of Dialectical Clarity)\n   governs severity floors for WR0 and related flags. Verify burden level before assigning Must-Fix.\n3. **Hard Prerequisite compliance.** On high-stakes runs, Field Reconnaissance MUST complete\n   before any Tier 2 evaluative pass. A declined Hard Prerequisite cannot proceed silently.\n4. **Firewall in argument context.** The engine diagnoses inferential structure; it does NOT author\n   claims, invent evidence, or complete the argument. See `core-editor/references/firewall.md`.\n5. **Premise-plausibility flags are not truth adjudications.** A Step-9 / GT8 premise flag records\n   that a load-bearing premise's *acceptability* is contestable, overloaded, or unearned; it never\n   rules the premise true or false, supplies missing evidence, or alters the warrant verdict. Any\n   output that adjudicates a flagged premise as true or false fails the Firewall check.\n6. **Register confirmation before calibration.** A generative run records writer confirmation,\n   consequence context, and the exact cash-out inventory in Argument_State §1. Silent inference is\n   a QA failure. Intake-declared high stakes forces `register=asserted`.\n7. **Stance verdicts classify moves, never claims.** `earned` is not evidence that a claim is true;\n   `unearned` is not evidence that it is false. Calibration occurs at Triage before the Deficit Lock,\n   and prescriptive cash-outs remain at asserted burden. The shape gate validates recorded\n   verdicts/effects and supplied cash-out joins; the auditor still owns completeness because an\n   omitted optional join or undeclared stance cannot be inferred mechanically from prose location.\n\n---\n\n*The nonfiction argument path has always been part of APODICTIC. This skill makes it named,\nexplicit, and bounded — so it can be selected, tested, and extended without touching the fiction path.*\n"
}