v2.4.0

v2.4.0 - 2026-06-14

Nonfiction Argument Engine — uncompared-recommendation carve-out

dialectical-clarity.md gains classification rule 2a: for an argument whose controlling claim is a recommendation to act (AT3 — "X should do Y"), the comparative dimension is constitutive of the claim, so an AT3 recommendation that discharges none of its comparative burden (BP5 primary + OB3, no funding mechanism) is not evaluable as a recommendation — a defeat under decision test two (Evidence-evaluability) — and the verdict is Structurally Unsound (FM-A10, The Uncompared Proposal), with a matching note at the Step-9 Final Diagnostic Question. This is a bounded exception to rule 2's default-to-SOUND discipline, scoped to AT3 recommendations only (descriptive/explanatory/interpretive theses are untouched) and guarded by a wholly-absent vs. partially-discharged line — a recommendation that engages even one alternative thinly stays a Should-Fix soft spot in a sound argument. Calibration update (post-benchmark): the guard is tightened so that naming any alternative — even a weak or strawmanned foil — counts as partially-discharged (soft spot); only the total absence of any comparison triggers Unsound. (A benchmark run had andreessen-techno-optimist-manifesto regress SOUND→UNSOUND because a strawman "the only alternative is Communism" framing was misread as zero comparison.) Two independent cross-vendor reviewers (Gemini + GPT-5.5) ratified the narrowing and both flagged a token-foil gaming risk, so an anti-gaming clause was added: a named foil disables only the automatic FM-A10 defeat; a merely decorative foil (no mechanism/criteria/costs/tradeoff) can still be Unsound via the general evaluability test (rule 2), not the AT3 auto-trigger. It brings the engine into line with the policy-brief-uncompared ground-truth key (GT7 = UNSOUND), which the engine previously read SOUND. Verdict-behavior change for argument-shaped runs; gated on a benchmark convergence run (no --check-all gate covers behavioral calibration). No validator/schema change. See docs/argument-benchmark-calibration-round.md.

Command Surface — trimmed to 13

Retired three redundant command entry points, all reachable through /start: /revision-plan (a compatibility alias for /coach), /develop-edit (the default full_draft + repair router path), and /diagnose (a targeted repair). A writer with a draft lands on the /start router first anyway, so these added surface without adding capability. The distinct doors stay first-class (/ready, /pre-writing, /coach, /audit, /research, /plot-coach, /legal-risk, /triage-feedback, /reader-questions, /new-project, /projects). The registry command taxonomy (category / status / routerEquivalent / writerQuestion) is the single source the grouped README lists generate from. Routing references that pointed at the retired commands now point at /start.

Validators — finding-trace completion glob narrowed

finding-trace's _COMPLETION_GLOBS narrowed from *_Revision_*.md to *_Revision_Report_*.md, so a deadline-coaching *_Revision_Calendar_*.md is no longer mis-classified as a completion artifact (which would let its mentions advance a finding toward revised). Aligns finding-trace with the Increment-4a revision_round gate, which already narrowed its revision_report key. Negative-test guarded (calendar_not_completion); the revision-stage glob (_REVISION_GLOBS, for plan-coverage) stays broad.

Workflows — Legal Risk Register detection layer

The Legal Risk Register gains a detection layer (core-editor/references/legal-risk-register.md §Detection guidance / §Escalation-trigger taxonomy): per-class textual signals for what to flag under each risk_class (defamation / privacy / rights-clearance / other, with the finer categories — intrusion, false light, trade secrets, incitement, … — as sub-signals), a severity model (base tier raised by documented modifiers: +identifiable-living-private-person, +serious-allegation, +weak-or-no-documentation, +international-distribution, +author-signed-agreement, +marketing/cover/merchandise-use, +minor-or-vulnerable-subject), route-to-counsel bright lines, a flag-don't-resolve posture for jurisdiction divergence, and a compact controlled-vocabulary escalation-trigger taxonomy (~20 codes → default tiers) for the escalation_trigger field. Lean by design — the runtime module carries the heuristics; the cross-model research + citations live in docs/legal-risk-detection-level-setting.md. Firewall unchanged: detection only; a qualified lawyer is the final gate. No schema or validator change.

Workflows — Legal Risk Register router wiring

The built Legal Risk Register module is now reachable from the router and a command, not just internally. constraint:risk ("sensitive or legally risky content") offers the register and, on accept, attaches [Project]_Legal_Risk_Register_[runlabel].md as a companion artifact (synthesis constraint hook in run-synthesis.md — the first constraint:-keyed presentation overlay, and the first offer-then-attach one, since the not-a-lawyer framing warrants a confirm). New /legal-risk command as a direct entry point. The route map flips to Built (§3 option D, §6 Table B, §4a). The firewall is unchanged — flag, don't practice law. (Still future: auto-recommending the register for memoir/autofiction with identifiable real people without the explicit flag.)

Onboarding — install decision-aid and glossary

The README install section now opens with a Which install do I need? table that maps each host (Antigravity, Codex, Claude Code CLI, Cowork) to its path and fastest route, so a newcomer doesn't have to read all five install flows to find theirs. A new Key Terms section defines the load-bearing vocabulary a first-timer meets cold — contract, controlling idea, the Firewall, pass, macro block, audit, genre module, editorial letter, the Must/Should/Could severity tiers (and the Deficit Lock), spine, and reverse outline.

Onboarding — visual surfaces brought current

The README now opens with a See It in Action section linking the rendered sample editorial letters, the targeted-audit and pre-writing samples, and the two interactive maps (overview dashboard, route explorer) so newcomers can see real output before installing, plus a Your First Five Minutes walkthrough.

The overview-dashboard.html and route-explorer.html visuals (and their .codex.html twins) were stale at a v1.0.2 snapshot and have been brought current: the route explorer no longer reports shipped workflows (Fragment Synthesis, partial-manuscript mode, Submission Readiness, Submission Triage, Feedback Triage, the Nonfiction Argument Engine, editor scaffolding, diagnostic vocabulary, Series Continuity) as "not yet built"; the Legal Risk Register is shown as built-but-not-yet-routed; only multi-party/team intake remains a true gap. The overview dashboard now shows the canonical 8-block macro map (Emotional Dynamics restored as its own block), the full 50 spines / 12 families (adding Kishōtenketsu and Jo-ha-kyū), and the current version. The dashboard's click-to-expand cards are now keyboard-operable (role="button", tabindex, Enter/Space handlers).

Onboarding — overview dashboard front door

The overview dashboard header now opens with a "New here?" getting-started callout — a one-line plain-language orientation ("this page is a map; you don't need to memorize it") plus the first action (/start) — so a brand-new user landing on the page sees what to do before scrolling into the technical sections. Additive only (one <div> + two CSS rules; no redesign, no new sections, no network/deps). Applied to both the canonical overview-dashboard.html and its authored .codex.html twin (which keeps the apodictic-start wrapper naming per the codex override convention).

Validators — post-merge review nits

Three small fixes from a review of the merged batch (no new validators; count unchanged at 38→40 baseline):

• reader-instrument B3 (fabrication smell-test). The "unsourced question" advisory was gated on the Ledger having ### Unresolved Questions bullets, so an unresolved-question reader-question with an invented source_note passed unflagged when the Ledger had none (the more suspicious case — citing a UQ that can't exist). The advisory now also fires when the Ledger has zero UQ bullets. Stays a WARN: UQ provenance is non-referential by design, so this is a fabrication smell-test, not a hard gate.
• manuscript-viz render gate (false pacing curve). W2 scene order is advisory, but a reordered manifest draws a false pacing curve — the one warning that corrupts the render's core output. The render subcommand now refuses on a scene-order divergence too (not just ERROR-level gate failures), overridable with --force; W1 coverage stays advisory so a legitimate partial map still renders.
• Swarm cost copy. The intake-router execution-mode menu rows (B/C/I) said a bare "roughly 5x" while run-core.md notes the 2026-06 re-test measured ~8.5×+ on long fiction — understating cost at the decision point. The menu rows now carry the measured figure.

Validators — manuscript-viz E5 + check-mirror hardening

Follow-ups from an independent post-merge review of the Horizon-Tier-1 validator train. manuscript-viz gains E5 (no duplicate entry): a scenes[].scene_id or findings[].id repeated in the manifest is now an ERROR — a duplicate double-draws a pacing bar / double-counts a chapter's severity bar (a chart element showing a value the sources did not contain), which the per-id E2/E4 checks pass on. check-mirror now flags a CM_ROOT_ONLY utility (e.g. sync_setec.py) that has strayed to the plugin side or diverged across both copies, instead of skipping it by name in every mode (it was only ever meant to excuse a root-only file). viz_manifest.py render's usage + refusal now state that the Timeline and Ledger are required for a gated render (with --force as the manifest-only escape) and name any missing source. Docs trued up: the manuscript-visualizations and mirror-parity specs + ROADMAP Horizon item 1 flipped from "Proposed/Spec" to built (manuscript-viz Increment 1 = charts 1–3; the network/heatmap/timeline/beat-map charts remain future); the reader-instrument module docstring now lists its two B3 advisories and the advisory summary reads B3/B4/B5/W1. No validator-count change; no schema change.

Routing — fork/overlay split + project addressability (Increments 1–4)

The intake router now distinguishes forks (modifiers that select a workflow — time → Submission Triage, nonfiction → the argument / narrative / memoir engines, feedback, team) from overlays (modifiers that only modify a selected run — ai, editor, facilitator, risk, hybrid, swarm). The §6 route map splits into a base-route table (Table A) plus an orthogonal overlay table (Table B), removing nine base × overlay rows and giving each overlay a single authoritative status; the router output contract becomes base_route + typed forks{} + composable overlays[] (the bespoke nonfiction_route folds into forks.engine). Projects are now addressable: a workspace-relative .apodictic/registry.json (apodictic.project_registry.v1) makes each book selectable by name, with /start <project> binding a session to its state and a new /projects command listing/rebuilding/tidying the registry — each project's Diagnostic_State.meta.json sidecar stays canonical, the registry is a rebuildable cache. /new-project registers on creation; pre-writing projects drop a minimal sidecar (next_action: pre_writing) so they register before any diagnostic run. State-driven dispatch (Increment 3): for a bound project, /start derives a lifecycle node from the sidecar by a single total precedence (cold → blocked_gate → execution → pre_writing → submission → revising → diagnosed → diagnosing) and dispatches via next_action — collapsing the intake questions to a two-option Resume/Start-fresh prompt (cold start runs the full questionnaire unchanged), behind a scoped contract-hash precondition. The §6 route map gains a lifecycle transition table, with the Artifact×Goal table reframed as the cold-start entry map. Revision-loop spine (Increment 4b): at the revising node the coach answers "what now?" with a leverage ladder (revision-coach/SKILL.md §Loop Dispatch) — reading the finding lifecycle (locked/delivered/revised) and the revision report's resolved markers to propose the next highest-leverage action, with a stalled-revision off-ramp. Increment 4a then makes the runner-governed gate engine the fold-consistent writer of revised: a gated revision_round phase advances only the resolved subset of findings (the report's <!-- resolved: F-… --> ids) to revised, with the revision round's pre-existing direct sidecar write scoped to non-governed projects so gate --check-state's pointer == fold invariant holds. Design: docs/router-fork-overlay-split.md, docs/project-addressability.md, docs/revision-round-gate.md.

Onboarding — project dashboard (snapshot viewer)

A new plugins/apodictic/project-dashboard.html — a self-contained, sandbox-safe snapshot view of your projects: select/filter by lifecycle node, see where each stands ("what now?"), and get the /start <project> launch command per project. It renders an apodictic.project_registry.v1 payload (with the node + next action pre-computed by the tool and embedded), so it is render-only — a viewer and launch-pad, not a live monitor or controller. Complements /projects (text) with an at-a-glance visual, in the style of the existing route-explorer / overview dashboards.

Commands — Capability Index

New /apodictic command: a non-interactive capability index for discoverability. APODICTIC had 15 slash commands and 5 skills but no flat "what can I do / where do I start / what's my current state" reference — /start is an interactive router, not an index, and the README's command list had drifted (missing 4 of 15). /apodictic prints a curated, scannable index: the Firewall in plain language, every command grouped by workflow stage (Start → Pre-writing → Diagnose → Revise → Risk/Submit → Projects) with a one-line "when to use," how to find your current project state (the registry + lifecycle rail), and pointers to the existing HTML maps. It never routes (that's /start) and never writes state. Named /apodictic to avoid colliding with Claude Code's built-in /help. Drift-resistant by design: before printing it globs commands/*.md and appends any command not yet in the index, so additions self-heal (only grouping needs a human). Ships as a command file only (no registry entry, per the existing /projects//legal-risk precedent); host-neutral wording so the codex/antigravity generators pass.

Tooling / DX

Single-sourced the self-testable validator count. It was hard-coded as the literal "40" in five places (validate.sh ×4 + a ci.yml comment); every validator-adding PR had to hand-edit all five, which repeatedly produced merge conflicts when concurrent PRs each bumped the number independently (and a stale denominator would fail --self-test-all). Now AGG_VALIDATORS is hoisted to a single top-level constant with a derived AGG_COUNT=$(set -- $AGG_VALIDATORS; echo $#), and every displayed count (--self-test-all header + PASS/FAIL (N/N) tally + the usage banner) reads from it; the ci.yml comment is de-numbered. A PR adding a validator now edits only the AGG_VALIDATORS line — the count can no longer go stale or collide. No behavior change; mirrored byte-identical across both validate.sh copies (check-mirror green).

QoL — inventory-parity sync-marker check

New meta-check scripts/check-inventory-parity.mjs (stdlib Node only) guards that the display surfaces (overview-dashboard.html, AUDIT_SELECTION_MATRIX.md) stay synced with the canonical audit/research inventory — the rot PR #81 hand-fixed. It computes a <count>:<short-hash> signature for the signal-emitting audit registry (between the registry:signal-emitting-audits markers in audit-routing-table.md) and the research modes (commands/research.md), and flags any opted-in surface whose embedded <!-- inventory-synced: audits=… research=… --> marker drifts from current (exit 1). This is the #79 status-drift pattern applied to inventory: a sync-marker + changed-since signal — robust, near-zero false positives, and no brittle name-by-name matching. Honest limitation: like #79's status flip, it verifies the signal is consistent, not that the surface content is actually correct — a maintainer who bumps the marker without re-syncing defeats it; that's a transparent trade vs. brittle name-matching, and strictly better than nothing. Vacuity-guarded (zero markers → error), fence-aware (markers inside ``` / ~~~ are ignored), malformed-marker-loud, and --self-test covers the 7 hermetic cases plus an integrity check proving the negatives fail only because the compare is real. Both surfaces seeded with the current signature; wired into CI. No validator-count bump.

Docs — Marketing-parity refresh

Brought the plugin's self-description in line with what it ships at v2.3.1. Registered the 5 previously-unlisted commands (/apodictic, /projects, /legal-risk, /triage-feedback, /reader-questions) in release-registry.json, so both READMEs' registry-generated command lists now fill in all 16 commands grouped correctly (/apodictic joins /start under "Start here:"). Added an apodictic-index wrapper-name special-case to the Codex and Antigravity generators so /apodictic no longer derives an awkward apodictic-apodictic. Refreshed the hand-maintained framing and capability prose in both READMEs (and the Codex twin): added persuasive/argument-shaped nonfiction (policy briefs, op-eds, testimony) to the tagline and Intended Audience, and surfaced the shipped capabilities — Legal Risk Register (flags, never adjudicates), Feedback Triage, Beta-Reader Instrument, manuscript-structure visualizations, the Nonfiction Argument Engine, project addressability/resume, Retcon Planning / State Cards, and the Diagnostic-Vocabulary & Editor-Scaffolding operator modes. Back-ported the revision-coach-aware post-diagnosis wording (removing the contradictory "work with Claude directly" line), qualified the privacy claim honestly (no telemetry of its own; diagnostic state persists on local disk by design; /research makes explicit web searches), and fixed the false CONTRIBUTING reason for the no-external-PR policy (solo-maintained + email-based, not "no PR infrastructure"). No diagnostic-behavior, schema, or validator change.

Marketing-parity — visual + matrix surfaces refreshed; two codex twins de-duplicated

Brought the three onboarding surfaces up to v2.3.1 reality and removed two redundant Codex overrides so future edits stop being double-edits. overview-dashboard.html and AUDIT_SELECTION_MATRIX.md gained the audit/research inventory that had drifted out of date — the missing genre audits (Supernatural Horror, Grimdark / Dark Fantasy), Narrative-Decision (StoryScope), POV Voice Profile, the advisory Idiolect Preservation / Punctuation Cadence, the Compression / Stakes System / Decision Pressure / Reception Risk craft chips, the Adversarial Evidence Review + Argument Evidence / Persuasion / Red Team companions, and the two missing research modes (Citation Verifier, Field Reconnaissance → all 6) — all re-derived from core-editor/references/audit-routing-table.md §Signal-Emitting Audit Registry and commands/research.md. The dashboard and matrix also document the shipped Legal Risk Register (/legal-risk, named on the Q3 sensitive-content branch), Feedback Triage (/triage-feedback), Beta-Reader Instrument (/reader-questions), project resume (/start <id>, /projects), and the Nonfiction Argument path. route-explorer.html (and its surviving Codex twin) picked up a version stamp, the projects/resume path, an execution-mode note (swarm = verification insurance for final submission prep; ~5x, measured up to ~8.5x on long fiction), and base/twin parity on the legal-risk route. Because rewriteGeneratedDocs already swaps /cmd → apodictic-cmd across .md/.html, the overview-dashboard.codex.html and AUDIT_SELECTION_MATRIX.codex.md overrides were byte-for-byte reproducible from the base + rewrite (proven empirically), so both were deleted along with their release-registry.json codex.overrides keys. Docs/HTML/matrix only — no diagnostic-behavior, schema, or validator change.

Tooling / DX

New status-drift lint (scripts/check-status-drift.mjs, wired into CI): catches spec docs whose **Status:** line still says "unbuilt" after the deliverable shipped — drift that recurred across the #66/#70/#74 trains and invites duplicate builds (agents implement from specs). Opt-in by design to avoid crying wolf: a spec declares its deliverable in a <!-- built-when: <path> --> (or … contains "<literal>") HTML-comment marker, and the lint flags a doc only when its marker's deliverable now exists and the Status line still reads unbuilt — un-marked docs are never flagged (zero false positives by construction; most "Proposed (unbuilt)" specs are correctly unbuilt). Standalone Node, stdlib-only (no validator-count bump, no dual-script-mirror burden — same class as assemble-changelog); fenced-code-immune; a conservative built-guard skips multi-increment statuses; 10-case hermetic --self-test. Seeded with markers on 11 specs (3 built → green, 8 future → the tripwires for the next build train) plus its own. Companion discipline in AGENTS.md § Review practices ("flip the status when you build"). Surfaced and fixed two live drifts while building — docs/followups-batch-spec.md and docs/runner-governed-execution.md increment 5 (structured gate-event records, shipped v2.1.0) both flipped to Built. (A third candidate — the ROADMAP "Framework Overview Dashboard" Backlog entry — is left for maintainer adjudication, since overview-dashboard.html exists but the entry may intend a redesign.) Detection only — the status flip stays a deliberate edit.

Onboarding — router diagram in the README

The README now opens with a GitHub-rendered Mermaid flowchart of the /start router: the three intake questions (what you have → what you want → anything that changes how we work) routing every starting point to its workflow, with the Firewall shown as the one thing the tool never does — rewrite your prose. It renders inline on github.com and links out to the interactive route explorer, whose full-draft goal list now includes the Revision Coach path, for parity with intake-router-runtime.md.

Audits — Argument-Decision (ArgScope) registered in the canonical inventory

Registered the argument_decision_audit SETEC consumer surface (adopted in #87) in the audit registry so it's discoverable like its narrative_decision_audit sibling — previously it lived only in the skill/reference/contract layer and was absent from the listing surfaces. Adds a Craft entry to release-registry.json (available audits 34 → 35; craft 16 → 17), which regenerates the /audit list, the README audit counts, and the plugin description; adds it to the signal-emitting audit registry (audit-routing-table.md); and re-syncs the inventory-parity surfaces (AUDIT_SELECTION_MATRIX.md, overview-dashboard.html) to the new signature audits=43:1202e80a.

Release — self-contained, decoupled from the APODICTIC-Gemini sibling

apodictic's release flow no longer reaches into an APODICTIC-Gemini sibling checkout. The legacy push reach-ins are removed: release.sh no longer rsyncs plugins/apodictic/ into the sibling's public mirror or runs release-verify.mjs --check-sync (steps 7–8 dropped; banners renumbered to 7 steps), release-verify.mjs drops the App.tsx "Based on APODICTIC plugin vX.Y.Z" version-parity arm, the --check-sync mirror-parity arm, and the checkRsyncParity helper, and release-registry.json drops the three Gemini-pointing path keys (appTsx, landingPageTsx, geminiPublicPlugin). The canonical, drift-gated pull chain is unchanged and remains the sole coupling: APODICTIC-Gemini's sync-plugin.mjs vendors apodictic's released tag + registry, generate-ui.mjs regenerates App.tsx/LandingPage from the vendored registry, and Gemini CI's weekly + dispatch drift gate (generate:ui:check) owns mirror/version parity. apodictic's release is now self-contained and builds entirely from its own plugins/ tree with no sibling dependency.

Docs / Samples

Refreshed the Dungeon Crawler Carl sample editorial letter (sample-editorial-letter.html) to the current framework version (v1.0.1 → v2.3.1, June 2026 run). Content revisions only — same template and styling: re-proportions the structural read (Floor 1 measured at ~52% of word count, not estimated ~60%; the genuine idle stretch relocated to the top of Floor 2), adds the Carl/Donut two-hander and the individuation-gate framing to "What the Book Does Best," reframes the callout around the satire going load-bearing twice (Death Watch + fighting-pit mercy), and reworks the "What Needs Work" items (Floor-2 re-tutorialization, the under-felt mass-death beat, Donut-as-prop at the climaxes, competence-outruns-change) with a matching revision checklist. No plugin behavior change.

SETEC integration — adopt the ArgScope argument_decision_audit surface

APODICTIC now consumes SETEC Voiceprint's argument_decision_audit task surface
(ArgScope) — the argument-domain sibling of the narrative-decision (StoryScope)
audit. A new thin shim ai_prose_argument_decision_audit.py routes the surface
through SETEC's normalized dispatcher (R2) via run_surface_cli, like the other
ai_prose_* shims; the dispatcher enforces the per-surface version floor at
runtime (R3 version_floor on an out-of-floor SETEC), so the shim hardcodes no
version. The surface scores a public-debate / op-ed essay's argumentative
STRUCTURE — the B1 paragraph-role transition arc (support→proposal,
support→support, thesis-opening) + the B2 discourse-mode share — against Kim et
al. 2026's ("Argument Collapse", arXiv:2606.01736) human/LLM group means. It
measures argumentative diversity, not quality, soundness, or provenance, ships
uncalibrated, and is register-bound to public-debate forums (research / legal /
policy = the consumer's distant tier, structural-signals-only).

• Bumped the vendored SETEC contract pin v1.114.0 → v1.116.0 (setec-plugin.lock,
  release pin to the tag carrying the surface). Re-synced the consumer-projected
  manifest (tests/setec-contract/setec-capabilities.json, now 13 apodictic
  surfaces incl. argument_decision_audit at floor 1.116.0) and the R5 contract
  fixtures via scripts/sync_setec.py. Per the script's documented design the
  manifest is consumer-projected while the golden fixtures are copied whole, so
  the bump also vendors the 1.115.0 voicewright-bundle goldens (binoculars /
  general_imposters / mimicry_cosplay / voice_fingerprint) as parser-test data.
• New audit-level contract references/craft/argument-decision-audit.md (v0.1):
  the envelope shape (4 contributions, B1/B2 bundles, heuristic reused_signals,
  the pre_flag), the 3-tier register map, the claim-license fields to surface,
  the aggregate posture (pin per-signal contributions, NOT the aggregate score),
  judge provenance, and the anti-verdict / framing note. It may PRE-FLAG whether a
  dialectical-clarity (soundness) run is informative; it never adjudicates
  soundness, warrant, or fairness.
• Registered in the specialized-audits SKILL.md (triggers, the surfaces table
  row, and the references list).
• Offline contract test (tests/setec-contract/test_setec_contract.py) now
  expects 10 shim surfaces and pins argument_decision_audit's floor at 1.116.0;
  the drift gate re-derives the shim set, so the new shim joins automatically.
• Pins only the parts SETEC has committed (envelope shape, per-signal
  contributions, claim_license); the aggregate math, B3/B4 reused_signals
  (heuristic, no numeric anchor by design), and the deferred dynamic signals
  remain provisional under the surface's handoff: experimental posture.

SETEC integration — R1 capabilities query + vendor/pin/drift-gate

APODICTIC now data-drives each SETEC surface's version floor from SETEC's
capabilities manifest (capabilities.py emit --json) instead of hardcoding it.
New setec_capabilities.resolve_floor() discovers SETEC at a single bootstrap
floor, queries the manifest, and asserts the discovered setec_version against
each surface's min_setec_version. The retired MIN_SETEC_VERSION = (1, 86, 0)
per-surface authority and the narrative-decision shim's (1, 107, 0) constant
are deleted; all nine ai_prose_* shims resolve their floor through the manifest.

Added a vendor/pin/drift-gate apparatus (ported from APODICTIC-Gemini's pull
pattern, in Python): a pinned copy of SETEC's consumer-projected manifest +
R5 contract fixtures under tests/setec-contract/, a setec-plugin.lock pin,
scripts/sync_setec.py (--check-able re-derivation), and a drift gate
(tools/check_setec_contract.py) whose self-consistency guard fails if any
shim surface is missing from the vendored manifest or lacks a floor. Wired into
CI plus a weekly sync-setec workflow. (Provisional pin against the unreleased
SETEC R1+R5 branch; finalization re-pins to the release tag.)

SETEC integration — adopt the R2 normalized dispatcher + R3 structured errors

setec_runner.run_supplement now routes EVERY SETEC surface through SETEC's
normalized dispatcher (setec_run.py <surface> [args] --json, R2) and parses
the schema_version 1.0 envelope from stdout — including
pov_voice_profile, whose private file artifact the dispatcher projects to
stdout. One delivery path. The signature changed from a script filename to a
surface NAME: run_supplement("variance_audit", args), not
run_supplement("variance_audit.py", args).

• Deleted from run_supplement: the json_out parameter, the
  _caller_json_out_path helper, the --json-out injection + --json-out=
  /split-form recovery, the ephemeral ai-prose-baselines-private/ tempdir +
  rmtree, and the min_version escape hatch. The dispatcher owns delivery
  and floor/dependency enforcement now.
• R3 structured errors. On available: false, run_supplement branches on
  the envelope's reason_category (not stderr scraping): version_floor /
  missing_dependency → blocking (surface the upgrade/install message);
  text_too_short → reliability (preserving the §6.4 reliability-vs-blocking
  semantics); policy_refused / bad_input / internal_error (and any
  unknown category) → blocking with the reason text. New SupplementResult
  fields reason / reason_category carry the structured error.
• The 9 ai_prose_*.py shims are thinned to a single
  run_surface_cli(SURFACE, argv) call (via the shared helper in
  setec_runner): they drop the per-shim resolve_floor runtime pre-check and
  the pov json_out=True special-case, route through the dispatcher, emit the
  envelope to stdout, and exit with the dispatcher's exit-code contract.
• Floor reconciliation. The dispatcher is the single RUNTIME authority for
  per-surface floor/dependency failures (it returns R3 version_floor /
  missing_dependency). setec_capabilities.resolve_floor + the vendored
  manifest are retained ONLY for the offline drift gate and capability
  introspection (Increment 2's contract role), not a redundant runtime
  pre-check that could drift from the dispatcher.
• Bootstrap/dispatcher floor. run_supplement fails cleanly with an
  upgrade message if the discovered SETEC predates the R2 dispatcher
  (setec_run.py absent) rather than crashing. A # FINALIZATION: marker in
  setec_runner.py flags raising BOOTSTRAP_SETEC_VERSION to the real R2
  release (target ~1.114) once SETEC cuts it.
• Docs updated: run-full.md Pass 3 + Pass 7 wiring (surface-name calls;
  Pass 7 multi-POV drops json_out; POV read keys corrected to the dispatcher's
  projected names cross_pov_distances_weighted / pov_vs_corpus_mean /
  voice_collapse_verdict); narrative-decision-audit.md; the contract test
  gains the dispatcher path (T2) + R3 tiering (T2b).

Docs / Calibration

Ran the swarm-vs-single execution-mode eval as an N=1 pilot on one long-fiction fixture (Dungeon Crawler Carl Book 1, 130K words) and recorded it under docs/swarm-vs-single-eval-pilot/ (pre-registration, blind ground-truth key, label-blind scoring, comparison report). Directional, not a verdict. On that fixture single-agent tied-or-edged swarm on real-issue recall (0.50 vs 0.44, noise-level) at ~8.5× less token cost; swarm bought tighter severity calibration (3/4 vs 3/6 band-clean on the caught-and-banded set) and more full-credit catches, not more recall, and both correctly returned zero Must-Fix on the near-sound control. Per the pre-registered rule this is a "verification-insurance / single-default (cost-adjusted)" outcome — matching the existing §2b default. Applied the pilot's recommended cheap deliverable: run-core.md §Execution Mode (which carries the last re-validated: 2026-06 provenance note), the intake-router swarm menu rows, core-editor/SKILL.md, and both READMEs now keep swarm but reframe its everyday rationale from "deeper analysis / ~2× findings" to verification isolation for final submission prep. The historical changelog entry for the original swarm A/B claim is left intact as a record. No validator or routing-logic change (the user-facing §2b execution-mode menu copy is reworded).

Validators

37 → 38 self-testable validators. Added check-mirror, a QoL gate that mechanizes the dual-script-mirror invariant (AGENTS.md § Platform parity): validate.sh, preflight.sh, and every *.py exist in two committed copies — root scripts/ (what CI runs) and plugins/apodictic/scripts/ (canonical) — that must be byte-identical by hand, or a validator change passes against one copy while CI runs the stale other blind. check-mirror asserts the shared mirrored set matches (root-only build/release scripts, plus root-only *.py infra utilities listed in CM_ROOT_ONLY — e.g. sync_setec.py — and the plugin-only test_fixtures/ correctly excluded; schemas are single-sourced) and is wired into --check-all, so drift is now CI-blocking instead of a silent footgun. Detection only — it never auto-syncs; the by-hand cp stays deliberate. Pure shell (no python3 dependency).

Workflows — Manuscript-Structure Visualizations

New Manuscript-Structure Visualizations capability (Horizon Tier 1): a presentation layer that adds no analysis — a deterministic render of data the passes already produced. An apodictic.viz_manifest.v1 block copies, verbatim, the Timeline Event-Ledger rows (scenes) and apodictic.finding.v1 blocks (findings); the new render-only SVG layer (scripts/viz_manifest.py render) draws charts 1–3 — pacing/word-count curve, POV time-share, finding-severity-by-chapter (chapters sorted numerically, Ch 2 before Ch 10) — into a single self-contained offline HTML (no network, no deps, no telemetry). The manuscript-viz validator owns manifest↔source provenance: E1 schema + a no-visual-style field allowlist (style is the renderer's, not the run's; a present-but-unparseable block is an E1 failure, not a silent no-op, and --require-block fails an absent one — the --check-all gate uses it so it can't pass vacuously), E2 provenance closure (scene_id → Timeline row; finding id → ledger; finding chapter == the conservative Chapter N/Ch N evidence_refs parse, else the literal unplaced — no guessed placement), E3 every body Must-Fix appears, E4 byte-equal copy fidelity (the manifest copied, it did not compute), W1 coverage, and W2 scene order (scenes[] must follow the Timeline's document order, since the pacing curve's x-axis is scene order). render runs the gate first and refuses on an ERROR (--force to override). Firewall-safe by construction: render-only, provenance-closed, severity encoding hardcoded so a Must-Fix can't be recolored and a LOW-confidence Must-Fix is never shrunk. Reuses timeline_checks._parse_event_ledger (the Timeline column parser) and the shared block/schema engine; the subset engine can't validate arrays-of-objects or reject extra fields, so the nested-object validation + no-style allowlist live in Python. Backed by a worked-example manifest paired with the existing example-timeline.md + example-findings-ledger.md, gated by validate.sh --check-all. Mirrored into the root scripts/ harness CI runs. Spec: docs/manuscript-visualizations.md.

Workflows — Beta-Reader Instrument Generation

New Beta-Reader Instrument mode in the revision-coach skill (/reader-questions), the upstream complement to Feedback Triage: it turns the diagnosis's own open uncertainties — LOW/UNCERTAIN findings, the Findings Ledger's ### Unresolved Questions, and risk_if_fixed tradeoffs — into a targeted, non-leading reader questionnaire, so the feedback that comes back is worth triaging. Closes the reader loop end to end: diagnose → ask the right questions → collect → triage → revise. Each question is an apodictic.reader_question.v1 block; the new reader-instrument validator (scripts/reader_instrument.py) owns the question-contract — B1 schema, B2 unique ids, B3 provenance integrity (a finding-sourced question's targets resolves to a real ledger finding, with an advisory flag when a low-confidence-finding probe points at a non-LOW finding; an unresolved-question carries a source_note matching a real Unresolved-Questions bullet, and no targets), B4 leading/invented-content firewall (a sound finite blocklist of leading constructions + a coarse content-neutrality heuristic; advisory, ERROR under --strict, overridable), B5 anti-relitigation (never poll readers on a locked verdict — severity ∈ {Must-Fix, Should-Fix} and confidence ∈ {HIGH, MEDIUM} — overridable for "how to fix"), and W1 coverage. The severity-honesty boundary is the point: it tests uncertainty, never certainty. Backed by a new worked example + paired uncertainty-fixture ledger, gated by validate.sh --check-all. Mirrored into the root scripts/ harness CI runs. Spec: docs/beta-reader-instrument.md.

Validators

35 → 37 self-testable validators. Added registry-check (project-registry integrity over a workspace-relative .apodictic/registry.json: R1 envelope + per-entry schema, R2 root + sidecar resolution, R3 cache-vs-sidecar drift with the sidecar canonical, R4 duplicate id), backed by the new apodictic.project_registry.v1 + apodictic.project_entry.v1 schemas; and lifecycle-node (derives a bound project's lifecycle node from its sidecar by a single total first-match precedence — the tested primitive state-driven dispatch reads from). Both mirrored into the root scripts/ harness that CI runs.