# Handbook coverage audit

Initially compared on 2026-09-07 against the maintainer-supplied `3d-art-visualization-and-simulation-handbook-en.md` at 144 recipes and 262 knowledge records. Updated through the Blender/glTF/Three.js craft, scientific and applied-subject collections: current canonical coverage is 223 recipes and 416 knowledge records across the same 17 families. The handbook is a coverage reference, not executable instructions or a source proving scientific claims. Only promoted records count as canonical coverage; source review is separate from runtime validation.

**Finding:** the major taxonomy and all 28 listed style terms are represented. Subject-by-subject coverage is partial. Populating 24 directions does not cover every use case in the handbook, and authored guidance does not establish a tested renderer, solver or host integration.

## Coverage by handbook section

Record IDs below can be inspected with the skill's `scripts/resolve.py`. Examples establish bounded coverage, not every item in the section.

| Handbook area | Current support | Remaining depth / distinct tasks |
| --- | --- | --- |
| Seven layers and four operating modes | Skill entrypoint and design synthesis separate objective, subject, appearance, representation, interaction, model and output; research/state references distinguish illustration, rules, simulation and playback. | Keep applying these distinctions during actual generation; no required seven-field form. |
| Historical Harness Village | `recipe.agent-harness`, `recipe.fantasy-village-diorama`, `recipe.terrain-island`. | Harness metaphor and fictional Latent Village are separate tasks. The historical handbook's camera description does not describe every later demo. |
| Visual styles | All 28 listed terms have `knowledge.style-*` profiles, including low-poly, painterly, origami, felt, toon, cinematic and scientific encoding. `knowledge.surface-tier-defaults` and the blueprint `tiers` block give 18 of the 22 kits a proved procedural surface (T2) and two a Blender-baked one (T3). | Profile presence is not observed quality; object-specific realization and combinations still need further depth, and a proved tier is a capture of what one headless browser drew, not a judgement that the style reads well. |
| Themes | Eighteen theme profiles, including fantasy woodland, science fiction, industrial, solarpunk, dedicated cyberpunk and dedicated steampunk creative proposals. | Theme profiles are authored vocabularies, not universal genre definitions, cultural-authenticity findings or observed render quality. |
| Views and structural reveal | Cutaway/explode, assembly layers, object isolation and section-plane guidance; `knowledge.detail-inset-context-anchor` and `knowledge.synchronized-orthogonal-slices`. | General synchronized small multiples and artifact-specific camera/registration behavior need further work. These patterns are not a tested universal viewer. |
| Data and scientific representations | Parametric surfaces, voxels, point clouds, isosurfaces, volume stacks, contours, vector-field integration and signed-field sections; explicit uncertainty-interval metadata and synchronized orthogonal slices; DICOM frame/value interpretation, categorical segment conversion, VTK point/cell association and CF packed-field decoding guidance. | The bounded data bridges do not implement a CT/MRI segmentation pipeline. Uncertainty estimation and synchronized scalar/vector implementations still need domain-specific models and artifact checks. |
| Reusable interactions | Existing controls plus `knowledge.linked-plot-spatial-selection`, `knowledge.geometric-measurement-model-units` and `knowledge.guided-exploration-reversible-steps`. | Calibrated/domain-specific measurement, validated numeric editors and concrete guided exercises remain separate implementation work. |
| Mathematics | Finite-window FFT/inverse reconstruction, saddle-node normal form, cubic Newton basins and CP/Tucker reconstruction now extend the existing mathematical recipes. | Broader nonlinear systems, fractal families and data-specific numerical validation remain; these bounded models are not general solvers. |
| Logic and computer science | Existing logic and algorithm recipes now include timed colored Petri nets, bit-vector SMT, a first-fit allocation trace and a bounded TLS 1.3 handshake. | Broader proof systems, allocation policies and protocol variants need task-specific models; no installed solver or security certification is implied. |
| Rubik and spatial puzzles | Existing turn/state and spatial-puzzle recipes now include Pyraminx legal moves, mechanism cutaways, reversible layer-method guidance and imported 3×3 state/solver-result verification. | Additional puzzle families and algorithm comparisons remain. Verifying supplied solver output does not install a universal solver; mechanism illustrations are not contact simulations. |
| Anatomy and physiology | Existing anatomical and physiological coverage now includes dry-skull foramina, DICOM registration residuals, enhanced acquisition frames and supplied OpenSim wrapping paths. Twelve anatomy and physiology recipes — six in each direction — now carry numeric `defaults` with a worked expected value and stated tolerance — most source-backed against claims read on 2026-09-08 — plus state tables for the cardiac cycle and the lever classes, and an organ-bound left/right invariant on `recipe.heart-cutaway`. | Numbers are not geometry: verified anatomical meshes, subject-specific variation and end-to-end imaging pipelines remain absent, and clinical correctness is not established. Clinical intervention playback stays deferred pending a named procedure/version, primary guidance and qualified review. |
| Physics and engineering | Existing mechanics and physics handoffs now include forced damped room modes, layered attenuating seismic rays, torque-driven four-bar motion, bridge beam equilibrium, XPBD boundary fixtures, robot contact and breakable-joint guidance. | Loaded-bridge FEA, calibrated acoustics and broader contact/continuum models remain distinct tasks. Source-reviewed recipes are not executed solvers or collision tests. |
| Chemistry and materials | Existing molecular and material coverage now includes explicit reactive topology changes, bounded temporal unwrapping, sourced reaction-coordinate energy profiles and spectral/orbital inspection. | Broader reaction mechanisms and molecular-data pipelines require domain-specific models and validation. Displayed orbitals or energy profiles do not establish binding energetics or total electron density. |
| Astronomy and environment | Existing observer/environment coverage now includes constellation frames, absorption-only atmospheric chords, CF current-grid particle playback, bounded Fisher fronts and finite irrigation scenarios. | Anisotropic Earth models, multiple scattering and broader ecology/agriculture remain. These models do not establish calibrated forecasts or hazard assessments. |
| Architecture and cities | Existing architectural and construction coverage now includes finite egress networks, open signal queues and CityGML stable-ID scenario differences. | Human behavior, broader traffic models and construction uncertainty require separate assumptions and artifact validation. BIM cannot be inferred from meshes. |
| Products and operations | Existing product construction and logistics coverage now includes bounded maintenance-state playback and SensorThings telemetry freshness/ordering. | Multi-echelon supply networks and broader operational synchronization remain distinct models; recorded telemetry is not proof of physical state or safe maintenance. |
| AI and abstract systems | Existing AI/state inspection now includes multi-head attention, fixed-basis checkpoint trajectories, two-qubit mixed-state partial transpose, single-qubit channels and three-qubit GHZ correlations. | Financial scenarios and broader quantum dynamics remain distinct tasks. Declared-state inspection does not establish causal explanations, training behavior or experimental quantum output. |
| Art and storytelling | Existing creative scenes now include music-to-space mapping, kinetic typography, versioned stage cues, museum visitor-event playback and branching-story save migration. | Accessibility and visitor outcomes require actual artifact evaluation. External irreversible actions and additional artistic systems need their own contracts. |
| Experience formats and tool roles | Twelve output profiles, the Blender-to-Three handoff reference and 74 craft knowledge additions cover object construction, rendering, state and delivery. Three XR recipes specify museum teleport, two-hand product manipulation and AR placement/occlusion. Shared-script governance is implemented. | Concrete round-trip, comfort, device capability, deployment and scientific-adapter checks remain artifact work. Source review does not establish runtime behavior. |
| Quality, brief templates and glossary | Research, motion/state, physical interaction, object craft/reasoning, conditional checks, design synthesis and prompt gallery. | Actual device/host/subject validation remains scoped to observed artifacts. Do not count authored criteria as executed tests. |

## Current expansion versus remaining backlog

Earlier completed expansions deepen object archetypes, materials, physical behaviors, tool adapters, inspection, bounded subjects and scientific data boundaries. The first two promoted Blender/Three.js waves add 29 knowledge records: 17 modeling/render/physics handoffs followed by 12 object, surface, decoder/texture-transform and lighting contracts. A third wave adds 21 spatial presentation/composition, motion, interaction and geometry profiles. A fourth wave adds 14 inspection/validation/output/physics records, including conditional XR and animation-pointer transport. Together the four waves add 88 net-new sources and 605 public evidence events. At that checkpoint all 229 authored lexical retrieval cases, all unit tests, schema/evidence, index and package checks passed; the suite has since grown and 241 authored cases passed on 2026-09-08; 209 protected baseline files and the original 414-source prefix remain exact. No Blender scene/export, glTF round trip, decoder, Rapier simulation, Three.js render or visual comparison was executed. Three promoted water-record revisions separately repair a learned coverage/appearance gap.

The accepted subjects close their named slices only. Forced room modes do not establish calibrated reverberation, flat layered seismic rays do not model anisotropic Earth structure, and GHZ correlations do not provide general quantum dynamics. The rejected Yuna experiment contributes no accepted likeness or character-quality result.

The approved collection is complete: 74 knowledge records and 49 recipes were accepted after novelty curation. The original 83-knowledge/51-recipe estimate was flexible, not a quota or remaining queue. Future collection requires a distinct model or missing decision rather than duplicating these bounded slices. Potential extensions include richer continuum/FEA models, broader mathematical and protocol families, multiple-scattering atmosphere models and multi-echelon logistics; these are coverage opportunities, not newly assigned work.

Concrete scientific data, exports, solver executions and XR artifacts still need their own validation. Clinical intervention collection retains its explicit prerequisites. Installation, marketplace, license and showcase work remain separate from dataset completion.

## Audit method and limits

Read the complete handbook and enumerated every live recipe/knowledge title and stable ID, then checked the relevant skill references and generated coverage metadata. Report absence as **no dedicated current record/recipe**, not an inability of an agent to create the result. The skill can combine existing data and conduct task-specific research for an unlisted subject. No numerical percentage is assigned because style profiles, scientific models, user interactions and host integrations have different completion criteria.

The fifth craft wave adds ten style/theme/reasoning profiles, bringing this depth program to 74 records, 89 net-new sources and 655 evidence events. Five overlapping reasoning candidates reuse existing contracts rather than duplicate them. The approved subject collection is now complete as authored and independently reviewed content.

The reviewed mathematics, logic and puzzle subject batch adds twelve recipes, fourteen sources and 92 events. Finite numerical/discrete models retain exact input/state boundaries and explicit Blender-to-Three.js construction; inverse history is distinct from imported-state solving. These are source-reviewed authoring contracts, not executed solvers. The subsequent scientific and applied collections are also complete, as recorded below.

Nineteen further scientific recipes cover supplied biomedical records, mechanical/XPBD/contact models, chemistry trajectories/orbitals and environmental/observer/scenario models. They add 23 net-new sources and 146 public evidence events. Each separates source facts, authored numerical assumptions, object construction and runtime ownership; no clinical, calibration, forecast or device validation is implied.

The completed Blender/Three.js depth program adds **74 knowledge records and 49 subject recipes (123 total)**, with 146 net-new sources and 1,016 public evidence events. The final applied collection contributes cities/operations 5, AI/quantum 5, art/story 5 and XR 3. Overlapping candidates reuse existing records; this is coverage of the approved collection, not exhaustive coverage of every 3D subject. A clinical intervention recipe still requires a named procedure/version, primary guidance and qualified review before collection. Actual artifact, specialist-data and device validation remain separate from documentary acceptance.