Research: faction form language and role readability for agent-built hulls, for ruling hull-art-communicates-form

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

docs/research/2026-10-18-faction-form-language-prior-art.md

@@ -0,0 +1,130 @@
+<!-- Serves ruling aetheria-release:ruling:hull-art-communicates-form (Operator, 2026-10-18). Eyes pass eyes-ag6-faction-form, session self-2026-10-18-ag6. Facts with pointers; no recommendation. Web access date for every URL: 2026-10-18. Search tool returned summaries; items marked (search-summary) were not read at the primary page. Mind read with view/query only; nothing admitted. -->
+
+# Faction form language for agent-built hulls: prior art and internal evidence (Eyes facts file, 2026-10-18)
+
+Ruling served: `aetheria-release:ruling:hull-art-communicates-form`. Operator quote (abridged): agent-built hulls should communicate the ship's intended form; "A Zhestokost ammo tender shouldn't look the same as a Pirate fighter, and with this generate-from-schematic approach there's no way to differentiate factions visually"; nice low-poly models are acceptable "as long as we do have a couple of fully-produced hero assets to compare them with"; "I just don't want to gate hull *variety* on my own manual throughput". The ruling leaves mechanism and per-faction form language open, keeps raw Tripo output out, and supersedes `hull-programmer-art-reads-as-stand-in`.
+
+Extends, does not repeat, `F:\Projects\Aetheria\docs\research\2026-10-18-agent-built-hull-art-prior-art.md` (LLM 3D workflows, ship generators, low-poly style, mesh-probe runs).
+
+## A. How games specify faction visual language
+
+Evidence quality is uneven: the best primary sources are one CCP dev blog, one Stardock dev diary and two Valve papers; most others are search summaries. Few studios publish a written shape-vocabulary document; the public record is mostly dev diaries.
+
+### A1. EVE Online (CCP)
+
+- CCP dev blog, "Iconic Ships Reimagined - The Minmatar Rupture" (2017), https://www.eveonline.com/news/view/iconic-ships-reimagined-the-minmatar-rupture (fetched, answered by a summarizing model; paraphrase). Minmatar is specified by a shape vocabulary plus a construction idea: aggressive wedge front, high rear with a transom, chunky segmented plates with exposed structural beams and dark gaps, bulk made largely of empty space (reference: spaced armour on WWII vehicles), warm orange window glow, oversized engines. Caldari is the contrast: smooth, uniform, enclosed shells, few exposed electronics, sparse windows. Gallente is mentioned only as voluminous hull and structure. The post says the redesign risked overlapping Caldari because both used large flat shapes, which pushed Minmatar toward exposure and negative space. It says the redesign looked too monolithic from front and top, which led to angled gun-port covers. Tech 1 hulls stay near the original silhouette; Tech 2 may deviate more. It does not address legibility at small size.
+- Amarr and Gallente design language: not found in this pass (search returned only this post).
+
+### A2. Sins of a Solar Empire II (Stardock)
+
+- Vasari dev diary, https://www.stardock.com/games/article/533639/sins-ii-behind-the-scenes---designing-the-vasari-empire-pt-1 (fetched). No formal style guide described; consistency comes from an art-director brief ("rounded, blade-like forms"), convex armoured shapes tapering to points, ridged organic surfaces, segmented plating, curves set against sharp edges and glowing conduits, thrusters echoing spider weaponry, wasp/beetle warning markings. Thumbnail silhouettes are explored first, tailored per role (nimble attacker, heavy bruiser, massive support). Palettes vary by ship and are partly chosen to place team colours; no fleet-wide colour rule is stated.
+- Pirate dev diary, https://forums.sinsofasolarempire2.com/532947/sins-ii-behind-the-scenes-avast-ye-buccaneers-pirate-concepts-ahead (direct fetch refused, HTTP 403; content from search-summary): pirates repurpose stolen TEC hulls; spikes, studs and reinforced armour are the signature modifications, with shared styling keeping the fleet cohesive.
+- TEC and Advent design language: not found. Original Sins concept artists: not found.
+
+### A3. Starsector (Fractal Softworks)
+
+- Three "design types" (low tech, midline, high tech) are a gameplay-and-style axis: low tech is high armour with ballistic/missile mixes, midline mixed, high tech speed and shields (search-summary of https://starsector.wiki.gg/wiki/Ships and https://fractalsoftworks.com/tag/ship-design).
+- "Uniquifying the Factions, Part 1" (2022-03-18), https://fractalsoftworks.com/2022/03/18/uniquifying-the-factions-part-1 (direct fetch refused, HTTP 402; search-summary only): factions with the strongest identity are those that restrict themselves to one tech level or ship type for doctrinal reasons; a faction fielding much the same ships as others with small stat tweaks feels muddy. The summary is about gameplay identity; sprite-level shape rules were not found.
+- Modder-side sprite style guidance for Starsector: not found.
+
+### A4. Homeworld (Relic)
+
+- Fan-site claim (unverified, search-summary): Aaron Kambeitz designed the Taiidan ships and Rob Cunningham the Kushan; Homeworld 2's Vaygr and Hiigaran differ more than the original's Kushan and Taiidan. https://www.parkablogs.com/index.php/content/book-review-art-of-homeworld , https://en.wikipedia.org/wiki/Homeworld_2.
+- The Parka review (fetched) names the races and says the book (The Art of Homeworld) holds the design commentary but gives no principles or silhouettes. A fan review is quoted by search as calling the ships "utilitarian, asymmetrical". The design commentary itself was not located online.
+
+### A5. Other titles
+
+- Ashes of the Singularity II dev journal #8 (Stardock): each faction gets its own materials, shaders and animation under shared lighting and camera distances; the UEF leans on blocky silhouettes and low-reflectivity weathered surfaces; rationale is that hundreds of units fill the screen so strong silhouettes and consistent materials help read the crowd; blocks at distance, detail up close. https://www.stardock.com/games/article/539126/ashes-ii-dev-journal-8-the-actors-of-war-on-stage-pt-1 (search-summary). Stardock's "Designing Alien Units": Seraphim and Substrate separated by palette and weapon effects, https://www.stardock.com/games/article/487404/dev-journal-designing-alien-units (search-summary).
+- Battlefleet Gothic: described by fan/review sources as Imperial cathedral ships with arches, spires and an aquila prow, broadside macrocannons, versus Ork hulks and roks; official design material not located. https://goonhammer.com/battlefleet-gothic-part-3-the-best-expansion-to-the-best-game-gw-ever-made , https://pcgamer.com/battlefleet-gothic-armada-unleashes-the-imperium (search-summary, secondhand).
+- Stellaris ship sets: modders select a graphical_culture/shipset and ship_kinds per species; ship sections (bow/mid/stern, "core") are the modular unit in community mods; base-game sets are the recommended modding templates. Steam Workshop comments (fan/anecdote): https://steamcommunity.com/sharedfiles/filedetails/comments/2866658484 (search-summary). No official shape-language material found.
+- Star Citizen: CIG describes visual language as a deliberate part of brand/manufacturer creation (an Inside Star Citizen episode, https://massivelyop.com/?p=293041, search-summary); the rules themselves were not found.
+- Elite Dangerous: no official per-manufacturer design language found; a forum thread (opinion) says older ships have atmospheric shaping and newer ones modern-sci-fi shaping, https://forums.frontier.co.uk/threads/thoughts-on-ship-shapes.639903 (search-summary, forum opinion).
+- FTL: not searched. Gap.
+
+### A6. How such languages are specified, as the sources show them
+
+- Shape vocabulary stated as a short list of forms per faction (EVE Minmatar: wedge front, transom, exposed beams; Vasari: blade-like convex tapers; Ashes UEF: blocky). Contrast against a neighbouring faction is used as a rule (EVE Minmatar vs Caldari).
+- Construction idea or reference object (EVE: WWII spaced armour; Vasari: spider and beetle motifs).
+- Modification vocabulary for scavenged/derived factions (Sins pirates: spikes, studs, reinforced armour on captured hulls).
+- Gameplay-axis style tiers (Starsector design types) carry visual style partly through the role mix.
+- Palette as a separate layer from shape (Sins: per-ship palettes tied to team colour placement; Stardock aliens: palette plus effects).
+
+## B. Role readability by silhouette and proportion
+
+- Valve, "Illustrative Rendering in Team Fortress 2" (Mitchell, Francke, Eng, NPAR 2007) and "Stylization with a Purpose" (GDC 2008): class differentiation is the stated reason for the style; each class is designed to have a distinct silhouette that stays identifiable without lighting, built from body proportions, weapons and clothing; value contrast and simple forms carry readability. https://cdn.fastly.steamstatic.com/apps/valve/2007/NPAR07_IllustrativeRenderingInTeamFortress2_Slides.pdf , https://cdn.fastly.steamstatic.com/apps/valve/2008/GDC2008_StylizationWithAPurpose_TF2.pdf , https://wiki.teamfortress.com/wiki/Illustrative_Rendering_in_Team_Fortress_2 (search-summary; PDFs not opened). These concern characters, not ships; no ship-specific equivalent was found.
+- Sins II Vasari diary (A2): silhouette thumbnails per role before detail.
+- Ashes II (A5): silhouette plus consistent material for crowd reading.
+- Concept-art practice (anecdote, search-summary): a trading-card capital-ship sheet began as black silhouettes and aimed at readability at fleet scale, https://jordan_taylor_arts.artstation.com/projects/V2rn95 ; a concept-art overview says fighters tend to show aggressive silhouettes and weapons, transports large cargo holds and robust structure, https://try.gocheetah.com/?p=12134 (SEO page, low quality); a Frontier forum thread observes a recurring spearhead-with-split-prongs capital-ship shorthand across games, https://forums.frontier.co.uk/threads/capital-ship-design.324746 (forum opinion).
+- Blog guidance (opinion, search-summary): players recognise threats by outline before colour or interior detail; most readability failures are value problems, check in greyscale: https://bugnet.io/blog/readable-game-art-why-clarity-beats-detail , https://rocketbrush.com/blog/shape-language-in-game-character-design-how-to-make-characters-readable-and-consistent.
+- EVE Rupture post (A1): front/top views read as monolithic until angled gun-port covers broke them up.
+- No GDC talk specific to spaceship role silhouettes was located.
+
+## C. Producing many consistent variants from a style
+
+- Kit/part libraries: commercial kitbash packs show the pattern at the asset level (snap/stack pieces; separate bodies, wings and weapons; colour/wear variants for faction), https://superhivemarket.com/products/600-hard-surface-piece-kitbash-set , https://superhivemarket.com/products/130-sci-fi-hard-surface-kitbash-set (vendor pages, search-summary). One vendor notes the packs target offline renders and may need adjustment for gameplay. This is marketing, not studio practice.
+- Section-based assembly: Stellaris community mods use bow/mid/stern(core) section slots (A5, fan mods). Starsector/Starfield-style attach-point builders are covered in the earlier note (B2).
+- Procedural variation within a style: the earlier note records a1studmuffin/SpaceshipGenerator and Kinnear-Kaplan greebling. 3D-GPT's reported limit (diversity bounded by the procedural generator and prompt expressiveness, second-hand) is in that note. No No Man's Sky parts-library source was found (search returned only fan/mod pages).
+- Reference-conditioned generation: Sloyd's custom style takes a reference image as the style guide and a text prompt for the object; reported example: identical prompts rebuilt one dungeon in two moods by swapping the reference image; the article lists no failure modes. https://sloyd.ai/blog/how-to-style-consistent (fetched; vendor). A Virtuall guide recommends separating fixed style rules ("constants") from per-asset variables, with turnarounds and orthographic views for 3D assets, https://virtuall.pro/blog/how-to-create-ai-game-assets-that-match-your-art-style (search-summary, vendor). Ludo.ai says it does not train on your art, so a locked house style rests on reference images (search-summary).
+- LLM or agent loops that keep a style via a style sheet plus a parts library plus render-and-compare against a hero asset: none located. Searches returned no source describing such a loop; the nearest are render-and-critique loops for Blender code (earlier note A1/A3) and BlenderGym's finding that verifier scale helps.
+- Failure modes reported: sameness from a bounded generator (3D-GPT, second-hand); Starsector's "same ships with tweaks feels muddy" (A3, search-summary); EVE's Minmatar/Caldari overlap when both leaned on large flat shapes (A1, fetched); style carried only by colour/markings over shared geometry is what the kit-pack vendors describe (A, vendor). Internal measured evidence on style leakage and drift is in the Internal section (Prompt lab rule 3).
+
+## Internal
+
+### I1. Stated aesthetics of the factions that matter
+
+Source for shapes, finish, palette: `F:\Projects\AetheriaLore\Aetheria\Brainstorming\Faction Flavor and Visual Identity.md` (sections by faction; "Silhouette and construction" and "Materials, color, and markings" paragraphs). Doctrine and play quirks: `F:\Projects\AetheriaLore\Aetheria\Game Design\Faction Play.md`. Lore: `F:\Projects\AetheriaLore\Aetheria\Worldbuilding\Pre-Elysium\Factions\Powers\...`. Per-faction prompt-ready blocks and registers: worktree `F:\Projects\AetheriaLore-wt-ship-language\Aetheria\Brainstorming\Ship Design Language.md` (branch eureka/ship-language-registers, status "brainstorming, not canon"; later rules "agent-promoted, pending operator review").
+
+Cast (rulings `aetheria-release:ruling:demo-cast`, `demo-content-bar`, `demo-start-hull-longinus`): Pirates allied; Zhestokost antagonist and boss faction; Lucent and AU featured neutrals; "none of the four has authored ship hulls today"; the player starts in an existing brand hull (Longinus). Miss Terri's is not in the demo cast.
+
+Per faction, brief:
+- **Zhestokost** (heavy ballistics, militarised Martian state). Silhouette: "Low, broad hulls built from stepped rectangular masses"; thick overlapping armour around recessed machinery; barrels, ammunition feeds, towing points, replaceable plates "remain easy to identify"; weight from deep seams, not surface clutter. Finish: rough cast metal, dark steel, ceramic armour; charcoal, iron gray, dull oxide red, cream serial numbers, block lettering. Identity "should survive in silhouette without relying on red paint or political symbols." Manufacturing: lore says advantage comes from coordinating foundries, stockpiles, supply routes and repair crews (`Zhestokost.md` line 35). Play: warships fly in a column around an ammunition tender and fall back to rearm (`Faction Play.md` Zhestokost quirk; a "Quartermaster (tender)" prompt exists in Bench 2, per `Ship Prompt Bench 3.md` line 36). Register: serious.
+- **Pirate Coalition** (allied). "An identifiable donor hull with asymmetrical additions, cut-down modules, adapted hardpoints, exposed cable routes, and bolted salvage plates. Each major addition has a plausible mounting and supply path." Finish: several manufacturer finishes interrupted by primer, bare welds, heat discolouration, a crew accent colour, crossed-out serials, emoji marks, repeated crew markings. Lore: loose network of crews, "captured industrial components", salvage, yards in hidden docks (`Pirate Coalition.md`). Doctrine: want cargo not death, hit drives (`Faction Play.md` line ~337). Register: Pirate, "wild and maximalist: a stolen donor hull piled with additions"; Bench 3 pirate prompts are built on named stolen donors (Welcome Mat from an AU miner, Jumble Sale from an R&D wedge). Needs a Faction record (`Faction Play.md` line 335; ruling demo-cast "need a catalog record").
+- **Lucent Media** (spectacle). Compact central body in overlapping shells, swept prongs, upflared rear fins, "detached-looking side plates held on visible supports", deep openings exposing a dark chassis; gaming-mouse vocabulary (palm covers, ribs, knurled wheels). Palette by component: golden yellow, white or pale lavender, cobalt blades, red-orange tips, navy interior, teal pods. Concept anchor: `D:/Google Drive/Gamecult/Aetheria/Concepts/Ships/lucent_media/transparent_03.png` (not read; outside the repos). Play: one headliner duels while others hold; outline changes when radiator shutters open.
+- **Aeronautics Unlimited** (vanilla all-rounder). "Cylinders, rounded rectangles, simple arcs, and compact wedges combined into balanced working machines"; tool sockets, lifting points, trusses, removable modules; chalk white, slate blue, bare aluminium, ochre safety panels. Lore: builds shipyards, habitats, extraction, salvage; works with Orbital Forge (`Aeronautics Unlimited.md` lines 1, 29). Play: haulers flee, escorts fire dumbfire swarms. Hulls named "Jason" in loadout notes.
+- **Finch Cybernetics**. "Continuous, tapered forms with the fit and articulation of a joint or sensory organ"; flush apertures, recessed seams; pearl ceramic, satin titanium, smoky membranes, celadon indicators; play: fragile shadowing ships.
+- **Rossum & Douglas ("R&D")**. "The flying wedge": shallow triangular/trapezoidal hull, blunt nose, wide engine face, flat dorsal planes, bevelled edges, small recessed windows; range scales the same outline longer, wider or thicker; light gray, blue-gray, navy, off-white with one stripe. Intended to read as "thoroughly unremarkable production ship".
+- **NiteLife Energy**. Full flowing volumes interrupted by crisp slots, flat collars, abrupt cuts, docking reservoirs; midnight blue, petroleum teal, lime. Distinguished from Finch: Finch is continuous, NiteLife deliberately interrupts curves.
+- **Aya Collective**. Rounded protective frames around replaceable cells, branching fluid paths, accessible couplings; terracotta, cream, deep green, charcoal, copper; supplied in part by Orbital Forge yards.
+- **Megiddo**. Nested protective volumes, shutters, separate service sections; blue-black, warm pale ceramic, aged metal, silver inlays. No Faction record (`docs/content-batch-one.md` section 9, per follow-up `content-batch-one-reconcile`).
+- **Alakrita**. Long creased bodies, swept blades, thin layered fins, aligned engines; ivory lacquer, wine red, black glass, bronze/gold inlay. Prestige supplier under Lucent. Its Longinus and LonginusX hulls exist (follow-up `demo-cast-hulls`).
+- **Orbital Forge**. No entry in Faction Flavor; mentioned as an independent yard supplier (Aya line 458; AU lore line 29). Its own lore page `Pre-Elysium\Factions\Powers\Minor\Orbital Forge.md` was not read. Cavity-building and salvage methods through open standards (AU page).
+- Others in Faction Flavor: Sol Dominion (axial spine, repeated module bands), Cryonix, Lightsail, Cetacean Navigators, Adrasteia, Ewan Hart, Death Monkey Explosives (own intact hull with exposed serviceable machinery), Miss Terri's, Corriedales, Framgång.
+
+Registers (Ship Design Language.md "Registers" table): Corriedales, Miss Terri's, Framgång high whimsy; Lightsail, Ewan Hart, Cetacean whimsy turned up; Lucent spectacle; Zhestokost, AU, Finch, R&D, NiteLife, Alakrita, Adrasteia, Sol Dominion, Cryonix, Aya, Megiddo serious; Pirates wild; DME reserved.
+
+### I2. Prompt-lab findings on keeping a faction read (internal measured evidence)
+
+From `Ship Design Language.md` "Rules From the Prompt Lab, Rounds 5 to 8" and Bench 4 (24 FLUX 3 renders, agent-promoted, pending operator review):
+- Rule 3: faction blocks carry livery and register only; shape language in a faction block "leaks onto every hull of that faction" (18 renders, 10 ships, 7 factions held the faction read on livery-only blocks). Failure: Lucent's identity lives in shapes, so Key Light went generic with motifs cut (r7a) and recovered when the ship prompt carried the gaming-mouse layout (r8a).
+- Rule 4: wear belongs to the faction block (16 renders, 4 factions).
+- Rule 1: vacuum hulls have no aero surfaces (21 of 22 renders held); a Pirate racer donor with no other identity grew a wing; livery identity fixed it on the next four.
+- Rule 2: weapons appear only where named; named counts drifted until each gun was given a position.
+- Craft rules for image-to-3D in `Faction Ship Concept Prompts.md` ("Prompting for Image-to-3D"): one familiar object as analogy supplies layout ("Headliner and Quiet Sense worked"); thick, closed, symmetric; four or fewer counts, each tied to a place; three or four colour zones each tied to a part.
+- Global operator-side facts: the operator adds thruster ports, attitude jets and radiators herself (`Ship Design Language.md`, "Thrust is named only where the concept needs it").
+These measure image-model prompts, not mesh authoring; no render-and-compare against a mesh hero asset has been run in this repo.
+
+### I3. Existing meshes and possible hero assets
+
+Shipped in the Unity project (`F:\Projects\Aetheria\Assets\Models\`, sizes from `ls`): `Longinus.fbx` 1.1 MB (Alakrita; hulls Longinus and LonginusX per follow-up `demo-cast-hulls`), `Djinni.fbx` 1.9 MB (Rossum & Douglas per the same follow-up), `Turret.fbx` 0.36 MB (Zhestokost turret per the same follow-up), `Zenith_Blender.fbx` 18 MB and `Zenith_Pad.fbx` 0.7 MB (AU's Zenith, a station), `mine.fbx`, `primitives.fbx`, spheres, `CatmullGrid.fbx`, `Discobolus` statue, Plugins/LowPoly_AsteroidsPack (asteroids only). Per the follow-up, the only ships among these are Longinus, LonginusX and Djinni. Operator quote on Longinus: "Honestly I'd be happy to never see that Longinus model again" (ruling `demo-player-start-hull`, since superseded by `demo-start-hull-longinus` because the Djinni was too strong).
+Operator sources (`F:\Projects\Aetheria\Asset Sources\`, names and sizes only; .blend files not opened): `Alakrita1.blend` 96 MB, `Adrasteia1.blend` 24 MB (+ `Adrasteia1_ContourRoute.blend`, `adra1.png`), `Ships.blend` 47 MB, `Turret.blend`, `Zhestokost.psd` 44 MB and `Zhestokost.png` 8 MB, `Lucent.psd`, `Mouse.psd`, `mine.blend`, `Royal_Edges.blend`.
+Not in tree: `GameData/Mods` does not exist in the repo checkout; the release map says the operator's five packages are to arrive there (`aetheria-release-map.md` line 42, `demo-cast-hulls` r3, waiting on `mods-in-tree`). Keepers from the concept pipeline (ruling `hull-concept-keepers-2026-10-08`): Bison, Squatter, Oboz r8f, Shchit, Key Light r8a, Tattletale, each going Tripo, Blender, package; their faction assignment was not established here (Key Light is a Lucent concept per the gaming-mouse layout in rule 3). The Tripo fixture `F:\Projects\Aetheria\tools\blender\tests\fixtures\tripo-hull.glb` exists as the add-on's bake fixture (`addon-tripo-fixture`). Which of these are "fully produced" is not recorded in the files read; the operator has said she is "rebuilding the nonsense geometry and unwrapping anyway" for Tripo-derived hulls (`hull-paint-mask-texture`).
+
+### I4. Rulings in force bearing on this (titles; quoted where read)
+
+- `hull-art-communicates-form` (this ruling). Supersedes `hull-programmer-art-reads-as-stand-in` and, before it, `hull-programmer-art-stopgap`.
+- `hardpoints-and-floating-greebles`: "Two separate systems": hardpoints from one standard size set, molded into the hull by boolean; a library of floating greebles to decorate, independent of fitted gear.
+- `schematic-and-model-are-separate-spaces`: schematic authored for gameplay; "No schematic fact is derived from model geometry and no model geometry from schematic cells"; the two are linked only by mount name; no consistent scale between ships.
+- `shippable-part-bar` (brush-tree): "clean unbroken flow, part-coherent low-distortion UVs, one-segment bevels with face-weighted normals". `representation-per-hull`: subd limit surfaces for continuous bodies, primitives for mechanical hulls. `openscad-target`: trees authored in OpenSCAD-shaped text. `art-tools-not-cad` (2026-10-14): "drop the cad thread, let's stick to art tools". Titles only read for these three; bodies not opened.
+- `mesh-probe-exploit-not-measure`: "I'm not interested in measuring Opus's spatial reasoning performance, in interested in exploiting it."
+- `demo-cast`, `demo-content-bar` (one player-usable hull per cast faction plus role hulls "haulers, tenders, miners"; follow-up `demo-role-hulls` says no spec owns them and quantity comes from faction-play doctrine and roles), `content-bar-one-per-concept` (13 hulls for the full game), `boss-hulls-exclusive`, `package-carries-product` (hulls ship as first-party packages with a product record), `demo-material-scope` (baked maps, a mask, one hull shader), `hull-paint-mask-texture` (per-ship livery mask on UV0), `hull-runtime-materials`, `ship-authors-need-no-unity`, `hardpoint-work-held-for-walkthrough`.
+- Release map: section "Stand-in hulls (SI)" at `aetheria-release-map.md` line 11272 is marked superseded; "its package path, anchors and replace-by-real-art lifecycle are evidence for the re-map; its visual grammar and cut sequence are not live."
+
+## Known gaps
+
+- No primary source for Homeworld race design language (Art of Homeworld book not read), Amarr/Gallente (EVE), TEC/Advent (Sins), FTL, Star Citizen or Elite manufacturer rules, Battlefleet Gothic official material, Starsector sprite-level style rules.
+- Direct fetch refused for the Starsector blog and Sins II pirate post; both are search-summary only.
+- Valve TF2 PDFs and EVE/Stardock items beyond the fetched pages were not opened. Valve material is about characters, not ships.
+- No source found for the exact mechanisms the ruling asks about: per-faction kit libraries driven by an LLM, style sheet plus parts library plus render-and-compare against a hero asset, or measured style-drift rates in agent-built 3D. Searched with several phrasings.
+- Silhouette-at-small-size evidence for ships is mostly blog/forum opinion.
+- Internally: Orbital Forge, Pirate Coalition-as-catalog-faction, and Megiddo have no ship concepts in the files read; the Lucent concept PNG, keeper concept images, and Bench renders (`Renders` dir) were not viewed; which assets count as "fully produced" is not recorded; the operator's five packages are not in the tree; Zhestokost, Pirate and AU lore pages were only grepped, not read in full; `Ship Prompt Bench 3/4` entries for Pirates and Zhestokost tenders were seen by heading only.
+- Pointers marked search-summary were not verified at the primary page.