Call out that the tabletop and engine package split is not yet landed, add a status note to tabletop.md, and drop the retired WIP banner in bgm-loader.md.
Add the command/dialog split to the state-model and interactions docs, declare role dialog and zone facing in the format spec, and let setups declare which dialogs power open interactions.
Describe the format's state model: components as constants, setup as the shape of states, path-to-stack by facing (part affordance x zone restriction), and the stacks-on-paths anchoring scope cut.
Group the bgm spec cluster under docs/bgm and move dev logs and plans under docs/status, add an overview index, and update cross-references in the docs, README, and source comments.
Replace the single dominant top light with evenly-spaced key lightformers
so a laid-flat card's top face isn't blown out and its underside is lifted
by a warm bounce.
Lay flat card/tile/token meshes so the scene-level contact shadow has a
surface to project onto, and size the shadow plane to the measured content
bounds instead of a fixed 22-unit default that left small objects shadowless.
Give deck carousel cards their own soft radial shadow so the fanned arc reads
as floating over a floor.
Replace the hand-placed light rig with an Environment built from Lightformers, giving MeshStandardMaterial objects consistent PBR lighting and richer color. Keep ContactShadows, which needs no light.
TextureLoader leaves colorSpace as NoColorSpace, so card, tile, token, model, and part art is uploaded as linear and then double-decoded in the shader, washing out contrast. Mark each texture sRGB so the GPU decodes it once, correctly.
The EffectComposer disables three's tone mapping, leaving colors flat and washed out. Add ACES filmic tone mapping plus a small brightness lift so cards match their source images.
Replace the counter with a sliding-dot pagination: dots slide and fade
as the active card changes, a counter above the bar shows the current
index, and round first/last and prev/next buttons flank the strip.
Scrolling over the dots steps through the deck. Also fix the camera fit
so it frames the active card front-on instead of the carousel center.
Fill the viewport with a fixed sidebar of highlightable type-filter
icons over an independently scrolling tree, and let the viewer expand
into a full-height scene.
Fit the camera in a frame callback so the active card has settled into
its arc slot first, and aim it straight at the card's front face instead
of keeping the initial side angle. Measure the card's width and widen the
arc radius so wide cards no longer clip their neighbors.
Replace the global TTS Workshop header on mod routes with a ModHeader showing the thumbnail, game name, and download/setup links styled like the Search/BGM nav.
Include the Workshop title and preview image in the items response so the mod header survives a page refresh, not just navigation from search. Resolve both from the same Steam API call.
three.js does not auto-declare uniforms added via onBeforeCompile, so the
injected uMapRepeat/uMapOffset were undeclared and the card shader failed
to compile, rendering cards transparent. Prepend the declarations to the
vertex shader.
Show the game name and thumbnail in the header, float the action buttons and object GUID to the right, and keep the viewer icon, label, and GUID on one line.
Move the card sprite repeat/offset out of per-card texture clones and
into a uniform injected into the material shader. Cards now share the
deck sheet (one GPU upload), the shader, and the geometry, with only
per-card material uniforms differing, so navigating a deck no longer
re-uploads the sheet on every step.
Add a DeckViewer that frames the active card in a deck carousel. Scene gains a fit prop so the shared bounds behavior can be disabled and refit manually.
Chips highlight object types to show only those in the tree; with none highlighted, everything is shown. Filtered nodes keep their original index path so selection stays stable.
Add a /pdf proxy route that forces Content-Disposition inline so PDFs render instead of downloading, and a web viewer that embeds the document in an iframe with a standalone open-in-new-tab link.
Replace the CommandHost with a handler registry keyed by message type.
Message is a permissive base type the host extends; CommandResult is a
discriminated union on the type suffix. Commands and orchestrators are
the same async shape taking a RunContext with an AbortSignal; runCommand
emits :done/:cancel/:error.
Messages are a discriminated union with generic command/result types;
the command host is dropped for a handler registry; commands and
orchestrators are the same async shape taking a cancellable RunContext.
New @tts/engine package: the pure message layer that drives bgm games.
MessageQueue snapshots and drains on tick; TriggerRegistry matches by
type/id with runtime enable/disable; runOrchestrator suspends on wait
until a matching message is processed. Engine ties them together behind
the CommandHost contract that @tts/tabletop will implement.
Define the message/queue/tick model, the three handler kinds (trigger,
orchestrator, command host), and the @tts/engine package split. Cross-link
from bgm-commands and bgm-tabletop, and add the engine to the architecture
package table.
A definition is identified by role.type and named role.type.lang
(package.lang for packages). Code blocks declare it via role= on the
info string; real files by their filename. Blocks and files without a
matching role.type are ignored. file= still overrides the auto-name.
Relative asset and $variants paths now resolve against the markdown
file's directory (baseDir) rather than the virtual name. Same role.type
with different ids groups under one name; duplicate type#id still
errors. Migrate fixtures and example games to role=.
Parse role.type#id from the fence info string and merge it into each
parsed object, erroring on conflicts with the content. id on the info
string can't combine with $variants. Auto-named blocks now hash their
content instead of the info string, so identical blocks dedupe.
Add bgm-commands.md covering async commands with ok/cancel/error
results, per-invocation run contexts, fire-and-forget vs self-managed
waiting, and tap interaction with part-local trigger points. Link it
from the tabletop design and plan docs, and record the decisions in
decisions.md.
Expand $variants to accept an array of csv paths, concatenating their
rows. Detect a path by a .csv suffix on the first line instead of a
newline, so inline csv and paths are self-documenting and the rule
applies uniformly to single values and array elements.
Replace the face/back boolean with a three-state facing (face, back,
standing) seeded from setup placements. Parts now orient about the
anchor at the resting face/edge, so back-down parts sit on the table
instead of below it and standing parts rest on their bottom edge.
Convert poker and carcassonne setups to the ordered placement list. Deal the deck/draw first, then move the flop cards and opening tiles out onto their paths.
Replace the path->parts setup map with an ordered list of placements, each moving its parts to a path. Entries apply in order, so a part listed in a later placement ends up on that placement's path.
Replace the path->part-list store with a per-part map keyed by part id, so each placed part carries its path, stack index, and face. Derive each path's ordered children for stacking, and flip face-down parts in PartPlacement.
The default include is now resolved relative to the package
declaration's directory, so the explicit root-relative patterns are
redundant. Drop them in favor of the default.
The default include matched every yaml across the games root, so a
package could absorb a sibling game's defs (e.g. two game#main setups).
Resolve patterns against the package declaration's directory instead;
a leading slash marks a pattern as root-relative.
Replace the dark surface plane with a grey background and a spotlight,
size the contact shadow to the table, and scale its blur and fade with
the shadow plane. Disable depth test on the stacking curve overlay.
Fix the tangent angle at the curve's start and rotate cards to follow
the curve direction. Add a stacking curve visualization shown with
surface debug.
Candidates now carry their own stacking strategy, inheriting the route's
when absent, so the default tilt applies consistently. Drop the now
redundant explicit tilt from the poker deck.
Apply the same tilt to every stacked part and rotate it about its local
Y axis instead of Z. Default tilt to 1 degree when a stacking strategy
is present, and arc the poker deck along the surface's bottom edge.
Normalize the games root to POSIX separators so the HMR path guard
matches on Windows, and watch the real on-disk markdown files (which
defMap.files omits in favor of their virtual code blocks) so edits
trigger a re-collect instead of a manual restart.
Replace the per-part lift with a local Y-axis tilt that fans the stack,
and add zStart/zEnd to ramp the stack's height across the curve so it
arches in 3D. Update the poker deck and docs accordingly.
Add a toggleable surface plane, bounds outline, and constant-size HTML
label. Scale placement and mount anchors to world units so parts land on
the surface, and rotate parts to face up.
A piece can appear on more than one path of the same surface (e.g. the
poker deck expands to every card while the flop also places the ace), so
the render key must include the path to stay unique.
Wire the tabletop library into the setup detail route: seed the store
from the setup, resolve the surface mount tree, and render world and HUD
surfaces. Add serializedToPackage to convert the vite-emitted package.
Implement the tabletop library's game state store, setup seeding with
bare-type expansion, surface mount tree resolution, part placement, and
the stacking positioning process with a dependency-free SVG path helper.
Wire the public API and add unit plus vite integration tests.
Lift the trace-to-shape geometry into @tts/mesh (traceToShape/
traceToUvBounds), parameterized by scale so the web token viewer and
@tts/tabletop share one implementation. Move the CORS proxy HTTP helpers
(assetUrl, resolveAssetUrl, traceImage) into a new @tts/http package, and
make @tts/tabletop's ErrorBoundary the single source used by the web app.
This removes the duplicated ErrorBoundary, assetUrl, tabletopHttp, and
trace-to-shape code from apps/web and packages/tabletop.
Vite 8 (rolldown) changed buildStart to take (this, options), so the
test's one-arg call failed typecheck. Add an ambient declaration for the
fixture's virtual bgm modules so its entry typechecks in the editor.
Add a SerializedPackage type for the JSON the plugin emits, and tests
that assert the maps serialize to plain objects, that the packages module
returns every package, and that buildStart watches def files. Add a
second fixture package to exercise multi-package collection.
Rename the plugin's virtual modules to virtual:bgm/packages and
virtual:bgm/package/<id> so they read as plugin-provided modules rather
than real packages. Update the unit and build integration tests.
Move the bgm vite plugin into the package so it can be tested in
isolation from the web app. Serve each package as a JSON module,
serializing its maps, and verify resolution through a real vite build
against a self-contained fixture.
Parse yaml/json/toml and markdown code blocks into packages, expanding
$variants via typed-csv and collecting parts, surfaces, and setups by
include patterns. Ships zod validation, a vitest config, and 16 tests.
Model ColorDiffuse in shared types and multiply each viewer's material color by the per-object tint, including textured faces and loaded custom models.
Model TTSObjectTransform in shared types and convert each object's position, rotation, and scale into three.js placement, with per-class base-size corrections and a lay-flat rotation for extruded meshes.
Add a /mod/:id/setup page that lays out every renderable object in a single shared scene. Export object-facing mesh wrappers from the viewers and share geometry and materials across objects via module-level caches.
Document the separated front/back/walls extrusion, material-level back
flips, CardID sprite selection from the parent deck, index-path tree
selection, and the Bounds-inside-Suspense camera fit.
Refactor `TokenMesh` to use a custom `useTrace` hook that suspends
during the alpha channel tracing process. This allows the `Bounds`
component in `Scene` to correctly fit the camera to the token's
geometry once it has been fully loaded and traced.
Cards in a deck share the deck's GUID, so GUID-based selection highlighted
every card with that GUID and rendered the first match. Key selection by the
node's unique index path instead.
Split extrudeShapeParts into front, back, and walls so each face can carry
its own material. Flip back textures left/right on the material so they are
not mirrored, and slice card faces from the deck sprite sheet via CardID.
Resolve deck config from the parent deck, which is authoritative over a
card's own CustomDeck. Add unit tests for card resolution, sprite UVs, and
the flip helper.
The compiled server runs Node's native ESM loader, where the CommonJS
namespace object nests the real exports under 'default', so
'import * as ClipperLib' left JoinType undefined and crashed with
offset. Match the existing vtracer pattern and require() the package
instead.
hypercross