Interface ToolpathRendererOptions

Camera projection (#150, DD-009 D3); default 'perspective'.

Count-based tick override (tests/deterministic hosts): exactly N chunks per tick. Default (unset): TIME-budgeted ticks — chunks build until ~8 ms of work has elapsed (§8: no main-thread stall > 16 ms during incremental build).

Logical-core budget for pool sizing — navigator.hardwareConcurrency (browser) or the cgroup CPU quota (Node/sidecar). Injected so the renderer stays platform-free; omitted → conservative default.

Factory for a geometry-build worker (DD-028). A bundler passes the batteries-included geometry-worker.js; tests inject a synchronous fake. Omitted → the pool never engages.

Injectables for tests / exotic hosts.

Framing target (#306/#6): 'all' extrusion (default) or the printed 'object' (excludes skirt/prime).

Parallel tube-geometry build across a worker pool (DD-028). 'auto' (default) sizes a capability- and memory-aware pool and engages it only for tube builds above a cost threshold (small builds and lines stay on the synchronous main-thread path); 'off' forces the synchronous path; a number pins the worker count (still bounded by the memory budget). Geometry is byte-identical either way. No effect without a createGeometryWorker factory (a bundler supplies one via the batteries-included worker entry; tests inject a stub).

Max bytes of tube geometry the parallel build may hold in flight at once (DD-028 memory backpressure). The pool shrinks so workers × maxChunkBytes ≤ this, bounding peak transient working set proactively (a cgroup OOM is uncatchable). Defaults to a fraction of the tube byte budget.

Interaction-aware quality (#306/2, DD-020): 'auto' reduces detail while the camera moves. Default 'off'.

Preserve the WebGL drawing buffer so the canvas can be read back after a render returns (toDataURL / convertToBlob / readPixels). Off by default (interactive rendering is faster without it); the headless still-render path (#132) turns it on so a single render is capturable.

During-parse preview presentation (#60 curtain): 'lines' (default, stream the progressive line preview), 'hold' (keep progress but reveal only the final coloured build — a single clean reveal), or 'off' (no progressive geometry or preview progress; consumer drives its own loading treatment). Does not change the final representation.

§4.3 quality tier; 'auto' (default) picks tubes ≤ 1 M segments, else lines.

Fidelity POLICY (DD-023 §4 D6), distinct from the geometry quality tier (lines/tubes) — what the user/admin WANTS, not what the client is running on (capabilityHint). Modulates the two large-file ceilings within the chosen geometry mode:

• 'full': render the COMPLETE representation — no every-Nth decimation, full-radial continuous tubes, no budget-driven tubes→lines fallback (only the per-chunk vertex safety net still applies, as an honest last-resort). The hard job of failing gracefully when a client genuinely can't is a later phase; today 'full' attempts the full build.
• 'adaptive' (default): the capability-aware auto path — auto decimation + the tubeByteBudget cross-section coarsening, every reduction disclosed. Reproduces today's behaviour.
• 'fast': explicitly trade fidelity for responsiveness — render as flat lines.

Render the scene after each incremental build tick (progressive-build feedback). Default true — the interactive viewer wants to watch geometry appear. A headless one-shot still sets this false: it builds geometry to completion and renders once at the end, avoiding the dozens-to-hundreds of discarded full-scene renders a large build would otherwise do (each an MSAA software rasterization of a growing tube mesh — the dominant cost of a big still in software WebGL). No visual difference; buildComplete and all build events still fire.

Bounded declarative theme (#153, DD-009 D4); omitted fields keep the default look.

Tube profile parameters (tubes mode only).

CPU byte budget bounding tube memory (RR-006): above it the tube cross-section is coarsened (fewer sides) — never dropping segments — and, only if the minimum cross-section still exceeds it, the build degrades to flat lines. Default ~450 MB (safe in a 2 GB render cgroup). Raise on a memory-rich host for higher-poly tubes on huge files.