nordisk/facett

facett-pipeline — crate card

Reverse-engineered (from source — verify, don't trust the name)

• One-line pitch: A dataflow / Spark pipeline-DAG viewer + job progress, themed and animated (Falsett throughput packets, per-stage status glow, throughput labels).
• Does: PipelineView uses facett_core::layered for producer-left columns and a persistent facett_flow::FlowScene for GPU/CPU geometry. Engine-agnostic: the canonical input is the neutral PipelineModel (in model.rs, with Stage/StageStatus/StageMetrics/PipelineEdge/PipelineMetrics) produced by a PipelineAdapter — so a new engine is a new adapter, not a new viewer. Back-compat surface (new/with_progress/counts/Status/JobProgress, navigation and state_json["dag"]) preserved. Status and consumer throughput map to separate Falsett tables; zero throughput suppresses packets. The deprecated flowsim overlay explicitly owns animation while active, avoiding duplicate packets.
• StageStatus::Unknown(UnknownReason): "we could not determine this node's state, and here is why" — distinct from Idle ("we know it has not started"), and rendered amber + dashed + ? rather than the idle grey, so an unobserved node can never read as a fake green. Idle is still the default; both status enums are #[non_exhaustive]. StatusTally is the five-way count that actually sums to the node count (counts()'s legacy four-tuple does not).
• Ships two in-crate adapters: RunaPipelineAdapter (feature runa-adapter, maps dwarves_runa::Pipeline) and PromoteLadderAdapter (always built; maps a local PromotionLadder of slots/rungs so a promotion ladder renders on this same viewer — see Njord/.nornir/deployer-layering-decision.md). Neither drags an engine dependency into facett.
• Key deps: facett-core (layered layout, effects/glow, nav, Facet, theme), facett-flow (persistent geometry, boxes/arrows/status/rate; optional wgpu), egui, serde_json. Dev-only: knut-pipelines (real Spark/SDP adapter drives the test matrix; tolerated normal+dev dep cycle).
• Shape: lib; small (lib.rs + model.rs); ships a pipeline_demo example (glow).
• Observed role in the workspace: Re-exported by the umbrella facett; consumed by knut for Spark/SDP pipeline visualization, with shared Falsett rendering.

Falsett migration verification (2026-10-04)

• Actual workspace: 65 tests passed (44 unit, 3 accessibility/determinism, 1 accessibility-priority, 6 flow-lane, 6 callchain matrix, 2 real knut connector and 3 robot/pixel tests). The 2,050-node test keeps every visible rectangle while limiting semantic buttons to 2,000; a separate 2,100-node case keeps late selected/focused nodes named.
• Existing flowsim pixel tests passed 3/3 on the Mac GPU: live overlay changed 522 pixels, zero-rate overlay changed zero, and advancing the clock changed 366 pixels.
• Real Metal callbacks pass the GPU integration test in the actual workspace: status-only reconstruction uploads 64 attribute bytes, keeps geometry/target identity and changes pixels; clock/camera changes upload no geometry. The new tests/snapshots/flow_gpu.png reference was captured and compared in the actual workspace on Apple M2 Pro / Metal within 2 levels per channel; its README records provenance. Other hardware comparisons have not run.
• wgpu compilation and library Clippy with -D warnings passed. The actual workspace manifest/GPU-install gates also pass (3 feature tests and 5 installation tests). These component results do not certify every workspace feature or complete application.

Fill in (one line each — only the human knows these)

• Primary consumer(s):
• Why it exists (vs the obvious alternative):
• Hard constraint (airgap / no-C-deps / no-FFI / pure-Rust / etc.):
• Release cadence & independence (rides a parent? own repo someday?):
• Status (production / prototype / experimental / stubbed):
• Must-NOT-change invariant:
• Biggest known gap or TODO:
• Anything a newcomer would get wrong about it: