viz: coherent operator notation across the logical and circuit views

[balegas]

Operator-notation fixes so the same operator reads the same in the logical view and the dbsp circuit view. Verified live in the running visualizer (Playwright).

1. Aggregation Σ → γ

The visualizer already uses relational-algebra operator symbols — σ (filter), ⋈ (join), π (projection). Aggregation was the odd one out, labelled Σ, which denotes summation, not the aggregation operator. γ is the standard grouping/aggregation symbol.

• Legend: Σ aggregate → γ aggregate (logical), Σ fold → γ fold (circuit)
• Node tags: Σ FOLD · STATE → γ FOLD · STATE, DBSP COUNTS · Σ STATE → DBSP COUNTS · γ STATE
• Engine (introspection.rs): the circuit-view fold label dropped its Σ prefix — it emitted Σ COUNT(*) while the logical view showed COUNT(*). Both views now show COUNT(*); the γ identity lives in the node tag (γ FOLD · STATE).
• README legend + a CSS comment

The Σ that denotes genuine summation is kept — Σ of Z-set weights, matching rows (Σ weights), and the fold formulas fold(Σ value·w over σ(where)). A γ node now reads as the aggregation operator implemented as a fold that computes Σ value·w.

2. Subquery inner set: arrange → distinct in the logical view

The logical-view inner-set node was tagged IN-SET ARRANGE, while its circuit-view counterpart (op-distinct) is DISTINCT — and the logical node's own formula (distinct(π(proj)·σ(where)·inner)) and sublabel (distinct project_id) already said distinct. So a reader tracing the subquery inner set saw two names.

Unify on distinct — the DBSP operator (paper Table 4.2 spells distinct, not the RA symbol δ; the blog's subquery diagram agrees):

• Logical legend IN-set arrange → IN-set distinct, tag IN-SET ARRANGE · STATE → IN-SET DISTINCT · STATE
• Circuit legend arrange (state) → distinct (state)
• README + two internal comments

arrange stays only for op-arrange — the genuinely different physical node (a subquery shape's feed set / a family's params index).

Not changed: distinct never became δ

DBSP notates distinct as the word, so the visualizer keeps it as distinct in both views (matching the blog diagram). No δ is introduced.

Verification

• vitest run apps/pipeline-viz/src/build-graph.test.ts apps/pipeline-viz/src/build-circuit.test.ts → 15 passed
• Rebuilt engine + drove the running visualizer (Playwright): logical and circuit views both render γ FOLD and distinct, the circuit fold label is now COUNT(*) (was Σ COUNT(*)), and every kept Σ is genuine summation.

🤖 Generated with Claude Code