Data workbench & EDA studio (/data · /studio)

Shipping since v0.7.70. Both surfaces share the same line-level permissions: charts are not data proxies — every read goes back through the server-authorized query path.

• /data workbench — look at data: line-scoped live mirror, MES console, export/batch ledger;
• /studio EDA workbench — use data: charts, pipeline post-processing, grid dashboards and a set of enterprise-grade EDA analyzers.

Data workbench (/data)

Three tabs, always line-scoped:

Tab	What it shows
Mirror	WS live mirror by line + time window (15 min / 1 h / 4 h / custom) + aggregation; scalars, array profiles and CCD images side by side
MES	MES console: interface messages and receipts per line
Ledger	Bulk export job ledger: every export produces manifest + merged.csv with trackable job status

EDA studio (/studio)

Charts, templates, dashboards

• Seven chart types: trend / bar / scatter / table / kpi / histogram / pipeline (kpi and table are single-series);
• Two visibility levels: private (owner only) and line (bound to a line, read governed by line grants);
• Templates save a query configuration as a reusable asset; the board (/studio/board) arranges charts into a grid cockpit;
• Board cards re-query on every render; pipeline snapshot charts are read-only replays on the board.

Golden phase reference

Build a reference profile on a 0–100% phase axis from historical good batches, overlay the running batch and output a deviation score — "how close does this batch run to the golden sample", in one number.

FDC / SPC / OCAP / VM

Analyzer	What it does
FDC reference bands	Extract per-batch features and judge against reference bands, as background jobs (run + jobs query)
SPC control charts	Nelson rules 1–8 / I-MR / EWMA / CUSUM, with recompute
Commonality	Cross-batch common-factor search
OCAP dispositions	Disposition workflow records after an out-of-limit event
Virtual metrology (VM)	Model-predicted quality values; OOD (out-of-distribution) is rejected outright instead of guessed
Derived datasets	Freeze cleaned/aggregated results into new datasets for downstream analysis

Runs support labels, giving the analyzers human anchors for good/bad batches.

Pipeline charts (/studio/pipeline)

Pipeline charts chain "fetch → code post-process → plot" into one step:

1. Online fetch (same authorization as /daq/query) or CSV import;
2. Code post-processing: transform in the browser (filter, derived columns, resampling);
3. Plot & online viewer: rendered with ECharts; the viewer supports zoom and PNG export;
4. Save: as a template (configuration only), or as a pipeline snapshot chart (see the B4 boundary below).

B4: charts store parameters, not data

Ordinary charts store only query parameters (nodes, time window, aggregation…); every open re-queries through the same authorization as /daq/query — no data copies, no way around line permissions.

The single exception is the pipeline snapshot, a deliberately bounded escape hatch:

• private visibility only, must carry a line_id that matches the chart;
• the rows in options are a client-supplied saved view, capped at 2000 rows;
• reads still go through the chart's owner / line visibility; Studio shows snapshots with a read-only badge and no overwrite;
• non-pipeline charts allow only template-style configuration references in options — any other data/extension key is a 400.

REST & agent surface

Everything the studio does is exposed as /api/workshop/eda/** REST routes (charts / dashboards / templates / golden / features / spc / ocap / commonality / vm / derived / run-segments / runs). On the agent side, data tools such as daq_query / daq_export read the same data plane, with references (nodes, lines, batches) passing the unified preflight (tool-verify: argument shape + reference ownership).

Acceptance

• End-to-end: scripts/e2e-eda-pipeline.mjs, 24/24, including B4 negatives (non-pipeline chart carrying data keys → 400; snapshot line_id mismatch → 400; missing rows → 400);
• Unit: tests/chart-pipeline.test.ts (series-to-rows reduction, dual-shape point compatibility).