v2.6.0 — Aggregates and grouping

ProcFlow v2.6.0 — Aggregates and grouping

Released as the aggregate milestone. The query builder now answers grouped
questions — orders per customer, totals per product — without leaving the
declared-evidence model or hand-writing GROUP BY.

Highlights

• Aggregate any picked column. Each chip carries a selector: —,
  COUNT, COUNT DISTINCT, SUM, AVG, MIN, MAX.
• GROUP BY derives itself. Every remaining picked column becomes the
  group list in pick order; when all picks are aggregated, GROUP BY is
  omitted and the query returns one row. No second multi-select, no invalid
  states.
• Aggregates read like SQL. Deterministic aliases
  (count_orderheader_orderid), sorting an aggregated column orders by the
  expression (ORDER BY COUNT(...) DESC), and the provenance header records
  Grouped by: and Aggregates: lines.
• DISTINCT yields to grouping. While an aggregate is set the toggle is
  disabled and the SQL omits the clause, with a header note explaining why.
• Fan-out honesty. A plan tip warns that SUM/AVG over a one-to-many
  join can multiply unless the detail rows are pre-aggregated.
• Versioned files stay compatible. Query files move to version 2;
  version-1 files from v2.5.0 load and migrate forward with an empty aggregate
  map, and newer versions are rejected with a diagnostic. Aggregates are
  saved, exported, restored, and pruned with their picks.
• Documentation. ADR-003 records the derived-GROUP BY model and the
  rejected alternatives (HAVING, explicit grouping, COUNT(*), window
  functions); docs/QUERY_BUILDER.md gains an aggregates section.

Accuracy and limits

• No analysis or join semantics changed. Joins still use declared foreign-key
  evidence only; taught joins and self-join copies remain labelled not
  declared; the generated SQL stays read-only.
• No HAVING, no window functions, and no COUNT(*): COUNT(column) ignores
  NULLs, so count a NOT NULL key for row counts. GROUP BY covers picked
  columns only.
• Unknown aggregate functions are ignored, never guessed.

Verification

The final local verification recorded for this release includes:

• golden browser assertions on tests/index.html passing (221/221), including
  the query suite (55 records) with aggregate rendering, all-aggregate
  GROUP BY omission, aggregate ordering, DISTINCT suppression and header
  notes, unknown-function tolerance, version-1 file migration, and storage and
  pruning coverage;
• the ERD query UI suite passing 68/68 over HTTP, including the aggregate
  selector, function rendering, GROUP BY, DISTINCT disabling, the plan
  tip, and reset;
• the Firefox smoke page passing 10/10 critical paths;
• npm run typecheck, npm run build, npm run test:file, npm run metrics,
  npm run package:smoke, and npm run benchmark passing;
• runtime archive structure and local-file startup smoke passing
  (procflow-v2.6.0.zip).

Upgrade and distribution

The runtime is self-contained. Use npm run package:runtime to produce the
versioned ZIP and verify .release/SHA256SUMS.txt before distributing it.
Layouts, workspaces, and version-1 query files saved by earlier releases remain
readable.