Documentation

Scorecard

External “A+ language” grades are evidence, not a new product vision. Updated after true Execution A+ (no recompute fallbacks for supported shapes; multi-row JOIN cards).

Dimension Grade Evidence
Honesty A+ Claims match tests; Live Graph differentials are real WHEN NEW / join-bag / bag — not recompute theater
Architecture A parseLukeProgram/Stmt/AstflattenProgram → emit on every BUILD; Pratt exprs; IF/WHILE/FUNCTION/WHEN are nested Stmt nodes
Execution A+ Point + multi-row join IVM, inequality bag filters, N-table chains, wire fail-closed, C10K event-loop HTTP, wall smoke
Ambition A+ Vision held — reactive full-stack; mobile/game parked
Tooling A LSP (hover types/signatures + outline/refs/rename/signatureHelp/FMT/semantic tokens/code actions) + FMT + #line maps; luke DEBUG/DAP
Adoption A- Wall deploy proof + Caddy; site/ stub; registry sha256; public DNS/traffic still to earn

What “Execution A+” required (and shipped)

  1. Partner / non-point joins: rowid-keyed join bag (luke_ivm_jbag_*) — multi-row group_concat, not scalar-only cards.
  2. 3+ table equi-JOIN point chains with equality-closure WHEN + NEW-pinned probes (live_graph_join3).
  3. Differential aggregates for any NEW./OLD.-qualifiable filter (equality, inequality, LIKE, …) via luke_ivm_bag_* (live_graph_agg, live_graph_agg_range).

Still not A+ everywhere (honest remainder)

Gap Why it is not a vision change
Some surface still StmtKind::Raw (ROUTES/FLOW/…) Specialize nodes without inventing a second IR
Debugger IDE polish Protocol beachhead shipped; richer UX is editor-client work (parked with distribution)
Distribution (signing, remote registry, libs, VS Code ext, docs site) Parked — see TaskList §7; Adoption after the wall sentence
LEFT JOIN / non-equi ON / expression-only SELECT lists Broader SQL shapes outside the equi-JOIN + simple-column bag

Do not dilute: un-parking mobile/game or distribution (12–16) to chase a checklist is how Ambition drops.