2026-05-16 — arc 199 REJECTED + D1 refactored to clean call form

Opened arc 199 (parametric-keyword expressiveness in defmacro) earlier same day on the back of Stone D1’s verbose call form. DESIGN sketch ran four-questions on three candidates; Candidate 1 (expand-time :wat::ast::parametric-keyword constructor) led.

Then user asked me to investigate existing substrate machinery. Findings:

• :wat::core::keyword/from-string (src/check.rs:11931) — String → keyword Value (adds : prefix; rejects :-prefixed input)
• :wat::core::keyword/to-string (src/check.rs:11923) — keyword → String (strips : prefix)
• :wat::core::string::concat (src/check.rs:4653) — variadic String concat
• Computed unquote at macro expand time — arc 143 slice 2 (src/macros.rs:1010+). When ~(:keyword/op args...) appears in a defmacro template, the expander substitutes macro params into the expression, calls crate::runtime::eval AT EXPAND TIME, then value_to_watast converts the result to a WatAST node landing at the ~(...) position.
• value_to_watast (src/runtime.rs:8815) — Value::wat__core__keyword(k) → WatAST::Keyword(k) is the working conversion.

Production precedent: arc 143 slice 6’s :wat::runtime::define-alias macro (wat/runtime.wat:22-29) uses the EXACT pattern — ~(:wat::runtime::rename-callable-name ...) at expand time. In production since arc 143 shipped (2026-05).

Arc 199 REJECTED 2026-05-16. DESIGN.md inscribed with REJECTED header; original DESIGN text preserved as historical artifact (per feedback_inscription_immutable).

Stone D1 refactored same-day

Macro signature changes:

• Before: (run-threads :Receiver<I> :Sender<O> factory client-fn) — caller spells out full channel wrappers
• After: (run-threads :I :O factory client-fn) — caller passes just type args; macro constructs :Receiver<I> / :Sender<O> at expand time via computed-unquote

Test call site updates: :rust::crossbeam_channel::Receiver<wat::core::String> → :wat::core::String. Test green; baseline preserved at 4.

Macro dialect note (Clojure-style)

• ~ = unquote
• ~@ = unquote-splicing
• , = whitespace literal (commas are visual separator only, like Clojure)

Some substrate docs use classical Clojure , notation when DESCRIBING quasiquote semantics. The actual wat source uses ~.

D2/D3 unblocked

With arc 199 rejected and D1 on the clean shape, D2 (multi-factory) and D3 (panic cascade) build on the cleaner call form directly. Stone E (run-processes) similarly unblocks.

Lesson captured

Before opening a substrate arc, investigate existing substrate machinery for the pattern in question. Arc 199’s DESIGN sketch spent cycles on four-questions across three candidates for a non-problem. The fix: grep + read the relevant primitives FIRST.

Discipline anchor: feedback_assertion_demands_evidence — “the substrate is missing X” needs evidence the substrate doesn’t have X. The user’s intuition (“we solved symbols in macros already”) was correct; the orchestrator’s opening of arc 199 was the reflex feedback_no_new_types exists to catch — applied at substrate-arc level, not just at within-arc type/verb level.

Upstream of feedback_no_new_types: don’t open new substrate arcs without proving the existing substrate doesn’t already solve it.

---