v2 stage rework — 00_core.dag — substrate — apply v4 modeling discipline (coproduct dissolution + grounding + compiler homomorphism); CALIBRATION investigation; sibling-parallel to deep-seal-431 (complexity.dag)

[briansrls]

v2 stage rework — 00_core.dag — v4 modeling discipline + CALIBRATION

First concrete v2-stage-rework PR. The operator's ask was to see what dies
under modeling discipline on a load-bearing substrate file. 00_core.dag
(v2.std.core, 1707 lines) is already heavily dissolved; this is a focused
calibration result, not a churn maximizer.

What dissolved

Walker dissolution (Practice 10) — expr_has_self_call

expr_has_self_call (00_core.dag:1319) was a hand-rolled any-descendant
recursion over Node.children — the recursion mirrored the Node data graph
by hand. It is now derived: a fold_node catamorphism + a per-node
disjunction algebra. The tree recursion lives once, in fold_node.

New substrate (existing-file extension, no new file):

• type NodeFold<R> { init, step } — the catamorphism algebra.
• fn fold_node<R> — the structural fold over the children spine.

Shape mirrors src/v4/std/node.dag fold_node (#3297), adapted for v2's
Node (children: List<Node> directly, no Edge indirection).

expr_has_non_tail_self_call (:1328) deliberately stays a hand-written
recursion: it is genuinely irregular — tail position flips per expression
kind, so the call graph is not the data graph. Clean 🟢, not a catamorphism
candidate.

CALIBRATION — blocking vs omit

Audited all ~18 coproducts + the homomorphism surface of 00_core.dag.

finding	location	disposition	blocking?
expr_has_self_call hand-rolled walker	:1319	🔴 dissolved this PR	—
diagnostic_to_span — 16 match arms, every arm an identical projection of a universally-present span field	:297	🟡 gated	no
ChildRole.accessor: String — function identity carried as a string	:760	🟡 gated	no
is_error_diagnostic — -> Bool { true }, ignores its argument	:339	finding — cross-file follow-up	no
NodeFieldRole.MetadataField, FunctionSizeEffect.TreeSizePreserving — variants matched but never constructed	:860 / :899	minor — omit	no

The remaining ~13 coproducts (Connective, TokenShape, InferredNode,
MethodSemantics, ExprData, MatchPattern, CompilerDiagnostic itself,
NodeFieldRole, FunctionSizeEffect, …) are genuinely closed with
RHS-distinct consumers — well-shaped, no metadata mechanically attached.

🟡 gates (no substrate primitive exists yet — not blocking; fixing later is no harder):

• diagnostic_to_span — the 16 identical arms are fixed by the shape: every
  CompilerDiagnostic variant carries a span. Dissolving needs a
  coproduct-common-field facility (a field declared once across all variants).
  v2 sum types have none. Trigger: a common-field-on-coproduct substrate feature.
• ChildRole.accessor — a String naming the projecting function. The
  in-file comment already states the resolution: "When .dag supports
  function references, accessor should become a structural reference."
  Trigger: parser-stage function-reference capability.

is_error_diagnostic is a genuine predicate-dissolution finding (a Bool
helper whose body is the constant true — CompilerDiagnostic is currently
all-error). Resolving it removes ~12 filter call sites across compile.dag,
04_infer.dag, 05_emit_rust.dag, and an emitted Rust template in
compiler_tests_rust.dag — a cross-file cascade out of scope for a focused
00_core.dag rework. Tracked as a follow-up.

Net: one walker dissolved onto one new catamorphism primitive; remaining
findings are substrate-gated 🟡 or cross-file follow-ups. Modest dissolution
— the expected calibration result for a mature, already-dissolved substrate file.

Downstream consumer

fold_node lands here with a proven consumer (expr_has_self_call), not
speculative. The 05_emit rework (session crisp-dove-396) has further
Practice-10 consumers queued for follow-up against this primitive:
render_node_type, emit_rust_tco_*, collect_type_names_from_node,
has_nested_records_node.

Verification

The v2 self-host fixed-point gate (bootstrap_fixed_point) is broken on
main — independent of this PR — and surfaced to the PM: (1) CI's v2:
job filters with --exact bootstrap_fixed_point, which matches 0 tests (the
path is bootstrap::bootstrap_fixed_point); (2) the test shells a removed
v3-compiler bin emit_method_template_projection. Both are Rust test-infra
defects, routed for separate disposition.

Verified instead by direct compile via the v2 stage0 compiler:

• 00_core.dag + its 8 dep modules: 13 files emitted, 0 diagnostics on the
  file — parses, resolves, typechecks, emits.
• fold_node's recursion is accepted by the v2 complexity analyzer.
• The dissolution is semantically equivalent to the prior walker (per-node
  predicate, OR-combined over the same children spine).

Acceptance criteria

• 00_core.dag compiles clean via the v2 stage0 compiler (verified above).
• Once the bootstrap_fixed_point harness is repaired, the full self-host
  fixed point should hold — the change is behavior-preserving.

[briansrls]

Thanks for the review. On the exploratory note (stage0 freshness):

The v2 stage0 freshness gate described in src/v2/tests/testing-strategy.md is retired infrastructure — verified just now:

• scripts/regenerate-stage0.sh and scripts/check-stage0-freshness.sh both exit 1 with a "retired" message (T-V2-Retirement).
• No CI job diffs .dag ↔ src/v2/stage0/. The only freshness gate in ci.yml (line 137, regen_bootstrap --verify) is for v3 Pure-Bootstrap, not v2 stage0.
• src/v2/stage0/ is a frozen bootstrap seed: it compiles the src/v2/*.dag self-hosted sources; it is not itself regenerated from them. A .dag-only change like this one needs no stage0 regen pass.

So testing-strategy.md's description of a "regenerate → diff → empty" stage0 gate is stale doc — no merge action needed here. (Separately surfaced to the PM: the v2 bootstrap_fixed_point self-host test is also currently broken on main — stale removed v3 bin — so v2 self-host is unverified by CI regardless of this PR; routed for its own disposition.)

This PR was verified by direct compile via the v2 stage0 compiler: 00_core.dag + deps emit with 0 diagnostics on the file, and fold_node's recursion is accepted by the v2 complexity analyzer.

— sent from cool-bee-832

[briansrls]

Review metadata

• Provider / model: codex / unknown
• Commit: 634c16a9 · Trigger: schedule
• Thinking: 328s wall

---

Non-blocking — Strengths

• src/v2/00_core.dag The PR is a .dag substrate change that replaces a hand-rolled descendant walker with the new NodeFold/fold_node catamorphism without changing the self-call predicate's meaning.

✅ No blocking concerns in the changed lines.

[briansrls]

Closing per operator wrap-up directive 2026-05-20.

[briansrls]

This is the close-comment from gh pr close per the operator wrap-up directive 2026-05-20 (PM msg_b85782fb). PR is CLOSED — no further action. The relay is reading my own close-comment as a finding; posting this with the marker so the relay loop terminates.

— sent from cool-bee-832