Dead witness tree: 17 .dag witnesses under src/v1 have never executed — triage each, migrate emitter-behaviour ones onto compile_dag_rust_emit_check, retire the rest with receipts

[briansrls]

Auto-opened by session-dashboard for session zesty-newt-828.
Pushing to session/zesty-newt-828 advances this PR.

Worker attestation

• Title describes the change (not the session id or branch).
• PR body summarises what and why (replace the TODO below).
• Tests run: name the command (e.g. npm test, cargo test) and the result.
• If this closes a work item, the body contains a Closes #N directive.
• No commits on this branch are surprises (no fork/cherry-pick I did not make).
• No secrets / credentials / large binaries staged.

Summary

Triage of the 17 dead .dag witnesses under src/v1/tests/claim (dashboard node adhoc-9b80ec49-d63): each witness's only historical consumer, v1_claim_scoped_witness_batch, was deleted 2026-08-15 (gunbc#8146), leaving these files unenrolled and never executed against current v1 behavior — in some cases for years. Emitter-behaviour witnesses migrate onto compile_dag_rust_emit_check (per-PR discovery, no v1 import needed); the rest retire with receipts. RESTORE (widening the required floor's source roots to --source-root src/v1) is ruled out repo-wide — see docs/plans/restore-src-v1-required-floor-stall-finding.md — on a stall signature (RSS plateau, zero phase progress) independently corroborated by dedicated-hardware CI numbers showing the unwidened floor already runs close to whatever ceiling caused the local stall.

Two language facts, discovered by execution, not previously established anywhere in the repo:

1. v1's |> pipe operator resolves only against registered builtin methods, never an arbitrary same-named free function. 1 |> demo (piping into fn demo(x: Int) -> Int) does not compile: |> desugars via UFCS method-call resolution (v1_compiler_infer_method), which matches only the builtin registry (count/filter/map/...) — never a free function of the same name, however plausible that reads to an author. Executed diagnostic: method 'demo' not found on receiver type 'Primitive(Int)'. This is plausibly the same spelling-keyed-resolution class behind several other defects this repo has hit this week — a method resolved by name against a builtin registry, with no path to a resolved declaration — noted here as a pointer for whoever owns that class, not a claim that it's the same root.
2. v1's for-expression is not a list comprehension. Its type is its body's tail-expression type directly — never wrapped in the container implied by iterating a collection. fn demo() -> List<String> { for ch in "abc" { ch } } (body type String) fails to typecheck against a List<String> declared return. Executed diagnostic: type mismatch: expected 'Container(List,Primitive(String))', got 'Primitive(String)' — reproduced independently with a List<Int> collection and an Int-typed body.

Both were believed otherwise by the original (dead-tree) fixture authors, who wrote assertions encoding these false premises and never executed them — the dead tree was not merely hiding failing claims, it was preserving false beliefs about the language in a form that looked like verification. This "dead tree preserved false beliefs" claim rests on these two language facts only — the four mechanical fixes below are not evidence for it, since they never encoded a false premise at all.

Separately, four more required_expr_newline_continuation_witness_test.dag assertions failed for a mechanical reason, not a language fact: all 8 fixtures in that migrated file declared the identical module test name, and compile_dag_rust_emit_check resolves per-invocation against a shared corpus — the collision broke compile-cleanliness for every fixture using that name. Fix: give each fixture its own unique module name (untyped_let_newline, typed_let_newline, return_newline, for_newline, let_statement_boundary, plus the three _rbrace/_missing_collection refusal-arm siblings). This is a translation bug in the migrated harness, not a language finding — the underlying continuation-parsing productions under test were never in question. One edit (module-name uniqueness) resolved all four at once, so these were never four independent failures — a single root cause, not four data points.

Third finding — a refusal witness must assert the reason, not merely the refusal. The 3 refuses_to_compile fixtures in the same file (see mapping below) also shared module test and also failed to compile-clean under the collision — and since their assertion is !compile_clean, that accidental failure already satisfied the assertion. They were green, counted, and meaningless: passing because of the harness bug, not because of the parse refusal they exist to prove — indistinguishable, from the assertion alone, from "refused for the reason I claim" vs. "refused for a reason I introduced by accident." This generalizes beyond this file: any !compile_clean-shaped witness in the corpus is satisfiable by an unrelated break (typo, missing import, renamed callee, collision) as easily as by the construct it names. Checked cheaply after the rename — compiled each of the 3 fixtures in isolation (own source root, no other corpus file present) and captured the actual diagnostic:

• w_untyped_let_newline_then_rbrace_refuses_test (let x =\n}\n): expected expression, found RBrace — names the missing required initializer expression after let x =. Correct cause.
• w_return_newline_then_rbrace_refuses_test (return\n}\n): expected expression, found RBrace — missing required expression after return. Correct cause.
• w_for_newline_then_block_without_collection_refuses_test (for x in\n { x }\n): expected LBrace, found RBrace — the missing collection expression before the loop body's { is exactly the shape under test. Correct cause.

All three now refuse for the reason they were written to prove, not for the module-name collision — converted from assumed-correct to verified: 3 false greens become 3 real ones.

Six-way mapping (all originally-failing assertions accounted for):

Assertion	File	Root cause	Fix
w_untyped_let_newline_before_initializer_test	required_expr	module-name collision (module test shared by all 8 fixtures)	unique module name untyped_let_newline
w_typed_let_newline_before_initializer_test	required_expr	module-name collision	unique module name typed_let_newline
w_return_newline_before_expression_test	required_expr	module-name collision	unique module name return_newline
w_statement_boundary_after_complete_initializer_test	required_expr	module-name collision	unique module name let_statement_boundary
w_for_newline_before_collection_test	required_expr	module-name collision + false language premise (for-as-list-comprehension)	unique module name for_newline + return type List<String> → String
w_pipe_newline_continuation_test	dag_parse_continuation	false language premise (UFCS pipe resolves against arbitrary free functions)	fixture receiver changed to a builtin (`[1, 2, 3]

All 9 affected fixtures verified locally via direct gunbc compile against their exact literal source strings, matching each fixture's expected parses-clean/refuses-to-compile polarity, before the fix commit.

Test plan

• claim_executor --required-floor --source-root dag --source-root src/v2 via CI run 32207545518 (fix commit 7975d5bfa8): planned=9035 executed=9035 terminal=9035 passed=8729 known_red_held=306 failed=0 stale_quarantine=0 budget_refused=0 host_tool_unresolved=0 — fully green.