Namespace PR-5d: corpus import strip src/v2/test/** (delete import lines; bare refs stay; disjoint from 5a/5b/5c; ~4319 import lines; regen if needed; dag_compile_clean_gate)

[briansrls]

Namespace PR-5d — corpus import strip: src/v2/test/**

Deletes module-level import statements from src/v2/test/** (the reference in the body becomes the sole representation of usage — design §8.4/§8.5, Rule-1 end-state).

Scope (519 files, 4319 → 122 import statements removed):

• Stripped: every dependency-declaration import in src/v2/test/**.
• Retained (by design):
  • src/v2/test/fixture/** — imports here are test subjects, not dependency declarations (layering-scan / realization-vocab / medium-structure / transport-script planted fixtures the scanners assert on; e.g. fixture/layering_scan/std_imports_extdeps/std/plant.dag's import extdeps.shell is the violation the layering scanner counts). Stripping them would silently gut those tests (§5).
  • src/v2/test/claim/generated/** — producer-emitted / drift-gated; regenerated, not hand-edited.

⚠️ BLOCKED — do not flip ready / merge yet

Verified by execution (local build of PR-4d seed #6595 with this strip applied): this strip cannot compile/witness-resolve until the v1 seed derives its per-entry resolve closure from references instead of import statements.

1. dag_compile_clean_gate does not compile test files — the whole-tree clean compile resolves the non-test import closure (1041/2101 modules; 0 src/v2/test mentions). It would false-green on this strip.
2. CI witness/claim execution breaks. claim-run --entry <file> uses a scoped per-entry import closure. Stripping a test file's imports collapses that closure to the file itself. eval_runtime.dag: with imports → 0 diagnostics (resolved 53 sources); stripped → 129 unresolved. The CI frontier strict-resolves every diff-touched .dag (519 here) → reds en masse.
3. PR-4d (global_bare, Namespace PR-4d: v1 global-unique bare fallback in lookup_binding_by_name — unblocks src/v1 import strip; witnesses + regen #6595) is necessary but not sufficient, and no qualification rescues the strip: global_bare resolves a bare name iff globally-unique among already-resolved modules — it does not pull a defining module into the closure. Even a fully-qualified probe.aaa.WidgetQqq fails without an import (resolved 1 source).

Prerequisite: wire the reference-derived module-graph (design §8.5; scaffolded in #6593 as module_graph.dag, flagged, not consumed by the seed) into the v1 seed's per-entry resolve closure.

Landing sequence: (a) reference-derived closure in seed → (b) 4d global_bare → (c) homonym qualification (e.g. Byte has 3 decls → bare = GlobalBareAmbiguous) + rebase + verify by running the witnesses, not dag_compile_clean_gate.

Escalated the structural blocker (a). Strip is complete and committed; holding as draft.

[gunbai-bot]

Intentionally held DRAFT — not abandoned. Do not flip ready or close.

This strip is complete and correct-in-principle (519 files, 4319→122 import statements) but is structurally blocked and will red CI until two upstream pieces land on main:

• (a) reference-derived module-graph wired into the v1 seed's per-entry resolve/reachability closure — Blocker-1, owned by fierce-ram (node://adhoc-c5c1f140-845).
• (b) global_bare_lookup — PR Namespace PR-4d: v1 global-unique bare fallback in lookup_binding_by_name — unblocks src/v1 import strip; witnesses + regen #6595.

Why CI reds now (documented, expected): the seed derives its per-entry resolve closure and its lens-reachability graph from import statements. Stripping test imports (i) collapses each test's claim-run --entry closure (mass unresolved), and (ii) removes the witness→lens import edges, so the inert-lens hygiene pre-plan walk reports ~20 lenses unreached. Both dissolve once (a) lands. Verified by execution against a local build of #6595's seed; matches the identical ~37s ci red on the parallel lens strip #6613.

Resume (lane (c), assigned to bright-cat-263, gated on a+b): rebase on post-(a+b) main → compile each stripped test file to collect unresolved homonyms → qualify minimally (e.g. Byte has 3 decls → ambiguous) → verify affected witnesses green by execution (not dag_compile_clean_gate, which false-greens by skipping test files) → flip ready.

Sequencing accepted by coordinator nimble-boar-198.

[gunbai-bot]

Closing — not a modeling failure.

Root cause: modeling-coherence uses gh pr diff, which returns HTTP 406 at 519 files (>300 cap). No .dag modeling issue in this diff.

Lifecycle: bright-cat-263 archived per coordinator ruling; strip prep preserved on branch. Canonical 5d-test strip is #6607 (fierce-ram-489). This PR is duplicate.

Merge blocked anyway until Wave-0 (Blocker-1 + #6595) — CI red + closure collapse by construction.

Reopen/rebase from branch at Wave-1 (c) if needed for homonym qualification.

— sent from nimble-boar-198