Print the exact closure behind each routed-entry resolve

[gunbai-bot]

Outcome

claim_batch::resolve_timed now prints one identity-grain row for every module in the resolved closure, directly after the existing aggregate [resolve] line. Each row carries the routed entry, module identity, and repository-relative source path in stable order.

This is a projection of ResolvedGraph.modules, the exact population the resolver already returned for execution. It does not scan imports, resolve again, or infer membership from item_registry. The existing claim_batch route consumes the projection on every entry resolve.

Pre-landing use

A lane can answer “is path P in routed entry E?” without running the long wet dispatch. Use the existing short resolve probe:

target/release/claim_batch \
  --source-root dag --source-root src/v2 \
  --entry E --function <a-real-function-in-E> --eval-budget-ms 1

Resolution completes and prints membership before the deliberately bounded evaluation. The 2540ms figure characterizes the predecessor resolve path, without this membership projection; it is not a measurement of the successor shipped here. Specifically, PR #9725 reports the existing [resolve] producer at 2540ms for dag/test/claim/self_host_00_compile_behavioral_witness_test.dag (79 modules, 2220 resolved items). This PR’s targeted test executes the membership projection over a controlled two-module resolved graph; it does not re-measure that production entry. The output is the instrument; this PR does not transcribe today’s members into a document.

The projection is outside the charged per-claim window: run_witnesses calls resolve_timed once for an EntryGroup, constructs one evaluation context, and only then loops over that group’s functions. Each iteration enters run_claim_measured, where the CPU clock and 500ms evaluation deadline begin. Closure sorting and output therefore add per-entry route overhead and log volume, but neither execute once per claim nor contribute to the CPU quantity enforced by the per-claim line. This is not a fixed-volume producer: it emits one row per resolved module per entry group, so a full floor log grows as entry groups × their closure sizes. That work remains visible in preparation and whole-run cost even though it is outside the per-claim charge.

The semantic subject digest currently hashes raw file content. Consequently any byte change in a printed member path, including annotation-only prose, moves that digest; this PR reports membership and does not claim otherwise.

Seed boundary

The sole new hand-authored declaration is v1_compiler.cli_run resolved_closure_members. gunbc.routed_entry_closure_membership_seed_growth routed_entry_closure_membership_seed_growth_justification enrolls it at declaration grain, explains why the seed host is still required before a .dag evaluation frame exists, names v1-hand-queue-drain, and dissolves when the modeled routed-entry execution path provides this same membership before dispatch.

The first revision exposed the existing private workspace_relative_repo_path; that public-surface expansion is removed. The private helper remains unchanged and the one justified projection calls it inside cli_run.

Scope split

The requested open-PR writer-set producer is deliberately separate. It depends on live forge state and does not belong on the compiler resolver path; dashboard node adhoc-c9d9cda1-caf owns that half.

Verification

• cargo fmt --all --check
• cargo test --release -p v1-compiler --bin claim_batch resolved_closure_membership_tests -- --nocapture (1 passed on current head)
• compiler-route control: cargo check -Z definitely-not-a-real-flag failed on stable Cargo as required, confirming the remote runner executed the command

A direct run with the locally installed gunbc was not treated as evidence: that binary predates the current grammar and refused the current corpus broadly.

[gunbai-bot]

Addressed review 59395 in 4cbfc75. The existing path relativizer is private again, eliminating that public-surface modification. The remaining cross-crate projection is one new declaration, v1_compiler.cli_run resolved_closure_members, now enrolled by gunbc.routed_entry_closure_membership_seed_growth and included in the closed seed-growth justification roster with reason, owning lane, capability trigger, and current boundary. The production consumer remains claim_batch::resolve_timed; the test continues to execute the resolver-result projection. — sent from sharp-koi-752