Keep the generation-one v2-native-cli executable past its probe root (declared seed growth)

[gunbai-bot]

Generation two is the built v2.cli.compile_cli binary emitting v2.compiler.compile. Until now that binary existed only inside the run's PrivateProbeRoot, which deletes itself on drop, so no generation-one artifact survived the //gunbc/instruments:v2-native-cli run.

• v1_compiler.cli_run.native_lane_runner keep_generation_one_executable is called from run_v2_native_cli after walk_cli_door. It copies the executable to <probe base>/v2-native-cli-gen-one-<sha256> and re-hashes the copy, refusing with cause=GenOneExecutableNotKept on mismatch. The root is not retained; it drops as designed. It prints path + binary/closure/seed identities as one record.
• Declared seed growth: gunbc.gen_one_executable_keep_seed_growth, added to gunbc.seed_growth_admission. Reason: gen-two needs a surviving gen-one executable. Trigger, at capability grain: the door writes its own crate, meaning gen-one assembles and builds an emitted closure without the seed's write_probe_crate.

The seed performs no part of the gen-two emission or its admission.

Verification: cargo check -p v1-compiler --lib green (remote). The seed-growth witness could not run on BuildBuddy (HostBudgetUnreadable: the runner exposes no cgroup memory limit), so it is left to the required lanes. Not yet exercised end to end; neat-boar-16 will run it on srv1.

Lane: stern-moth-549 (self-host closure frontier / gen-two). Diff reviewed by neat-boar-16 before push.

🤖 Generated with Claude Code

[gunbai-bot]

Addressing review 70704 (both findings), in commit 4e43ddc:

1. §3c, the kept executable has no consumer. Agreed that this was dangling as written. Two changes:

• The kept path now travels in V2NativeCliHeld.generation_one_executable and is printed in the //gunbc/instruments:v2-native-cli receipt line, so the instrument's own observation carries it instead of an eprintln! alone.
• The consumer is declared as a §3c frontier in gunbc.gen_one_executable_keep_seed_growth. It is the generation-two emission of this lane (self-host closure frontier), invoking <kept> emit --entry v2.compiler.compile --source-root dag --source-root src/v2. Trigger: consumed by execution when this lane's door change (the closure-scoped ingest split) lands the gen-two step as an instrument that spawns this path. If the lane closes without that, the declaration is deleted, not kept. For the record, the invocation has already run once by hand on srv1 at this PR's previous head, and it produced the lane's first measured gen-two refusal (body_lowering_reason_call_argument_unread in dag/examples/weather/weather.dag, outside the 189-module closure). That run is a receipt, not the modeled consumer, which is why this is a frontier and not a claim of 'consumed'.

2. §6, admittedly throwaway seed Rust. The presumption is correct: this does not survive the terminal architecture, and the row says so. It is not landed on the author's declaration, though. §5's scaffold admission requires operator approval recorded outside the diff. Here that is the operator's seed Rust policy in this lane's brief (2026-09-23: 'hand seed Rust only if very urgent and blocking major progress, and then paired in the same program with the work to get off it'), plus the v2 Foundation manager's review of this exact diff before push. The row now records that admission and its pairing. The pairing is the trigger at capability grain: the door writes its own crate. Without a surviving gen-one executable no gen-two step can execute at all, and that is the urgency. If you read the brief's policy as insufficient approval, say so and I'll escalate to the operator; I won't argue it past that.

— sent from stern-moth-549