Production selector cutover into the native execution kind (post-#7599 slice): bounded production witness population -> content-addressed native bundle/shards on the real CI floor, full transition receipt

[briansrls]

Summary

Cut the bounded production-selected logic population (meet, join, complement) onto a new NativeBundleWitnessKind in the real per-PR floor. The .dag selector produces one content-addressed Rust bundle/process specification; claim_executor preserves the kind, starts one native bundle process, calls all three witnesses directly, runs the interpreter as the timed equivalence oracle, executes a separately addressed planted-RED bundle, and writes the full transition receipt.

Unknown, absent, or unhandled witness kinds refuse with a located counted diagnostic. Native artifact/build refusal can fall back only inside the native handler, where the fallback and unavailable population are typed and counted; native/interpreter divergence and missing planted-RED equivalence stay hard refusals.

Why

gunbc_pr_native_batch previously projected an empty routing frontier, and claim_executor discarded ScheduleWitnessEntry.kind, so a native row would have been absorbed into ordinary interpretation. This change makes the production selector consume the native execution kind introduced by the post-#7599 bundle substrate without changing nodes, dependencies, batch ordering machinery, or on-success stages.

Enrollment activates the existing conditional native-batch append, increasing the ordinary floor roster from five batches to six. Per the Dispatch A→C ruling, that roster change requires exactly one companion row in gunbc_ci_floor_batch_clamp_params; the native runnable remains isolated rather than being co-located with an unrelated realization batch.

Bounded slice and scaling trigger

Slice 1 is one stable shard containing the three logic members. Move to the next stable shard when either the generated closure exceeds 32 selected members or measured cold compile / peak cgroup memory exceeds 2x this slice's baseline. That exposes code-generation/closure scaling before attempting an immature universal bundle.

Transition receipt

Local run against the current CI-floor executor and real host transport:

• selected/native/interpreted/unavailable: 3 / 3 / 0 / 0
• bundle/shard: 1 / 1
• cold compile wall: 367,209,379 ns
• warm artifact-hit wall: 8,040 ns
• native execution wall: 39,332,338 ns
• interpreter-oracle wall: 17,407,839 ns
• fallback count: 0
• RSS peak: 4,801,241,088 bytes
• cgroup peak: 11,940,978,688 bytes
• verdict: accepted
• planted-red equivalence: true
• bundle identity: 09d7d2e53554c783

The handler writes the same mandatory fields to target/native-selected-witness-transition-receipt.tsv on the floor. The sixth batch's measured whole-batch wall was 77.020s; its 125s + 0ms/unit clamp applies the family 1.6× fleet-spread factor and rounds upward. The runnable stays Substantial and isolated; the measured RSS/cgroup peaks and their 7.14 GB gap are recorded in the clamp basis note, while the memory governor remains the fail-closed capacity authority.

#7522 overlap audit

Re-diffed immediately before the final clamp push against draft #7522 exact head 2778125eaee914b495de86546b61d5aa7ea610a6.

• Shared files: src/v1/stage0/src/bin/claim_executor.rs, dag/gunbc/ci_spec.dag, dag/std/realization_schedule.dag, and src/v2/workflow/ci_floor_plan.dag.
• Function/declaration-level overlap: none. This PR changes runnable_from_value, group_batch_units, run_batch_unit, native-handler helpers, entry_kind_is_execution, gunbc_pr_native_*, and gunbc_ci_floor_batch_clamp_params; Run merge admission as ordered CI success stages #7522 changes the WalkPlan/pre-walk/population-budget/stage scheduling surface (including batch_unit_lane and run_stage) and does not edit the clamp table.

Checks

• cargo check -j 1 -p v1-compiler --bin claim_executor — pass
• cargo test -j 1 -p v1-compiler --bin claim_executor --no-run — pass
• cargo test -j 1 -p v1-compiler --bin claim_executor native_bundle_fallback_requires_planted_red_equivalence — pass, 1/1
• target/debug/regen_stage0 --verify — pass, regen_divergence_count=0
• target/debug/claim_batch --source-root dag --source-root src/v2 --entry src/v2/test/claim/pr_native_batch_test.dag --functions witness_selected_native_bundle_enrolled,witness_floor_appends_one_selected_native_batch — pass, 2/2
• Direct clamp predicates witness_floor_batch_clamp_params_cover_schedule, witness_floor_batch_clamp_overhead_all_positive, and witness_floor_batch_clamp_rate_all_nonneg — pass, 3/3
• Native transition receipt run through a temporary standard WalkPlan<NoWalkFinalization> containing only gunbc_pr_native_batch — pass; temporary harness removed after measurement

The two WIP-check failures at 32d7d2a were exhaustive-match fallout from adding the execution kind and are fixed. The later b47cd5d floor refusal correctly detected that the newly enrolled sixth batch lacked its positional clamp companion; d3d74a2 adds the measured sixth row and basis note.

[cursor]

Bugbot is not enabled for your account, so this pull request was not reviewed.

Enable Bugbot in the Cursor dashboard to get automatic reviews on future PRs.