ci(profile): stop regenerating committed schedules

[andrii-lz]

Summary

Stop regenerating tracked schedule tables twice in every profile-benchmark matrix job. The benchmark now consumes the tables committed at each revision; the dedicated schedule-drift CI guard remains the source of truth.

Why

In the fresh slow CI job, the visible schedule-generation step took 1m43s. The merge-base build then ran the same generator again inside its worktree before compiling its profile binary.

Expected improvement

• Removes two generator invocations per matrix cell.
• Expected net saving: roughly 2–3 minutes per cell after accounting for dependencies that the following profile build must compile itself.
• No benchmark/runtime behavior change; both revisions still use their own committed generated tables.

Validation

• git diff --check

Stack

1. This PR: remove redundant schedule regeneration.
2. Add narrow profile feature groups.
3. Switch each benchmark matrix group to its narrow feature.

[github-actions]

Documentation blast radius (advisory)

These regions may need doc/spec/book updates based on changed paths.
This is not a merge gate. See docs/documentation.md.

Changed files in this PR: 1

ci-tooling

CI workflows and repo scripts

Code paths touched:

• .github/workflows/profile-bench.yml

Consider updating:

• docs/ci-test-timing.md
• docs/documentation.md
• specs/ci-test-timing.md

---

Per-PR checklist: spec Status / acceptance criteria; book owning page; AGENTS.md if contracts changed; archive spec after fold.

[github-actions]

CI test timing

• Report generated: 2026-08-07T00:21:53Z.
• Source: 716c1cd on codex/profile-bench-skip-regeneration.
• Workflow run: 31133541320.
• Main baseline: c9ca8e9.

Run summary

Wall s	Main wall s	Main Δ	Ratio	Tests	Skipped	Failed	Status
320.0	302.0	+6.0%	1.06x	1278	0	0	ok

Wall time spans 2 parallel nextest slice shards.

Slowest tests

Rank	Duration s	Test
1	12.0	akita-planner::schedule_params::tests::pruned_mixed_search_matches_unpruned_traversal_and_is_canonical
2	9.2	akita-pcs::akita_e2e::dense_d64_snap_regen_prove_verify_nv24
3	7.3	akita-planner::schedule_params::tests::mixed_nv36_benchmark_policy_selects_minimum_setup_schedule
4	6.5	akita-planner::schedule_params::tests::uniform_suffix_dp_matches_unpruned_exact_cutover_search
5	5.7	akita-sis-estimator::search_mode_parity::parallel_exhaustive_matches_serial_exhaustive_smoke
6	5.4	akita-prover::kernels::linear::tests::chunking::q128_many_blocks_digits_chunk_instead_of_unsafe_block_parallel
7	4.9	akita-pcs::single_poly_e2e::single_dense_nv18
8	4.4	akita-pcs::setup::d64_dense::large_setup_batch_passes
9	4.3	akita-sis-estimator::search_mode_parity::exhaustive_search_is_at_least_as_good_as_local_minimum_smoke
10	3.3	akita-pcs::setup::d128_dense::large_setup_nv_passes
11	3.1	akita-pcs::setup::d64_dense::same_size_passes
12	2.9	akita-prover::protocol::sumcheck::relation_range_image::tests::stage2_large_odd_dense_prefix_matches_padded_reference
13	2.8	akita-prover::protocol::sumcheck::relation_range_image::tests::stage2_large_odd_sparse_boolean_prefix_matches_padded_reference
14	2.8	akita-pcs::batched_aggregated_e2e::aggregated_mixed_dense_and_onehot_under_dense_cfg
15	2.7	akita-pcs::scheme::tests::single::fp128_degree_one_batched_proof_roundtrip_is_stable
16	2.7	akita-pcs::batched_aggregated_e2e::non_zk_aggregated_cases::aggregated_dense_nv17_batch4
17	2.7	akita-pcs::scheme::tests::onehot::multi_group_root_allows_precommitted_arity_above_final_group
18	2.6	akita-pcs::setup::d128_dense::large_setup_batch_passes
19	2.4	akita-pcs::scheme::tests::single::verify_rejects_malformed_v_dimension_without_panicking
20	2.3	akita-pcs::setup::d64_dense::large_setup_nv_passes

Regressions vs main

No per-test regressions above the threshold.

New slow tests

No new tests ≥30s vs main baseline.

[github-actions]

Benchmark Report

• Latest run: 694d3fa
• Message: ci: retrigger checks after GitHub outage
• Ref: codex/profile-bench-skip-regeneration
• Workflow run: run 31133541339 attempt 1
• Report generated: 2026-08-07T00:24:12Z.
• Main baseline: 0f49d0a from the merge-base benchmarked on this runner.
• Binary: target/release/examples/profile.
• Memory: maximum resident set size from /usr/bin/time on the benchmark process.

Status	Workload	Setup contribution	Setup and preparation	Setup vector size	Prepared NTT cache size	Verifier NTT cache size	Commit	Prove	Verify	Peak process RSS	Proof size
ok	Fp32 - nv28Onehot256 - D=128	direct	0.043 s +4.58% vs main	36.0 MiB +0.00% vs main	144.0 MiB +0.00% vs main	1.2 MiB +0.00% vs main	0.106 s +14.42% vs main	1.532 s -0.22% vs main	33.7 ms +1.23% vs main	460.8 MiB +0.04% vs main	77,834 bytes +0.00% vs main
ok	Fp64 - nv28Onehot256 - D=128	direct	0.036 s -3.03% vs main	40.0 MiB +0.00% vs main	120.0 MiB +0.00% vs main	0.9 MiB +0.00% vs main	0.071 s -2.40% vs main	1.087 s +0.27% vs main	32.1 ms +1.10% vs main	562.2 MiB +0.54% vs main	83,596 bytes +0.00% vs main
ok	Fp128 - nv24Dense - D=64	direct	0.142 s -0.49% vs main	215.0 MiB +0.00% vs main	537.5 MiB +0.00% vs main	1.4 MiB +0.00% vs main	1.770 s +3.29% vs main	1.207 s -0.60% vs main	16.3 ms +1.81% vs main	1572.4 MiB +0.32% vs main	83,562 bytes +0.00% vs main
ok	Fp128 - nv26Onehot256 - D=64 - Tensor	direct	0.658 s +0.89% vs main	1024.0 MiB +0.00% vs main	2560.0 MiB +0.00% vs main	1.4 MiB +0.00% vs main	0.117 s +1.89% vs main	1.255 s +0.12% vs main	33.6 ms -1.51% vs main	4136.3 MiB -0.24% vs main	85,080 bytes +0.00% vs main
ok	Fp128 - nv32Onehot256 - D=64	direct	0.206 s -0.20% vs main	320.0 MiB +0.00% vs main	800.0 MiB +0.00% vs main	1.4 MiB +0.00% vs main	1.236 s +0.79% vs main	1.190 s +0.65% vs main	23.3 ms -0.34% vs main	1704.8 MiB +0.24% vs main	84,972 bytes +0.00% vs main
ok	Fp128 - nv32Onehot256 - D_a=256D_b=128D_d=128 - MixedD256ToD64	direct	0.139 s -1.51% vs main	128.0 MiB +0.00% vs main	608.5 MiB +0.00% vs main	1.4 MiB +0.00% vs main	0.978 s +0.09% vs main	1.316 s -0.72% vs main	21.0 ms +8.10% vs main	1410.0 MiB -0.11% vs main	85,080 bytes +0.00% vs main
ok	Fp128 - nv30Onehot256 - Batched4 - D=64	direct	0.207 s +0.12% vs main	320.0 MiB +0.00% vs main	800.0 MiB +0.00% vs main	1.4 MiB +0.00% vs main	1.146 s -1.24% vs main	1.194 s -0.39% vs main	24.4 ms -0.30% vs main	1701.3 MiB +0.12% vs main	84,970 bytes +0.00% vs main
ok	Fp128 - nv32Onehot256 - Batched4 - D=64 - MultiGroup	direct	0.427 s -0.20% vs main	1032.0 MiB +0.00% vs main	1075.0 MiB +0.00% vs main	1.4 MiB +0.00% vs main	3.215 s +1.16% vs main	1.645 s +1.19% vs main	29.8 ms +5.48% vs main	3030.6 MiB +0.09% vs main	85,093 bytes +0.00% vs main
ok	Fp128 - nv32Onehot256 - Batched4 - D=64 - MultiGroup	recursive	3.903 s -0.21% vs main	1032.0 MiB +0.00% vs main	1075.0 MiB +0.00% vs main	1.4 MiB +0.00% vs main	3.169 s +0.57% vs main	3.450 s -0.34% vs main	26.1 ms -4.94% vs main	3788.3 MiB +0.31% vs main	89,873 bytes +0.00% vs main
ok	Fp128 - nv32Onehot256 - Batched4 - D=64 - MultiGroupW8R2	recursive	1.889 s -0.04% vs main	430.0 MiB +0.00% vs main	1075.0 MiB +0.00% vs main	1.4 MiB +0.00% vs main	3.522 s +0.43% vs main	10.830 s +0.19% vs main	37.9 ms -5.50% vs main	3579.5 MiB +0.14% vs main	95,904 bytes +0.00% vs main
ok	Fp128 - nv32Onehot256 - D=64 - MultiChunkW2R2	direct	0.207 s -0.45% vs main	320.0 MiB +0.00% vs main	800.0 MiB +0.00% vs main	1.4 MiB +0.00% vs main	1.254 s +1.89% vs main	1.556 s -0.06% vs main	24.7 ms +1.18% vs main	1846.1 MiB +0.06% vs main	85,331 bytes +0.00% vs main
ok	Fp128 - nv32Onehot256 - D=64 - MultiChunkW4R2	direct	0.223 s -0.40% vs main	344.0 MiB +0.00% vs main	860.0 MiB +0.00% vs main	1.4 MiB +0.00% vs main	0.428 s -0.75% vs main	1.820 s +0.07% vs main	28.2 ms +2.66% vs main	2024.0 MiB +0.21% vs main	85,946 bytes +0.00% vs main
ok	Fp128 - nv32Onehot256 - D=64 - MultiChunkW8R2	direct	0.223 s +0.70% vs main	344.0 MiB +0.00% vs main	860.0 MiB +0.00% vs main	1.4 MiB +0.00% vs main	0.428 s +0.17% vs main	2.584 s +0.54% vs main	30.3 ms -1.80% vs main	2266.0 MiB +0.09% vs main	86,215 bytes +0.00% vs main

Negative deltas are improvements for time, memory, and proof size.

Terminal response component breakdown

Workload	Folded response (z)	Opening values (e)	Inner-commitment values (t)	Total terminal response
Fp32 - nv28Onehot256 - D=128	21,406 bytes	3,072 bytes	24,576 bytes	49,054 bytes
Fp64 - nv28Onehot256 - D=128	21,460 bytes	7,168 bytes	28,672 bytes	57,300 bytes
Fp128 - nv24Dense - D=64	21,854 bytes	7,168 bytes	28,672 bytes	57,694 bytes
Fp128 - nv26Onehot256 - D=64 - Tensor	21,848 bytes	7,168 bytes	28,672 bytes	57,688 bytes
Fp128 - nv32Onehot256 - D=64	21,852 bytes	7,168 bytes	28,672 bytes	57,692 bytes
Fp128 - nv32Onehot256 - D_a=256D_b=128D_d=128 - MixedD256ToD64	21,848 bytes	7,168 bytes	28,672 bytes	57,688 bytes
Fp128 - nv30Onehot256 - Batched4 - D=64	21,850 bytes	7,168 bytes	28,672 bytes	57,690 bytes
Fp128 - nv32Onehot256 - Batched4 - D=64 - MultiGroup	21,861 bytes	7,168 bytes	28,672 bytes	57,701 bytes
Fp128 - nv32Onehot256 - Batched4 - D=64 - MultiGroup	21,825 bytes	7,168 bytes	28,672 bytes	57,665 bytes
Fp128 - nv32Onehot256 - Batched4 - D=64 - MultiGroupW8R2	21,868 bytes	7,168 bytes	28,672 bytes	57,708 bytes
Fp128 - nv32Onehot256 - D=64 - MultiChunkW2R2	21,875 bytes	7,168 bytes	28,672 bytes	57,715 bytes
Fp128 - nv32Onehot256 - D=64 - MultiChunkW4R2	21,850 bytes	7,168 bytes	28,672 bytes	57,690 bytes
Fp128 - nv32Onehot256 - D=64 - MultiChunkW8R2	21,831 bytes	7,168 bytes	28,672 bytes	57,671 bytes

The z column includes its per-segment length prefixes and Golomb payload; e and t are raw field bytes. These three columns sum exactly to the serialized terminal response.

Detailed schedule and proof-size breakdowns by fold level are available in the uploaded report.md benchmark artifact.