claude
4474814fa7
Architect drift review (cycle a3785a6..b3b7115) found the cycle's raw switch incomplete on two high items; both fixed forward here rather than minuted as residue, since #319 builds on this namespace: - The real/campaign executor still minted WRAPPED manifest.params and manifest.defaults (AC2 overclaim in the feat commit body). The mint seams now reshape to raw on every route (member.rs raw_bound_defaults, runner.rs manifest_space mirror of the family.rs pattern); the fix surfaced a latent regression in persist_campaign_traces, which keyed point_from_params by exact wrapped name and is now translation-aware (same recipe as reproduce_family_in). Old on-disk families stay replayable (C29; reproduce reads both shapes). - The walkforward --axis intake silently accepted wrapped names — the silent alias fork 2 rejected. It now runs the same refuse_wrapped_synthetic_axes preflight as the plain synthetic sweep (new e2e: aura_walkforward_synthetic_blueprint_refuses_a_wrapped_form _axis_name); family builders translate raw names onto wrapped SweepBinder slots on all three entry points. C24's "retired from the surface" claim is thereby made true instead of softened. - Low items: stale graph_construct docstring updated; new graph_params_and_sweep_list_axes_are_line_identical pins the two discovery surfaces against format drift. - Bench infra: aura-bench's campaign_sweep surface seeded wrapped axis literals and was correctly refused by the new intake — seeds converted to raw; full bench rerun: all 5 surfaces fingerprint OK, deltas within load noise (report-only, no baseline update needed). Library-unit tests calling family builders directly keep wrapped input (valid: translation is idempotent on exact wrapped names; intake-level refusal is the user contract). Suite + clippy verified green post-fix. refs #328
aura-bench
Wall-clock benchmarks with committed baselines (issue #251, phase 1). Catches hot-path regressions the E2E suite structurally cannot see: every surface pins its inputs and asserts a determinism fingerprint, so a fast-but-wrong binary fails the bench instead of pinning a bogus baseline.
Running
cargo run --release -p aura-bench -- run # measure + compare, report-only
cargo run --release -p aura-bench -- run --surface engine_throughput
cargo run --release -p aura-bench -- pin # rewrite committed baselines
Debug builds refuse to measure (exit 2) — wall-clock numbers from a debug build measure the profile, not the code.
Surfaces
| surface | workload | metrics |
|---|---|---|
engine_throughput |
10M synthetic bars through an SMA-cross harness (library) | wall_s, bars_per_s |
ingest_throughput |
24 synthetic archive months x 20 fresh drains via the ingest window API (library) | wall_s, bars_per_s |
campaign_sweep |
8-cell sweep-only campaign, release aura binary on a scratch project |
wall_s, cpu_percent, peak_rss_mb |
campaign_heavy |
same cells; sweep → gate → walk-forward → 1000-resample bootstrap | wall_s, cpu_percent, peak_rss_mb |
cli_fixed_cost |
spawn floor (aura --help) + minimal data-only project run |
help_ms, run_ms |
Discipline
- Quiet box: the driver warns at 1-min loadavg > 2.0. Close heavy jobs first.
- 1 discarded warmup + 3 measured reps (median). Reps must agree on the fingerprint — cross-rep drift is reported as nondeterminism, not averaged.
- Baselines are per-host context (
hostblock inside each JSON): comparison on a different host prints an informational banner and stays report-only. - Timing drift NEVER fails the run (report-only phase, #251 cadence decision); a NOTICE line flags drift >= 10%. A fingerprint mismatch exits 1.
- Re-pin (
pin) deliberately, after an accepted behaviour or performance change; the baseline diff is reviewed like any other commit.
Deferred (follow-up on the tracker)
- Instruction-count backend (iai-callgrind) once valgrind is installed on the measuring host — slots in as a sibling backend without reshaping baselines.
- Native-crate project-load sub-cost (dylib SHA-256) — needs a build step in the scratch project.
- Thresholded failure, once recorded variance data justifies thresholds.