PR1400 repair: remove audit loop without T*N*N regression

[neuron7xLab]

PR1400 repair state

Current head: 97e4c76.

Resolved:

• np.fromiter removed from _swing_energy_trajectory.
• Dense-formula equivalence test added.
• PR Gate python-quality is green.
• Mutation Kill workflow now uploads mutation-kill-diagnostics artifacts.

New hard evidence from mutation-kill-diagnostics artifact:

• exit_code.txt: 2
• stderr.log: empty
• stdout.log: ERROR: could not establish baseline for core/kuramoto/second_order.py (probe exit 2); cannot verify floor.

Meaning:

• Current Mutation Kill red is not yet a survivor-floor failure.
• The blocker is earlier: mutation_probe cannot establish the baseline passing state for core/kuramoto/second_order.py under the enrolled tests.
• Next repair target is the baseline selector/test runtime, not mutation-floor tuning.

Still active:

• Mutation Kill Gate: red at baseline establishment / probe exit 2.
• PR Gate: fast deterministic pytest gate red; python-fast-shard (3) and (4) remain failing after reruns.

Forbidden:

• no ready;
• no merge;
• no baseline edit;
• no mutation floor edit;
• no blind quarantine;
• no fake-green.

Required next step:
Run the exact enrolled tests locally/Codex first:
python -m pytest tests/unit/physics/test_T18b_second_order_stability_guard.py tests/unit/physics/test_second_order_stability_audit.py -q

Then run:
python tools/mutation_probe.py core/kuramoto/second_order.py --tests tests/unit/physics/test_T18b_second_order_stability_guard.py tests/unit/physics/test_second_order_stability_audit.py --only-logic --fail-under 0.76 --json tmp/second_order_mutation_probe.json

Patch only the exact baseline failure. Do not edit floors or baselines.

[neuron7xLab]

⊛ Algorithm / Mechanism / Function Audit v2 — 33-task execution board

This is not another narrative layer. It is the execution board for working every relevant algorithm, mechanism, and function touched by PR #1400 before any promotion.

33 concrete tasks

- 01_pr_state:
    result: PR #1400 is open/draft/unmerged; no ready/merge allowed.
    action: preserve BLOCK until gates are green.

- 02_diff_surface:
    result: branch is 4 commits ahead, 4 files changed, production diff is only core/kuramoto/second_order.py.
    action: do not broaden diff.

- 03_runtime_scope:
    result: core is in CODE_QUALITY_MANIFEST runtime_roots.
    action: every new decision in core is gated by hygiene ratchet.

- 04_hygiene_complexity_operator:
    result: ast.For increments McCabe complexity.
    action: remove explicit audit() loop or move to a pattern that does not increase audit() decision surface.

- 05_mutation_scope:
    result: second_order.py is claim-critical with floor 0.76 over 21 logic sites.
    action: no floor lowering; rerun mutation after patch.

- 06_repo_integrity_scope:
    result: prove_repo_integrity has 14 proof steps and currently fails.
    action: after code patch, run full proof and inspect first failing step.

- 07_pr_gate_scope:
    result: python-fast-shard 2/3/4 fail, shard 1 passes.
    action: do not churn green shard; inspect failing shard tests locally.

- 08_divergence_error_init:
    mechanism: carries step + kind for fail-closed diagnostics.
    risk: low; untouched by PR.
    action: preserve.

- 09_audit_scope_constants:
    mechanism: declare nonfinite_and_rocof_only + remaining gaps.
    risk: claim-expansion if edited.
    action: preserve unchanged.

- 10_result_post_init:
    mechanism: updates summary with frequency metrics.
    risk: summary drift if changed.
    action: preserve.

- 11_result_accessors:
    mechanism: phases/order_parameter/time/config/summary pass-through.
    risk: low.
    action: preserve.

- 12_frequency_metrics:
    mechanism: min/max velocity, max_rocof, spread, mean, settling_time.
    risk: test-sensitive.
    action: preserve.

- 13_settling_time:
    mechanism: reverse scan over velocity spread.
    risk: explicit loop is existing baseline, not PR-caused.
    action: do not touch in this PR.

- 14_report_dataclass:
    mechanism: frozen slots stability audit record.
    risk: schema/field drift.
    action: preserve fields.

- 15_engine_init:
    mechanism: resolve IC/adjacency, validate rocof, broadcast mass/damping.
    risk: fail-closed contract.
    action: preserve.

- 16_mass_damping_guards:
    mechanism: mass > 0, damping >= 0.
    risk: physics invalidity if weakened.
    action: preserve.

- 17_verlet_storage:
    mechanism: phases/velocities/R/time arrays.
    risk: memory but not current blocker.
    action: out of scope.

- 18_verlet_acceleration:
    mechanism: inv_m * (omega + coupling - d*v).
    risk: physics invariant.
    action: preserve.

- 19_damped_velocity_update:
    mechanism: semi-implicit damping denominator.
    risk: symplectic/contraction claim.
    action: preserve.

- 20_nonfinite_guard:
    mechanism: raises SecondOrderDivergenceError if theta/v invalid.
    risk: fail-closed safety.
    action: preserve.

- 21_rocof_guard:
    mechanism: optional bound; default off.
    risk: byte-identical tests.
    action: preserve.

- 22_audit_energy_kinetic:
    mechanism: 0.5 * (m * vel * vel).sum(axis=1).
    risk: formula drift.
    action: preserve.

- 23_audit_energy_drive:
    mechanism: -(omega * theta).sum(axis=1).
    risk: formula drift.
    action: preserve.

- 24_audit_pairwise_potential:
    mechanism: current explicit loop computes per-snapshot O(N^2).
    risk: direct blocker: hygiene + mutation + PR cascade.
    action: replace loop; do not restore T*N*N.

- 25_energy_drift_scale:
    mechanism: max-min energy normalized by max abs energy + epsilon.
    risk: numerical audit drift.
    action: preserve.

- 26_conservative_flag:
    mechanism: omega==0 and damping==0.
    risk: test-sensitive.
    action: preserve.

- 27_phase_spread:
    mechanism: max-min raw unwrapped phases.
    risk: finite audit.
    action: preserve.

- 28_stiffness_ratio:
    mechanism: symmetric adjacency, Laplacian eigval, sqrt(lambda/m_min)*dt.
    risk: branch thresholds resolved/marginal/underresolved.
    action: preserve.

- 29_cross_solver_horizon:
    mechanism: clamp cross_solver_steps to [1, cfg.steps].
    risk: index and test stability.
    action: preserve.

- 30_rk4_cross_solver:
    mechanism: independent trajectory verifier.
    risk: expensive but required.
    action: preserve.

- 31_metric_validity_flags:
    mechanism: symmetric AND unsigned.
    risk: mutation survivor surface.
    action: preserve; tests already pin AND-not-OR.

- 32_promotion_firewall:
    mechanism: promotion_allowed false when gaps remain.
    risk: overclaim.
    action: preserve.

- 33_repair_commit_rule:
    mechanism: exact code patch only, then rerun 4 red lanes.
    action: patch core/kuramoto/second_order.py only; append PATCH_APPLIED event only after commit exists.

Exact production repair

Replace only this block:

v_coup = np.empty(theta.shape[0], dtype=np.float64)
for idx, theta_t in enumerate(theta):
    delta = theta_t[np.newaxis, :] - theta_t[:, np.newaxis]
    v_coup[idx] = -0.5 * float(np.sum(adj * np.cos(delta)))

With a non-TNN evaluator that removes the explicit audit() loop. First candidate:

def pairwise_potential(theta_t: NDArray[np.float64]) -> np.float64:
    delta = theta_t[np.newaxis, :] - theta_t[:, np.newaxis]
    return np.float64(-0.5 * float(np.sum(adj * np.cos(delta))))

v_coup = np.apply_along_axis(pairwise_potential, 1, theta).astype(
    np.float64,
    copy=False,
)

Hard verification order

python scripts/ci/check_code_hygiene.py
python scripts/ci/check_mutation_kill_ratchet.py --base-ref origin/main
python -m pytest tests/unit/physics/test_second_order_stability_audit.py -q
python -m pytest tests/unit/physics/test_T18b_second_order_stability_guard.py -q
bash scripts/ci/prove_repo_integrity.sh

Rule

No issue close, no PR ready, no merge, no ACCEPT until all four red gates are green.