test_block_step_sync: locate block_stepper without importing it

The module-level import of CelerisLab.lbm.block_stepper loaded
pycuda.driver into the shared test process, unlike the other CPU host
tests (which load mocked copies via AST). find_spec gives the source path
without the import. Test content and assertions unchanged.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

docs/validation/evidence/physical-750e8a4-20261020/GROK-PHYSICAL-HANDOFF.md

@@ -1,76 +0,0 @@
-待 Grok 独立复核：750e8a4 本轮六例物理验证
-==========================================
-
-固定候选 `750e8a483ec47dcf472e528edc1cf2936f0e7e83`，工作树 `/home/frank14f/CelerisLab-release-prep-20261020` 干净。六例全部完成、退出码 0、无超时；调度器和六个子进程均已消失，最终 `nvidia-smi` 无计算进程。206 个源码/测试/文档哈希与启动 manifest 一致，各例运行前后 source map 相同，原生结果与外部摘要一致；V2 原生源码捕获及 runner 哈希也一致。V4 采用外部 manifest 绑定，没有冒称 runner 内部具备同样 provenance。没有新实验、阈值变更、源码变更、main 合并或公开发布。
-
-六例状态及进程墙钟秒数：V2 zero refined `periodic` 511.409，zero uniform `periodic` 325.254；fixed refined `driven_diagnostics_complete` 530.173，fixed uniform 同状态 333.016；V4 refined `NOT_STEADY` 315.714，uniform `NOT_STEADY` 114.183。墙钟包含初始化和观测开销，仅用于执行记录，不能作正式性能比较。
-
-**V2 zero：MRT、unshifted，均达到 t*=300。** 每例 3480 条 history；refined 1.2M steps，uniform 2.4M steps。各提取 14 个 t*>250 的周期，末五周期全部原生周期门槛通过，各自与冻结参考比较通过。
-
-| 周期指标 | refined | uniform | 原门槛 |
-|---|---:|---:|---:|
-| 末五周期范围 | 281.663875–298.232508 | 281.590556–298.153468 | 起点>=250，5 个连续周期 |
-| period min / max | 3.311073 / 3.316868 | 3.311314 / 3.314196 | 每周期严格在 (2,5) |
-| period ptp/mean | 0.00174873 | 0.000869972 | <=0.005 |
-| 末周期距最后力采样 | 1.757492 | 1.836532 | [0,5] |
-| 连续周期最大间隙 | 0 | 0 | <1e-8 |
-| 最小 Cl peak-to-peak | 2.003137 | 1.998667 | >=0.2 |
-| Cd_min / Cd_max 跨周期 range | 8.14323e-5 / 1.39714e-4 | 2.60406e-4 / 8.66420e-5 | 每项<=0.005 |
-| Cl_min / Cl_max 跨周期 range | 1.96279e-4 / 1.15290e-4 | 1.59490e-4 / 1.82477e-4 | 每项<=0.005 |
-| dp_min / dp_max 跨周期 range | 5.50774e-4 / 5.87194e-4 | 8.12699e-4 / 7.99311e-4 | 每项<=0.005 |
-| 周期质量均值 ptp/abs(mean) | 3.61945e-8 | 8.51505e-9 | <=1e-4 |
-| 最大周期均值 flux_relative | 6.35676e-6 | 5.22752e-5 | <=0.005 |
-
-末周期实际值：refined period=3.312817588、St=0.301857852、Cd=[3.160863689,3.234990541]、Cl=[-1.019161308,0.984224152]、dp=[2.404928344,2.499337133]；uniform period=3.311313790、St=0.301994937、Cd=[3.161468623,3.231458726]、Cl=[-1.016677997,0.982020469]、dp=[2.420541455,2.513318511]。
-
-| 末周期参考误差 | refined | uniform | 原门槛 |
-|---|---:|---:|---:|
-| Cd_min / Cd_max 相对误差 | 0.00107520 / 0.00235425 | 0.000884026 / 0.00125993 | 每项<=0.02 |
-| Cl_min / Cl_max 绝对误差 | 0.00212432 / 0.00234646 | 0.00460763 / 0.00455014 | 每项<=0.03 |
-| dp_min / dp_max 相对误差 | 0.00638037 / 0.00306172 | 0.0000703422 / 0.00251519 | 每项<=0.02 |
-| St 相对误差 | 0.0000558054 | 0.000509970 | <=0.02 |
-| Cd / Cl / dp waveform RMS | 0.00472589 / 0.00216188 / 0.0117979 | 0.00317231 / 0.00337428 / 0.00382206 | 描述指标，不是 gate |
-
-`matched_case_pass=false` 原样保留：四次 V2 调用均选择单 case，runner 只有 `--case both` 的两结果原生比较分支才赋值这一字段。它不等于 zero 的各自周期或参考比较失败，也不能由执行者自行改为 true；本轮没有运行该 paired-case 原生比较。fixed 没有官方 DFG 参考 verdict，完成执行不等于物理验收。
-
-**Fixed Jet 健康与 CV。** 两例达到 t*=300，五个 250–300 诊断窗均完整。原健康规则为 physical leaf 数据及 impulse 有限、rho_min>=0.5、rho_max<=1.5、Mach<=0.2。所有 3480 条 history 均有限，观测数值如下。
-
-| 指标 | refined | uniform |
-|---|---:|---:|
-| 全序列 rho min / max | 0.994218661 / 1.004163150 | 0.993391573 / 1.004678726 |
-| 全序列 max Mach | 0.0422925943 | 0.0421810197 |
-| mass t*=0.5 / t*=300 | 0.8931541481 / 0.8931214909 | 0.8936680602 / 0.8936524063 |
-| 上述端点质量相对变化 | -3.65640e-5 | -1.75164e-5 |
-| t*=290–300 mass ptp/mean | 8.40337e-5 | 6.31000e-5 |
-| 290–300 Cd / Cl 均值 | 3.184020005 / -0.580990812 | 3.182612828 / -0.582235721 |
-
-CV 各 501 个端点，间距 t*=0.02，积分区间 290–300（root time 40000）。排除首端点之前的 impulse，只累加后 500 个区间；root reaction impulse refined=(254.721600438,-46.479264945)，uniform=(254.609026243,-46.578857642)。两条矩形轮廓 refined 为 (48,44,160,120)/(44,36,180,128)，uniform 的 x 坐标均加 0.25。以原式 `J_CV=-ΔP-∫outward_momentum_flux dt`，残差为 `J_CV-J_raw`，系数除以 `duration*0.5*U_root²*D_root=80`。这里是 wall+aperture 原始流体对物体反作用，不是已修正的 nozzle thrust。
-
-| 动量残差系数 (x,y) | refined | uniform |
-|---|---|---|
-| 内轮廓、stride1 | (2.24958e-4,-1.00611e-4) | (1.78190e-4,-7.24811e-5) |
-| 外轮廓、stride1 | (7.88038e-5,-4.35879e-5) | (-8.30841e-5,-1.99246e-5) |
-| 外减内 | (-1.46155e-4,5.70231e-5) | (-2.61274e-4,5.25565e-5) |
-| stride2 相对 stride1 最大分量差 | 1.18312e-5 | 2.19402e-5 |
-| stride4 相对 stride1 最大分量差 | 4.44884e-5 | 6.26044e-5 |
-
-CV 未定义验收门槛，以上仅为离线积分诊断。质量记录缺少 raw `mass_into_fluid` Jet 源账本，无法扣除质量源判断闭合。`ΔM+∫outward_mass_flux dt` 仅是表观质量源：refined 内/外=-0.0172316912/-0.646617293 root mass；uniform=3.519017515/3.575029022。相对初始 CV 库存分别 (-2.37505e-6,-5.74501e-5)/(4.84891e-4,3.17573e-4)。轮廓间质量差异保留供审查，不能据此宣称质量闭合通过。
-
-**V4：SRT，无 Jet，t*=60。** uniform 480k/refined 240k steps，各 120 个样本。末窗 t*=50–60 共 21 点；drift 是原 runner 的 `abs(OLS slope per t*)*10`。
-
-| case / 指标 | 末窗 range | 原 range 上限 | 末窗 drift | 原 drift 上限 |
-|---|---:|---:|---:|---:|
-| refined Cd | 9.86614e-4 | 0.002 | 4.68834e-6 | 0.0004 |
-| refined Cl | 6.95374e-5 | 6e-5 | 3.28188e-6 | 1.2e-5 |
-| refined p_diff | 1.45066e-4 | 4e-5 | 9.28999e-6 | 8e-6 |
-| uniform Cd | 1.19061e-3 | 0.002 | 1.60374e-4 | 0.0004 |
-| uniform Cl | 7.93289e-5 | 6e-5 | 5.92289e-6 | 1.2e-5 |
-| uniform p_diff | 1.27157e-4 | 4e-5 | 2.80735e-5 | 8e-6 |
-
-两例均有 Cl range、p_diff range/drift 超限，保持 `steady=false,status=NOT_STEADY,verdict=null`。末窗均值 refined=(Cd 5.588876919,Cl 0.010732806,p_diff 0.117811546)，uniform=(5.586439289,0.010680915,0.117402607)。原区间 Cd=[5.57,5.59]、Cl=[0.0104,0.011]、p_diff=[0.1172,0.1176]；refined 压差均值在区间外，uniform 均值在区间内，但均不能越过稳态 gate。
-
-120 点全序列（t*=0.5–60）存在初始瞬态和反复起伏：refined Cd/Cl/p_diff 的增量符号变化次数为 82/80/82，uniform 为 79/78/81；末窗仍为 13/13/14 和 12/13/11。末窗有符号 OLS slope per t* refined=(-4.68834e-7,-3.28188e-7,-9.28999e-7)，uniform=(-1.60374e-5,+5.92289e-7,-2.80735e-6)。这只是当前采样分辨率下的非单调时序描述，不能确定振荡频率、物理/数值根因，也不是发散或瓶颈结论。完整 JSON 保留全序列及逐 10 t* 窗统计。边界偏差仍为 pressure-NEQ/neq_extrap outlet 与 local Zou-He inlet，不能称为原始 do-nothing DFG 边界。
-
-**复核边界与证据。** 本轮不是发布通过：短套件仍为 767 passed/24 failed/1 skipped；V1 历史 pin 阻塞，V3 四例原 TG dry-run 均被 frozen-source guard 拒绝、尚无 CUDA 计算或 TG 结果；未接纳待修订 `56004f44`，没有修改 dtype/NumPy/数值核心。下一步由 Claude 提交修订和解释、Grok 独立复核，再由集成任务接受明确通过审查的 SHA。
-
-原始证据根 `/home/frank14f/CelerisLab-release-evidence-20261020/physical-750e8a4/followup-queue`。`final-physical-digest.json` 约 99 KB，含全部六例原生小摘要、门槛、CV 抽稀结果、源/输出哈希、进程检查和限制；后处理脚本 `final_physical_digest.py` 不导入或运行模拟器。原始 history/log 只保留在 Horsea，没有上传巨量日志。

docs/validation/evidence/physical-750e8a4-20261020/final_physical_digest.py

@@ -1,94 +0,0 @@
-"""Read-only postprocessing of the six existing, authorized simulation outputs."""
-import csv
-import hashlib
-import json
-import math
-from pathlib import Path
-import subprocess
-import datetime
-import numpy as np
-
-Q = Path('/home/frank14f/CelerisLab-release-evidence-20261020/physical-750e8a4/followup-queue')
-R = Path('/home/frank14f/CelerisLab-release-prep-20261020')
-SHA = '750e8a483ec47dcf472e528edc1cf2936f0e7e83'
-def read(p):
-    return json.loads(p.read_text())
-def v4_cases(data):
-    """DFG Re20 result.json: schema 1 is a bare case list, schema 2 is {schema, provenance, cases}."""
-    if isinstance(data, list): return data
-    if isinstance(data, dict) and data.get('schema') == 2: return data['cases']
-    raise ValueError('unknown DFG Re20 result schema')
-def sha(p):
-    return hashlib.sha256(p.read_bytes()).hexdigest()
-def cmd(*a):
-    return subprocess.check_output(a, text=True).strip()
-def span(a):
-    return float(np.ptp(a))
-def finite(x):
-    if isinstance(x, dict): return all(finite(v) for v in x.values())
-    if isinstance(x, list): return all(finite(v) for v in x)
-    return not isinstance(x, (int,float)) or math.isfinite(x)
-def stats(rows, key):
-    t=np.array([v['tstar'] for v in rows]); v=np.array([r[key] for r in rows])
-    d=np.diff(v); nz=np.sign(d[d!=0])
-    return dict(samples=len(v),first=float(v[0]),last=float(v[-1]),min=float(min(v)),max=float(max(v)),mean=float(np.mean(v)),range=span(v),ols_slope_per_tstar=float(np.polyfit(t-t[0],v,1)[0]),increment_sign_changes=int(np.count_nonzero(nz[1:]!=nz[:-1])),total_variation=float(np.abs(d).sum()))
-
-state=read(Q/'state.json'); manifest=read(Q/'manifest.json'); execution=read(Q/'execution.json')
-out=dict(schema=1,utc=datetime.datetime.now(datetime.timezone.utc).isoformat(),sha=SHA,evidence_root=str(Q),postprocessor_sha256=sha(Path(__file__)),scope='Offline read-only digest; no new simulation, gate edit, acceptance promotion, or performance claim',verification={},cases={})
-v=out['verification']; v['all_six_completed']=len(state['jobs'])==6 and all(x['status']=='COMPLETED' for x in state['jobs'].values())
-v['all_exit_zero_no_timeout']=all(x['exit_code']==0 and not x['timeout'] for x in state['jobs'].values())
-v['head']=cmd('git','-C',str(R),'rev-parse','HEAD');v['git_status_porcelain']=cmd('git','-C',str(R),'status','--porcelain');v['clean_fixed_head']=v['head']==SHA and not v['git_status_porcelain']
-v['tree']=cmd('git','-C',str(R),'rev-parse','HEAD^{tree}');v['src_tree']=cmd('git','-C',str(R),'rev-parse','HEAD:src')
-v['tracked_source_map_matches_manifest']=all(sha(R/p)==h for p,h in manifest['files_sha256'].items());v['tracked_source_map_count']=len(manifest['files_sha256'])
-v['execution_final_source_matches_manifest']=all(execution['source_after'][k]==manifest[k] for k in ['sha','tree','src_tree','git_status','git_diff_sha256','files_sha256'])
-pids=[state['scheduler_pid']]+[x['pid'] for x in state['jobs'].values()]
-v['tracked_processes']={}
-for pid in pids:
-    p=subprocess.run(['ps','-p',str(pid),'-o','pid=,ppid=,stat=,comm='],capture_output=True,text=True)
-    v['tracked_processes'][str(pid)]=p.stdout.strip() or 'absent'
-v['no_tracked_process_running']=all(s=='absent' or ' Z' in s for s in v['tracked_processes'].values())
-v['gpu_compute_processes']=cmd('nvidia-smi','--query-compute-apps=gpu_uuid,pid,process_name,used_gpu_memory','--format=csv')
-out['hashes']={p.name:sha(p) for p in [Q/'manifest.json',Q/'execution.json',Q/'state.json',Q/'queue-plan.json',Q/'receipt-preflight.json',Q/'authorized_followup_queue.py',Q/'v4_external_timing.py']}
-for label,sj in state['jobs'].items():
-    p=Q/label; r=read(p/'result.json'); c=out['cases'][label]=dict(status=sj['native_status'],exit_code=sj['exit_code'],timeout=sj['timeout'],process_wall_seconds=sj['wall_seconds'],pid=sj['pid'],gpu_uuid=sj['gpu_uuid'],source_unchanged=r['source_unchanged'])
-    c['source_matches_manifest']=all(r['source'][k]==manifest[k] and r['source_after'][k]==manifest[k] for k in ['sha','tree','src_tree','git_status','git_diff_sha256','files_sha256'])
-    c['external_manifest_hash_matches']=r['external_manifest_sha256']==sha(Q/'manifest.json')
-    c['hashes']={str(z.relative_to(p)):sha(z) for z in [p/'command.json',p/'result.json',*sorted((p/'output').rglob('*.json')),*sorted((p/'output').rglob('*.csv')),*sorted((p/'output').rglob('*.jsonl'))]}
-    if label.startswith('v4'):
-        n=v4_cases(read(p/'output/result.json'))[0]; rows=n['rows']
-        captured=r['native_results'][0]
-        c['native_receipt_matches_external_result']=all(n[k]==z for k,z in captured.items() if k not in ['last_row','rows_count']) and captured['last_row']==rows[-1] and captured['rows_count']==len(rows)
-        c['external_receipt_history_handling']='V4 external receipt stores row count and last row; full native rows are separately hashed'
-        c['native_all_finite']=finite(n)
-        c['native']={k:z for k,z in n.items() if k not in ['rows','lattice']}
-        c['full_series']={k:stats(rows,k) for k in ['cd','cl','p_diff']}
-        c['ten_tstar_windows']={str(a):{k:stats([z for z in rows if a<=z['tstar']<=a+10],k) for k in ['cd','cl','p_diff']} for a in range(0,60,10)}
-        c['last_window']=dict(tstar=[50,60],metrics={k:stats(rows[-21:],k) for k in ['cd','cl','p_diff']})
-        c['health']=dict(rho_min=min(z['rho_min'] for z in rows),rho_max=max(z['rho_max'] for z in rows),max_mach=max(z['max_mach'] for z in rows))
-        continue
-    case=label.rsplit('-',1)[1]; base=p/'output'/case;n=read(base/'summary.json');plan=read(p/'output/plan.json')
-    c['native_receipt_matches_external_result']={k:z for k,z in n.items() if k!='history'}==r['native_summary']
-    c['external_receipt_history_handling']='V2 external receipt omits the history path; complete history.csv is separately hashed'
-    c['native_all_finite']=finite(n)
-    c['native_runner_hash_matches']=plan['runner_sha256']==sha(R/'tests/validation/run_refined_dfg.py')
-    sm=plan['source_manifest'];c['native_source_capture_matches']=sm['checkout_revision']==SHA and not sm['checkout_dirty'] and all(sha(R/f)==h==manifest['files_sha256'][f] for f,h in sm['sha256'].items())
-    c['native_source_capture_file_count']=len(sm['sha256'])
-    c['native']={k:z for k,z in n.items() if k not in ['history','cycles','jet_diagnostic_windows']}; c['comparison']=read(p/'output/comparison.json');c['gates']=plan['gates'];c['reference_gates']=plan['reference_gates'];c['diagnostic_windows']=n['jet_diagnostic_windows']
-    rows=[{k:float(z) for k,z in row.items()} for row in csv.DictReader((base/'history.csv').open())];late=[x for x in rows if x['tstar']>=290]
-    c['health']=dict(rows=len(rows),tstar=[rows[0]['tstar'],rows[-1]['tstar']],all_finite=finite(rows),rho_min=min(z['rho_min'] for z in rows),rho_max=max(z['rho_max'] for z in rows),max_mach=max(z['max_mach'] for z in rows),mass_first=rows[0]['mass'],mass_last=rows[-1]['mass'],mass_change_fraction=(rows[-1]['mass']-rows[0]['mass'])/rows[0]['mass'],mass_late_range_fraction=span([z['mass'] for z in late])/np.mean([z['mass'] for z in late]))
-    if n['cycles']:
-        tail=n['cycles'][-5:]; periods=[z['period'] for z in tail]; masses=[z['mass_time_mean'] for z in tail]
-        c['periodic_actual']=dict(cycles_total=len(n['cycles']),tail_cycles=len(tail),tail_start=tail[0]['start'],tail_end=tail[-1]['end'],period_min=min(periods),period_max=max(periods),period_mean=float(np.mean(periods)),period_relative_range=span(periods)/np.mean(periods),last_cycle_age=rows[-1]['tstar_force']-tail[-1]['end'],consecutive_max_gap=max(abs(a['end']-b['start']) for a,b in zip(tail,tail[1:])),minimum_lift_amplitude=min(z['cl_amplitude'] for z in tail),extrema_ranges={k+'_'+e:span([z[k+'_'+e] for z in tail]) for k in ['cd','cl','dp'] for e in ['min','max']},mass_relative_range=span(masses)/abs(np.mean(masses)),max_cycle_flux_relative=max(z['flux_relative'] for z in tail),last_cycle={k:z for k,z in tail[-1].items() if not k.endswith('_waveform')})
-    cv=[json.loads(x) for x in (base/'cv_history.jsonl').read_text().splitlines()];t=np.array([x['time_root'] for x in cv]);inv=np.array([x['inventory'] for x in cv]);flux=np.array([x['outward_flux'] for x in cv]);j=np.array([x['interval_impulse_root'] for x in cv[1:]]).sum(axis=0);duration=t[-1]-t[0];norm=duration*.5*.01**2*40
-    cvout=c['cv']=dict(samples=len(cv),all_finite=finite(cv),tstar=[cv[0]['tstar'],cv[-1]['tstar']],duration_root=duration,coefficient_normalization=norm,raw_body_reaction_impulse_root=j.tolist(),first_preceding_impulse_excluded=True,contours_root=plan['cv_observation'][case+'_contours_root'],mass_qualification='Apparent mass source only: raw Jet mass_into_fluid ledger was not recorded, so mass-source-subtracted closure cannot be assessed',acceptance='offline diagnostics only; no CV pass threshold was defined',strides={})
-    for stride in [1,2,4]:
-        idx=np.arange(0,len(t),stride);assert idx[-1]==len(t)-1
-        integral=np.trapz(flux[idx],t[idx],axis=0); delta=inv[-1]-inv[0];jcv=-delta[:,1:]-integral[:,1:];res=jcv-j
-        cvout['strides'][str(stride)]=dict(samples=len(idx),momentum_delta_root=delta[:,1:].tolist(),outward_momentum_integral_root=integral[:,1:].tolist(),cv_reaction_impulse_root=jcv.tolist(),residual_impulse_root=res.tolist(),residual_coefficient=(res/norm).tolist(),apparent_mass_source_root=(delta[:,0]+integral[:,0]).tolist(),apparent_mass_source_fraction_initial_inventory=((delta[:,0]+integral[:,0])/inv[0,:,0]).tolist())
-    a=np.array(cvout['strides']['1']['residual_coefficient']);cvout['contour_residual_coefficient_difference']=(a[1]-a[0]).tolist();cvout['coarsening_coefficient_difference']={s:(np.array(cvout['strides'][s]['residual_coefficient'])-a).tolist() for s in ['2','4']}
-v['all_six_source_unchanged']=all(c['source_unchanged'] and c['source_matches_manifest'] and c['external_manifest_hash_matches'] for c in out['cases'].values())
-v['all_six_native_receipts_match']=all(c['native_receipt_matches_external_result'] and c['native_all_finite'] for c in out['cases'].values())
-v['all_v2_native_captures_match']=all(c['native_runner_hash_matches'] and c['native_source_capture_matches'] for k,c in out['cases'].items() if k.startswith('v2'))
-out['limitations']=['V1 historical source pin remains blocked','V3 original TG source guard blocked all four dry-runs before CUDA; no TG result or acceptance','Short suite remains 767 passed, 24 failed, 1 skipped; not a release pass','Matched_case_pass remains false because each invocation selected one case; native paired-case comparison branch did not execute','Fixed Jet CV is diagnostic; mass source ledger absent; no physical acceptance','V4 both NOT_STEADY with original ranges/drifts/bands; boundary deviations retained','No formal performance claim: timings include observation/profiling and startup is separately recorded']
-dest=Q/'final-physical-digest.json';dest.write_text(json.dumps(out,indent=2,allow_nan=False)+'\n')
-print(json.dumps(dict(output=str(dest),bytes=dest.stat().st_size,sha256=sha(dest),verification=v),indent=2))

src/CelerisLab/lbm/block_stepper.py

@@ -204,11 +204,21 @@ class PeriodicBlockStepper:
                 raise
 
     def step(self, n=1):
-        """Advance n steps: halo refresh, then core pull/collide, per step."""
+        """Advance n steps: halo refresh, then core pull/collide, per step.
+
+        All n steps are enqueued in order on the one stream and the host waits
+        once before returning (same launches and bits as n step(1) calls).
+        Asynchronous device errors surface at that single wait, so they are not
+        attributed to the step that caused them; call step(1) repeatedly to
+        localize one. On any failure the instance is failed and step_count keeps
+        its entry value.
+        """
         if isinstance(n, bool) or not isinstance(n, Integral) or n < 0:
             raise ValueError("n must be a nonnegative integer")
         with self._lock, self.ctx.activate():
             self._require("ready")
+            if n == 0:
+                return
             kw = dict(block=self._block, stream=self._stream)
             try:
                 self._status = "advancing"
@@ -224,9 +234,9 @@ class PeriodicBlockStepper:
                     self._step(self._current, self._spare, np.uint32(self.layout.halo),
                                self._tables.get("flags", np.uintp(0)), np.uintp(0),  # no kind
                                grid=self._grid, **kw)
-                    self._stream.synchronize()
-                    self._current, self._spare = self._spare, self._current
-                    self._step_count += 1
+                    self._current, self._spare = self._spare, self._current  # host handles only
+                self._stream.synchronize()
+                self._step_count += n
                 self._status = "ready"
             except BaseException:
                 self._status = "failed"

src/CelerisLab/lbm/refinement_stepper.py

@@ -822,27 +822,41 @@ class FixedPatchStepper:
                             grid=((pairs + self._block[0] - 1) // self._block[0], 1), **kw)
 
     def step(self, n=1):
-        """Advance G→G by n base intervals: level l takes 2**l hops per interval."""
+        """Advance G→G by n base intervals: level l takes 2**l hops per interval.
+
+        All n intervals are enqueued in order on the one stream, and the host waits
+        once, before returning; step(0) launches and waits for nothing. On return
+        the device work is complete, step_count and an open window cover all n
+        intervals, and the raw-exchange ledger holds the final interval only.
+
+        Asynchronous device errors surface at that single wait, so they are not
+        attributed to the interval that caused them; call step(1) repeatedly to
+        localize one (same launches, same bits). On any failure the instance is
+        failed, the exchange ledger and window are invalid, and step_count keeps
+        its value from entry to the failing call.
+        """
         if isinstance(n, bool) or not isinstance(n, Integral) or n < 0:
             raise ValueError("n must be a nonnegative integer")
         with self._lock, self.ctx.activate():
             self._require("ready")
+            if n == 0:
+                return
             root = self._levels[0]
             kw = dict(block=self._block, stream=self._stream)
             try:
                 self._status = "advancing"
+                self._exchange_valid = False
                 for _ in range(n):
-                    if self._observe_raw_exchange:
-                        self._exchange_valid = False
+                    if self._observe_raw_exchange:  # stream-ordered: holds the last interval only
                         for b in self._level_boundaries:
                             if b is not None:
                                 cuda.memset_d8_async(b.interval_integral, 0,
                                                      len(b.tables["cell"]) * 32, self._stream)
                     self._hop(0, root.current, root.spare, **kw)
-                    self._stream.synchronize()
-                    root.current, root.spare = root.spare, root.current
-                    self._step_count += 1
-                    self._exchange_valid = self._observe_raw_exchange
+                    root.current, root.spare = root.spare, root.current  # host handles only
+                self._stream.synchronize()
+                self._step_count += n
+                self._exchange_valid = self._observe_raw_exchange
                 self._status = "ready"
             except BaseException:
                 self._window_valid = False

tests/unit/test_block_step_sync.py

@@ -0,0 +1,104 @@
+"""Host contract of PeriodicBlockStepper.step(n): one stream wait per call (mocked driver)."""
+import ast
+from contextlib import nullcontext
+import importlib.util
+from pathlib import Path
+import types
+from unittest.mock import Mock
+import unittest
+
+import numpy as np
+
+from CelerisLab.config import LBMConfig
+from CelerisLab.lbm.blocks import BlockLayout2D
+
+
+def host_module():
+    """block_stepper with pycuda/compiler replaced by mocks; no GPU context."""
+    # Locate the source without importing it: importing block_stepper loads pycuda.driver,
+    # which CPU-only import-boundary tests (test_build_contracts) must never see.
+    path = Path(importlib.util.find_spec("CelerisLab.lbm.block_stepper").origin)
+    tree = ast.parse(path.read_text())
+    tree.body = [node for node in tree.body if not (
+        isinstance(node, ast.Import) and any(a.name == "pycuda.driver" for a in node.names)
+        or isinstance(node, ast.ImportFrom) and node.module == "cuda")]
+    module = types.ModuleType("CelerisLab.lbm._test_host_block_stepper")
+    module.__package__ = "CelerisLab.lbm"
+    module.__dict__.update(cuda=Mock(), compiler=Mock())
+    exec(compile(tree, str(path), "exec"), module.__dict__)
+    return module
+
+
+def backend():
+    host, log = host_module(), []
+    host.cuda.mem_get_info.return_value = (1 << 40, 1 << 40)
+    host.cuda.mem_alloc.side_effect = lambda size: Mock(size=size)  # distinct buffers
+    module = Mock()
+    module.get_global.return_value = (Mock(), 20)
+    module.get_function.side_effect = lambda name: Mock(
+        side_effect=lambda *a, **k: log.append((name, a[:2])))
+    host.compiler.load_module.return_value = module
+    cfg = LBMConfig(nx=16, ny=16, viscosity=.1)
+    run = host.PeriodicBlockStepper(cfg, BlockLayout2D.build(16, 16, 8, 8),
+                                    context=Mock(activate=nullcontext, sm_arch="sm_70"))
+    run._stream.synchronize.side_effect = lambda: log.append(("sync", ()))
+    run.initialize(np.full((9, 16, 16), 1 / 9))
+    log.clear()
+    return run, log
+
+
+class BlockStepSync(unittest.TestCase):
+    def test_step_n_waits_once_and_matches_n_single_steps(self):
+        many, log_many = backend()
+        many.step(3)
+        single, log_single = backend()
+        for _ in range(3):
+            single.step(1)
+        names = [name for name, _ in log_many]
+        self.assertEqual(names.count("sync"), 1)
+        self.assertEqual(names[-1], "sync")
+        self.assertEqual(names.count("BlockStep"), 3)
+        # Same launch order and buffer roles as three step(1) calls.
+        strip = lambda log: [(n, tuple(id(x) for x in a)) for n, a in log if n != "sync"]
+        roles = lambda run, log: [(n, tuple({id(run._current): "c", id(run._spare): "s"}.get(x, "-")
+                                            for x in ids)) for n, ids in strip(log)]
+        self.assertEqual([n for n, _ in strip(log_many)], [n for n, _ in strip(log_single)])
+        self.assertEqual(roles(many, log_many), roles(single, log_single))
+        self.assertIsNot(many._current, many._spare)
+        self.assertEqual((many.step_count, single.step_count), (3, 3))
+
+    def test_step_zero_launches_and_waits_for_nothing(self):
+        run, log = backend()
+        run.step(0)
+        self.assertEqual((log, run.step_count), ([], 0))
+
+    def test_failure_keeps_entry_step_count(self):
+        run, log = backend()
+        run.step(1)
+        run._stream.synchronize.side_effect = RuntimeError("device fault")
+        with self.assertRaises(RuntimeError):
+            run.step(2)
+        self.assertEqual((run.step_count, run._status), (1, "failed"))
+
+    def test_launch_failure_mid_batch_fails_closed_without_waiting(self):
+        run, log = backend()
+        calls = []
+        step = run._step
+
+        def failing(*args, **kwargs):
+            calls.append(1)
+            if len(calls) == 2:
+                raise RuntimeError("launch failure")
+            return step(*args, **kwargs)
+
+        run._step = failing
+        with self.assertRaises(RuntimeError):
+            run.step(3)
+        self.assertNotIn("sync", [name for name, _ in log])
+        self.assertEqual((run.step_count, run._status), (0, "failed"))
+        with self.assertRaises(RuntimeError):
+            run.download()
+
+
+if __name__ == "__main__":
+    unittest.main()

tests/unit/test_step_sync.py

@@ -0,0 +1,170 @@
+"""Host contract of FixedPatchStepper.step(n): one stream wait per call (mocked driver).
+
+All n base intervals are enqueued in order on the backend stream and the host
+waits once before step returns. Launch order and buffer arguments must equal n
+step(1) calls; completion metadata, raw-exchange and window ledgers, and failure
+invalidation are published only after that wait.
+"""
+from contextlib import nullcontext
+from unittest.mock import Mock
+import unittest
+
+import numpy as np
+
+from CelerisLab.config import LBMConfig
+from tests.unit.test_raw_exchange import body
+from tests.unit.test_refinement_three_level_multipatch import host_module
+
+BODY_LEVEL, HOPS = 3, 8  # body on L3: 2**3 hops per base interval
+
+
+def backend(window=False):
+    """Four-level host backend; every launch, memset and stream wait goes to one log."""
+    host = host_module()
+    log, allocations = [], []
+
+    def allocate(size):
+        allocations.append(Mock(size=size))
+        return allocations[-1]
+
+    module = Mock()
+    module.get_global.return_value = (Mock(), 20)
+    module.get_function.side_effect = lambda name: Mock(
+        side_effect=lambda *a, **k: log.append((name, a, k)))
+    host.compiler.load_module.return_value = module
+    host.cuda.mem_alloc.side_effect = allocate
+    host.cuda.memcpy_dtoh.side_effect = lambda array, pointer: array.fill(0)
+    host.cuda.memset_d8_async.side_effect = lambda *a: log.append(("memset", a, {}))
+    run = host.FixedPatchStepper(
+        LBMConfig(nx=48, ny=48, viscosity=.02), context=Mock(activate=nullcontext, sm_arch="sm_70"),
+        refinements=[[(6, 6, 26, 26)], [(20, 20, 44, 44)], [(56, 56, 72, 72)]],
+        level_blocks=[(4, 4)] * 4, level_objects=[(), (), (), (body(),)],
+        observe_raw_exchange=True, observe_window_exchange=window)
+    run._stream.synchronize.side_effect = lambda: log.append(("sync", (), {}))
+    run.initialize(*[np.full(level.shape, 1 / 9, np.float32) for level in run._levels])
+    log.clear()
+    return run, log, allocations
+
+
+def signature(log, allocations):
+    """Event names with device-pointer arguments replaced by allocation index."""
+    index = {id(a): i for i, a in enumerate(allocations)}
+    return [(name, tuple(index.get(id(a), "-") for a in args)) for name, args, _ in log]
+
+
+def names(log):
+    return [name for name, _, _ in log]
+
+
+class StepSync(unittest.TestCase):
+    def test_step1_waits_once_after_all_launches(self):
+        run, log, _ = backend()
+        run.step(1)
+        self.assertEqual(names(log).count("sync"), 1)
+        self.assertEqual(names(log)[-1], "sync")
+        self.assertEqual(names(log)[0], "memset")
+        self.assertEqual(names(log).count("BlockStep"), 1 + 2 + 4 + 8)
+        self.assertEqual(run.step_count, 1)
+        run.close()
+
+    def test_step_n_equals_n_step1_launches_with_one_wait(self):
+        single, single_log, single_alloc = backend()
+        for _ in range(3):
+            single.step(1)
+        batched, batched_log, batched_alloc = backend()
+        batched.step(3)
+        expected = [e for e in signature(single_log, single_alloc) if e[0] != "sync"]
+        got = signature(batched_log, batched_alloc)
+        self.assertEqual(got[-1][0], "sync")
+        self.assertEqual(names(batched_log).count("sync"), 1)
+        self.assertEqual(got[:-1], expected)  # same order, same ping-pong buffers
+        self.assertEqual(names(batched_log).count("memset"), 3)
+        self.assertEqual(batched.step_count, single.step_count)
+        index = [{id(a): i for i, a in enumerate(alloc)} for alloc in (single_alloc, batched_alloc)]
+        self.assertEqual(index[1][id(batched._levels[0].current)],
+                         index[0][id(single._levels[0].current)])
+        single.close()
+        batched.close()
+
+    def test_step0_launches_and_waits_for_nothing(self):
+        run, log, _ = backend()
+        run.step(2)
+        log.clear()
+        run.step(0)
+        self.assertEqual(log, [])
+        self.assertEqual(run.step_count, 2)
+        value = run.read_raw_exchange(0, level=BODY_LEVEL)  # step(0) keeps the ledger valid
+        self.assertEqual((value["time_start"], value["time_end"]), (1, 2))
+        run.close()
+
+    def test_completion_metadata_covers_all_intervals(self):
+        run, _, _ = backend()
+        with self.assertRaisesRegex(RuntimeError, "completed"):
+            run.read_raw_exchange(0, level=BODY_LEVEL)
+        run.step(3)
+        value = run.read_raw_exchange(0, level=BODY_LEVEL)
+        self.assertEqual((value["time_start"], value["time_end"], value["sample_count"]),
+                         (2, 3, HOPS))  # step(n) retains only the final base interval
+        run.step(2)
+        self.assertEqual(run.read_raw_exchange(0, level=BODY_LEVEL)["time_end"], 5)
+        run.close()
+
+    def test_window_accounting_across_batched_calls(self):
+        windows = []
+        for calls in ((1, 3), (4,), (1, 1, 1, 1)):
+            run, log, _ = backend(window=True)
+            run.step(1)
+            run.start_exchange_window()
+            log.clear()
+            for n in calls:
+                run.step(n)
+            self.assertEqual(names(log).count("BlockAccumulateExchange"), 2 * 4 * HOPS)
+            self.assertEqual(names(log).count("sync"), len(calls))
+            out = run.read_exchange_window(0, level=BODY_LEVEL)
+            windows.append((out["time_start"], out["time_end"], out["sample_count"]))
+            run.close()
+        self.assertEqual(windows, [(1., 5, 4 * HOPS)] * 3)
+
+    def test_failed_wait_publishes_nothing(self):
+        run, _, _ = backend(window=True)
+        run.step(2)
+        run.start_exchange_window()
+        run._stream.synchronize.side_effect = RuntimeError("device failure")
+        with self.assertRaisesRegex(RuntimeError, "device failure"):
+            run.step(3)
+        self.assertEqual(run.step_count, 2)  # value at entry to the failing call
+        self.assertFalse(run._exchange_valid)
+        self.assertFalse(run._window_valid)
+        for read in (lambda: run.read_raw_exchange(0, level=BODY_LEVEL),
+                     lambda: run.read_exchange_window(0, level=BODY_LEVEL), run.download,
+                     lambda: run.step(1)):
+            with self.assertRaisesRegex(RuntimeError, "failed"):
+                read()
+        run._stream.synchronize.side_effect = None
+        run.close()
+
+    def test_launch_error_mid_batch_skips_wait_and_fails(self):
+        run, log, _ = backend()
+        run.step(1)
+        log.clear()
+        hop, calls = run._hop, []
+
+        def failing(index, *args, **kw):
+            if index == 0:
+                calls.append(1)
+                if len(calls) == 2:
+                    raise RuntimeError("launch failure")
+            return hop(index, *args, **kw)
+        run._hop = failing
+        with self.assertRaisesRegex(RuntimeError, "launch failure"):
+            run.step(3)
+        self.assertNotIn("sync", names(log))
+        self.assertEqual(run.step_count, 1)
+        self.assertFalse(run._exchange_valid)
+        with self.assertRaisesRegex(RuntimeError, "failed"):
+            run.read_raw_exchange(0, level=BODY_LEVEL)
+        run.close()
+
+
+if __name__ == "__main__":
+    unittest.main()