Add private single-attempt judging orchestration

roadscore/EVENT.md

@@ -107,3 +107,9 @@ Native muted replay `normal_1789795416` is running without continuous UI video e
 `normal_1789795416` passed the instrumented native gate through final-segment EOF: 254.1 seconds captured, six accepted fresh generation jobs, zero accepted-music holds, zero underflows, zero emergency fallbacks, no worker failure, zero output flags and all blocks muted. Camera accepted 5,088 frames; path/lane drawing, navigation and ten curve activations were observed. All request timestamps were causal. Maximum source-clock drift was 24.671 ms. Continuous UI video encoding was disabled; the overlay snapshot and UI audit remain available. This single pass does not prove that encoding caused the earlier timing failure.
 
 The local private evidence is `results/event_night_one/prism_power45/`: preparation metadata, native audit, overlay and captured score. The reusable post-run audit correctly rejects the earlier full-route failure and has three focused negative-evidence tests. It does not claim human musical approval, physical speaker/Bluetooth validation or modeld coexistence. Phase 2 has not yet begun; the native prerequisite is now met. No upload or push was performed.
+
+### Private judging batch setup
+
+After local checkpoint `f75a98181f`, all eleven unique submissions returned native route-file listings. Private complete-route caching has begun. The private manifest fixes anonymous labels, route-derived seeds, common Prism/45 W/muted settings and objective excerpt rules before any judging playback. Exact identities and acquisition logs remain ignored/private.
+
+Checkpoint `963d069778` adds optional deterministic sampling: separate preparation/continuation seed streams, request sequence independent of wall-clock job filenames, and rejection of a mismatched resident preparation. Legacy sampling remains unchanged unless explicitly configured. Three seed tests and six composer tests pass. Submission A's first attempt is preparing; the single-attempt runner waits for its matching preparation and complete ordinary route cache, then performs muted native replay. It refuses automatic reruns. No listening scores or winner exist. Range-disambiguation entries remain blocked from launch until their metadata decision is recorded.

roadscore/tools/judging_run.py

@@ -0,0 +1,93 @@
+"""Private single-attempt native judging orchestration. No musical selection or uploads."""
+import argparse
+import json
+import os
+from pathlib import Path
+import subprocess
+import time
+
+ROOT = Path(__file__).resolve().parents[1]
+PYTHON = '/usr/local/venv/bin/python'
+
+
+def save(path, value):
+    temp = path.with_suffix('.tmp')
+    temp.write_text(json.dumps(value, indent=2))
+    os.chmod(temp, 0o600)
+    temp.replace(path)
+
+
+def main():
+    parser = argparse.ArgumentParser()
+    parser.add_argument('manifest', type=Path)
+    parser.add_argument('--label', required=True, choices=list('ABCDEFGHIJK'))
+    parser.add_argument('--resume-preparation', action='store_true')
+    args = parser.parse_args()
+    manifest = json.loads(args.manifest.read_text())
+    row = next(r for r in manifest['submissions'] if r['label'] == args.label)
+    if row.get('range_resolution') == 'pending metadata':
+        raise SystemExit('Resolve submitted range before launching this entry')
+    out = ROOT / 'results/community_judging'
+    out.mkdir(exist_ok=True, mode=0o700)
+    ledger = out / f'official_{args.label}.json'
+    if ledger.exists():
+        record = json.loads(ledger.read_text())
+        if not args.resume_preparation or record['phase'] != 'preparing' or record['seed'] != row['seed']:
+            raise SystemExit('Existing attempt preserved; refusing an automatic rerun')
+    else:
+        record = {'label': args.label, 'seed': row['seed'], 'attempt': 1, 'phase': 'preparing',
+                  'started_wall': time.time(), 'configuration': manifest['configuration'], 'manual_reruns': 0}
+        with ledger.open('x') as stream:
+            json.dump(record, stream, indent=2)
+        os.chmod(ledger, 0o600)
+    env = os.environ.copy()
+    env.update(ROADSCORE_GENERATION_SEED=str(row['seed']), AM_POWER_LIMIT='45')
+    service = ROOT / 'prototype/worker_service.py'
+    try:
+        if not args.resume_preparation:
+            subprocess.run([PYTHON, str(service), 'stop', '--composer', 'ace'], check=True)
+            subprocess.run([PYTHON, str(service), 'start', '--composer', 'ace', '--profile', 'prism'], env=env, check=True)
+        deadline = time.monotonic() + 1560
+        while True:
+            state = json.loads((ROOT / 'generated/ace_worker_state.json').read_text())
+            if state.get('generation_seed') == row['seed'] and state.get('phase') == 'READY':
+                break
+            if state.get('phase') == 'Stopped' or time.monotonic() > deadline:
+                raise RuntimeError('Preparation stopped or timed out; original attempt retained')
+            time.sleep(5)
+        initial = json.loads((ROOT / 'generated/ace_initial.json').read_text())
+        if initial.get('generation_seed') != row['seed']:
+            raise RuntimeError('Prepared seed mismatch')
+        save(out / f'preparation_{args.label}.json', initial)
+        # Require the ordinary complete cache, without inspecting musical/road events.
+        import sys
+        sys.path.insert(0, str(ROOT / 'prototype'))
+        from route_library import cache_complete
+        deadline = time.monotonic() + 3600
+        while not cache_complete(ROOT / 'routes' / row['route']):
+            if time.monotonic() > deadline:
+                raise RuntimeError('Cache not ready within one hour; no replay launched')
+            time.sleep(10)
+        before = set((ROOT / 'results').glob('normal_*'))
+        record.update(phase='replaying', replay_started_wall=time.time())
+        save(ledger, record)
+        command = [str(ROOT.parent / 'onroad'), '--routeid', row['route'], '--roadscore',
+                   '--composer', 'ace', '--profile', 'prism', '--muted',
+                   '--start', str(row.get('start_seconds', 0)), '--duration', str(row.get('duration_seconds', 86400))]
+        with (out / f'console_{args.label}.log').open('wb') as log:
+            result = subprocess.run(command, env=env, cwd=ROOT.parent, stdout=log, stderr=subprocess.STDOUT)
+        runs = sorted(set((ROOT / 'results').glob('normal_*')) - before)
+        record.update(phase='finished' if result.returncode == 0 else 'failed', exit_code=result.returncode,
+                      run_paths=[str(r) for r in runs], ended_wall=time.time())
+        for run in runs:
+            subprocess.run([PYTHON, str(ROOT / 'tools/event_replay_audit.py'), str(run)], stdout=subprocess.DEVNULL, check=True)
+    except Exception as error:
+        record.update(phase='failed', error_type=type(error).__name__, error=str(error), ended_wall=time.time())
+        raise
+    finally:
+        save(ledger, record)
+    print(args.label, record['phase'], flush=True)
+
+
+if __name__ == '__main__':
+    main()