feat: seed 996 runs across 3 models from hosted bench results

Import script converts JSON exports to seed SQL.
claude-sonnet-4: 170 runs (96.5% pass rate)
deepseek-chat: 251 runs (84.9%)
gemini-2.5-flash: 575 runs (75.0%)
Bump bench-cli to 0.5.7 (infra fixes).

Author: vitas

demo/postgres/import-json.py

@@ -0,0 +1,88 @@
+#!/usr/bin/env python3
+"""Convert bench JSON exports to seed SQL."""
+import json
+import sys
+import os
+
+SCRIPT_DIR = os.path.dirname(os.path.abspath(__file__))
+SCENARIOS_JSON = sys.argv[1] if len(sys.argv) > 1 else os.path.expanduser("~/git/evidra-bench/runs/export/bench-scenarios.json")
+RUNS_JSON = sys.argv[2] if len(sys.argv) > 2 else os.path.expanduser("~/git/evidra-bench/runs/export/demo-runs.json")
+OUTPUT = os.path.join(SCRIPT_DIR, "seed-data.sql")
+
+
+def q(v):
+    """SQL-quote a value."""
+    if v is None:
+        return "NULL"
+    s = str(v).replace("'", "''")
+    return f"'{s}'"
+
+
+def n(v, default=0):
+    """Numeric value."""
+    return str(v) if v is not None else str(default)
+
+
+def b(v):
+    """Boolean value."""
+    return "true" if v else "false"
+
+
+with open(SCENARIOS_JSON) as f:
+    scenarios = json.load(f)
+
+with open(RUNS_JSON) as f:
+    runs = json.load(f)
+
+with open(OUTPUT, "w") as out:
+    out.write("-- Auto-generated seed data from bench JSON exports.\n")
+    out.write("-- Idempotent: ON CONFLICT DO NOTHING / DO UPDATE.\n")
+    out.write("SET client_min_messages = warning;\n\n")
+
+    # Scenarios
+    out.write(f"-- Scenarios ({len(scenarios)})\n")
+    for s in scenarios:
+        out.write(
+            f"INSERT INTO bench_scenarios (id, category, title, track, level, tags, chaos, evidra_enabled) "
+            f"VALUES ({q(s.get('id'))}, {q(s.get('category',''))}, {q(s.get('title',''))}, "
+            f"{q(s.get('track',''))}, {q(s.get('level',''))}, {q(s.get('tags',''))}, "
+            f"{b(s.get('chaos'))}, {b(s.get('evidra'))}) "
+            f"ON CONFLICT (id) DO UPDATE SET category=EXCLUDED.category, title=EXCLUDED.title, "
+            f"track=EXCLUDED.track, level=EXCLUDED.level, tags=EXCLUDED.tags, "
+            f"chaos=EXCLUDED.chaos, evidra_enabled=EXCLUDED.evidra_enabled;\n"
+        )
+
+    # Runs
+    out.write(f"\n-- Runs ({len(runs)})\n")
+    for r in runs:
+        out.write(
+            f"INSERT INTO bench_runs (id, tenant_id, scenario_id, model, provider, adapter, "
+            f"evidence_mode, passed, duration_seconds, exit_code, turns, memory_window, "
+            f"prompt_tokens, completion_tokens, estimated_cost_usd, checks_passed, checks_total, "
+            f"checks_json, metadata_json, created_at, tool_server, tool_server_version, scenario_version) "
+            f"VALUES ({q(r.get('id'))}, 'default', {q(r.get('scenario_id'))}, "
+            f"{q(r.get('model'))}, {q(r.get('provider',''))}, {q(r.get('adapter','bench-cli'))}, "
+            f"{q(r.get('evidence_mode','none'))}, {b(r.get('passed'))}, "
+            f"{n(r.get('duration_seconds',0))}, {n(r.get('exit_code',0))}, "
+            f"{n(r.get('turns',0))}, {n(r.get('memory_window',-1))}, "
+            f"{n(r.get('prompt_tokens',0))}, {n(r.get('completion_tokens',0))}, "
+            f"{n(r.get('estimated_cost_usd',0))}, {n(r.get('checks_passed',0))}, "
+            f"{n(r.get('checks_total',0))}, {q(r.get('checks_json',''))}, "
+            f"{q(r.get('metadata_json',''))}, {q(r.get('created_at','2026-03-29T00:00:00Z'))}, "
+            f"{q(r.get('tool_server',''))}, {q(r.get('tool_server_version',''))}, "
+            f"{q(r.get('scenario_version',''))}) ON CONFLICT (id) DO NOTHING;\n"
+        )
+
+lines = sum(1 for _ in open(OUTPUT))
+print(f"Written {OUTPUT} ({lines} lines)")
+print(f"  Scenarios: {len(scenarios)}")
+print(f"  Runs: {len(runs)}")
+
+# Summary by model
+from collections import Counter
+model_counts = Counter(r['model'] for r in runs)
+model_passed = Counter(r['model'] for r in runs if r.get('passed'))
+for model in sorted(model_counts, key=model_counts.get, reverse=True):
+    total = model_counts[model]
+    passed = model_passed.get(model, 0)
+    print(f"  {model}: {total} runs, {passed} passed ({100*passed//total}%)")