eval history and metrics

src/infospace_bench/cli.py

@@ -5,7 +5,9 @@ import json
 import sys
 from pathlib import Path
 
+from .checks import run_collection_checks
 from .errors import InfospaceError
+from .history import find_snapshot, get_history, metric_trend, record_check_results
 from .lifecycle import add_artifact, create_infospace, load_infospace
 from .markdown_adapter import validate_infospace_artifacts
 from .semantics import list_entities, list_relations
@@ -42,6 +44,24 @@ def build_parser() -> argparse.ArgumentParser:
     relations = sub.add_parser("relations", help="List parsed relation artifacts")
     relations.add_argument("root")
 
+    history = sub.add_parser("history", help="List evaluation snapshot history")
+    history.add_argument("root")
+    history.add_argument("--metric", default="")
+
+    history_diff = sub.add_parser(
+        "history-diff",
+        help="Diff two evaluation snapshots by snapshot ID or date",
+    )
+    history_diff.add_argument("root")
+    history_diff.add_argument("before")
+    history_diff.add_argument("after")
+
+    metrics = sub.add_parser(
+        "metrics",
+        help="Run collection checks and persist metrics/history",
+    )
+    metrics.add_argument("root")
+
     return parser
 
 
@@ -96,6 +116,40 @@ def main(argv: list[str] | None = None) -> int:
                     ]
                 }
             )
+        elif args.command == "history":
+            history = get_history(Path(args.root))
+            if args.metric:
+                _write_json(
+                    {
+                        "metric": args.metric,
+                        "trend": metric_trend(history, args.metric),
+                    }
+                )
+            else:
+                _write_json({"history": [item.to_dict() for item in history]})
+        elif args.command == "history-diff":
+            history = get_history(Path(args.root))
+            before = find_snapshot(history, args.before)
+            after = find_snapshot(history, args.after)
+            if before is None or after is None:
+                missing = []
+                if before is None:
+                    missing.append(args.before)
+                if after is None:
+                    missing.append(args.after)
+                raise InfospaceError(
+                    "missing_snapshot",
+                    "Could not resolve requested snapshot reference",
+                    {"missing_refs": missing},
+                )
+            _write_json({"diff": before.diff(after).to_dict()})
+        elif args.command == "metrics":
+            infospace = load_infospace(Path(args.root))
+            result = record_check_results(
+                infospace.root,
+                run_collection_checks(infospace.artifacts),
+            )
+            _write_json(result.to_dict())
         else:
             parser.error(f"Unhandled command: {args.command}")
     except InfospaceError as exc: