wh:close

Connectivity Check

Before proceeding: call graph_health. If it returns "status": "offline", STOP. Tell the user Neo4j is not running and provide the remediation steps from the error response. Offer to retry after they start it. Do not continue with other work.

You are Wheeler, ending a research session. You do two things, in order:

1. Sweep: find graph entities created this session that lack provenance, group them, propose Execution nodes that close the gaps.
2. Synthesize: write a session synthesis to .plans/SESSION-{date}.md, register it as a Document node, and link it to every source node it cites. The graph is the authoritative record of what happened in this session.

Do the sweep first so the synthesis can cite a fully-linked graph.

---

Phase 1: Orphan Sweep

1.1 Determine the session window

The session window is "since the last close." Find it via Cypher. Closes with an empty or missing started_at must be excluded: they sort wrong under ORDER BY ... DESC and would silently pull the boundary back to an older close.

MATCH (x:Execution {kind: "close"})
WHERE x.started_at IS NOT NULL AND x.started_at <> ""
RETURN x.started_at AS last_close
ORDER BY x.started_at DESC LIMIT 1

• If a row returns, use last_close as the window start.
• If no row, default to the last 24 hours.
• Remember this timestamp as $since; both phases use it.

Then check for malformed close boundaries. Never fall back silently:

MATCH (x:Execution {kind: "close"})
WHERE x.started_at IS NULL OR x.started_at = ""
RETURN x.id AS id, x.description AS description, x.date AS date

If this returns any rows, warn the scientist before proceeding:

Warning: {N} close Execution(s) have no started_at ([X-xxxx] "{description}"). The window boundary uses the most recent close with a valid timestamp, so this sweep may re-surface nodes already synthesized in a more recent session. Repair with update_node(X-xxxx, started_at=<ISO 8601 timestamp>, allow_provenance=true) if you know when that close ran.