Storage and migration problems

Incompatible state

Current state uses manifest.json with format nerve-home, version 1. Nerve does not automatically downgrade, reset malformed state, or open unknown/future versions. The explicit legacy importer accepts released Nerve 0.26 homes only when their checksummed ledger ends at 0012-remove-workers; a modified or partial ledger is rejected. Stop all Nerve processes and preserve the complete directory before testing a fresh NERVE_HOME.

Legacy history appears missing

The server-owned v2-home migration is selective; desktop only supplies confirmation and progress UI. A released post-0012 home is converted directly to current storage without first running migration 0013. Validated settings, the custom provider/model catalog, recognized credentials, projects and agents needed by retained conversations, semantic conversation/run history, referenced payloads, and plans are imported. Logs, caches, task process state/logs, daemon metadata, TLS identity, and project allow trust remain only in the timestamped backup by design.

Do not delete the backup. It contains the unchanged original post-0012 layout and is the recovery source if imported data needs investigation.

Credential import failed

Startup can continue with the complete backup intact and ask for authentication. This is nonfatal because credentials may be machine-bound or undecryptable. Reauthenticate in the new profile.

Migration aborts

Malformed settings, catalog data, daemon metadata, or startup journals fail closed. Before commit, failures restore the original legacy home; interrupted journals are recovered on the next startup. Read the reported paths and fix/copy data rather than deleting VERSION, ledger, lock, or journal files.

Retired internal archives disappeared

Storage migration 0008 permanently removes committed internal migration archives, pre-dense event archives, the retired in-home Electron profile, and other obsolete paths. It does not remove timestamped whole-home ~/.nerve-bk-* backups or current project/conversation state.

Cleanup appears stuck

Cleanup cancellation is observed between targets, not inside every operation. Search index replacement cannot be interrupted mid-target. Wait for the current target and inspect storage-operation status/logs.