Creation Parameters

Some parameters of a store are chosen when it is created and never change afterwards: they are written into the store itself (its GRAPH file, its container), every copy of the store carries them, and no API, command or option changes them. Everything else is an open option: given to every open, it may differ from one open to the next and is stored nowhere.

graphersal store info <dir> lists the creation parameters under "fixed at creation"; Store::info / Store::inspect return them (StoreInfo::store_id, chunk_bytes, damage_policy, created_at, format_version, space for a single file), the dev server's GET /api/store and the Store menu show them.

$ graphersal store create data/ --from modern --on-damage continue
Created the store data/: graph 01a11cfc-..., commit 0, 1 snapshot(s); damage policy continue (fixed for its life).
$ graphersal store info data/
...
fixed at creation (never changed):
  store id      01a11cfd-...
  created       2026-10-08T19:27:26Z
  chunk size    1048576 bytes
  on damage     continue (damage found while open is reported, commits continue)
  backend       directory (format version 2)

The parameters fixed at creation

The graph id is not a parameter: it identifies the current lineage. It is set when the store is created, an in-place rollback starts a new one in the same store, and a fork or repair has its own (its lineage chain names the parent). The store id identifies the store itself: a fork, repair or conversion is another store (a new id), so it never continues the original's incremental backups (Backups).

Damage policy

Damage can appear on disk while a store is open. A backup (it reads every checksum of what it copies), a checkpoint (its merge reads the last snapshot and the WAL after it) and verify find it; the damage policy decides what happens then:

• maintenance (the default): the open store turns read-only with the damage as the reason. Commits, marks, checkpoints, prune, rollback and compaction are refused (PersistError::DamageFound); queries go on; a backup from memory saves the intact graph; the store writes nothing more to the damaged disk.
• continue: the store keeps accepting commits; the damage stays reported (store info, the Store menu, a warning in the log) for as long as it is open.

It does not apply to damage found when the store is opened: a store that cannot be loaded intact always opens in maintenance mode. See Damage, Maintenance and Repair.

Why fixed: the policy is a decision about the data (may it change while its files are known to be damaged?), not about one process. A fork, a repair or a backup is the same data, so it keeps the decision; an open never overrides it. An open given another policy ignores it (StoreOptions::with_damage_policy is used only when a store is created), and the CLI refuses --on-damage for an existing store with another policy instead of ignoring it silently. A store whose GRAPH files are both lost cannot tell its policy: a repair of it gets the default.

The storage type of the graph (Store<TraversalGraph> or a host's Store<S>, Your Own Storage) is NOT a creation parameter: the files do not depend on it, so any storage type opens any store.

Open options (per open, stored nowhere)

A new creation parameter is added to the first table (and to store info); see CLAUDE.md.