Batch-tidying extracted memories

rsrs extract can import session summaries. Without a parent, summaries become roots. This page describes how to inspect and archive them in batches with resort. Reattaching cannot exactly restore an originally parentless record.

The problem

By default, imported summaries use --importance important and become roots when no parent is specified. A bulk import can add hundreds of roots at once. These roots are not reported as orphans: audit counts an orphan only when a parent id dangles, and a root is not an orphan, so audit can be all green while the tree stays fragmented.

Why the standard tidy does not fix it

The built-in tidy commands do not attach an unparented imported entry automatically:

Command	Behavior
classify --plan	Local only; performs no semantic grouping, so it usually proposes nothing
classify --tree --auto	Requires an available classification backend; inspect it with classify-config

Classify entries by working directory and keywords, then attach them with resort. Review each entry first.

Deterministic classification

The flow assigns roots to archive anchors, creates anchors, then attaches children with resort --spec. Only resort itself previews by default; creating anchors writes data. Back up before writing.

Prerequisites

Confirm the installed options before starting. This page does not pin a version; check the local binary:

rsrs --version
rsrs --help
rsrs prompt --json

Then inspect the local store and tidy state:

rsrs status
rsrs resort --status
rsrs tree-cure

Read data_dir from rsrs status and use that path in later commands. Do not copy paths from another device.

Restricted and sandbox hosts

On a restricted or sandbox host, append --client-only to every command on this page so it connects to the existing authenticated local runtime. Do not use --direct. If the connection or authentication fails, report it to the host owner instead of changing the runtime lifecycle.

rsrs --client-only status
rsrs --client-only resort --status
rsrs --client-only tree-cure

Step by step flow

Select targets

Choose the batch. --since-resort selects entries created since the last tidy; --since 1900 sets an early lower bound. --limit truncates results, and old entries with an empty creation time are excluded.

rsrs list --since-resort --limit 500 --json
rsrs list --since 1900 --limit 5000 --json

rsrs list --json includes full records in details; use rsrs show <id> --json to inspect an entry.

Diagnose the tree

Collect the current tree signals before changing anything:

rsrs audit --json
rsrs tree-cure
rsrs defrag

Review same-title groups and defrag clusters entry by entry; similarity alone does not establish duplication. Check for true duplicates, misplaced or broken links, and scattered roots.

Back up the data

Export a plaintext snapshot to a persistent directory before writing. Use a new path each time: export overwrites existing files. Do not use a temporary directory. The export contains every record in plaintext, so restrict access to the directory (for example, a user-only ACL) and encrypt it when appropriate.

rsrs export C:\Users\<user>\rsrs-backups\memories_YYYYMMDD.json

rsrs export ~/rsrs-backups/memories_YYYYMMDD.json

Create archive anchors

Create one archive container per project or topic under the target tree root. Attach every anchor with --parent; never store an anchor as a bare root. --parent accepts an id, an eight-character prefix, or a tree name.

rsrs remember "<purpose of this archive>" \
  --title "Session archive: <project>" \
  --type context --importance important \
  --parent <tree-root-or-id> --json

Keep the anchor title at or under 80 characters. An anchor is an archive container, not a new catalog node; create a new catalog node only with explicit agreement and root-create --yes.

Define classification rules

Decide the anchor for each scattered root from its working directory and keywords, and order the rules from specific to general so a specific project is tested before a shared word. When the working directory is often missing, fall back to title and body keywords.

Archive anchor	Match rule (title, body, or working directory)	Notes
Session archive: project A	Unique keywords for project A
Session archive: project B	Unique keywords for project B
Session archive: other	Matches none of the above	Fallback, evaluated last

Attach in batches

resort --spec accepts an id, an eight-character prefix, or a tree name for both id and parent. It is a dry run by default; add --go to write.

{
  "ops": [
    { "id": "<child-short-id>", "parent": "<anchor-short-id>" }
  ]
}

rsrs resort --spec '<JSON>' --json
rsrs resort --spec '<JSON>' --go --json

A dry run reports status=pending and moved=N. The write reports status=ok, moved=N, and counter=reset. Split large operations into batches: the safe size depends on the JSON and the platform's command-line limit. Check each batch before writing. PowerShell may drop quotes in a JSON argument, so prefer passing arguments as a list through a script.

Verify the result

Re-run the checks and inspect each anchor:

rsrs audit --json
rsrs tree-cure
rsrs resort --status
rsrs tree --outline --json

Acceptance criteria: roots drops clearly, lone leaf falls to or near zero, orphans=0, and the TIDY counter is reset. For anchors created empty in this pass, the child count must equal the number of roots attached to them; for anchors that already had children, compare the increase in child count before and after the run instead of the total.

Diagnosis signals

Signal	Source	Interpretation
Root count	rsrs tree-cure	Imported summaries that have no parent
Lone leaf	rsrs tree-cure	A candidate for mounting under an existing node
Orphans	rsrs audit --json	Dangling parent id; a root is not counted here
Same-title groups	rsrs audit --json	Review each record; a shared title does not prove duplication
Truncated titles	rsrs audit --json	Titles cut off during import; review before editing
Maximum depth	rsrs audit --json	Overall tree depth for context
Defrag clusters	rsrs defrag	Similarity clusters require individual review before merging

Command cheat sheet

Purpose	Command
Select a batch	rsrs list --since-resort --limit 500 --json
Select from an early date	rsrs list --since 1900 --limit 5000 --json
Diagnose	rsrs audit --json, rsrs tree-cure, rsrs defrag
Back up	rsrs export <persistent-dir>/memories_YYYYMMDD.json
Create an anchor	rsrs remember "<purpose>" --title "Session archive: <project>" --importance important --parent <tree-name-or-id> --json
Preview an attach	rsrs resort --spec '<JSON>' --json
Apply an attach	rsrs resort --spec '<JSON>' --go --json
Reset the counter	rsrs resort --reset
Read a node	rsrs show <id> --json
Read a subtree	rsrs tree --from <id> --depth 3 --json

Single command helper script

The skills repository contains memory-batch-tidy/scripts/tidy_extracted.py, but do not run it without review: preview creates anchors, repeated runs overwrite the plaintext export at the same path, and an unrelated same-title anchor can be reused. The script does not chunk operations or save a per-entry undo plan. Review targets manually and use distinct protected backup paths.

Common pitfalls

1. Preview first. resort does not write by default; confirm status=pending and moved=N before adding --go.
2. resort --go resets the TIDY counter and reports counter=reset. This is expected and no separate --reset is needed.
3. The same title does not prove duplication. Review each same_title_groups entry.
4. Similarity does not prove duplication. Review defrag clusters and their context before merging.
5. Back up and record children before merging. remember --merge-ids deletes the old entries; inspect the result and re-attach any orphaned children.
6. Quoting. Passing a JSON string with quotes to a native program through PowerShell can drop the quotes; pass arguments as a list instead.
7. Console encoding. Windows output encoding varies by environment; set PYTHONIOENCODING=utf-8 when needed.
8. Oversized batches. Split the spec to fit the platform limit, and keep exports and backups in a persistent directory.

Rollback

An attachment can be reversed by pointing each child back to its recorded original parent with another resort call. Pointing a child at the target tree root does not restore an originally parentless entry, so it is not an exact rollback.

An empty anchor can be removed with rsrs forget <anchor-id>, and a soft-deleted entry can be restored with rsrs restore <full-id>.

rsrs import mints new ids and appends copies; it does not restore in place. Alternatively, create a database snapshot with rsrs snapshot create before writing. Restore does not roll back the server and quarantines sync; read the snapshot instructions first. Keep a backup because a recoverable revision is not guaranteed for every change.

Related pages

Getting started, CLI commands, CLI JSON, Skills repository