---
digest: >
  v0.36.1 is a patch release of one Change Request, CR153. Session
  maintenance used to read Pi's session directory and import every session a
  home did not already track, so on a machine with more than one Mirror home
  each home copied the other's Pi conversations and titled them again through
  the provider. Maintenance now imports no Pi session: a home's Pi
  conversations are the ones the Pi extension logged into it, as they
  happened, and `PI_SESSIONS_DIR` is no longer read. Conversations an earlier
  version imported stay where they are, and this note shows how to list them.
  Update every install that has a home. No migration.
---

# v0.36.1 — A Home Keeps Only Its Own Pi Sessions

**Date:** 2026-10-21

## Highlights

- **Session maintenance imports no Pi session**
  ([CR153](../project/refinement/rs010-cv22-oracle-and-port-hygiene/cr153-a-home-s-pi-backfill-imports-another-home-s-sessions.md)).
  It used to read every file in Pi's session directory and import each one the home
  did not track, every session another Mirror home on the same machine had logged
  among them. A home's Pi conversations are now exactly the ones the Pi extension
  logged into it, as they happened.
- **Nothing in Mirror reads Pi's session files.** Not your history from before Mirror,
  not sessions you run without it, and not another home's. `PI_SESSIONS_DIR` is inert:
  a `.env` that sets it keeps working, and it changes nothing.
- **What was already imported stays.** This release removes the cause, not the copies;
  [Upgrading](#upgrading) shows how to list them.
- **Update every install that has a home.** A home is safe from a neighbour's
  maintenance only once the install that neighbour runs carries this release.

## Where We Started

The import is older than the idea of two homes on one machine. It existed to fill a
gap: a Pi session the extension never logged reached the home anyway, at the next
maintenance. v0.32.0 narrowed it to sessions newer than the home, so a new home would
not swallow a lifetime of Pi history. Nothing scoped it to the home.

CV21.DS5.US3's live route exposed it. A Claude Code cell ran on the real `HOME` against
a scratch home born that evening, and two seconds into its first session, maintenance
imported four of the person's Pi sessions from that day, from both of their real homes,
and titled them through the provider. The route closed its own hole the same evening;
the product's was CR153.

Measured before planning, read-only, on the maintainer's two homes, one a copy of the
other since July: each had taken the other's Pi conversations all along, 543 between
them, and both were still doing it. Titling them took 411 provider calls. No memory
came from any of them, because extraction needs a journey and an import carries none,
but every one listed as the newest conversation, since an import's start is the moment
it was imported. The gap the import existed to fill came up once in three months.

## What Changed

**The decision.** Three ways out were weighed: a claim recorded at session start, so a
home imports only sessions it saw begin; a mark written into each Pi session file, so a
home imports only files carrying its name; or no import at all. The claim would have
kept the one recovery in three months for a table, a migration, and a claim lifecycle.
The mark would have left every home reading every other home's files at every start.
Retiring the import is the only option in which no home reads Pi's files at all, and
it gives up almost nothing
([decision](../project/decisions.md#session-maintenance-imports-no-pi-session-a-home-holds-the-pi-conversations-it-logged-live)).

**The change.** The Pi backfill, the import seam only it still used, and v0.32.0's
birth floor are deleted, and a retired-surface guard keeps them deleted. The tests came
first: three cases — a session in Pi's default directory, one under `PI_SESSIONS_DIR`,
and the full `session-start` the SessionStart hooks run — failed on the old code, each
importing a session another home had logged, and pass now. The Navigator then ran the
published 0.36.0 beside the change on the same session: the old version took it and
the new one did not. On the real Pi history the new one took nothing, and a live Pi
session against a scratch home was logged as before.

### What you will notice

- The maintenance report has four lines: `Closed stale conversations`, `Retitled pending
  conversations`, `Extracted pending conversations`, and `Finished deferred close tails`.
- If you have run more than one home on this machine, each may hold Pi conversations
  the other logged. They are not removed.
- A Pi session you run without Mirror's extension never appears in Mirror. Run Pi with
  the Mirror package, and its sessions are logged as they happen.

## Upgrading

```bash
mirror runtime update          # or: npm install -g mirror-mind
```

Update every install that has a home: the package, and each clone. There is no
database migration, and no runtime wiring changes.

To list the Pi conversations an earlier version imported into a home, read-only:

```bash
sqlite3 ~/.mirror-minds/<user>/memory.db \
  "SELECT id, title, started_at FROM conversations
    WHERE interface = 'pi' AND julianday(started_at) >= julianday(ended_at)
    ORDER BY started_at"
```

An import gives itself away: it started when it was imported, after its last message.
The home's dated backups and its pre-write backup hold the same rows.

## Conscious Exclusions

- **Removing what was imported.** A command that lists and removes those conversations
  is [CR154](../project/refinement/rs010-cv22-oracle-and-port-hygiene/cr154-a-home-cannot-list-or-remove-the-pi-conversations-it-imported.md).
- **A session whose logging never reached the core** is now simply missing from the
  home, with an error line in `mirror-logger.log` and nowhere else. `runtime diagnose`
  reading those lines is
  [CR155](../project/refinement/rs010-cv22-oracle-and-port-hygiene/cr155-runtime-diagnose-does-not-read-the-pi-extension-s-error-lines.md).
- **A home made by copying another's database** is the same home twice — the same ids,
  the same history. The docs saying so is
  [CR156](../project/refinement/rs010-cv22-oracle-and-port-hygiene/cr156-a-home-made-by-copying-another-s-database-is-the-same-home-twice.md).
- **Importing a Pi session on request.** If it is wanted, it is a feature with its own
  story, not something maintenance does in the background.
- **The transcript backfill** matches a session's replies by time window — the same
  shape, inside one home. That is
  [CR151](../project/refinement/rs010-cv22-oracle-and-port-hygiene/cr151-the-transcript-backfill-writes-a-session-s-replies-into-other-sessions-conversations.md),
  its own change.

## What We Learned

The import decided ownership from an absence: a session the home did not track was the
home's to take. That was true while a machine had one home, and it stopped being true
on the day a second home appeared, with nothing to mark the day.

The leak was also invisible to the check that should have caught it. The first look at
the real homes asked by journey, and an import has no journey. Measured by construction
instead — an import starts after its own last message — the same homes held 543. A
measurement taken before planning turned "consider pulling it first" into a fact about
the maintainer's own data.

## Next Horizon

CR154: a home lists, and on request removes, the Pi conversations an earlier version
imported. Then CV21.DS5.TS4 and DS11, and CV21's own release.
