Introducing a new way to review your PRs.

Read the announcement

stablyai / orca

Orca lets you run a fleet of bots with your own subscription (e.g. Claude Code, Grok, Agy, Codex). Available on desktop, mobile and remote runtime.

View on GitHub

34 documents · Updated Nov 20, 2026

docs / desktop-persistence-and-lifecycle

Desktop persistence and lifecycle

Desktop persistence gives Orca one durable profile for workspace state, settings, UI choices, automations, bots, and recovery records. The main process validates this state, restores it into local and host-owned surfaces, and coordinates writes and shutdown so a restart does not create a second owner or lose a confirmed change. This page complements Desktop app platform, Desktop renderer experience, Relay protocol and transport, Browser automation and embedded web, and Command-line and headless runtime.

Persistence model

The persistence model separates durable profile facts from short-lived views and refetchable data. A shared authority checks ownership and revision before state becomes live.

Profile authority

The profile authority is the one accepted source for a profile while Orca runs. Desktop and Orcad startup use the same boundary. SQLite is the writable profile source, while legacy JSON remains a compatibility and recovery source. The live store becomes visible only after the writer owns the admitted startup revision.

References: src/…/profile-state-startup-authority.ts• L35-49 src/…/profile-state-live-store-factory.ts• L12-43 src/…/profile-state-store-factory.ts• L59-78

Profile database safety

The profile database becomes durable state only after integrity, schema, and profile identity checks pass. The open path validates the database before it changes its journal or schema, so an unknown or incomplete database is not changed during inspection.

An unsupported newer schema can be opened read-only for inspection, while writable startup rejects it until the runtime is upgraded. Writable mode also requires WAL support, full sync, foreign-key checks, and hardened database files so an accepted profile has a strong write boundary.

References: src/…/profile-state-database.ts• L29-45 src/…/profile-state-database.ts• L46-121 src/…/profile-state-database.ts• L192-204

The authority rejects a stale revision instead of allowing an older process to replace newer state. It supports selective domain writes for normal changes and complete checkpoints when a full profile boundary is needed. Automation runs remain part of the same revisioned profile.

References: src/…/profile-state-writer-worker-client.ts• L11-27 src/…/profile-state-revision-readmission.ts• L6-20

Workspace ownership

Workspace state describes the working surface that should survive a restart. It includes workspaces, terminal tabs and layouts, browser and editor records, focus, and restore hints. Host partitions keep local state separate from SSH or runtime-host state, so the same worktree identity on two hosts is not treated as one owner.

References: src/…/workspace-session-partitions.ts• L21-60 src/…/workspace-session-snapshot-publication.ts• L35-45

Each host partition is written through a shared boundary. That boundary preserves runtime-owned records, removes closed tabs, and keeps frozen workspace rows. A sparse or stale renderer snapshot therefore cannot erase a live terminal, a frozen workspace, or another host's state.

References: src/…/workspace-session-snapshot-publication.ts• L40-50 src/…/wsl-move-session-freeze.ts• L27-62

Workspace identity and lineage

Workspace identity links a worktree or folder workspace to its host incarnation and parent history. Host-qualified metadata and lineage records prevent a local row from overwriting a remote row with the same raw ID. A child has one active parent record, and cross-store links move or release with that child.

References: src/…/worktree-identity-metadata.ts• L33-53 src/…/worktree-identity-metadata.ts• L225-280 src/…/metadata-lineage-operations.ts• L243-269

Lineage writes are scoped to the host that owns the identity bucket. Removing a workspace also releases its session owners and lineage links, so later startup does not restore a workspace that no longer has a valid catalog owner.

References: src/…/cross-store-lineage-persistence.ts• L50-79 src/…/metadata-lineage-operations.ts• L134-209 src/…/folder-workspace-operations.ts• L276-288

Settings and UI state

Settings and UI state preserve the choices that make the next session continuous. They cover terminal behavior, layout and host visibility, sidebar and sorting choices, onboarding, notices, feature-use records, browser defaults, and saved quick commands. Updates are routed through the profile write scheduler instead of creating an independent state owner.

References: src/…/profile-preferences.ts• L65-199 src/…/profile-preferences.ts• L256-280

Load and update paths normalize values. Invalid input gets a safe value, explicit opt-outs remain explicit, and a repair is marked for a later durable save. Paired clients cannot use a partial or stale UI snapshot to restore all-host visibility or revive dismissed tips. Refetchable GitHub data uses a sidecar so each refresh does not force a full profile write.

References: src/…/ui-state-read.ts• L38-88 src/…/ui-state-update.ts• L62-80 src/…/ui-state-update.ts• L105-252 src/…/settings-update.ts• L58-76

Automation state

Automation persistence keeps scheduled definitions beside their run history. Owner checks keep a change for one host from changing an automation owned by another host. Batch changes use one profile write, and repeated events fold into an unfinished run instead of creating duplicate runs.

Enablement holds are runtime gates rather than durable state, so a restart does not restore a temporary pause or re-enable transition. A loaded schedule saved on another host clock is reseeded in its schedule zone before it can dispatch a false missed run.

References: src/…/automation-persistence.ts• L104-147 src/…/automation-persistence.ts• L169-251 src/…/automation-definition-update.ts• L51-81 src/…/automation-run-occurrence-folding.ts• L22-65 src/…/automation-zone-reseed.ts• L14-52

Bot state

Bot records are durable profile rows. A bot save replaces or appends one identity and a removal checks both bot identity and host identity. Each change schedules the bot domains and revokes any publication fence for the affected agent session so an old session cannot republish removed data.

References: src/…/bot-persistence.ts• L18-51 src/…/normalize-loaded-profile-state.ts• L47-58 src/…/normalize-loaded-profile-state.ts• L150-157

Protected values

Protected values are encrypted at rest and validated before they affect a service. They include provider cookies, proxy credentials, browser session links, and SSH ownership data. If the secret store is unavailable, existing ciphertext stays preserved instead of becoming an empty value. Explicit clearing removes the retained value.

References: src/…/loaded-state-parsing.ts• L117-179 src/…/ui-state-update.ts• L66-71 src/…/settings-update.ts• L58-76

Disposable cache state

GitHub issue and pull-request data is a disposable cache for the next launch. It is kept outside the primary profile so polling does not create a full profile write. The cache imports an older sidecar only after shape checks, writes changed rows, and refuses to overwrite unreadable data in the current session.

References: src/…/github-cache-database.ts• L22-37 src/…/github-cache-database.ts• L40-99 src/…/loaded-state-parsing.ts• L267-285

Startup restoration

Startup turns an admitted profile into a usable desktop. It loads and repairs profile data first, then restores host partitions and live terminal work only when ownership and process evidence support the restore.

Profile loading

Profile loading classifies available sources before it reads them. It applies defaults, migrations, protected-value handling, ownership cleanup, automation migration, bot normalization, and cache separation in one load path. Any repair is marked dirty so the next save makes the loaded state durable.

References: src/…/loaded-state-parsing.ts• L80-193 src/…/loaded-state-parsing.ts• L203-285 src/…/normalize-loaded-profile-state.ts• L39-58

Migration preserves user intent across changes in defaults and data shape. Cohort choices and explicit opt-outs are fixed before later defaults can change their meaning. Automation owners, legacy run context, and time-zone state are also repaired before the profile becomes live.

References: src/…/loaded-cohort-migrations.ts• L17-52 src/…/loaded-cohort-migrations.ts• L99-130 src/…/loaded-automation-migrations.ts• L7-41

Missing, malformed, and damaged data are different cases. The load path does not replace unreadable input with a new default profile when retained authority evidence exists. A disposable cache may be skipped, but primary profile evidence remains subject to recovery.

References: src/…/loaded-state-parsing.ts• L194-201 src/…/normalize-loaded-state-collections.ts• L21-46 src/…/profile-state-recovery-required.ts• L28-99

Settings restoration

Settings restoration merges valid saved choices with current defaults and converts older shapes into the current meaning. The same intent-preserving rules apply to loaded state and later updates, which keeps a repaired value stable after a restart.

References: src/…/normalize-loaded-global-settings.ts• L35-39 src/…/normalize-loaded-global-settings.ts• L91-208 src/…/normalize-loaded-ui-state.ts• L30-45

Session hydration

Session hydration joins the profile authority, host services, and renderer. The renderer receives settings and persisted UI first. Host-owned workspace partitions become available as their hosts are ready, while main-owned topology remains authoritative during the handoff.

References: src/…/workspace-session-partitions.ts• L21-60 src/…/workspace-session-snapshot-publication.ts• L100-139

Saved terminal identifiers are attach hints, not proof that a process still exists. Restoration checks ownership, pane and tab membership, identity generations, retirement records, and process evidence. If those facts do not agree, the system repairs or starts a binding instead of reviving a closed pane.

References: src/…/workspace-session-snapshot-publication.ts• L50-100 src/…/terminal-binding-recovery.ts• L1-120 src/…/saved-local-terminal-ids.ts• L30-73

The following diagram shows how saved state becomes a live workspace.

flowchart LR
    A[Admitted profile] --> B[Validated and repaired state]
    B --> C[Settings and persisted UI]
    C --> D[Local and host workspace partitions]
    D --> E[Renderer hydration]
    E --> F[Topology and pane ownership]
    F --> G[Terminal evidence check]
    G --> H[Reattach or start a live session]
    E -. critical failure .-> I[No-save recovery]

References: src/…/profile-state-live-store-factory.ts• L12-43 src/…/workspace-session-snapshot-publication.ts• L35-139

Host workspace transfer

Host workspace transfer moves a bounded set of durable state from one execution host to another. It carries catalog rows, worktree metadata, workspace sessions, automation history, UI routing, and selected terminal history. Unclear ownership blocks a move rather than choosing a host by guess.

References: src/…/orcad-catalog-commit-domains.ts• L4-26 src/…/orcad-source-workspace-session-eligibility.ts• L1-120

For a WSL move, the planned scope is checked again at start. The move accepts only local projects whose paths belong to the target distro. It records mapped drives that the manifest uses, rewrites host paths into the destination form, and refuses a stale plan or an ambiguous path instead of silently dropping a project.

References: src/…/wsl-move-driver-plan.ts• L203-320 src/…/wsl-move-scope.ts• L53-111 src/…/wsl-move-manifest-export.ts• L113-224 src/…/wsl-move-mapped-drives.ts• L81-99

The move fences the source, pauses selected workspaces, proves terminal exit, and records each phase before the next phase starts. Large history and restore data use a separate checked blob channel. The destination commits only after the manifest and blobs match. Desktop stores then rewrite one at a time, and the journal records each completed rewrite so a restart retries only the missing work.

References: src/…/wsl-move-quiesce.ts• L97-223 src/…/wsl-move-driver-transfer.ts• L53-140 src/…/wsl-move-desktop-rewrite.ts• L126-159 src/…/wsl-move-cutover-fixture.ts• L11-75

Durable writes

Durable writes protect profile state from bursty UI updates, concurrent owners, stale snapshots, secret-store outages, and process failure. Normal changes can wait briefly, but lifecycle changes use an acknowledged barrier when later work depends on the write.

Writer ownership

The live store sends profile changes to one writer that owns the database connection. The writer applies a complete state, selected domains, or automation run history and returns the new revision. If the required writer or backup support is unavailable, writable startup fails instead of using a partial persistence path.

References: src/…/profile-state-store-factory.ts• L33-78 src/…/profile-state-writer-worker-client.ts• L6-63 src/…/profile-state-writer-worker-entry.ts• L40-87

The writer can stop after an idle period and reopen from its last acknowledged revision. Reopen failures use bounded backoff. A revision conflict is terminal and is reported to the lifecycle owner instead of being retried as if it were a worker failure. Shutdown and replacement wait for the worker exit event before another owner can start.

References: src/…/profile-state-writer-reopen-policy.ts• L17-65 src/…/profile-state-writer-connection.ts• L102-115 src/…/profile-state-writer-thread.ts• L4-39

Save barriers

The save scheduler combines short bursts and bounds how long changed state stays only in memory. A write generation prevents an older snapshot from publishing after a newer one. An awaited flush drains admitted work and reports whether the required generation reached durable storage.

References: src/…/write-scheduling.ts• L25-26 src/…/write-scheduling.ts• L127-191 src/…/write-flush-barriers.ts• L145-209

Profile publication uses a durable staging step and does not weaken atomic no-overwrite behavior when the file system cannot provide the needed operation. A failed primary mutation freezes further writes rather than committing only part of the edit. Backups use a temporary database and must complete, validate, and publish as a self-contained recovery point.

References: src/…/primary-state-write-runtime.ts• L34-72 src/…/profile-state-database-publication.ts• L16-31 src/…/profile-state-backup-worker.ts• L23-91

Backup publication

Backups provide self-contained recovery points for the profile. They run outside the main runtime path, fail if the worker does not complete, and remove temporary files after the attempt. The profile writer drains backup work before a final checkpoint completes.

References: src/…/profile-state-backup-worker.ts• L34-91 src/…/profile-state-backup-worker-entry.ts• L8-33 src/…/write-flush-barriers.ts• L83-113

Cache writes

Routine GitHub cache writes wait for a cadence and retry a busy or failed write later. A quit or explicit barrier writes immediately, then closes the cache handle when maintenance or quit owns the profile. This cache may be stale after a crash because the next poll can refill it, but unread cache data is never replaced by an empty snapshot.

References: src/…/github-cache-persistence.ts• L29-43 src/…/github-cache-persistence.ts• L44-69 src/…/github-cache-persistence.ts• L113-147

Shutdown coordination

Shutdown has a checkpoint stage and a teardown stage. The renderer stages its final view quickly, then main freezes new ownership changes, completes durable writes, stops runtime resources, and releases profile ownership last.

Renderer checkpoint

The renderer cannot wait for asynchronous work during unload. It sends a final snapshot that main rebases against fields owned by runtime services, such as closed panes, sleep records, and terminal binding state. This keeps a last view from replacing newer main-owned facts.

References: src/…/workspace-session-snapshot-publication.ts• L40-50 src/…/workspace-session-snapshot-publication.ts• L100-139

Profile shutdown checkpoint

Main starts the profile checkpoint while it still owns the profile. It waits for the pane grid to finish its last flush, then asks the store for a full checkpoint without closing the writer. The returned completion is the handoff to teardown, and a failed checkpoint is reported instead of being hidden.

References: src/…/profile-shutdown-checkpoint.ts• L3-14 src/…/write-flush-barriers.ts• L83-119

The checkpoint is safe only after its durable work is acknowledged. If checkpoint preparation or hydration fails, the desktop keeps the last known profile and enters no-save recovery instead of saving fallback state over it.

References: src/…/profile-shutdown-checkpoint.ts• L8-14 src/…/write-flush-barriers.ts• L83-113 src/…/profile-state-startup-recovery-dialog.ts• L26-65

Teardown barrier

Main uses one bounded teardown path for normal quit, operating-system signals, launcher disconnects, startup failure, and headless runtime stop. It stops new writes, drains profile writes and backups, closes the writer after its exit is observed, and releases profile admission only after pending writers can no longer change the profile.

References: src/…/write-flush-barriers.ts• L87-119 src/…/profile-state-writer-connection.ts• L92-137 src/…/profile-state-writer-thread.ts• L4-49

Recovery behavior

Recovery keeps uncertain state visible and preserves the evidence needed to make a safe choice. It prefers an explicit recovery decision over a silent merge or an invented success.

Profile recovery

Profile recovery handles unreadable state, conflicting profile sources, unsupported schemas, revision conflicts, and unavailable write support. Retained authority evidence prevents startup from treating a missing database as a new profile. Diverged SQLite and JSON copies are presented as a choice; the system does not merge them automatically.

References: src/…/profile-state-recovery-required.ts• L28-99 src/…/profile-state-startup-recovery-dialog.ts• L107-154

Whole-profile rollback requires maintenance ownership, validates a self-contained backup, quarantines the current files, and publishes the replacement durably. The failed profile remains available for inspection if the replacement does not open.

References: src/…/profile-state-database-recovery.ts• L49-109

Section repair can restore damaged domains from readable backups while keeping intact domains current. It refuses changed approvals and reopens the result through the normal authority path. If the repaired profile still cannot open, the quarantined original remains the basis for recovery.

References: src/…/profile-state-section-repair.ts• L54-123 src/…/profile-state-section-repair.ts• L137-201

Interrupted host transfer

An interrupted host transfer resumes from its durable journal, not from in-memory progress. Journal phases, manifest digests, rewrite ledgers, and host receipts show whether the safe result is reset, waiting for the host, or finish-after-commit. An unreadable journal keeps the affected projects fenced.

References: src/…/wsl-move-cutover-journal.ts• L20-25 src/…/wsl-move-cutover-journal.ts• L71-129 src/…/wsl-move-orchestrator-step.ts• L45-69

The move chain advances one phase at a time. A later move extends the last finished move for the same distro, while a stale source row remains retained instead of being copied a second time. If mapped drives, the host, or the planned scope changed, the move pauses or asks for a new plan.

References: src/…/wsl-move-cutover-journal.ts• L86-158 src/…/wsl-move-delta-plan.ts• L95-220 src/…/wsl-move-retained-source.ts• L78-145 src/…/wsl-move-driver-source-drives.ts• L87-116

The transfer state machine follows this order:

flowchart LR
    A[Plan and check blockers] --> B[Fence and pause workspaces]
    B --> C[Prove terminal exit]
    C --> D[Durable quiesced journal]
    D --> E[Stage manifest and blobs]
    E --> F[Recheck and commit host]
    F --> G[Rewrite desktop stores]
    G --> H[Durable rewrite ledger]
    C -. failure .-> I[Reset or keep fence]
    F -. lost reply .-> J[Ask host for receipt]

References: src/…/wsl-move-quiesce.ts• L188-223 src/…/wsl-move-driver-transfer.ts• L100-140 src/…/wsl-move-desktop-rewrite.ts• L85-106

Migration blob recovery

Large terminal history, shell history, restore hints, and automation run history use resumable blob staging instead of enlarging the signed manifest. The destination checks descriptor identity, size, and digest before commit. After a durable catalog commit, installation can repeat safely; failed installation leaves the verified bytes for the next startup.

References: src/…/orcad-migration-blob-staging.ts• L79-155 src/…/orcad-migration-blob-staging.ts• L185-240 src/…/orcad-catalog-commit-domains.ts• L4-26

Session and automation import

Imported workspace sessions merge only non-conflicting records. Focus stays with the destination desktop, while duplicate terminal, browser, tab, and group identities stop the import. Imported automations keep their row identity, but their next run is reseeded for the destination schedule zone and their run owners must exist.

References: src/…/orcad-destination-workspace-session.ts• L38-55 src/…/orcad-destination-workspace-session.ts• L78-114 src/…/orcad-destination-automation-state.ts• L31-60

Change checklist

When these files or sections are changed, remember to consider these:

  • A new profile domain needs one revision rule, one load normalizer, one selective-write path, one complete-checkpoint path, and one recovery source.
  • A new workspace field must declare its host and writer. Renderer snapshots must not erase main-owned or frozen fields.
  • A settings default change must preserve explicit opt-outs and persist any one-time decision before later UI changes can alter the evidence.
  • A protected field needs a validator, retained-ciphertext behavior when the secret store is unavailable, and an explicit clear path.
  • A writer lifecycle change must preserve the last acknowledged revision across idle stop, reopen, failure backoff, maintenance, and close.
  • A profile database opening change must keep validation ahead of journal or schema changes, reject newer schemas for writes, and preserve full-sync durability settings.
  • A host transfer change must update the catalog, sessions, UI routing, automation owners, history descriptors, blob receipts, desktop rewrites, and retention marks as one lifecycle.
  • A journal phase change must preserve manifest binding, monotonic phase order, append-only rewrite records, and fail-closed chain checks.
  • A terminal restore change must keep the distinction between a saved hint, verified process evidence, a consumed hint, and an unknown remote result.
  • A shutdown change must keep renderer staging, pane-grid flushing, checkpoint acknowledgement, writer freeze, backup draining, and profile release in order.
  • A bot or automation identity change must keep host ownership, publication fences, run ownership, and the matching dirty profile domains aligned.
  • A cache migration must preserve sidecar precedence, unreadable-data protection, import receipts, and handle closure during maintenance or quit.
  • A WSL move change must keep mapped-drive scope, path rewriting, manifest digest, destination identity, retained rows, and the durable rewrite ledger aligned.
  • A cutover fixture change must keep the manifest digest, destination identity, scope lists, dormant state, and journal key aligned with the move contract.

Important failure modes

TriggerConsequenceGuard
A renderer snapshot omits a main-owned or frozen workspace fieldA live pane or moved workspace can disappear after restartRebase snapshots against prior state and preserve runtime-owned fields
A writer reopens from an unacknowledged revisionA retry can overwrite newer profile stateReopen only from the last acknowledged revision and reject conflicts
A secret-store read fails during saveA cookie, proxy, or ownership lease can become emptyKeep existing ciphertext and validate plaintext before use
A load repair is not marked dirtyThe same damaged value returns on every launchMark repairs during load and publish them through the normal save path
A profile database is missing while authority evidence remainsStartup can create a false new profile or hide recoverable dataRefuse initialization and require an explicit recovery choice
A profile database is changed before its schema and integrity are checkedA malformed or unsupported file can be made harder to recoverValidate the database before journal or schema changes and keep newer schemas out of writable startup
A migration journal is unreadable or forksStartup cannot prove which host owns the moving projectsFail closed and keep the affected scope fenced
A host transfer commits but a desktop rewrite stops part wayHost state and desktop state can describe different ownersRecord each durable rewrite and retry only missing stores
A mapped drive changes between preview and move startThe move can copy a project from the wrong locationRe-read relevant drives and refuse a changed or unverifiable scope
A blob is installed before its catalog commit is durableRestart can lose the link between received bytes and moved stateValidate and commit blobs at the catalog boundary
A remote host loses the reply after commitA reset or duplicate server can corrupt ownershipKeep the move fenced until the host proves its receipt
A writer closes before its thread exit is observedAnother owner can start while the old writer still actsWait for the worker exit event before releasing ownership
Shutdown closes ownership before the pane grid and profile checkpoint completeThe last UI state can remain only in memory or be replaced by an older snapshotWait for the checkpoint promise while the current process still owns the profile
A cache sidecar or row store is unreadable but the session saves an empty cacheNewer refetchable data can be lostDisable cache persistence for that session and let a later poll refill it
A bot or automation update ignores its host ownerOne host can change or republish another host's recordCheck the owner fence and mark the matching profile domains
A cutover fixture uses a manifest digest for different scope or destination dataMove tests can accept an invalid journal or miss a contract changeBuild the fixture digest from the same scoped manifest and destination identity used by cutover

Key files

FileWhy to read it
src/…/profile-state-startup-authority.tsShows the shared startup authority for desktop and Orcad.
src/…/profile-state-live-store-factory.tsShows when the live store becomes visible after writer admission.
src/…/profile-state-database.tsShows database validation, schema admission, migration, and full-durability settings.
src/…/profile-state-writer-connection.tsShows writer requests, revisions, faults, close, and observed exit.
src/…/profile-state-writer-thread.tsShows the exit barrier used before writer replacement.
src/…/profile-state-writer-reopen-policy.tsShows retry and backoff behavior after an idle writer failure.
src/…/loaded-state-parsing.tsShows the high-level load, decrypt, normalize, migrate, and cache flow.
src/…/normalize-loaded-profile-state.tsShows how catalogs, sessions, bots, lineage, automations, and host rows become valid state.
src/…/normalize-loaded-ui-state.tsShows UI migration, defaults, and repair persistence.
src/…/profile-preferences.tsShows the durable boundary for settings, UI state, onboarding, and cache state.
src/…/ui-state-update.tsShows safe UI updates, stale-client merges, and protected-value clearing.
src/…/workspace-session-snapshot-publication.tsShows safe workspace snapshot publication and restore preparation.
src/…/workspace-session-partitions.tsShows host-partition validation and ownership separation.
src/…/worktree-identity-metadata.tsShows host-qualified worktree identity and metadata ownership.
src/…/metadata-lineage-operations.tsShows workspace parent links and cleanup when a workspace is removed.
src/…/automation-persistence.tsShows automation ownership, definitions, and run-history writes.
src/…/bot-persistence.tsShows bot row replacement, removal, and publication-fence invalidation.
src/…/github-cache-database.tsShows disposable cache rows, sidecar import, and unreadable-data handling.
src/…/github-cache-persistence.tsShows routine, retry, barrier, and fenced cache writes.
src/…/write-scheduling.tsShows domain-aware save scheduling and bounded debounce.
src/…/write-flush-barriers.tsShows pending, checkpoint, and final flush barriers.
src/…/primary-state-write-runtime.tsShows write freezing after a failed durable mutation.
src/…/profile-shutdown-checkpoint.tsShows the pane-grid and full-profile checkpoint boundary before teardown.
src/…/profile-state-backup-worker.tsShows isolated backup publication, completion checks, and cleanup.
src/…/profile-state-database-recovery.tsShows validated whole-profile rollback and quarantine.
src/…/profile-state-section-repair.tsShows section-level repair while preserving intact domains.
src/…/wsl-move-driver-plan.tsShows host readiness, scope planning, stale-plan checks, and interrupted-move settling.
src/…/wsl-move-manifest-export.tsShows the bounded state exported into a signed host-transfer manifest.
src/…/wsl-move-scope.tsShows which local projects can move and why others are refused.
src/…/wsl-move-cutover-journal.tsShows durable move phases, bindings, and fail-closed recovery.
src/…/wsl-move-desktop-rewrite.tsShows post-commit desktop rewrites and retry of missing stores.
src/…/wsl-move-retained-source.tsShows how source rows stay available for later reconciliation.
src/…/wsl-move-startup.tsShows startup recovery and retry for interrupted host moves.
src/…/orcad-destination-workspace-session.tsShows conflict-safe session import and destination focus ownership.
src/…/orcad-destination-automation-state.tsShows automation and run-history import checks.
src/…/orcad-migration-blob-staging.tsShows resumable large-data transfer and digest checks.
src/…/wsl-move-cutover-fixture.tsShows the test move shape used to bind scope, destination, dormant state, and manifest integrity.