Ownership rollout gates
The bounded whole-Project transfer is production-accepted. The dated local tooling below has its own narrower scope.
Local tooling — October 4, 2026
This page describes local integration tooling, not deployed transfer support. Namespace ownership remains separate from creator attribution, project filing, collaborator grants and current management authority. Team secret scope does not become a resource owner. Legacy namespace IDs remain opaque.
The read-only inventory lives in
dreamlake-server/scripts/ownership/inventory.mjs. It reads inclusion-only
metadata projections, caps each collection at 100,000 rows, and refuses to
produce a partial report. Output contains aggregate counts only: no resource
IDs, names, bodies, share tokens, RTC keys, secret references or ciphertext.
It never updates ownership, memberships, credentials or schema.
Run against a disposable owned local fixture after installing server dependencies:
This tool has no production connection discovery or fallback to DATABASE_URL.
Live inventory needs a separately reviewed explicit target adapter. Reads occur
separately, not in a transaction snapshot; repeat immediately before migration.
Do not interpret an empty fixture as evidence about real resources.
Inventory and reconciliation
Inspect current personal/organization owners, resources without active namespaces, ambiguous namespace ownership, organizations without an active OWNER, departed note/project creators, public visibility and explicit grants. Inspect source credential reference tenant/version binding without decrypting. Local-provider references cannot prove tenant binding from persisted metadata; credential broker tests remain required.
This source revision uses project-tree resource mounts and node-based
BindrMember tags. The inventory also recognizes legacy direct resource
memberships. Unknown resource types and malformed membership shapes require
review rather than a guessed owner. Distinct project affiliations are counted
as zero, one or multiple; duplicate tags do not invent additional projects.
Cross-namespace references require explicit policy review and grant validation.
inspect(rows) produces the report; inventory(db) performs bounded reads.
reconcile(before, after) reports collection count drift and unresolved
quarantine/duplicate findings. It always leaves permission equivalence
unverified. Counts alone cannot prove preservation of body/history/media,
source bytes/lineage, credential access or collaboration revocation.
Quarantine here is a report category requiring manual review, not a database
mutation or an automatic remediation. The additive A-schema backfill is described below. Resource storage, permission
and journal migrations remain gated on the corresponding adapters.
Transfer enablement and rollback
rollout.mjs assesses supplied evidence offline. It is not a deployment gate
wired to the server, and does not enable transfers. Evidence must identify
reviewed A–D refs and a complete enumerated fleet of compatible API, worker
and RTC writers. An individual upgraded instance is insufficient.
Require all four directions: personal to organization, organization to personal, organization to organization and personal to personal. Validate source-owner approval plus current destination acceptance, including old-owner denial on list/search/count/preview/download and live sessions. RTC durable revocation remains an independent release gate.
Evidence must cover roles and removed membership, direct team secret scope, zero/one/multiple project affiliations, share/public grant revocation, collisions, active jobs, credential tenant binding, idempotent retries, stale previews, crashes around external effects, cancellation, deleted destinations, quota and storage failures, concurrent authority changes, cleanup/restore and offboarding. Use signed-in disposable resources and independently verify denial paths. Browser coverage must include create in org, scoped vault, source transfer, a two-project memo transfer, project transfer with excluded dependencies and offboarding. Missing product surfaces stay with their A–D owners.
Rollback first disables new requests. Pending operations must terminate; pre-commit staging must abort and restore source availability; committed operations must roll forward. Unknown commit boundaries require investigation. Revert writer code/schema only after a complete journal inventory proves that all operations are terminal with verified cleanup. Never restore an old owner merely because an operation needs recovery.
Release evidence and client documentation
API, UI, CLI and SDK command examples and support matrices must come from reviewed implemented interfaces. This tooling does not invent unreleased commands. Project/note restore, terminal source deletion, vault retirement and organization/team quarantine need separate tested lifecycle evidence.
Record local validation, CI, merge, deployment and signed-in live acceptance separately. PR events trigger CI normally; never poll or weaken gates. Deployment requires an explicit ship instruction and production drift review.
No public skill impact for this inventory-only increment: it adds internal read-only development tools and does not change Notes/CLI product operations or any source action guide. Run docs generation/freshness checks; regenerate relevant public skills from committed owning docs when A–D public interfaces and examples are integrated. Do not publish a skill claiming deployed support from these offline checks.
Integration update — October 5, 2026
E integrates A workspace e32e0e4a, B 3a431fa9, C 6a38217c and D
97a57434 prepared refs, plus A/B/C UI, CLI and SDK foundations and D browser
draft preservation. A's actual routes use /namespaces/:namespace/ownership;
owner IDs remain opaque. C registers project/source preview-only adapters.
Their execution hooks refuse cutover and all execution readiness flags stay
false. No memo transfer adapter is registered.
B now serializes shared Vault release with actor/org revocation, supports privileged metadata-only copy, policy and audit operations, and explicitly reseals into supported destinations while retaining the source. C's bounded manifest inspection covers selected resources, filing detachment, collisions, active jobs and destination Connection preflight. D's local RTC test passes old-room revocation on two instances plus restart; the browser retains drafts without replaying old operations. These proofs resolve the earlier failed RTC subgate and missing C/B interface evidence. None establishes an end-to-end resource transfer.
backfill.mjs exports backfillAuthorityDefaults(db, { apply, batchSize }) for
a reviewed runner; it has no command that discovers a live database. Dry-run is
the default. Batches of at most 1,000 rows fill only missing User/Organization
authorityEpoch and Namespace ownershipAuthorityFence with zero. Missing-field
CAS preserves concurrent increments, BSON IDs and owner fields. Ambiguous
namespace ownership is reported without repair. Restart selects remaining
missing defaults; quarantined rows require disposition before further migration.
This backfill neither creates indexes nor enables adapters. Verify the exact
unique journal indexes separately, and reconcile counts, grants, affiliations
and authority reports before and after migration.
The inventory now includes aggregate journal activity, missing defaults,
stranded leases and unknown operation states. It never fetches executor tokens.
Rollback uses A's durable ownerCommitted field; an executor claim remains a
blocker even when its operation appears terminal. An operator must prove the
old executor and external work quiescent before clearing its exact token.
Rollout proof records require a full source commit, HTTPS evidence URL, artifact SHA-256 and recording time. Transfer/scenario/RTC/adapter proofs must be marked resource evidence; synthetic journal proofs cannot substitute. Enumerated writer IDs must match the fleet snapshot exactly. These checks validate evidence structure, not URL accessibility or artifact truth; reviewers must independently inspect the linked evidence and current deployment fleet.
Remaining boundaries are specific. B 3a431fa9 blocks source-retiring moves
without a complete consumer registry/acknowledgement and blocks customer-KMS
destinations without policy-aware reseal recovery. C 6a38217c still needs an
atomic owner/structure/grant cutover, writer and parent epoch coverage, storage
staging/recovery, name/quota reservations and project snapshot restore. D
97a57434 needs mutation/upload/media and issued download-capability fencing,
actual memo transfer adapter and abort/roll-forward recovery, filing/grant
reconciliation and retained lifecycle authority. A e32e0e4a needs fleet-wide
writer fencing and an operator recovery procedure for crashed executors;
organization/team deletion remains unsupported until resource disposition,
quarantine and successor restore authority exist.
Local checks include native Mongo backfill/reconciliation, executor restart,
post-CAS cleanup failure, source/destination journal approval and stale-epoch
denial; scoped vault release denial; retired-room draft resync; and focused
UI/CLI/SDK tests. These are test-fixture results, not signed-in browser or
production acceptance. See the committed dated integration evidence in
dreamlake-server/ci/ownership-e/2026-10-05/.
Whole-Project activation and production acceptance
The bounded Project-plus-owned-Notes adapter has its own scope, activation and crash-claim procedure. Production acceptance passed on October 14, 2026, with compatible RTC, exact application release and full old-writer drain receipts. Execution is disabled by default for new deployments and does not enable Source, standalone Memo or other resource adapters. One actor owned both fixture namespaces; real Unicode/S3, external detach, share settings, historical aliases and completed replay passed. Independent-recipient and NoteFile/provider production matrices remain unverified. There is no TTL executor-claim takeover. Earlier dated local-tooling sections remain evidence of their original scope, not the current release boundary.