# Pilot scope: MapleJockey internal file-recovery rehearsal

17 November 2026 | Pilot ID: MJ-FILE-PILOT-001 | Internal synthetic workload

## Objective and ownership

Demonstrate a repeatable, reversible file export/restore process with independently retained source evidence, explicit failure detection and useful post-restore checks. Max Foulkes is the founder sponsor. The operator and reviewer are recorded per run; no external customer, partner commitment or paid engagement is implied.

The workload is the **MapleJockey synthetic research-file set**, fixture MJ-SYNTHETIC-FILES-001: eight invented files, including six text/Markdown/CSV/JSON files and two opaque binary/SVG files. The latter have byte-integrity checks only; application behaviour remains Not tested. The authoritative paths, byte lengths and hashes are generated from the selected fixture at run time. Record them in the baseline rather than assuming that a hard-coded count proves completeness.

The default target is **60 seconds**, a demonstrator threshold rather than a business-approved recovery time objective. The functional data task reads metadata, parses four CSV observations and calculates an expected total value of 40. This task adds useful evidence beyond matching checksums; it does not establish every possible use of the files.

## Included

- One immutable source snapshot and a portable JSON bundle.
- One restored candidate compared with the separately retained source baseline.
- Per-file path, byte-length and SHA-256 comparison; complete accounting for missing, extra and changed files.
- Supported text, JSON and CSV checks. An opaque binary file can pass byte-integrity checks while its application behaviour remains Not tested.
- A failure exercise and a rollback exercise, with each attempt retaining its own outcome.
- A metadata report with target, measured elapsed time, scope, checks, failures and unresolved dependencies.
- A browser workbench and a separately recorded isolated disk exercise. Each environment is identified in its evidence.

## Limits and exclusions

The workbench limits a run to **200 files, 5 MiB per file, 20 MiB total and 32 MiB for a portable JSON bundle**. Unsafe paths, conflicting names, malformed bundles and invalid content encodings are rejected. These are prototype guardrails, not a statement of provider limits or production scalability.

No customer records, personal information, credentials, human-participant data, confidential research, controlled goods or classified/protected defence information are included. No source account is deleted or altered. No accounts are created, no subscription is purchased, no external messages are sent and no service is published.

The following remain outside this local result:

- Supplier upload, download, synchronisation, tenant region and server-side deletion.
- Identity, SSO, account recovery, administrator restrictions, allowed/denied roles and shared-link revocation.
- Version history, deleted-item history, cloud-native document behaviour, comments and application integrations.
- Creation/modification timestamps, extended attributes, operating-system ACLs and symbolic links.
- Backup retention, key custody, actual decryption authority, support access and legal entity/jurisdiction arrangements.
- Production scale, sustained transfer performance and a business-approved recovery time or recovery point objective.

The [earlier Dropbox-to-Sync guide](../../index.html#guide/guide-dropbox-sync) proposes a separate 30-file provider exercise under 100 MB. The smaller local workbench is a preparatory proof, not execution of that provider guide. Its own limits and measured file count govern this run.

## Baseline and acceptance rules

Record the source baseline before creating the portable bundle. Keep it separate from the imported bundle and do not replace it with the bundle's own assertions. A hash demonstrates a content comparison, not who created the content or whether a source is authentic.

For the bounded byte-recovery result, require all intended source paths to be present with matching byte lengths and hashes, zero unexplained extra files and no silent path rewriting. Every included file must have an integrity outcome. Functional status is reported separately: required supported checks must pass, and unsupported or unperformed tasks remain Not tested.

Set a target before execution and record what the timer covers. A fast local pass does not establish a provider migration rate. Any plan, baseline or scope change requires a new run; it cannot retrospectively approve the previous result.

## Failure and rollback

Run at least one changed-content or missing-file case and require a failed integrity result. A failure case succeeds only as a **detection test**; it is never a successful recovery.

Rollback is limited to discarding the new restored candidate and returning to the unchanged source baseline. For the isolated disk exercise, remove only the run-owned destination directory after recording evidence, then verify the source remains unchanged. Do not make claims about files already downloaded elsewhere or about deletion propagation in a supplier synchronisation client.

## Deliverables and authority

The pilot produces the workbench, source baseline, portable synthetic bundle, observed local exercise results, an evidence report, test results and the acceptance form. All commercial prices and previous contracted budgets remain unapproved hypotheses.

This build is authorised local work. Moving actual customer data, accepting contracts, purchasing services, publishing publicly or changing production requires a concrete later scope. Resolve the contracting/IP entity before external contracting; that question does not block this synthetic rehearsal.

Pilot materials. Pilot workspace · [Recorded disk checks](../evidence/disk-rehearsal.json) · [Acceptance worksheet](acceptance.html).
