MapleJockey pilot: independent acceptance and input-safety review

ยท Download Markdown

17 November 2026 | Local synthetic file-migration rehearsal

Result

The reviewed core, report bridge and recorded disk exercise meet the bounded local acceptance criteria. No unresolved material issue remains in those reviewed paths. This is a code and evidence review, not a penetration test, signed attestation or completed supplier migration.

The independent executable suite contains 15 passing tests in negative.test.mjs. The separate inspect_disk_evidence.mjs script independently checks the retained disk source, cleanup state and report boundaries. The final wizard source and its 18-check event/rendering harness were also reviewed. Browser layout, keyboard behaviour, file chooser behaviour and native downloads are not verified by this review.

What was independently checked

Recorded disk evidence

The inspected run restored eight synthetic files, totalling 1,392 bytes, into a fresh local directory. Its record contains 14 passed checks, zero failed checks and six Not tested checks. The latter comprise two unsupported file types and four provider/control boundaries.

The recorded functional checks use restored JSON to locate the restored CSV, parse four observations, calculate a total of 40, and preserve accented, multiline, quoted and formula-like text. These checks add evidence beyond a matching export manifest. They do not establish all application behaviours.

Three altered-bundle cases were correctly rejected or failed comparison: changed bytes under an unchanged hash, changed bytes with attacker-recomputed hashes, and a self-consistent bundle missing a file. Successful detection of these failures is distinct from successful recovery.

The reviewer independently recalculated all eight retained source hashes using node:crypto and confirmed they match the recorded source manifest. The run-owned restored directory was absent after cleanup. The source files remained available. This verifies retained bytes and current cleanup state, not a cryptographically authenticated historical sequence.

See ../results/disk-rehearsal.json and ../results/PILOT_RESULT.md for the exact run ID, elapsed time and measured phase timings. The elapsed result is specific to this small local dataset and environment; it is not a supplier transfer rate or business-approved recovery objective.

Issue found and corrected

The first report bridge could mark functional usability Partly supported when every file type was unsupported and every format check was Not tested. It also grouped unrelated integrity/timing failures into the functional finding.

The bridge now derives functional findings only from actual format/domain checks. An all-unsupported selection produces an Undetermined functional finding and says that no supported file-reading check executed. Integrity and timing findings use their own checks. The independent suite includes this regression.

Review boundaries

The wizard preflights source count, individual/total bytes and imported bundle size before reading file contents. It retains the independent source snapshot, invalidates old results after scope changes and rejects stale asynchronous import completion. Failed or oversized imports cannot retain a previous successful result. Plain-text previews use textContent and a 32 KiB limit; SVG, HTML and binary files are not actively rendered. The event/rendering harness covers these main boundaries but is not a substitute for a native browser check. The exclusions field now uses the same Markdown escaping path as the rest of the report.

The transfer bundle is unencrypted and deliberately contains full file bytes. Metadata-only reports are a separate export. This workbench is not an encrypted evidence vault or a multi-user security boundary.

The disk runner creates a fresh private run directory, writes files exclusively, rejects symbolic links on readback and removes only its own candidate directory during rollback. The exercised environment is Linux; this review does not establish a complete operating-system, filesystem or browser compatibility matrix.

No real Dropbox or Sync account was accessed. No live upload, download, regional setting, sharing role, version history, supplier backup, key service or account recovery was tested. The documentation-reviewed provider migration guide retains that status.

Business acceptance and manual content review remain separate from automated success. Blank sign-off fields are not approvals. Imported reports and assessment histories are editable assertions and require independent verification before reliance.

Reproduce

From the project package root with an existing Node 22 or later installation:

node --test pilot/review/negative.test.mjs
node pilot/review/inspect_disk_evidence.mjs

The disk evidence inspection uses the retained source directory referenced in the latest result. In a packaged copy without the original temporary run directory, it checks results/source-fixture instead and explicitly reports historical cleanup as not reobserved. Run node pilot/run_disk_rehearsal.mjs to create a new isolated synthetic run when needed. Each run has its own scope and timings; do not copy a prior pass into a changed workload.