Browser-based graphical shell and control plane for a Linux server you own — real terminal, file manager, live metrics and BYOK AI in one mobile-first workspace. Next.js 16 · React 19 · node-pty · Tailwind v4 · shadcn/ui. Public alpha, self-hosted.
Nakama parity evidence
The frozen v0.1 snapshot preserves every requirement, clause, package, stage, acceptance journey, provider, format, transport and platform from the approved draft. Its complete parsed JSON has a fixed SHA-256 in the validator. Record progress in the implementation map. Changing identifiers while retaining counts, or removing a format or contract, does not preserve this snapshot.
The source receipts identify 100 distinct files at the two pinned commits. A source hash proves which bytes were read. It does not prove that a feature works or that a current provider account supports it.
Development checks
Run bun run parity:nakama and bun run test:parity. They are also part of
bun run check. Regenerate the current reference with
node scripts/gen-nakama-parity.mjs after changing the implementation map.
Each requirement is pending, partial, or implemented. Partial rows need code, tests, and explicit remaining scope. Implemented rows also need a verification owner and no remaining scope. Neither status alone establishes runtime parity.
Full parity candidate gate
From the exact clean candidate checkout, run:
bun run parity:nakama:release --evidence /absolute/path/to/manifest.json --commit EXACT_HEAD_SHA
The gate rejects a different HEAD, modified tracked files, untracked files,
index entries marked skip-worktree or assume-unchanged, and
source/test mappings that are not regular blobs in that candidate's Git tree.
Ignored dependencies, build outputs, symlinks and submodules cannot stand in for
committed implementation evidence. All 64 requirements must be implemented.
The manifest has schema mso.nakama-parity-evidence.v1, candidate_commit,
working_tree_clean: true, a valid created_at timestamp, and an environment
with kind: "runtime" and nonempty strings for name, platform, and node.
It contains these arrays:
| Array | Required records |
|---|---|
gates |
typecheck, lint, tests, build, security, regression |
requirements |
Every NK ID, its exact acceptance references, and all eight DOD dimensions |
acceptance |
Every A01–A30 journey |
compatibility |
Every ID and test dimension from requiredVariants() in the variant module |
artifacts |
Identified, nonempty evidence files with byte count and SHA-256 |
Each result record has id, status: "pass", candidate_commit, and a nonempty
array of artifact IDs. Requirements, acceptance journeys, compatibility records
and their dimensions also require mode: "runtime". Requirements include
acceptance as the exact set in the snapshot and dimensions as an array of
DOD result records. Compatibility records have a dimensions array containing
the exact adapter test dimensions. Empty expected dimension sets still use an
empty array. Additional or duplicate IDs do not count as coverage.
An artifact has id, path, bytes, and sha256. Its path is relative to
the manifest directory, contains no traversal or symlink, and names a regular
file of 1 byte through 20 MiB. Artifact verification currently requires Linux or
WSL with /proc/self/fd; unsupported platforms or unavailable descriptor pinning
block the gate. Every directory is opened from a pinned parent with O_NOFOLLOW.
The final file is opened once, checked through that descriptor, and read through
the same descriptor into a bounded buffer. A one-byte sentinel detects growth;
size or modification metadata changes during the read block the artifact.
Pathname replacement cannot redirect the read to another file. This is a checked
byte snapshot, not a filesystem-wide transaction or a physical storage guarantee.
Keep credentials and private tenant data out of public receipts; include only the
bounded evidence needed to assess the result.
The variant list derives providers, transports, harnesses, channels, formats, the knowledge ZIP container, platforms, and all baseline shell/network combinations. Merely supporting a model tool does not imply media, MCP resources, or prompts support. Additional platforms marked as optional extensions in the snapshot do not become a baseline installer claim.
What a passing manifest means
This validator checks completeness, candidate identity, file boundaries and integrity. It cannot establish whether a receipt's statements are true. Reviewers must inspect actual requests, responses, provider identities, delivery IDs, authorization denials and recovery outcomes before making a release claim.
Synthetic tests deliberately exercise the validator's success shape. They are not release manifests. Labeling fixture results as runtime results would still be false, even if someone could produce a structurally valid manifest. The full parity gate remains blocked while requirements or real acceptance work remain.