Live acceptance record
Validation on 14 September 2026 uses synthetic model scenarios, isolated test stores and one explicitly authorized private Discord channel. Tests are evidence for the behavior exercised, not a blanket guarantee.
Model runtime
Section titled “Model runtime”Muse Spark 1.3 Free through OpenCode Zen passed all six live runtime scenarios: greeting, quoted creative instruction, correction, exact member identity, grounded image reference and scoped category research. The live calls use Sophia’s Responses adapter with honest client identity and task session metadata. Log: storage/v5-muse-live.log.
The earlier GLM profile also passed all six. GPT-OSS generated unsupported service claims in a separate evaluation and remains a model-specific limitation. No model receives an automatic claim of equivalent behavior just because it is selectable.
Discord fixtures
Section titled “Discord fixtures”The user authorized creating a new private test channel. The exact guild, channel and message IDs remain in the local acceptance receipt rather than the public handbook.
Plain message send/edit, production card payloads for both pages/navigation states, and a native poll passed REST acceptance. All fixture messages were deleted; exact-ID fetches returned Unknown Message. The empty private channel remains. Local receipt: storage/v5-discord-acceptance-recheck.json.
The user waived further Discord visual interaction checks. Real gateway clicks and Discord desktop/mobile rendering remain unverified, and are not blockers to finishing this requested local implementation.
Containers
Section titled “Containers”Docker’s Linux engine is now responding and npm run sandbox:build built sophia-sandbox:5. No rejected destructive repair was retried. The previous engine-unavailable condition is resolved.
The opt-in container suite passed execution in Python, JavaScript and shell, analysis-library imports, workspace file transfer, read-only root enforcement, no outbound connection, no host credentials, symlink exclusion and per-operation timeout. The eight real card-script tests passed state mutation, replies, contextual inputs, send bounds, rendering helpers, errors, restricted script globals and oversized-state preservation. Logs: storage/v5-live-sandbox.log and storage/v5-artifact-live.log.
Video inspection additionally exercises a generated clip through the real owned workspace and timestamped decoder. Audio provider transcription remains covered by deterministic byte/provenance tests rather than a live speech-quality evaluation.
Documentation
Section titled “Documentation”The Starlight site separates user guides, installation and developer contracts. It generates references for all 78 capabilities and all registered commands. Builds check missing capability mappings; link validation checks generated HTML and anchors at both root and GitHub Pages subpaths.
Browser checks verified chapter navigation, production search for dreaming, phone-width reading/menu behavior and both color themes. Preview runs locally; this work does not deploy the site.
Repeat the checks
Section titled “Repeat the checks”Run npm run check for TypeScript and deterministic tests. Set LIVE_SANDBOX=1 and run npm run test:sandbox for local Docker tests. Run npm run test:live -- tests/live/ArtifactScript.live.test.ts for card scripts. Model tests require the explicit LIVE_MODEL_TESTS=1 switch, a selected LIVE_MODEL_PROFILE and locally loaded credentials; see testing. npm run docs:build and npm run docs:check validate the website.