Validation status
Validation is reported by evidence tier. Dates and counts below are the latest records checked into upstream/main; they are not a substitute for rerunning acceptance on a new commit, provider configuration or physical board.
Evidence ladder
repository source
→ Rust/App contract tests
→ simulator end-to-end
→ ESP32-P4 and ESP32-S3 release builds
→ physical boot/display/touch/storage per board
→ fresh Wi-Fi/provider/App end-to-endRecorded on 2026-08-24
- 57 Rust tests completed without failure.
- workspace Clippy passed with warnings denied.
- Pi Agent assets, shared View SDK, simulator, ESP32-P4 release firmware and ESP32-S3 release firmware built.
- core tests covered viewport propagation and scaling, minimum touch targets, App lifecycle, SQLite ownership and model/Tool contracts.
- physical P4 coverage includes the Agent, workspace, schedules, App install/update/uninstall, display/touch, Wi-Fi and direct model providers.
- physical S3 coverage includes boot, 480×800 logical scanout, GT911 touch, integrated Wi-Fi, workspace Tool Calls, ordinary App install and an Exa request.
Evidence remains board-specific
P4 has the broader reference-target history, including update and uninstall. S3 physical evidence proves the shared runtime and source App contract on its current board composition, but it does not inherit P4-only stress or lifecycle results. Long-running latency, display stability and memory-pressure testing remain separate acceptance work on both targets.
Current automated contract coverage
- View and Action LRUs retain the three most recent ordinary Guests.
- Tool routing selects each App's declared Action.
- App revisions coalesce at a foreground frame boundary.
- code-only update preserves data and native credentials.
- schema update rejects missing steps and preserves rows.
- boot finishes an approved update interrupted before activation.
- schedule success records only after the Action completes.
- Projection/SQLite errors propagate rather than becoming an empty View.
- uninstall removes App-owned routes, runtimes, schedules, credentials and data.
Not yet implied by those checks
- phone upload on every network topology;
- a fresh live provider call for every current commit;
- unattended long-duration memory pressure;
- physical acceptance of a future DeepSeek Harness build;
- a future third-board port or a web Playground.
How to report a new result
Record the exact commit, build command, board/host, provisioned backend, whether storage was erased, the observable pass/failure and the highest completed tier. Keep failures such as Wi-Fi association or DHCP timeout attached to the network-dependent acceptance instead of calling a successful boot complete end-to-end validation.
Checked-in evidence summary: README current validation ↗.