> ## Documentation Index
> Fetch the complete documentation index at: https://syscall.mintlify.site/llms.txt
> Use this file to discover all available pages before exploring further.

# Testing and validation

> What has been exercised, what CI proves, and which claims require a privileged operator run.

SYSCALL distinguishes automated checks, offline evidence tests, and privileged live capture. A green unit test is not a kernel PASS.

## Validation status

Last documented validation: **2026-11-13**.

| Area | Status | Evidence |
| - | - | - |
| Non-privileged quality gate | Pass | `scripts/check.sh` and GitHub Actions: formatting, Clippy, workspace tests, CLI and collector builds, eBPF compile and skeleton, desktop build and renderer tests, both frontend builds |
| Live kernel load and attach | Pass | Operator-run WSL2 e2e on 2026-11-10; all seven tracepoints attached |
| Identity integrity | Pass | Live e2e plus unit tests; raw, observer, and canonical identity recorded |
| Signal handling | Pass | SIGINT exited `130`, session finalized as `interrupted` |
| Filesystem activity | Pass | Live `openat` capture with raw paths, flags, results, errno, truncation and relative-path limits |
| Network activity | Pass | Live IPv4 and IPv6 `connect` attempts with `-111` |
| Behavioral analysis | Pass, offline | Deterministic rules over real temporary SQLite evidence; no privileged live claim |
| Desktop workbench | Pass | Release binary launched under WSLg; renderer and real-SQLite service tests pass |
| BTF layout in CI | Advisory | CI runner kernel differs from the validated WSL2 kernel; the local gate enforces the check |

Phase 5 live runs persisted:

* **72 events**: 64 file opens, 1 IPv4 connect.
* **315 events**: 2 connects, IPv4 and IPv6.

Both sessions verified `valid` with zero findings.

## Test layers

| Layer | What it proves | What it does not prove |
| - | - | - |
| Unit tests | Decoding, validation, identity, path, clock, and renderer logic | Kernel attach or live capture |
| Storage integration tests | SQLite schema, queries, pagination, verification, export behavior | Privileged recording |
| Analysis tests | Rule conditions, exclusions, confidence, determinism, read-only access | Live kernel behavior |
| Desktop service tests | Real temporary SQLite through the command-independent service layer | GUI interaction |
| Renderer tests | Routes, filters, screens, and IPC-facing behavior with controlled data | Real database semantics |
| eBPF compile and BTF check | Object sections and running-kernel layout compatibility | Successful load or attach |
| Privileged e2e | Load, attach, capture, identity, signals, ownership | Every workload or loss scenario |

## Automated checks

```bash theme={null}
./scripts/check.sh
```

Individual checks:

```bash theme={null}
cargo fmt --all --check
cargo clippy --workspace --all-targets -- -D warnings
cargo test --workspace
make -C ebpf clean check
scripts/check-btf-layout.sh
npm --prefix desktop test
npm --prefix desktop run build
npm --prefix ui run build
```

The workspace test command includes the Tauri crate and needs desktop development packages. Headless mode:

```bash theme={null}
cargo test -p syscall-desktop --no-default-features
```

## Privileged e2e

```bash theme={null}
cargo build -p syscall-cli
sudo scripts/e2e-root.sh
```

The script verifies:

* Invocation through sudo for a non-root user.
* BPF capability and visible lifecycle tracepoints.
* A recorded target tree is inspectable as the invoking user.
* SIGINT exits `130` and finalizes `interrupted`.
* The database is user-owned with mode `600`.

The Phase 5 manual capture is documented in [WSL2 compatibility](https://syscall.mintlify.site/operations/wsl.md).

## Desktop validation

The workbench validation covered:

* Release build with `tauri build --no-bundle`.
* WSLg launch at 1480×920 with software rendering.
* Opening a real recorded session read-only.
* Session search, process tree, timeline, filesystem, network, analysis, verification, and export screens.
* A 64,138-byte JSON export with mode `0600` and a rendered manifest.
* 30 renderer tests and 8 real-SQLite service tests.

The screenshots in the docs use the standalone renderer's deterministic preview data; they are not live evidence.

## What CI does not prove

* That eBPF loads on the runner.
* That tracepoints attach.
* That live filesystem or network activity is captured.
* That privilege dropping works on the target host.
* That the workbench renders correctly in a real window.
* That a database was authentic or untampered.

Those require the operator-run checks or external controls.

## Remaining validation gaps

* Native Linux host run beyond WSL2.
* Stress and sustained ring-buffer loss.
* Crash injection that leaves a `recording` session.
* Privileged CI with explicit provisioning.
* Automated GUI end-to-end testing.

See [Roadmap](https://syscall.mintlify.site/project/roadmap.md).


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.