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

# diag.offline

> Offline wifi / filtering / kinematics analysis of a teleop flight-recorder capture.

Analyzes a teleop flight-recorder capture (`axol teleop --teleop.record PREFIX`) offline — no hardware needed. Each suite isolates one stage of the teleop pipeline and answers one question, so jitter can be attributed deterministically instead of guessed at:

| Suite        | Question                                      | What it reports                                                                                                                                                                                                                                                                  |
| ------------ | --------------------------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| `wifi`       | Is the transport jittery?                     | VR frame inter-arrival median/p95/p99/max, gaps (> 100 ms) and burst frames — recovered from raw-pose change ticks, so hand motion can't fake them. A quiet IK tick interval alongside gaps points at the network; a noisy one points at the host starving the IK process.       |
| `filtering`  | What does the pose filter stack pass through? | 3–15 Hz band RMS per stage (raw VR pose → filtered pose → world EE target) per axis, the pass-through ratio, and the lag each stage adds. Mid-band content that reaches the EE target is what the arm is asked to reproduce — if it's high, tune the filter, not the motors.     |
| `kinematics` | Does the IK add motion the hand didn't make?  | EE pose set vs actual (the world target given to the solver vs FK of the solved joints, per axis), per-tick solve time (median/p95/max ms), per-joint churn (°/min — a restless null space churns without EE motion), and per-joint mid-band jitter with its dominant frequency. |

| Flag           | Description                                                                                                                                                 |
| -------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------- |
| `suite`        | `wifi`, `filtering`, or `kinematics` (required)                                                                                                             |
| `prefix`       | Flight-recorder prefix used with `--teleop.record`. A bare name resolves in `~/.almond/recordings/`; omit it entirely to analyze the newest recording there |
| `--save-run`   | Persist the analysis as a tuning-run artifact (`~/.almond/diagnostics/tuning/`) for the diagnostics UI                                                      |
| `--label TEXT` | Free-form note stored on the run artifact                                                                                                                   |

```bash theme={null}
axol teleop --teleop.record rec1         # record a session first
axol diag.offline wifi rec1
axol diag.offline filtering rec1 --save-run
axol diag.offline kinematics --save-run  # newest recording
```

The `kinematics` suite backs the dashboard's **IK** tuning tab: EE set-vs-actual charts per axis, the solve-time trace, and the per-joint churn scorecard.

See also [`diag.teleop-jitter`](/cli/diag-teleop-jitter) for the stage-by-stage band breakdown of one capture, and [`tune.motion`](/cli/tune-motion) for the hardware-side counterpart.