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

# Accepted hour

> The unit Avala delivers in a fleet learning program, and the rejection reasons Avala reports

Avala counts delivery in accepted hours. Raw capture time is not the unit. An hour counts only when it passes every check below.

## Definition

An accepted hour is one hour of sequence time that is:

1. **Manipulation-dense.** Hands or end effectors are working on the task for the counted time. Idle waiting, walking between rooms and appliance cycle time do not count. The brief states how waiting inside a task is handled.
2. **Hands visible.** The hands and the manipulated object stay in view through the manipulation segments.
3. **Calibrated.** The sequence has a valid calibration for its rig.
4. **Consented.** Consent covers every person and place in view. Privacy redactions required by the program are applied.
5. **Accepted.** The sequence passes the program's acceptance criteria and human review.

Accepted hours appear in each [coverage report](/docs/fleet-learning/coverage.md#the-coverage-report), broken down by coverage cell. Rejected time is reported next to it, with the reasons.

## How acceptance works

Every capture gets a machine verdict first. A reviewer then confirms or overrides it.

| Verdict | Meaning |
| - | - |
| `accept` | Every criterion passed. The capture goes to human review. |
| `quarantine` | A criterion could not be measured. A reviewer decides. |
| `reject` | A blocking criterion failed. The reason code is recorded. |

You can read verdicts and reasons with the MCP server. `get_result_acceptance` returns the verdict for one capture. `get_campaign_acceptance_summary` returns acceptance rates and the top rejection reasons for a capture campaign. See [Use Avala with AI](/docs/getting-started/mcp.md).

## Rejection reasons

Avala reports a reason for every rejected sequence.

### Capture checks

These reasons come from machine checks and reviewer decisions. They carry stable codes.

| Code | Reason |
| - | - |
| `hands_not_visible` | The hands leave the frame during manipulation. |
| `too_short` | The sequence is shorter than the task requires. |
| `too_long` | The sequence runs far past the authored task length. |
| `wrong_task` | The sequence shows a different task from the brief. |
| `bad_framing` | The camera points away from the work area. |
| `phone_moved` | The capture device moved or fell during the sequence. |
| `audio_missing` | The program requires audio and the audio track is missing. |
| `extraction_failed` | The file is truncated or corrupt and cannot be read. |
| `acceptance_frame_rate_drop` | The frame rate collapses for a sustained period. |
| `acceptance_shake_sustained` | Camera shake stays high for a sustained period. |
| `acceptance_camera_stopped` | A camera stops recording before the sequence ends. |
| `acceptance_missing_channel` | A sensor channel the program requires is absent. |

### Program checks

Avala also rejects sequences that fail the program definition. These appear in the coverage report with the sequence count and hours for each reason.

| Reason | Meaning |
| - | - |
| Not manipulation-dense | Too much of the sequence is idle, waiting or walking. |
| Calibration missing or invalid | No calibration exists for the rig, or the calibration fails its check. |
| Consent or privacy | Consent does not cover a person or place in view, or a required redaction failed. |
| Timestamp sync | Streams drift apart beyond the program tolerance. |
| Duplicate or overlap | Video fingerprinting finds the footage in this delivery, an earlier delivery or a reference set. |
| Staged scene | The scene fails the [scene-realism standard](/docs/fleet-learning/coverage.md#scene-realism-standard). |
| Over concentration limit | The sequence would push a location, operator or cell past its agreed limit. |
| Unsafe | The sequence contains an unsafe action or state. |

A sequence can be technically valid and still be excluded from accepted hours. Its [sequence metadata](/docs/fleet-learning/sequence-metadata.md) keeps technical validity and the behavioral outcome apart.