# An audio bank needs selection rules

From catalogue to the right file: roles, permissions and version checks. Six actual test cases using our own synthetic signal.

L00P.AI · 1.0 · 2026-09-26

https://www.l00p.ai/en/resources/series/bank-audio-potrzebuje-regul-wyboru/

## 01 / A shelf of recordings is not yet a bank

An agent finds a promising filename and inserts the file into a programme. The name does not reveal whether it is a master, a distribution copy, a raw stem or an old version. It does not establish the intended channel or permission to use the recording there. A useful bank answers these questions before the file is opened.

The internal imaging-bank guide we read distinguishes originals, masters and broadcast copies. Its machine-readable catalogue connects them through identifiers and provenance. We develop that method into a public template. We do not distribute station recordings or certify rights across the existing collection.

## 02 / Three file roles

The original preserves the input. The master supports further production. A distribution copy meets a particular channel's requirements. Similar-sounding files can serve these different roles, but an automation should not treat them as interchangeable.

A raw stem, a separate production layer, does not become a finished programme element merely by being copied into a folder. The record should state the role explicitly and identify the version to use. If that version is missing, selection stops; the agent does not substitute the master.

## 03 / A record that can be checked

A minimum record links a stable identifier and version to a file, its checksum, role, channel, duration and technical parameters. Separately, it references permission evidence, scope of use, check date and acceptance result. A public catalogue need not expose documents or personal details: a safe reference accessible to an authorized checker can suffice.

The fragment below is a field template, not clearance to use a recording. Empty values deliberately remain empty.

```json
{
  "id": "demo-001",
  "version": 1,
  "role": "distribution",
  "channel": "lab",
  "file": "demo.wav",
  "sha256": null,
  "permission_ref": null,
  "review_ref": null,
  "eligible": false
}
```

A hash alone confirms that bytes match a recorded digest. It does not establish authorship, licensing or quality. If someone can freely change both the file and catalogue, access controls and an approval history are also needed.

## 04 / Selection has an order

First check scope of use and purpose, then version and integrity, followed by parameters and acceptance. Only within the eligible set should you select by category, duration, language or tags. A relevant description does not override an earlier block.

![Selection order: use scope, role and channel, version and hash, parameters and acceptance, then recording selection.](https://www.l00p.ai/wydawnictwo/bank-audio-potrzebuje-regul-wyboru/wybor-z-banku-en-v1.svg)

Original method diagram, Codex / L00P.AI. Missing confirmation at any stage stops selection.

Before playback, check the same file that will actually be played. Replacement between verification and use invalidates the earlier result. Immutable versions and a release identifier can help; a spreadsheet entry alone does not provide that property.

## 05 / Six cases, one signal

On 26 September 2026 we ran a local trial using our own one-second 440 Hz tone: mono, 16-bit PCM, 48,000 Hz. The file contains 48,000 samples. The test catalogue contained six record variants referring to that signal, not six third-party recordings.

- Matching record: accepted for the laboratory only.
- Missing declared permission for the demo: rejected.
- Master role instead of distribution version: rejected.
- Incorrect expected hash: rejected.
- A channel other than the laboratory: rejected.
- Two declared seconds instead of one measured second: rejected.

We checked actual bytes and the WAV header. All five negative cases returned the expected rejection reason. This does not measure the security of an entire system. It excludes real-voice permissions, loudness normalization, listening quality and integration with playout software.

## 06 / Loudness belongs to a channel profile

The source guide describes a particular loudness target for its own stream. That is a setting of that bank, not a universal standard for every broadcaster, platform and short signal. For a new use, record the channel requirements, tolerances and method for measuring the finished copy.

Do not transfer a master's measurements to a converted file. Acceptance should refer to the final format and its hash. For very short elements, state measurement limitations explicitly. A missing reliable result is neither zero loudness nor an automatic pass.

## 07 / Withdrawal is a catalogue operation too

When permission expires or a defect emerges, block the record and record the reason. Check playback queues and exported copies. Removing an entry from the main catalogue does not guarantee removal from an earlier saved playlist. Preserve a trace of which version was used and when.

The related [article on accepting a render](/en/resources/series/dobry-render-to-jeszcze-nie-odebrany-material/) explains why creating a file is only one stage. To organize your own library, start with a few cleared recordings, one channel and an explicit set of selection conditions.

## 08 / What counts as evidence

Our basis is the complete internal bank guide, read on 26 September 2026, and the synthetic PCM trial described here. We did not verify claims about existing distribution copies or import compatibility in actual playout software. Those claims should not be treated as our acceptance results.

Codex produced the PL/EN text, original diagram and checker trial. There was no independent second-model review. The diagram proposes a method; the test is a narrow demonstration. Changes to channel requirements, permissions, files or the checker require renewed verification of the affected conditions.