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

# Evaluators

> How Podonos sources, qualifies, and filters the humans who rate your audio.

## Sourcing through vetted partners

Podonos does not run a single global pool. We work with multiple regional partners that supply qualified human evaluators in their target languages and demographics. This gives us:

* **Native speakers, not language-adjacent speakers.** Spain Spanish, not LATAM Spanish. UK English, not US English.
* **Geographic and demographic coverage.** Region, age band, and gender — we can stratify on any of these.
* **Redundancy.** When one pool is exhausted or in a different time zone, another partner backfills.

<Note>
  We do not disclose which partners supply which evaluators. From your perspective the pool is unified — you specify the target language and demographic constraints, and we deliver qualified evaluators.
</Note>

## Basic qualification (one-time onboarding)

When an evaluator first joins our system, they go through multi-stage onboarding that is more rigorous than typical crowd platforms:

<CardGroup cols={2}>
  <Card title="Hearing capability" icon="ear">
    Multi-frequency test across left and right channels. Evaluators identify directionality and detect tones at various frequencies. Rules out hearing loss that would compromise quality judgments.
  </Card>

  <Card title="Language proficiency" icon="language">
    Comprehension and production tasks in the target locale. We verify that an evaluator labeled as Spain Spanish actually speaks Spain Spanish — not Argentinian or Mexican.
  </Card>

  <Card title="Instruction-following" icon="list-check">
    Multi-step instruction tasks. People who skim instructions are filtered out at this stage.
  </Card>

  <Card title="Device & setup" icon="headphones">
    Headphone use is verified through audio probes, not self-report. Speaker users are rejected.
  </Card>
</CardGroup>

## In-session re-qualification

Passing onboarding once is not enough. Every time an evaluator joins a new session, we re-run a subset of these checks:

<Steps>
  <Step title="Acoustic environment measurement">
    We measure ambient noise through the device and verify the playback path. Sessions in noisy environments are paused or rejected. See [In-Session Quality Control](/reliability/quality-control#acoustic-environment-detection) for details.
  </Step>

  <Step title="Headphone re-detection">
    We play stereo and binaural test signals every session and confirm the listener is on headphones, not speakers — automatically.
  </Step>

  <Step title="Language refresh">
    Short comprehension cues confirm the evaluator is still operating in the right locale.
  </Step>
</Steps>

## Demographic filtering

You can filter or stratify evaluations by:

* **Locale** (en-us, en-gb, es-es, fr-fr, …)
* **Gender** and **age band**
* **Region** within a country (e.g., California vs. Texas)
* Any custom tag you attach to a query in your evaluation

After the evaluation completes, slice results by any subset of these in the Workspace dashboard. Tagging is documented in [Key Concepts → Tags](/keyconcepts#tags) and applied throughout the [Ranking](/ranking) page.