> ## Documentation Index
> Fetch the complete documentation index at: https://docs.heify.com/llms.txt
> Use this file to discover all available pages before exploring further.
# Evaluators
> Create and manage reusable quality-control rubrics that the AI uses to score transcriptions.
An **Evaluator** is a reusable quality-control template that defines the criteria the AI uses to audit a call transcript. You specify a set of scored criteria, assign weights, and the AI produces a structured pass/fail report for every transcription associated with this evaluator.
For the full data model and criteria reference, see [Evaluator](/core/evaluator).
***
## Your evaluators
Search, filter, and sort in real time by **name**, **description**, or **evaluator ID**. Sort by name or creation date (ascending/descending). Toggle between a card grid and a table layout.
Each evaluator displays **capability badges** at a glance:
| Badge | What it means |
| -------------- | -------------------------------------------------------- |
| Language name | Feedback language configured for the evaluator |
| **N Criteria** | Total number of evaluation criteria defined |
| **Critical** | Shown only when at least one Strict criterion is defined |
To delete multiple evaluators at once, hover over any card or row to reveal its checkbox, select the items you want to remove, and click **Delete selection** in the action bar that appears. The table header includes a **Select all** checkbox for the current filtered results.
Single-item deletion is done from the detail view, not from the list.
***
## Creating an evaluator
Click **New Evaluator** to open the creation form.
### General parameters
| Field | Required | Limit | Notes |
| --------------------- | -------- | --------- | -------------------------------------------------------------------------------- |
| **Evaluator Name** | Yes | 100 chars | Human-readable label — e.g. `Outbound Sales Audit 2025` |
| **Feedback Language** | No | — | Language the AI writes feedback in. Default: inferred from audio. 60+ languages. |
| **Description** | No | 250 chars | Short summary of the evaluator's purpose |
The **Name** cannot be changed after creation. Choose a meaningful, descriptive name before saving.
### Evaluation context *(optional)*
Provides situational context so the AI understands who is being evaluated and under what circumstances. Max 1000 characters.
Six preset templates are available to fill the field instantly:
| Preset | Context provided to the AI |
| -------------------- | -------------------------------------------------------------------------- |
| Call center | Support agent on an inbound call following internal protocols |
| Job interview | Candidate being assessed for role suitability and communication clarity |
| Training or coaching | Participant in a training session; comprehension and engagement |
| Sales meeting | Salesperson on a sales call; needs identification and objection handling |
| Customer follow-up | Account manager in a follow-up meeting; relationship quality and proposals |
| Custom | Blank — write your own |
Selecting a preset fills the textarea with a ready-to-edit prompt. Selecting **Custom** clears it.
### Evaluation criteria
Define up to **10 criteria**. A `N / 10 criteria used` counter is shown at the top of the section.
Each criterion requires:
* **Name** *(required, max 100 chars)* — human-readable label, e.g. `Corporate Greeting`
* **Type** — how the AI scores this criterion (see [Criteria types](#criteria-types))
* **Weight** *(0–100%)* — contribution to the final score. Strict criteria are always 0%
* **AI Instructions** *(required, max 2000 chars)* — natural-language description of exactly what to look for in the transcript
Weights across all non-Strict criteria **must sum to exactly 100%** before the evaluator can be saved. Use the **Balance Weights** button in the sticky bottom bar to distribute them evenly with a single click.
### Suggested Criteria panel
The right panel offers 8 pre-built criteria you can add with a single click:
| Criteria | Type | Default weight | What it evaluates |
| --------------------- | ------- | -------------- | --------------------------------------------------------------------- |
| Corporate Greeting | Boolean | 10% | Name + company + welcome at call start |
| Identity Verification | Strict | 0% | Two personal data points verified before sensitive info is shared |
| Empathy | Scale | 20% | Acknowledgement phrases used; no condescending tone |
| Active Listening | Scale | 15% | No interruptions; paraphrases key points; asks clarifying questions |
| Resolution | Boolean | 25% | Concrete solution offered or escalation with a defined next step |
| Objection Handling | Scale | 15% | Objections answered with data; proposal adapted to client needs |
| Sales Close | Boolean | 15% | Explicit close attempt made (e.g. "Shall we proceed?") |
| Formal Farewell | Boolean | 0% | Actions summarized; further help offered; waits for client to hang up |
The 7 non-Strict suggested criteria already sum to **100%** — they form a complete, immediately usable evaluator without any manual weight adjustment.
Once a criterion has been added, its button shows a checkmark and "Already added". It cannot be added twice.
### Quality tips
Describe concrete, verifiable actions the AI can detect in the transcript. Avoid subjective criteria.
**Poor:** `"Be friendly."`
**Good:** `"Greet the client by name and thank them for the call."`
Include negative examples in the AI instructions to detect violations and reduce false positives.
Example: *"The agent must not interrupt the client while they are speaking."*
Include expected phrases, scripts, or specific business situations the AI should recognize.
Example: *"The agent must mention the Premium Plus plan (\$29.99/month) and at least two of its core benefits."*
Split complex criteria into simpler, focused ones. Each criterion should evaluate a single observable behavior.
**Poor:** `"Greeted AND verified identity AND offered a solution."`
**Good:** Three separate criteria — one for each action.
***
## Criteria types
| Type | How the AI scores | Weight |
| ----------- | ------------------------------------------------------ | --------------------------------- |
| **Boolean** | Pass or Fail | Contributes to score via weight % |
| **Scale** | 1–5 based on degree of compliance | Contributes to score via weight % |
| **Strict** | Pass or Fail — failure **fails the entire evaluation** | Always 0% |
**Strict criteria are automatic disqualifiers.** A single failed Strict criterion marks the entire evaluation as failed, regardless of how well all other criteria were met. Use them only for non-negotiable compliance requirements — such as identity verification or required legal disclosures.
***
## Viewing and editing
Click any evaluator in the list to open its detail view.
**Read mode** shows three sections:
* **General Information** — name, feedback language, description, evaluation context, creation date
* **Critical Criteria** *(shown only when Strict criteria exist)* — red-bordered card listing each Strict criterion with its name and description
* **Evaluation Criteria** — each Boolean and Scale criterion with its name, type badge, weight, and AI instructions
Click **Edit** to enter edit mode (amber borders and header). You can update:
* Feedback language, description, and evaluation context
* Criteria: edit AI instructions, type, and weight; add new criteria; remove existing ones
**Name** is read-only in edit mode and cannot be changed after creation.
Use the sticky **Save changes** bar at the bottom to confirm, or **Cancel** to discard all edits. The bar also shows the current weight status and the **Balance Weights** button.
To delete an evaluator, use the trash icon in the detail view header. **This cannot be undone.** Deleting an evaluator does not affect transcriptions that were already processed with it.
***
## Cloning an evaluator
Click the **Clone** button (copy icon, in the detail view header alongside Edit and Delete) to duplicate an existing evaluator.
You are taken directly to the creation form with all fields pre-filled:
| Field | Pre-filled value |
| ---------------------- | --------------------------------------------------------------------------- |
| **Name** | Original name + `(Copy)` suffix — truncated if over the 100-character limit |
| **Feedback Language** | Copied from original |
| **Description** | Copied from original |
| **Evaluation Context** | Copied from original |
| **Criteria** | All criteria copied — names, types, weights, and AI instructions |
The clone is a new draft — nothing is saved until you click **Create Evaluator**. Edit any field before saving, or save immediately to create an exact copy.
***
Full data model and criteria reference
View performance insights for this evaluator