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

# The Quality hub: run the supervisor loop across evaluations, recordings, leaderboard, and practice

> Read the Quality landing page as a supervisor — coverage counters and trend bars on the hub, the policies behind it (AI Auto-QA, sampling, calibration), and the loop that ties recordings to evaluations, leaderboard, and Practice Studio.

# The Quality hub: run the supervisor loop

**Quality** in the dashboard is the supervisor front door for the whole QM suite. Individual guides cover each surface in depth — recordings, evaluations, leaderboard, practice — this guide covers the hub itself: what the landing page reads, which knobs feed it, and the loop that turns a recording into a coaching decision.

Every surface in the suite is supervisor-scoped. Owners, admins, and supervisors see the full rollup; other roles get a supervisor-only notice instead of the scorecards.

## 1. The four QA surfaces

The hub's sister-surface menu links each surface directly:

* **Quality → Recordings** — the recording and transcript library. Every captured recording joined to its diarised transcript, its latest linked QA score, and its recording-health verdict, searchable by transcript text. Play, scrub, and grade from the scorecard pane without leaving the call. See [Play, search, and grade recordings from the Recording Library](/guides/quality-recordings).
* **Quality → Evaluations** — the evaluation pipeline for live and post-call scoring. Author weighted scorecard forms, grade conversations, and run the acknowledge/appeal lifecycle — with AI auto-scored evaluations and agent self-evaluations feeding the same queue. See [QA Evaluations](/guides/quality-evaluations).
* **Quality → Leaderboard** — the team ranking computed on demand from QA scorecards, handled calls, and CSAT responses, with per-request custom scoring weights and queue filtering. See [Quality leaderboard](/guides/quality-leaderboard).
* **Quality → Practice** — Practice Studio roleplay. Agents rehearse a hard conversation against an AI that plays the customer, then read an automatic score with coaching feedback. Supervisor-authored scenarios stock the library. See [Train agents with AI roleplay in Practice Studio](/guides/practice-studio-roleplay).

Below the menu, the hub renders the per-agent scorecard picker — the roster-driven entry point to each agent's own scorecard page at **Quality → Agent scorecard**, where the per-agent trend, group breakdown, and recent outcomes live.

## 2. Reading the Quality landing tile

Two always-on counters sit on the hub before the point-in-time scorecard, and both answer "should I trust what the pipeline is doing right now."

**AI auto-QA coverage counters.** The coverage card answers "is the automation actually covering my calls." It renders the two generation gates (the AI Auto-QA switch on, and an active scorecard form), the share of eligible agent-attributed calls with an AI scorecard over the trailing window, and the AI scorecard quality bars — average score plus flagged-for-review share. When a gate is off, the card carries the next-step button ("Turn on AI auto-QA" or "Activate a scorecard form") straight to the surface that fixes it.

**QA score trend bars.** The trend card answers "is quality moving." Per-day average scores across time-windowed official evaluations — reviewer-authored plus AI auto-scored — walk a 7-, 30-, or 90-day window. Drill by agent or by scorecard form; clicking an agent or form row re-runs the same series on that cut. Score coloring matches the rest of the suite (green from 80, amber from 60, red below), so a trend bar, a KPI tile, and a per-agent card read as one color language.

Below the counters, the point-in-time scorecard rollup renders for its own window and channel filter: conversations scored, pass rate, average judge confidence, the by-channel comparison, the by-rubric breakdown, and the recent failures list for triage.

## 3. Policies a supervisor reaches for

Three policy surfaces decide what the pipeline scores and how heavy the human-review load is. Wire each one once; the hub counters tell you when they drift.

* **Settings → AI Auto-QA** — the tenant's automated scorer. Enable the toggle and set the flag threshold; the sweep scores completed agent-handled calls against your newest active evaluation form and stamps low scores for human review. The hub coverage card's "Turn on AI auto-QA" chip links here. See [AI Auto-QA configuration](/guides/qa-autoscore-settings).
* **Settings → QA Sample Assignment** — the weekly random sample that assigns a fixed quota of each agent's conversations to evaluators, distributed round-robin. Keeps per-agent comparisons fair instead of drifting on ad-hoc picks. See [QA sampling](/guides/qa-sampling-settings).
* **Quality → Evaluations → Calibration tab** — the same conversation graded by multiple reviewers, so reviewer scores align before they hit agents. Calibration sessions keep the "official" evaluations — and therefore the trend bars and leaderboard — comparable across reviewers. See [Build a contact-center QA program](/guides/quality-management-program).

Calibration auto-QA readiness maps directly onto the hub: when the calibration read is green, reviewer-authored and AI-authored evaluations land in the same ledger without grade inflation between them, and the trend and leaderboard stays meaningful.

## 4. The supervisor loop, end to end

Run the loop weekly:

1. **Recordings → autoscore.** The AI scorer covers every eligible call against your active form; the coverage counter shows the share it landed. Recordings it flags below your threshold surface for human review.
2. **Autoscore → evaluations.** Reviewers work the flagged and sampled queue, score against the rubric, and the evaluated agent acknowledges or appeals on the same ledger. AI and manual scores mix as one evaluation history.
3. **Evaluations → leaderboard.** Points accumulate from the scorecard stream plus handled calls and CSAT; the board re-ranks on open, so the ranking always reflects current data.
4. **Leaderboard → practice.** A rank drop or a rubric drift names the coaching target; a supervisor-authored scenario in Practice Studio gives the agent the rehearsal, scored automatically with feedback.

Step 4 is worked below.

## Worked example: the first week of a supervisor

Day one, open **Quality** and read top to bottom:

* Check the **AI auto-QA coverage** card first. If either gate is off, use the inline chip to fix it before anything else — the rest of the loop runs on automation coverage.
* Look at the **QA score trend** bars over a 30-day window. A downward walk names where to drill; a flat read means the manual and auto mix is stable.
* Drill **By agent** on the trend card to find the lowest-scoring agents, then open their per-agent scorecard from the roster picker at the bottom of the page.
* Open the **Recent failures** list on the hub for the triage queue — each row carries the excerpt the judge wrote.
* On **Evaluations**, check **Calibration** before trusting the leaderboard: reviewers who haven't calibrated score the same conversation differently.
* Close the week by reading the **Leaderboard** on the week window with default weights. If a custom weighting is in play, the chip at the top of the tab makes it explicit.

## Worked example: leaderboard drop to practice

An agent slides on the leaderboard this week. The pairing that works:

1. Open the **Leaderboard** and drill the agent's row to see which KPI slipped — QA score, handled calls, or CSAT.
2. If the QA score slipped, open **Quality → Agent scorecard** for the agent and read the group breakdown in the per-agent trend. The rubric section that fell names the skill gap (for example, "closing" scores dropped).
3. Open **Quality → Practice** and author a scenario for the failing pattern (for example, a difficult close with a price-objection customer). Set the passing score.
4. Assign the scenario to the agent. The agent rehearses until their best session clears the passing score; you review the session transcript and the automatic coaching feedback.

The common mistake is skipping step 2: a supervisor who sends the agent to Practice before reading the scorecard evaluation is guessing at the scenario. Practice sessions count toward practice metrics; they do not re-enter the evaluation ledger, so the extra drill never cancels a bad score on the already-graded conversation. Work the graded evaluation first, pair it to practice second, and the leaderboard recovers on real conversations — not on rehearsal volume.

## See also

* [Build a contact-center QA program](/guides/quality-management-program) — scorecard rubrics, calibration, and the ack/appeal lifecycle
* [QA workload management](/guides/qa-workload-management) — assign reviews to evaluators with quotas and due dates
* [Play, search, and grade recordings](/guides/quality-recordings) — the recording library worklist
* [Quality leaderboard](/guides/quality-leaderboard) — read the ranking and normalize for sample size
* [AI Auto-QA configuration](/guides/qa-autoscore-settings) and [QA sampling](/guides/qa-sampling-settings) — the two policy surfaces behind the coverage counter