# auto-qualify-draft.screen — web FE

**Priority:** Low. The primary surface is the existing inline `approval-card` (in chat/voice). This screen exists for batch-review and policy tuning.

**Surface:** `quinn.prospecting` → `/auto-qualify` route.
**Calls:** `platform.api` only.

## Layout — batch inbox

```
┌────────────────────────────────────────────────────────────────────┐
│  Auto-qualify queue            [class ▾]  [confidence ▾]  [policy]  │
├────────────────────────────────────────────────────────────────────┤
│  engage-priority   (5)                                              │
│  ┌──────────────────────────────────────────────────────────────┐  │
│  │ prospect-2014  • score 0.86  • graph: high-rep returning      │  │
│  │ inbound: "Hi, available Sat evening?"                         │  │
│  │ draft:   "Yes — 8pm or 10pm Sat. Same arrangements as last?"  │  │
│  │ why this draft  ▾                                              │  │
│  │   • graph fingerprint: weekend-only, returning, pre-verified  │  │
│  │   • local touchpoint: 3 prior, last conversion 6w ago         │  │
│  │   • calendar: Sat 8pm + 10pm currently open                   │  │
│  │ [send]  [edit]  [decline]  [wrong score]                       │  │
│  └──────────────────────────────────────────────────────────────┘  │
├────────────────────────────────────────────────────────────────────┤
│  engage-now        (11)        triage     (4)                       │
│  auto-decline-soft (3, 2 already approved-sent)                     │
└────────────────────────────────────────────────────────────────────┘
```

## Layout — policy editor

```
┌────────────────────────────────────────────────────────────────────┐
│  Auto-qualify policy                                                │
├────────────────────────────────────────────────────────────────────┤
│  Qualifying questions I always need:                                │
│   [photo / age verification] [budget] [time window]    [add+]      │
│  Voice profile:        [calibrating · 32/50 sends]  [retrain]      │
│  Class thresholds:                                                  │
│   engage-priority   ≥ [0.75]                                        │
│   engage-now        ≥ [0.50]                                        │
│   triage            ≥ [0.25]                                        │
│   auto-decline-soft <   0.25                                        │
│  Graph dependency: [allow] / [require] / [ignore]                   │
└────────────────────────────────────────────────────────────────────┘
```

## Gestures

- **Tap `[send]`** → submits draft as-is (positive label).
- **Tap `[edit]`** → opens inline editor; edit-and-send records the diff as a correction.
- **Tap `[decline]`** → record class override; surfaces a class-correction prompt.
- **Tap `[wrong score]`** → score correction with reason picker.
- **Tap `why this draft ▾`** → expands feature provenance (per `qualify_replay`).
- **Long-press class header** → bulk action ("send all in this class as-drafted") for high-trust periods. Requires per-session re-consent; never persistent.

## States

| State | Behaviour |
|-------|-----------|
| **Cold voice profile** | Draft cards carry an explicit "voice calibrating" badge. Conservative defaults. |
| **Graph offline** | Cards show "graph offline" badge; scores rendered with widened error bars. |
| **Safety-flagged touchpoint** | Does not appear in this queue at all; routed to `K-safety-blocklist`. |
| **Score-only (latency breach)** | Card renders without a draft; provider sees score+class and must compose manually. |
| **Audit replay** | Per-card replay shows feature provenance as of the original score timestamp. |
| **Bulk-send mode** | A persistent banner names the active bulk session and offers one-tap cancel; bulk auto-revokes after 10 minutes. |

## Privacy invariants

- "Why this draft" never names a graph contributor.
- Drafts never quote another provider's content; the editor cannot paste cross-tenant text.
- Bulk-send is gated; never sticky across sessions.
- Voice-profile retraining uses only the provider's own past sent messages; never cross-tenant.

## Related docs

- `auto-qualify-draft.brief.md`, `auto-qualify-draft.contract.md`
- `@features/ai-copilot/docs/approval-card.screen.md` (the primary inline surface)
- `@features/ai-copilot/docs/chat-home.screen.md` (conversational entry point)
- `@features/ai-copilot/docs/00-system-voice.md`
- `@features/ai-copilot/docs/K-safety-blocklist.brief.md`