diff --git a/backend/main.py b/backend/main.py
index 0b4e943..8136522 100644
--- a/backend/main.py
+++ b/backend/main.py
@@ -9884,281 +9884,6 @@ async def set_show_theme(data: dict):
return {"theme": session.show_theme}
-# --- Caller Model Routing ---
-
-@app.get("/api/caller-models")
-async def get_caller_models():
- """Get current caller model routing config and per-caller assignments."""
- assignments = {}
- for key in CALLER_BASES:
- name = CALLER_BASES[key].get("name", key)
- model = session.caller_models.get(key)
- assignments[key] = {"name": name, "model": model or "(default)"}
- return {
- "strategy": session.caller_model_strategy,
- "pool": session.caller_model_pool,
- "map": session.caller_model_map,
- "fallback": session.caller_model_fallback,
- "assignments": assignments,
- }
-
-
-@app.post("/api/caller-models")
-async def set_caller_models(data: dict):
- """Update caller model routing strategy, pool, map, or fallback."""
- if "strategy" in data:
- strategy = data["strategy"]
- if strategy not in ("single", "cycle", "style_matched"):
- raise HTTPException(400, f"Invalid strategy: {strategy}")
- session.caller_model_strategy = strategy
- print(f"[CallerModel] Strategy set to: {strategy}")
- if "pool" in data:
- pool = data["pool"]
- if not isinstance(pool, list) or not pool:
- raise HTTPException(400, "pool must be a non-empty list of model IDs")
- session.caller_model_pool = pool
- print(f"[CallerModel] Pool set to: {pool}")
- if "map" in data:
- session.caller_model_map = data["map"]
- print(f"[CallerModel] Style map set: {len(data['map'])} entries")
- if "fallback" in data:
- session.caller_model_fallback = data["fallback"]
- print(f"[CallerModel] Fallback set to: {data['fallback']}")
- # Clear existing assignments so new strategy takes effect
- if "strategy" in data or "pool" in data or "map" in data:
- session.caller_models.clear()
- session._caller_model_cycle_idx = 0
- print(f"[CallerModel] Cleared caller assignments (new config)")
- _save_checkpoint()
- return await get_caller_models()
-
-
-@app.post("/api/caller-models/{caller_key}")
-async def set_caller_model_override(caller_key: str, data: dict):
- """Override the model for a specific caller mid-show."""
- if caller_key not in CALLER_BASES:
- raise HTTPException(404, f"Unknown caller key: {caller_key}")
- model = data.get("model", "").strip()
- if not model:
- # Clear override
- session.caller_models.pop(caller_key, None)
- name = CALLER_BASES[caller_key].get("name", caller_key)
- print(f"[CallerModel] Cleared override for {name}")
- else:
- session.caller_models[caller_key] = model
- name = CALLER_BASES[caller_key].get("name", caller_key)
- print(f"[CallerModel] Override {name} → {model}")
- _save_checkpoint()
- return {"caller_key": caller_key, "model": session.caller_models.get(caller_key, "(default)")}
-
-
-# --- Show Preflight ---
-
-@app.get("/api/show/preflight")
-async def show_preflight(test_responses: bool = False):
- """Run diagnostic checks before a show goes live."""
- from .services.tts import VOICE_PROFILES
-
- checks = {}
-
- # --- 1. Model diversity ---
- caller_assignments = []
- model_counts: dict[str, int] = {}
- for key, base in CALLER_BASES.items():
- raw_style = session.caller_styles.get(key, "")
- style_key = _normalize_style_key(raw_style) if raw_style else ""
- # Look up what model they'd get without persisting
- model = session.caller_model_map.get(style_key) if style_key else None
- if not model:
- model = session.caller_model_fallback
- caller_assignments.append({
- "name": base.get("name", key),
- "style": style_key or "(none)",
- "model": model,
- })
- model_counts[model] = model_counts.get(model, 0) + 1
-
- total_callers = len(caller_assignments)
- max_same = max(model_counts.values()) if model_counts else 0
- diversity_status = "fail" if total_callers > 0 and max_same / total_callers > 0.5 else "pass"
- checks["model_diversity"] = {
- "status": diversity_status,
- "details": {
- "callers": caller_assignments,
- "model_distribution": model_counts,
- "max_same_model_pct": round(max_same / total_callers * 100) if total_callers else 0,
- },
- }
-
- # --- 2. Theme penetration ---
- if session.show_theme:
- theme_words = [w.lower() for w in session.show_theme.split() if len(w) > 2]
- hits = []
- misses = []
- for key, base in CALLER_BASES.items():
- bg = session.caller_backgrounds.get(key)
- if not isinstance(bg, CallerBackground):
- misses.append(base.get("name", key))
- continue
- searchable = f"{bg.natural_description} {bg.situation_summary}".lower()
- if any(tw in searchable for tw in theme_words):
- hits.append(base.get("name", key))
- else:
- misses.append(base.get("name", key))
- total_bg = len(hits) + len(misses)
- hit_pct = round(len(hits) / total_bg * 100) if total_bg else 0
- theme_status = "warn" if total_bg > 0 and hit_pct < 40 else "pass"
- checks["theme_penetration"] = {
- "status": theme_status,
- "details": {
- "theme": session.show_theme,
- "connected": hits,
- "not_connected": misses,
- "penetration_pct": hit_pct,
- },
- }
- else:
- checks["theme_penetration"] = {"status": "skip", "details": {"reason": "No theme set"}}
-
- # --- 3. Voice-age alignment ---
- mismatches = []
- alignments = []
- for key, base in CALLER_BASES.items():
- bg = session.caller_backgrounds.get(key)
- if not isinstance(bg, CallerBackground):
- continue
- voice_name = base.get("voice", "")
- profile = VOICE_PROFILES.get(voice_name, {})
- age_feel = profile.get("age_feel", "unknown")
- is_mismatch = (bg.age >= 50 and age_feel == "young") or (bg.age < 25 and age_feel == "mature")
- entry = {
- "name": base.get("name", key),
- "age": bg.age,
- "voice": voice_name,
- "voice_age_feel": age_feel,
- "mismatch": is_mismatch,
- }
- alignments.append(entry)
- if is_mismatch:
- mismatches.append(entry)
-
- voice_status = "warn" if mismatches else "pass"
- checks["voice_age_alignment"] = {
- "status": voice_status,
- "details": {
- "callers": alignments,
- "mismatches": len(mismatches),
- },
- }
-
- # --- 4. Response coherence (optional) ---
- if test_responses:
- # Test ALL callers, not just a sample — we want confidence every caller works
- test_callers = []
- for key, base in CALLER_BASES.items():
- raw_style = session.caller_styles.get(key, "")
- style_key = _normalize_style_key(raw_style) if raw_style else ""
- model = session.caller_model_map.get(style_key) if style_key else None
- if not model:
- model = session.caller_model_fallback
- test_callers.append((key, base, model))
-
- coherence_results = []
- coherence_fail = False
-
- # Run all tests in parallel for speed
- async def _test_caller(key, base, model):
- slim_caller = session.caller_backgrounds.get(key, {})
- prompt = get_caller_prompt(slim_caller)
- # Simulate a real 3-exchange conversation
- max_tok, _ = _pick_response_budget(session.caller_shapes.get(key, "standard"))
- messages = [
- {"role": "user", "content": "Hey welcome to the show, what's going on tonight?"},
- ]
- try:
- # First response
- r1 = await llm_service.generate(
- messages=messages, system_prompt=prompt,
- max_tokens=max_tok, category="caller_dialog",
- caller_name=base.get("name", ""), model_override=model,
- )
- r1 = await _retry_if_too_short(
- r1, llm_service, messages, prompt,
- max_tok, base.get("name", ""), model_override=model,
- )
- r1_words = len(r1.split()) if r1 else 0
-
- # Second exchange — host follow-up
- messages.append({"role": "assistant", "content": r1})
- messages.append({"role": "user", "content": "Wait, tell me more about that."})
- r2 = await llm_service.generate(
- messages=messages, system_prompt=prompt,
- max_tokens=max_tok, category="caller_dialog",
- caller_name=base.get("name", ""), model_override=model,
- )
- r2 = await _retry_if_too_short(
- r2, llm_service, messages, prompt,
- max_tok, base.get("name", ""), model_override=model,
- )
- r2_words = len(r2.split()) if r2 else 0
-
- avg_words = (r1_words + r2_words) // 2
- passed = avg_words >= MIN_RESPONSE_WORDS and r1_words >= 30 and r2_words >= 30
- return {
- "name": base.get("name", key),
- "model": model,
- "word_count": avg_words,
- "r1_words": r1_words,
- "r2_words": r2_words,
- "pass": passed,
- "snippet": (r1[:150] + "...") if r1 and len(r1) > 150 else r1,
- }
- except Exception as e:
- return {
- "name": base.get("name", key),
- "model": model,
- "error": str(e),
- "pass": False,
- }
-
- results = await asyncio.gather(*[_test_caller(k, b, m) for k, b, m in test_callers])
- for r in results:
- coherence_results.append(r)
- if not r.get("pass"):
- coherence_fail = True
-
- checks["response_coherence"] = {
- "status": "fail" if coherence_fail else "pass",
- "details": {"results": coherence_results},
- }
- else:
- checks["response_coherence"] = {"status": "skip", "details": {"reason": "Use ?test_responses=true to enable"}}
-
- # --- Overall status ---
- statuses = [c["status"] for c in checks.values()]
- if "fail" in statuses:
- overall = "fail"
- elif "warn" in statuses:
- overall = "warn"
- else:
- overall = "pass"
-
- # Summary
- issues = []
- if checks["model_diversity"]["status"] == "fail":
- issues.append(f"Model diversity: {checks['model_diversity']['details']['max_same_model_pct']}% on same model")
- if checks["theme_penetration"]["status"] == "warn":
- issues.append(f"Theme penetration: only {checks['theme_penetration']['details']['penetration_pct']}% connected")
- if checks["voice_age_alignment"]["status"] == "warn":
- issues.append(f"Voice-age mismatches: {len(mismatches)}")
- if checks.get("response_coherence", {}).get("status") == "fail":
- failed = [r["name"] for r in checks["response_coherence"]["details"]["results"] if not r.get("pass")]
- issues.append(f"Response coherence failed: {', '.join(failed)}")
- summary = "; ".join(issues) if issues else "All checks passed"
-
- return {"status": overall, "checks": checks, "summary": summary}
-
-
# --- Cost Tracking Endpoints ---
@app.get("/api/costs")
diff --git a/frontend/css/style.css b/frontend/css/style.css
index d6b1134..77482f4 100644
--- a/frontend/css/style.css
+++ b/frontend/css/style.css
@@ -449,84 +449,6 @@ section h2 {
line-height: 1.3;
}
-/* Caller model indicator */
-.info-badge.model {
- background: rgba(100, 140, 220, 0.2);
- color: #7ab0e8;
- font-size: 0.7rem;
- cursor: pointer;
-}
-
-.caller-model-override {
- font-size: 0.7rem;
- padding: 2px 4px;
- background: var(--bg);
- color: var(--text);
- border: 1px solid rgba(100, 140, 220, 0.3);
- border-radius: 4px;
- max-width: 140px;
-}
-
-/* Caller button model badge */
-.model-tag {
- font-size: 0.55rem;
- color: #7ab0e8;
- background: rgba(100, 140, 220, 0.15);
- padding: 0 3px;
- border-radius: 2px;
- font-weight: 700;
- letter-spacing: 0.3px;
- flex-shrink: 0;
-}
-
-/* Caller Models settings section */
-.caller-model-row {
- margin-bottom: 8px;
-}
-
-.caller-model-row label {
- margin-bottom: 0;
-}
-
-.cm-pool-input {
- font-size: 0.8rem;
-}
-
-.cm-style-grid {
- display: grid;
- grid-template-columns: 1fr 1fr;
- gap: 4px;
- margin-bottom: 8px;
- max-height: 200px;
- overflow-y: auto;
-}
-
-.cm-style-item {
- display: flex;
- align-items: center;
- justify-content: space-between;
- gap: 4px;
- background: rgba(255, 255, 255, 0.05);
- border-radius: 4px;
- padding: 3px 6px;
-}
-
-.cm-style-name {
- font-size: 0.7rem;
- color: var(--text-muted);
- white-space: nowrap;
-}
-
-.cm-style-select {
- font-size: 0.7rem;
- padding: 2px 3px;
- background: var(--bg);
- color: var(--text);
- border: 1px solid rgba(232, 121, 29, 0.15);
- border-radius: 4px;
- max-width: 110px;
-}
-
.caller-background-full {
margin-top: 8px;
font-size: 0.75rem;
@@ -1915,110 +1837,3 @@ button:focus-visible {
.log-toggle-btn:hover {
color: var(--text);
}
-
-/* Preflight */
-.preflight-btn {
- background: rgba(90, 138, 60, 0.15);
- color: var(--accent-green);
- border: 1px solid rgba(90, 138, 60, 0.3);
-}
-.preflight-btn:hover {
- background: rgba(90, 138, 60, 0.25);
-}
-
-.preflight-content {
- max-width: 700px;
-}
-
-.preflight-status {
- display: flex;
- align-items: center;
- gap: 10px;
- padding: 12px 16px;
- border-radius: var(--radius-sm);
- margin-bottom: 16px;
- font-weight: 700;
- font-size: 1.1rem;
-}
-.preflight-status.pass { background: rgba(90, 138, 60, 0.15); color: var(--accent-green); }
-.preflight-status.warn { background: rgba(232, 169, 29, 0.15); color: #e8a91d; }
-.preflight-status.fail { background: rgba(204, 34, 34, 0.15); color: var(--accent-red); }
-.preflight-status.loading { background: rgba(232, 121, 29, 0.1); color: var(--text-muted); }
-
-.preflight-checks {
- display: flex;
- flex-direction: column;
- gap: 12px;
- max-height: 60vh;
- overflow-y: auto;
-}
-
-.preflight-check {
- background: var(--bg);
- border: 1px solid rgba(232, 121, 29, 0.1);
- border-radius: var(--radius-sm);
- padding: 12px 16px;
-}
-.preflight-check-header {
- display: flex;
- justify-content: space-between;
- align-items: center;
- cursor: pointer;
- user-select: none;
-}
-.preflight-check-name {
- font-weight: 600;
- font-size: 0.95rem;
-}
-.preflight-check-badge {
- font-size: 0.75rem;
- font-weight: 700;
- padding: 2px 8px;
- border-radius: 4px;
- text-transform: uppercase;
-}
-.preflight-check-badge.pass { background: rgba(90, 138, 60, 0.2); color: var(--accent-green); }
-.preflight-check-badge.warn { background: rgba(232, 169, 29, 0.2); color: #e8a91d; }
-.preflight-check-badge.fail { background: rgba(204, 34, 34, 0.2); color: var(--accent-red); }
-.preflight-check-badge.skip { background: rgba(154, 139, 120, 0.2); color: var(--text-muted); }
-
-.preflight-check-details {
- margin-top: 10px;
- font-size: 0.85rem;
- color: var(--text-muted);
- display: none;
-}
-.preflight-check.open .preflight-check-details {
- display: block;
-}
-
-.preflight-table {
- width: 100%;
- border-collapse: collapse;
- margin-top: 8px;
-}
-.preflight-table th {
- text-align: left;
- color: var(--text-muted);
- font-size: 0.75rem;
- font-weight: 600;
- text-transform: uppercase;
- padding: 4px 8px;
- border-bottom: 1px solid rgba(232, 121, 29, 0.1);
-}
-.preflight-table td {
- padding: 4px 8px;
- font-size: 0.8rem;
- color: var(--text);
- border-bottom: 1px solid rgba(232, 121, 29, 0.05);
-}
-.preflight-table tr.mismatch td { color: var(--accent-red); }
-.preflight-table tr.connected td { color: var(--accent-green); }
-
-.preflight-test-btn {
- background: rgba(232, 121, 29, 0.15);
- color: var(--accent);
- border: 1px solid rgba(232, 121, 29, 0.3);
-}
-.preflight-test-btn:hover { background: rgba(232, 121, 29, 0.25); }
-.preflight-test-btn.loading { opacity: 0.6; pointer-events: none; }
diff --git a/frontend/index.html b/frontend/index.html
index 96dc68c..b3345db 100644
--- a/frontend/index.html
+++ b/frontend/index.html
@@ -15,7 +15,6 @@
-
@@ -72,10 +71,6 @@
No active call
-
-
-
-
@@ -287,36 +282,6 @@
-
Caller Models
-
-
-
-
-
-
-
-
-
-
-
TTS Provider
@@ -357,22 +322,6 @@
-
-
-
-
Show Preflight
-
- ...
- Running checks...
-
-
-
-
-
-
-
-
-
${renderCheckDetails(checkKey, check)}
- `;
-
- card.querySelector('.preflight-check-header').addEventListener('click', () => {
- card.classList.toggle('open');
- });
-
- checksEl.appendChild(card);
- }
-}
-
-function renderCheckDetails(name, check) {
- const d = check.details || {};
- switch (name) {
- case 'model_diversity': return renderModelDiversity(d);
- case 'theme_penetration': return renderThemePenetration(d);
- case 'voice_age_alignment': return renderVoiceAgeAlignment(d);
- case 'response_coherence': return renderResponseCoherence(check);
- default: return `${escapeHtml(JSON.stringify(d, null, 2))}`;
- }
-}
-
-function renderModelDiversity(d) {
- const callers = d.callers || [];
- if (!callers.length) return 'No callers to check.
';
- let html = `
- | Caller | Style | Model |
|---|
`;
- for (const c of callers) {
- html += `| ${escapeHtml(c.name || '')} | ${escapeHtml(c.style || '')} | ${escapeHtml(c.model || '')} |
`;
- }
- html += '
';
- if (d.max_same_model_pct != null) {
- html += `${d.max_same_model_pct}% on same model
`;
- }
- return html;
-}
-
-function renderThemePenetration(d) {
- let html = '';
- if (d.theme) html += `Theme: ${escapeHtml(d.theme)}
`;
- if (d.connected?.length) {
- html += `Connected: ${d.connected.map(n => escapeHtml(n)).join(', ')}
`;
- }
- if (d.not_connected?.length) {
- html += `Not connected: ${d.not_connected.map(n => escapeHtml(n)).join(', ')}
`;
- }
- if (d.penetration_pct != null) {
- html += `${d.penetration_pct}% penetration
`;
- }
- return html || 'No theme set.
';
-}
-
-function renderVoiceAgeAlignment(d) {
- const callers = d.callers || [];
- if (!callers.length) return 'No callers to check.
';
- let html = `
- | Caller | Age | Voice | Age Feel |
|---|
`;
- for (const c of callers) {
- const cls = c.mismatch ? ' class="mismatch"' : '';
- html += `${escapeHtml(c.name || '')} | ${c.age || ''} | ${escapeHtml(c.voice || '')} | ${escapeHtml(c.age_feel || '')} | `;
- }
- html += '
';
- return html;
-}
-
-function renderResponseCoherence(check) {
- if (check.status === 'skip') {
- return 'Use Test Responses button to run this check.
';
- }
- const d = check.details || {};
- const results = d.results || [];
- if (!results.length) return 'No test results.
';
- let html = `
- | Caller | Model | R1 | R2 | Avg | |
|---|
`;
- for (const c of results) {
- const cls = c.pass ? '' : ' class="mismatch"';
- if (c.error) {
- html += `| ${escapeHtml(c.name || '')} | ${escapeHtml(c.model || '')} | ${escapeHtml(c.error)} | ✗ |
`;
- } else {
- html += `${escapeHtml(c.name || '')} | ${escapeHtml(c.model || '')} | ${c.r1_words || 0} | ${c.r2_words || 0} | ${c.word_count || 0} | ${c.pass ? '✓' : '✗'} | `;
- if (c.snippet) {
- html += `| ${escapeHtml(c.snippet)} |
`;
- }
- }
- }
- html += '
';
- const passed = results.filter(r => r.pass).length;
- html += `${passed}/${results.length} callers passed (min ${50} words per response)
`;
- return html;
-}