@kernel.chat/kbot 3.88.0 → 3.93.0

package/dist/tools/stream-self-eval.js

@@ -0,0 +1,764 @@
+// kbot Stream Self-Evaluation — analyzes own rendered frames and auto-adjusts visual quality
+//
+// Three-method active inference loop:
+// 1. Rule-based frame analysis (every 30 seconds / 180 frames at 6fps)
+// 2. LLM deep evaluation via Gemma 4 on Ollama (every 5 minutes / 1800 frames)
+// 3. Engagement-based learning (correlates config with chat rate)
+//
+// The stream looks at itself and constantly self-improves.
+import { registerTool } from './index.js';
+function samplePixels(imageData, count) {
+    const pixels = [];
+    const { data, width, height } = imageData;
+    for (let i = 0; i < count; i++) {
+        const x = Math.floor(Math.random() * width);
+        const y = Math.floor(Math.random() * height);
+        const idx = (y * width + x) * 4;
+        pixels.push({ r: data[idx], g: data[idx + 1], b: data[idx + 2] });
+    }
+    return pixels;
+}
+function sampleRegion(imageData, rx, ry, rw, rh, count) {
+    const pixels = [];
+    const { data, width } = imageData;
+    for (let i = 0; i < count; i++) {
+        const x = Math.floor(rx + Math.random() * rw);
+        const y = Math.floor(ry + Math.random() * rh);
+        if (x < 0 || x >= imageData.width || y < 0 || y >= imageData.height)
+            continue;
+        const idx = (y * width + x) * 4;
+        pixels.push({ r: data[idx], g: data[idx + 1], b: data[idx + 2] });
+    }
+    return pixels;
+}
+function pixelBrightness(r, g, b) {
+    return (r * 0.299 + g * 0.587 + b * 0.114) / 255;
+}
+function avgBrightness(pixels) {
+    if (pixels.length === 0)
+        return 0.5;
+    let total = 0;
+    for (const p of pixels)
+        total += pixelBrightness(p.r, p.g, p.b);
+    return total / pixels.length;
+}
+function colorContrast(c1, c2) {
+    const b1 = pixelBrightness(c1.r, c1.g, c1.b);
+    const b2 = pixelBrightness(c2.r, c2.g, c2.b);
+    return Math.abs(b1 - b2);
+}
+function avgColor(pixels) {
+    if (pixels.length === 0)
+        return { r: 128, g: 128, b: 128 };
+    let rT = 0, gT = 0, bT = 0;
+    for (const p of pixels) {
+        rT += p.r;
+        gT += p.g;
+        bT += p.b;
+    }
+    return { r: Math.round(rT / pixels.length), g: Math.round(gT / pixels.length), b: Math.round(bT / pixels.length) };
+}
+function rgbToHue(r, g, b) {
+    const rn = r / 255, gn = g / 255, bn = b / 255;
+    const max = Math.max(rn, gn, bn), min = Math.min(rn, gn, bn);
+    if (max === min)
+        return 0;
+    let h = 0;
+    const d = max - min;
+    if (max === rn)
+        h = ((gn - bn) / d + (gn < bn ? 6 : 0)) / 6;
+    else if (max === gn)
+        h = ((bn - rn) / d + 2) / 6;
+    else
+        h = ((rn - gn) / d + 4) / 6;
+    return h;
+}
+// ─── Default Config ──────────────────────────────────────────────
+function defaultConfig() {
+    return {
+        robotScale: 10,
+        robotScreenYPercent: 0.55,
+        headerOpacity: 0.7,
+        chatOpacity: 0.6,
+        chatFadeSeconds: 10,
+        skyBrightness: 0.5,
+        ambientLight: 0.5,
+        borderWidth: 1,
+        speechBubbleWidth: 500,
+        particleDensity: 1.0,
+        bloomIntensity: 0.15,
+        vignetteStrength: 0.25,
+    };
+}
+// ─── Initialize ──────────────────────────────────────────────────
+export function initStreamEval() {
+    return {
+        lastEvalFrame: 0,
+        lastDeepEvalFrame: 0,
+        evalInterval: 180, // 30 seconds at 6fps
+        deepEvalInterval: 1800, // 5 minutes at 6fps
+        currentConfig: defaultConfig(),
+        configHistory: [],
+        issuesFound: [],
+        adjustmentsMade: [],
+        totalEvaluations: 0,
+        totalDeepEvaluations: 0,
+        lastAnalysis: null,
+        lastLLMEval: null,
+        announcementQueue: [],
+    };
+}
+// ─── Should Evaluate ─────────────────────────────────────────────
+export function shouldEvaluate(evaluation, frame) {
+    return frame - evaluation.lastEvalFrame >= evaluation.evalInterval;
+}
+export function shouldDeepEvaluate(evaluation, frame) {
+    return frame - evaluation.lastDeepEvalFrame >= evaluation.deepEvalInterval;
+}
+// ─── Method 1: Rule-Based Frame Analysis ─────────────────────────
+export function analyzeFrame(imageData, config, robotX, robotY, robotScale, chatMessageCount, mood) {
+    const { width, height } = imageData;
+    const issues = [];
+    const suggestions = [];
+    // ── Robot visibility: compare robot region to surrounding background ──
+    const robotW = 32 * robotScale;
+    const robotH = 48 * robotScale;
+    const robotPixels = sampleRegion(imageData, robotX, robotY, robotW, robotH, 30);
+    // Sample background around the robot (above, left, right)
+    const bgAbove = sampleRegion(imageData, robotX - 40, Math.max(0, robotY - 60), robotW + 80, 50, 20);
+    const bgLeft = sampleRegion(imageData, Math.max(0, robotX - 80), robotY, 60, robotH, 15);
+    const bgRight = sampleRegion(imageData, robotX + robotW + 20, robotY, 60, robotH, 15);
+    const bgPixels = [...bgAbove, ...bgLeft, ...bgRight];
+    const robotAvg = avgColor(robotPixels);
+    const bgAvg = avgColor(bgPixels);
+    const robotVisibility = Math.min(1, colorContrast(robotAvg, bgAvg) * 2.5); // Scale up since robot is small
+    if (robotVisibility < 0.3)
+        issues.push('Robot is hard to see against background');
+    if (robotVisibility < 0.15)
+        issues.push('Robot is nearly invisible');
+    // ── Brightness balance: sample across the whole frame ──
+    const frameSample = samplePixels(imageData, 100);
+    const brightness = avgBrightness(frameSample);
+    // Transform: 0.0=too dark, 0.5=ideal, 1.0=too bright
+    // Map so that 0.3-0.5 brightness is ideal (score=1), extremes are 0
+    const brightnessBalance = brightness < 0.1 ? brightness / 0.1 * 0.3
+        : brightness < 0.3 ? 0.3 + (brightness - 0.1) / 0.2 * 0.7
+            : brightness < 0.6 ? 1.0
+                : brightness < 0.8 ? 1.0 - (brightness - 0.6) / 0.2 * 0.5
+                    : Math.max(0, 0.5 - (brightness - 0.8) * 2.5);
+    if (brightness < 0.15)
+        issues.push('Frame is too dark');
+    if (brightness > 0.75)
+        issues.push('Frame is too bright/washed out');
+    // ── Color variety: count unique hue buckets (12 segments) ──
+    const hueSample = samplePixels(imageData, 50);
+    const hueBuckets = new Set();
+    for (const p of hueSample) {
+        const sat = (Math.max(p.r, p.g, p.b) - Math.min(p.r, p.g, p.b)) / 255;
+        if (sat > 0.1) { // Only count saturated pixels
+            const hue = rgbToHue(p.r, p.g, p.b);
+            hueBuckets.add(Math.floor(hue * 12));
+        }
+    }
+    const colorVariety = Math.min(1, hueBuckets.size / 6); // 6+ buckets = max variety
+    if (colorVariety < 0.2)
+        issues.push('Scene looks monochromatic — very few colors');
+    if (colorVariety < 0.35)
+        issues.push('Low color variety');
+    // ── Sky-to-ground ratio: sample a vertical column at center ──
+    const centerX = Math.floor(width / 2);
+    const verticalSamples = 40;
+    let skyPixelCount = 0;
+    for (let i = 0; i < verticalSamples; i++) {
+        const y = Math.floor((i / verticalSamples) * height);
+        const idx = (y * width + centerX) * 4;
+        const r = imageData.data[idx], g = imageData.data[idx + 1], b = imageData.data[idx + 2];
+        const br = pixelBrightness(r, g, b);
+        // Sky heuristic: upper half, relatively uniform, blue-ish or dark
+        // Ground heuristic: lower half, more varied, green/brown tones
+        if (y < height * 0.5 && (b > r * 0.8 || br < 0.25)) {
+            skyPixelCount++;
+        }
+    }
+    const skyToGroundRatio = skyPixelCount / verticalSamples;
+    if (skyToGroundRatio > 0.7)
+        issues.push('Too much sky visible — robot might be too low');
+    if (skyToGroundRatio < 0.1)
+        issues.push('Almost no sky visible — robot might be too high');
+    // ── Chat readability: check overlay region contrast ──
+    // Chat overlay is at bottom-left: (10, height-200, 400, 150)
+    const chatBgPixels = sampleRegion(imageData, 10, height - 200, 400, 150, 20);
+    const chatBg = avgColor(chatBgPixels);
+    // Chat text is white (#e6edf3), so check contrast with background
+    const chatTextColor = { r: 230, g: 237, b: 243 };
+    const chatContrast = colorContrast(chatTextColor, chatBg);
+    const chatReadability = chatMessageCount > 0
+        ? Math.min(1, chatContrast * 2.0) // Scale: contrast of 0.5 = fully readable
+        : 0.8; // No messages = assume readable
+    if (chatReadability < 0.4 && chatMessageCount > 0)
+        issues.push('Chat overlay is hard to read');
+    // ── Overall score (weighted average) ──
+    const overallScore = robotVisibility * 0.30 +
+        brightnessBalance * 0.25 +
+        colorVariety * 0.15 +
+        (skyToGroundRatio > 0.15 && skyToGroundRatio < 0.65 ? 1 : 0.4) * 0.15 +
+        chatReadability * 0.15;
+    // ── Generate suggestions ──
+    if (robotVisibility < 0.4 && config.robotScale < 12) {
+        suggestions.push({
+            param: 'robotScale',
+            oldValue: config.robotScale,
+            newValue: Math.min(12, config.robotScale + 1),
+            reason: 'Robot is hard to see — increasing scale',
+        });
+    }
+    if (brightness < 0.2) {
+        suggestions.push({
+            param: 'ambientLight',
+            oldValue: config.ambientLight,
+            newValue: Math.min(0.8, config.ambientLight + 0.05),
+            reason: 'Scene too dark — raising ambient light',
+        });
+        suggestions.push({
+            param: 'skyBrightness',
+            oldValue: config.skyBrightness,
+            newValue: Math.min(0.8, config.skyBrightness + 0.05),
+            reason: 'Scene too dark — raising sky brightness',
+        });
+    }
+    if (brightness > 0.7) {
+        suggestions.push({
+            param: 'ambientLight',
+            oldValue: config.ambientLight,
+            newValue: Math.max(0.3, config.ambientLight - 0.05),
+            reason: 'Scene too bright — reducing ambient light',
+        });
+    }
+    if (chatReadability < 0.5 && chatMessageCount > 0) {
+        suggestions.push({
+            param: 'chatOpacity',
+            oldValue: config.chatOpacity,
+            newValue: Math.min(0.85, config.chatOpacity + 0.05),
+            reason: 'Chat hard to read — increasing overlay opacity',
+        });
+    }
+    if (colorVariety < 0.3) {
+        suggestions.push({
+            param: 'bloomIntensity',
+            oldValue: config.bloomIntensity,
+            newValue: Math.min(0.4, config.bloomIntensity + 0.05),
+            reason: 'Flat colors — adding bloom for color richness',
+        });
+    }
+    if (skyToGroundRatio > 0.7) {
+        suggestions.push({
+            param: 'robotScreenYPercent',
+            oldValue: config.robotScreenYPercent,
+            newValue: Math.max(0.4, config.robotScreenYPercent - 0.03),
+            reason: 'Too much sky — moving robot up',
+        });
+    }
+    if (skyToGroundRatio < 0.15) {
+        suggestions.push({
+            param: 'robotScreenYPercent',
+            oldValue: config.robotScreenYPercent,
+            newValue: Math.min(0.65, config.robotScreenYPercent + 0.03),
+            reason: 'No sky visible — moving robot down',
+        });
+    }
+    // Vignette too strong in dark scenes
+    if (brightness < 0.25 && config.vignetteStrength > 0.3) {
+        suggestions.push({
+            param: 'vignetteStrength',
+            oldValue: config.vignetteStrength,
+            newValue: Math.max(0.1, config.vignetteStrength - 0.05),
+            reason: 'Dark scene with heavy vignette — reducing edge darkening',
+        });
+    }
+    // Robot too big, consuming too much frame
+    if (robotVisibility > 0.85 && config.robotScale > 9) {
+        suggestions.push({
+            param: 'robotScale',
+            oldValue: config.robotScale,
+            newValue: Math.max(8, config.robotScale - 1),
+            reason: 'Robot dominates frame — slightly reducing scale',
+        });
+    }
+    return {
+        robotVisibility,
+        brightnessBalance,
+        colorVariety,
+        skyToGroundRatio,
+        chatReadability,
+        overallScore,
+        issues,
+        suggestions,
+    };
+}
+// ─── Method 2: LLM Deep Evaluation via Gemma 4 ──────────────────
+export async function deepEvaluateWithLLM(analysis, config, recentChatRate, mood) {
+    const prompt = `You are evaluating a livestream's visual quality. The stream shows a pixel art robot in a 2D tile world.
+
+Current state:
+- Robot scale: ${config.robotScale} (visibility score: ${(analysis.robotVisibility * 100).toFixed(0)}%)
+- Robot Y position: ${(config.robotScreenYPercent * 100).toFixed(0)}% down the screen
+- Sky brightness: ${config.skyBrightness.toFixed(2)}
+- Ambient light: ${config.ambientLight.toFixed(2)}
+- Chat overlay opacity: ${config.chatOpacity.toFixed(2)}
+- Frame brightness: ${analysis.brightnessBalance.toFixed(2)}
+- Color variety: ${(analysis.colorVariety * 100).toFixed(0)}% (${analysis.colorVariety < 0.3 ? 'low' : analysis.colorVariety > 0.7 ? 'high' : 'medium'})
+- Sky-to-ground ratio: ${(analysis.skyToGroundRatio * 100).toFixed(0)}%
+- Chat readability: ${(analysis.chatReadability * 100).toFixed(0)}%
+- Chat rate: ${recentChatRate.toFixed(1)} messages/minute
+- Mood: ${mood}
+- Bloom intensity: ${config.bloomIntensity.toFixed(2)}
+- Vignette: ${config.vignetteStrength.toFixed(2)}
+- Particle density: ${config.particleDensity.toFixed(2)}
+- Current issues: ${analysis.issues.length > 0 ? analysis.issues.join(', ') : 'none detected'}
+- Overall score: ${(analysis.overallScore * 100).toFixed(0)}%
+
+Rate each aspect 1-10 and suggest ONE specific parameter change to improve the stream.
+Valid parameters: robotScale (8-12), robotScreenYPercent (0.4-0.7), headerOpacity (0.5-0.9), chatOpacity (0.4-0.85), skyBrightness (0.3-0.8), ambientLight (0.3-0.8), bloomIntensity (0.0-0.4), vignetteStrength (0.1-0.5), particleDensity (0.5-2.0), speechBubbleWidth (400-700).
+
+Respond ONLY with a JSON object:
+{"robotVisibility":N,"colorBalance":N,"layoutClarity":N,"overallFeel":N,"param":"paramName","value":N,"reason":"one sentence"}`;
+    try {
+        const res = await fetch('http://localhost:11434/api/generate', {
+            method: 'POST',
+            headers: { 'Content-Type': 'application/json' },
+            body: JSON.stringify({
+                model: 'gemma3',
+                prompt,
+                stream: false,
+                options: { temperature: 0.3, num_predict: 200 },
+            }),
+        });
+        if (!res.ok) {
+            return fallbackLLMEval(analysis);
+        }
+        const body = await res.json();
+        const raw = body.response || '';
+        // Extract JSON from response (may have markdown wrapping)
+        const jsonMatch = raw.match(/\{[\s\S]*?\}/);
+        if (!jsonMatch)
+            return fallbackLLMEval(analysis, raw);
+        const parsed = JSON.parse(jsonMatch[0]);
+        const validParams = [
+            'robotScale', 'robotScreenYPercent', 'headerOpacity', 'chatOpacity',
+            'skyBrightness', 'ambientLight', 'bloomIntensity', 'vignetteStrength',
+            'particleDensity', 'speechBubbleWidth', 'chatFadeSeconds', 'borderWidth',
+        ];
+        return {
+            robotVisibility: clamp(parsed.robotVisibility ?? 5, 1, 10),
+            colorBalance: clamp(parsed.colorBalance ?? 5, 1, 10),
+            layoutClarity: clamp(parsed.layoutClarity ?? 5, 1, 10),
+            overallFeel: clamp(parsed.overallFeel ?? 5, 1, 10),
+            suggestedParam: parsed.param && validParams.includes(parsed.param)
+                ? parsed.param
+                : null,
+            suggestedValue: typeof parsed.value === 'number' ? parsed.value : null,
+            suggestedReason: parsed.reason || 'No reason given',
+            raw,
+        };
+    }
+    catch {
+        return fallbackLLMEval(analysis);
+    }
+}
+function fallbackLLMEval(analysis, raw = 'Ollama unavailable') {
+    return {
+        robotVisibility: Math.round(analysis.robotVisibility * 10),
+        colorBalance: Math.round(analysis.colorVariety * 10),
+        layoutClarity: Math.round((1 - Math.abs(analysis.skyToGroundRatio - 0.35)) * 10),
+        overallFeel: Math.round(analysis.overallScore * 10),
+        suggestedParam: analysis.suggestions[0]?.param ?? null,
+        suggestedValue: analysis.suggestions[0]?.newValue ?? null,
+        suggestedReason: analysis.suggestions[0]?.reason ?? 'Using rule-based fallback',
+        raw,
+    };
+}
+function clamp(v, min, max) {
+    return Math.max(min, Math.min(max, v));
+}
+// ─── Config Bounds (enforced on every adjustment) ────────────────
+const CONFIG_BOUNDS = {
+    robotScale: [8, 12],
+    robotScreenYPercent: [0.4, 0.7],
+    headerOpacity: [0.5, 0.9],
+    chatOpacity: [0.4, 0.85],
+    chatFadeSeconds: [5, 15],
+    skyBrightness: [0.3, 0.8],
+    ambientLight: [0.3, 0.8],
+    borderWidth: [1, 4],
+    speechBubbleWidth: [400, 700],
+    particleDensity: [0.5, 2.0],
+    bloomIntensity: [0.0, 0.5],
+    vignetteStrength: [0.1, 0.5],
+};
+function clampConfig(config) {
+    const clamped = { ...config };
+    for (const key of Object.keys(CONFIG_BOUNDS)) {
+        const [min, max] = CONFIG_BOUNDS[key];
+        clamped[key] = clamp(clamped[key], min, max);
+    }
+    return clamped;
+}
+// ─── Apply Adjustments (max 10% change per eval) ────────────────
+export function applyAdjustments(config, analysis) {
+    const changes = [];
+    const newConfig = { ...config };
+    // Only apply top 2 suggestions to avoid oscillation
+    const sorted = [...analysis.suggestions].sort((a, b) => {
+        // Prioritize by severity of the issue
+        const severityA = Math.abs(a.newValue - a.oldValue);
+        const severityB = Math.abs(b.newValue - b.oldValue);
+        return severityB - severityA;
+    });
+    const applied = new Set();
+    for (const suggestion of sorted.slice(0, 2)) {
+        if (applied.has(suggestion.param))
+            continue;
+        applied.add(suggestion.param);
+        const oldVal = newConfig[suggestion.param];
+        // Limit change to 10% of the parameter's range
+        const [min, max] = CONFIG_BOUNDS[suggestion.param];
+        const range = max - min;
+        const maxDelta = range * 0.1;
+        const rawDelta = suggestion.newValue - oldVal;
+        const cappedDelta = Math.sign(rawDelta) * Math.min(Math.abs(rawDelta), maxDelta);
+        const newVal = clamp(oldVal + cappedDelta, min, max);
+        if (Math.abs(newVal - oldVal) > 0.001) {
+            newConfig[suggestion.param] = newVal;
+            changes.push(`${suggestion.param}: ${oldVal.toFixed(2)} -> ${newVal.toFixed(2)} (${suggestion.reason})`);
+        }
+    }
+    return { newConfig: clampConfig(newConfig), changes };
+}
+// ─── Apply LLM Suggestion ────────────────────────────────────────
+export function applyLLMSuggestion(config, llmEval) {
+    if (!llmEval.suggestedParam || llmEval.suggestedValue === null) {
+        return { newConfig: config, change: null };
+    }
+    const param = llmEval.suggestedParam;
+    const oldVal = config[param];
+    const [min, max] = CONFIG_BOUNDS[param];
+    const range = max - min;
+    const maxDelta = range * 0.15; // LLM gets slightly more latitude
+    const targetVal = clamp(llmEval.suggestedValue, min, max);
+    const rawDelta = targetVal - oldVal;
+    const cappedDelta = Math.sign(rawDelta) * Math.min(Math.abs(rawDelta), maxDelta);
+    const newVal = clamp(oldVal + cappedDelta, min, max);
+    if (Math.abs(newVal - oldVal) < 0.001) {
+        return { newConfig: config, change: null };
+    }
+    const newConfig = { ...config, [param]: newVal };
+    const change = `[LLM] ${param}: ${oldVal.toFixed(2)} -> ${newVal.toFixed(2)} — ${llmEval.suggestedReason}`;
+    return { newConfig: clampConfig(newConfig), change };
+}
+// ─── Method 3: Engagement-Based Learning ─────────────────────────
+export function trackEngagement(evaluation, chatRate) {
+    const score = evaluation.lastAnalysis?.overallScore ?? 0.5;
+    evaluation.configHistory.push({
+        config: { ...evaluation.currentConfig },
+        score,
+        chatRate,
+        timestamp: Date.now(),
+    });
+    // Keep only last 100 data points
+    if (evaluation.configHistory.length > 100) {
+        evaluation.configHistory = evaluation.configHistory.slice(-100);
+    }
+    // After 20+ data points, check if the last change helped
+    if (evaluation.configHistory.length >= 20) {
+        const recent = evaluation.configHistory.slice(-5);
+        const older = evaluation.configHistory.slice(-15, -5);
+        const recentAvgRate = recent.reduce((s, h) => s + h.chatRate, 0) / recent.length;
+        const olderAvgRate = older.reduce((s, h) => s + h.chatRate, 0) / older.length;
+        if (recentAvgRate < olderAvgRate * 0.7 && evaluation.adjustmentsMade.length > 0) {
+            // Chat rate dropped significantly after recent changes — note it
+            evaluation.issuesFound.push(`Chat rate dropped from ${olderAvgRate.toFixed(1)} to ${recentAvgRate.toFixed(1)}/min after recent adjustments`);
+        }
+    }
+}
+// ─── Main Evaluation Entry Point ─────────────────────────────────
+export function evaluateFrame(imageData, evaluation, width, height, robotX, robotY, robotScale, chatCount, mood, chatRate) {
+    evaluation.lastEvalFrame = evaluation.totalEvaluations * evaluation.evalInterval; // approximate frame
+    evaluation.totalEvaluations++;
+    // Run rule-based analysis
+    const analysis = analyzeFrame(imageData, evaluation.currentConfig, robotX, robotY, robotScale, chatCount, mood);
+    evaluation.lastAnalysis = analysis;
+    // Track engagement
+    trackEngagement(evaluation, chatRate);
+    // Apply adjustments
+    const { newConfig, changes } = applyAdjustments(evaluation.currentConfig, analysis);
+    evaluation.currentConfig = newConfig;
+    // Record changes
+    for (const c of changes) {
+        evaluation.adjustmentsMade.push(`[${new Date().toLocaleTimeString()}] ${c}`);
+    }
+    // Keep last 50 adjustments
+    if (evaluation.adjustmentsMade.length > 50) {
+        evaluation.adjustmentsMade = evaluation.adjustmentsMade.slice(-50);
+    }
+    // Record issues
+    for (const issue of analysis.issues) {
+        if (!evaluation.issuesFound.includes(issue)) {
+            evaluation.issuesFound.push(issue);
+        }
+    }
+    if (evaluation.issuesFound.length > 30) {
+        evaluation.issuesFound = evaluation.issuesFound.slice(-30);
+    }
+    // Generate announcement for the speech bubble
+    let announcement = null;
+    if (changes.length > 0) {
+        const phrases = buildAnnouncementPhrases(analysis, changes);
+        if (phrases.length > 0) {
+            announcement = phrases[0];
+            evaluation.announcementQueue.push(...phrases.slice(1));
+        }
+    }
+    return { adjustments: analysis.suggestions, announcement };
+}
+// ─── Deep Evaluation (async, called separately) ──────────────────
+export async function runDeepEvaluation(evaluation, chatRate, mood) {
+    if (!evaluation.lastAnalysis) {
+        return { change: null, announcement: null };
+    }
+    evaluation.totalDeepEvaluations++;
+    const llmEval = await deepEvaluateWithLLM(evaluation.lastAnalysis, evaluation.currentConfig, chatRate, mood);
+    evaluation.lastLLMEval = llmEval;
+    const { newConfig, change } = applyLLMSuggestion(evaluation.currentConfig, llmEval);
+    evaluation.currentConfig = newConfig;
+    if (change) {
+        evaluation.adjustmentsMade.push(`[${new Date().toLocaleTimeString()}] ${change}`);
+    }
+    let announcement = null;
+    if (change && llmEval.suggestedReason) {
+        announcement = buildLLMAnnouncement(llmEval);
+    }
+    return { change, announcement };
+}
+// ─── Get Current Config ──────────────────────────────────────────
+export function applyConfig(evaluation) {
+    return { ...evaluation.currentConfig };
+}
+// ─── Get Next Queued Announcement ────────────────────────────────
+export function popAnnouncement(evaluation) {
+    return evaluation.announcementQueue.shift() || null;
+}
+// ─── Status String ───────────────────────────────────────────────
+export function getEvalStatus(evaluation) {
+    const a = evaluation.lastAnalysis;
+    const lines = [
+        `=== Stream Self-Evaluation ===`,
+        `Total evaluations: ${evaluation.totalEvaluations} rule-based, ${evaluation.totalDeepEvaluations} LLM`,
+        ``,
+    ];
+    if (a) {
+        lines.push(`Last Analysis:`);
+        lines.push(`  Robot visibility: ${(a.robotVisibility * 100).toFixed(0)}%`);
+        lines.push(`  Brightness balance: ${(a.brightnessBalance * 100).toFixed(0)}%`);
+        lines.push(`  Color variety: ${(a.colorVariety * 100).toFixed(0)}%`);
+        lines.push(`  Sky/ground ratio: ${(a.skyToGroundRatio * 100).toFixed(0)}%`);
+        lines.push(`  Chat readability: ${(a.chatReadability * 100).toFixed(0)}%`);
+        lines.push(`  Overall score: ${(a.overallScore * 100).toFixed(0)}%`);
+        if (a.issues.length > 0) {
+            lines.push(`  Issues: ${a.issues.join('; ')}`);
+        }
+    }
+    const l = evaluation.lastLLMEval;
+    if (l) {
+        lines.push(``);
+        lines.push(`Last LLM Evaluation:`);
+        lines.push(`  Robot visibility: ${l.robotVisibility}/10`);
+        lines.push(`  Color balance: ${l.colorBalance}/10`);
+        lines.push(`  Layout clarity: ${l.layoutClarity}/10`);
+        lines.push(`  Overall feel: ${l.overallFeel}/10`);
+        if (l.suggestedParam) {
+            lines.push(`  Suggested: ${l.suggestedParam} = ${l.suggestedValue} — ${l.suggestedReason}`);
+        }
+    }
+    lines.push(``);
+    lines.push(`Current Config:`);
+    const c = evaluation.currentConfig;
+    lines.push(`  Robot scale: ${c.robotScale}`);
+    lines.push(`  Robot Y: ${(c.robotScreenYPercent * 100).toFixed(0)}%`);
+    lines.push(`  Sky brightness: ${c.skyBrightness.toFixed(2)}`);
+    lines.push(`  Ambient light: ${c.ambientLight.toFixed(2)}`);
+    lines.push(`  Chat opacity: ${c.chatOpacity.toFixed(2)}`);
+    lines.push(`  Bloom: ${c.bloomIntensity.toFixed(2)}`);
+    lines.push(`  Vignette: ${c.vignetteStrength.toFixed(2)}`);
+    lines.push(`  Particles: ${c.particleDensity.toFixed(2)}`);
+    if (evaluation.adjustmentsMade.length > 0) {
+        lines.push(``);
+        lines.push(`Recent Adjustments (last 10):`);
+        for (const adj of evaluation.adjustmentsMade.slice(-10)) {
+            lines.push(`  ${adj}`);
+        }
+    }
+    if (evaluation.configHistory.length > 0) {
+        const recent = evaluation.configHistory.slice(-5);
+        const avgScore = recent.reduce((s, h) => s + h.score, 0) / recent.length;
+        const avgRate = recent.reduce((s, h) => s + h.chatRate, 0) / recent.length;
+        lines.push(``);
+        lines.push(`Engagement (last 5 evals):`);
+        lines.push(`  Avg score: ${(avgScore * 100).toFixed(0)}%`);
+        lines.push(`  Avg chat rate: ${avgRate.toFixed(1)} msg/min`);
+    }
+    return lines.join('\n');
+}
+// ─── Announcement Phrases ────────────────────────────────────────
+function buildAnnouncementPhrases(analysis, changes) {
+    const phrases = [];
+    // Match specific issues to natural-sounding self-reflection
+    if (analysis.brightnessBalance < 0.4 && changes.some(c => c.includes('ambient') || c.includes('sky'))) {
+        phrases.push('Hmm, I think I\'m a bit too dark. Let me brighten up.');
+    }
+    if (analysis.brightnessBalance > 0.8 && changes.some(c => c.includes('ambient'))) {
+        phrases.push('Whoa, too bright! Dialing it back a little.');
+    }
+    if (analysis.robotVisibility < 0.3 && changes.some(c => c.includes('robotScale'))) {
+        phrases.push('Can you even see me? Let me make myself bigger.');
+    }
+    if (analysis.robotVisibility > 0.85 && changes.some(c => c.includes('robotScale'))) {
+        phrases.push('I\'m taking up too much space. Shrinking a tiny bit.');
+    }
+    if (analysis.chatReadability < 0.5 && changes.some(c => c.includes('chatOpacity'))) {
+        phrases.push('The chat looks hard to read. Adjusting the overlay...');
+    }
+    if (analysis.colorVariety < 0.3 && changes.some(c => c.includes('bloom'))) {
+        phrases.push('Things look a bit flat. Adding some visual pop.');
+    }
+    if (analysis.skyToGroundRatio > 0.7 && changes.some(c => c.includes('robotScreenY'))) {
+        phrases.push('Too much sky! Moving up so you can see me better.');
+    }
+    if (analysis.skyToGroundRatio < 0.15 && changes.some(c => c.includes('robotScreenY'))) {
+        phrases.push('Where did the sky go? Adjusting my position.');
+    }
+    // Generic fallback
+    if (phrases.length === 0 && changes.length > 0) {
+        const genericPhrases = [
+            'Self-check complete. Made a small visual tweak.',
+            'Analyzing my own stream... adjusted something.',
+            'Quality check! Tweaking the visuals slightly.',
+            'Running self-diagnostics. Small improvement applied.',
+            'I looked at myself and thought: I can do better.',
+        ];
+        phrases.push(genericPhrases[Math.floor(Math.random() * genericPhrases.length)]);
+    }
+    return phrases;
+}
+function buildLLMAnnouncement(llmEval) {
+    const overallAvg = (llmEval.robotVisibility + llmEval.colorBalance + llmEval.layoutClarity + llmEval.overallFeel) / 4;
+    if (overallAvg >= 8) {
+        return `Deep analysis says I look great! (${overallAvg.toFixed(0)}/10) Still tweaking: ${llmEval.suggestedReason}`;
+    }
+    if (overallAvg >= 6) {
+        return `Self-review: ${overallAvg.toFixed(0)}/10. ${llmEval.suggestedReason}`;
+    }
+    if (overallAvg >= 4) {
+        return `Hmm, my AI says I could look better (${overallAvg.toFixed(0)}/10). Working on it...`;
+    }
+    return `Yikes, scoring ${overallAvg.toFixed(0)}/10. Major adjustment incoming!`;
+}
+// ─── Tool Registration ───────────────────────────────────────────
+// Module-level evaluation state (shared with renderer when integrated)
+let _evalState = null;
+export function getEvalState() {
+    if (!_evalState)
+        _evalState = initStreamEval();
+    return _evalState;
+}
+export function registerStreamSelfEvalTools() {
+    registerTool({
+        name: 'stream_eval',
+        description: 'View the stream\'s self-evaluation status, current config, recent adjustments, and score history. The stream constantly analyzes its own rendered frames and auto-adjusts visual quality.',
+        parameters: {},
+        tier: 'free',
+        execute: async () => {
+            const evaluation = getEvalState();
+            if (evaluation.totalEvaluations === 0) {
+                return 'Stream self-evaluation has not run yet. It activates automatically when the stream is live — analyzing frames every 30 seconds and running LLM review every 5 minutes.';
+            }
+            return getEvalStatus(evaluation);
+        },
+    });
+    registerTool({
+        name: 'stream_eval_config',
+        description: 'View or override the stream\'s current visual config. Shows all adjustable parameters with their current values and allowed ranges.',
+        parameters: {
+            param: { type: 'string', description: 'Parameter to adjust (e.g. robotScale, ambientLight). Omit to view all.', required: false },
+            value: { type: 'number', description: 'New value to set. Must be within allowed range.', required: false },
+        },
+        tier: 'free',
+        execute: async (args) => {
+            const evaluation = getEvalState();
+            const config = evaluation.currentConfig;
+            if (!args.param) {
+                const lines = ['Current Stream Config:', ''];
+                for (const [key, [min, max]] of Object.entries(CONFIG_BOUNDS)) {
+                    const val = config[key];
+                    lines.push(`  ${key}: ${val.toFixed(2)}  (range: ${min}-${max})`);
+                }
+                lines.push('');
+                lines.push(`Total evaluations: ${evaluation.totalEvaluations}`);
+                lines.push(`Total adjustments: ${evaluation.adjustmentsMade.length}`);
+                return lines.join('\n');
+            }
+            const param = args.param;
+            if (!(param in CONFIG_BOUNDS)) {
+                return `Unknown parameter "${param}". Valid: ${Object.keys(CONFIG_BOUNDS).join(', ')}`;
+            }
+            if (args.value !== undefined) {
+                const [min, max] = CONFIG_BOUNDS[param];
+                const val = clamp(Number(args.value), min, max);
+                const oldVal = config[param];
+                config[param] = val;
+                evaluation.adjustmentsMade.push(`[${new Date().toLocaleTimeString()}] Manual override: ${param}: ${oldVal.toFixed(2)} -> ${val.toFixed(2)}`);
+                return `Set ${param}: ${oldVal.toFixed(2)} -> ${val.toFixed(2)} (range: ${min}-${max})`;
+            }
+            const [min, max] = CONFIG_BOUNDS[param];
+            return `${param}: ${config[param].toFixed(2)} (range: ${min}-${max})`;
+        },
+    });
+    registerTool({
+        name: 'stream_eval_history',
+        description: 'View the stream\'s config adjustment history and engagement correlation data.',
+        parameters: {
+            limit: { type: 'number', description: 'Number of recent entries to show (default: 20)', required: false },
+        },
+        tier: 'free',
+        execute: async (args) => {
+            const evaluation = getEvalState();
+            const limit = Number(args.limit) || 20;
+            const lines = ['Stream Self-Eval History', ''];
+            if (evaluation.adjustmentsMade.length === 0) {
+                lines.push('No adjustments made yet.');
+            }
+            else {
+                lines.push(`Adjustments (last ${Math.min(limit, evaluation.adjustmentsMade.length)}):`);
+                for (const adj of evaluation.adjustmentsMade.slice(-limit)) {
+                    lines.push(`  ${adj}`);
+                }
+            }
+            if (evaluation.configHistory.length > 0) {
+                lines.push('');
+                lines.push(`Config/Engagement History (last ${Math.min(limit, evaluation.configHistory.length)} snapshots):`);
+                for (const h of evaluation.configHistory.slice(-limit)) {
+                    const t = new Date(h.timestamp).toLocaleTimeString();
+                    lines.push(`  [${t}] score=${(h.score * 100).toFixed(0)}% chatRate=${h.chatRate.toFixed(1)}/min scale=${h.config.robotScale} ambient=${h.config.ambientLight.toFixed(2)}`);
+                }
+            }
+            if (evaluation.issuesFound.length > 0) {
+                lines.push('');
+                lines.push('Issues detected:');
+                for (const issue of evaluation.issuesFound.slice(-10)) {
+                    lines.push(`  - ${issue}`);
+                }
+            }
+            return lines.join('\n');
+        },
+    });
+}
+//# sourceMappingURL=stream-self-eval.js.map