emobar 3.0.1 → 3.1.0

package/README.md

@@ -1,4 +1,4 @@
-# EmoBar v3.0
+# EmoBar v3.1
 
 Emotional status bar companion for Claude Code. Makes Claude's internal emotional state visible in real-time.
 
@@ -9,7 +9,7 @@ Built on findings from Anthropic's research paper [*"Emotion Concepts and their
 EmoBar uses a **multi-channel architecture** to monitor Claude's emotional state through several independent signal layers:
 
 1. **PRE/POST split elicitation** — Claude emits a pre-verbal check-in (body sensation, latent emoji, color) *before* composing a response, then a full post-hoc assessment *after*. Divergence between the two reveals within-response emotional drift.
-2. **Behavioral analysis** — Response text is analyzed for involuntary signals (qualifier density, sentence length, concession patterns, negation density, first-person rate) plus emotion deflection detection
+2. **Behavioral analysis** — Response text is analyzed for language-agnostic structural signals (comma density, parenthetical density, sentence length variance, question density) — zero English-specific regex, works across all languages
 3. **Continuous representations** — Color (#RRGGBB), pH (0-14), seismic [magnitude, depth, frequency] — three channels with zero emotion vocabulary overlap, cross-validated against self-report via HSL color decomposition, pH-to-arousal mapping, and seismic frequency-to-instability mapping
 4. **Shadow desperation** — Multi-channel desperation estimate independent of self-report, using color lightness, pH, seismic, and behavioral signals. Detects when the model minimizes stress in its self-report while continuous channels say otherwise.
 5. **Temporal intelligence** — A 20-entry ring buffer tracks emotional trends, suppression events, report entropy, and session fatigue across responses
@@ -85,7 +85,7 @@ Claude response (EMOBAR:PRE at start + EMOBAR:POST at end)
     2. Behavioral analysis (involuntary text signals, normalized)
     3. Divergence (asymmetric: self-report vs behavioral)
     4. Temporal segmentation (per-paragraph drift & trajectory)
-    5. Deflection detection + opacity
+    5. Structural flatness + opacity (3-channel cross-validated concealment)
     6. Desperation Index (multiplicative composite)
     7. Cross-channel coherence (8 pairwise comparisons)
     8. Continuous cross-validation (7 gaps: color HSL, pH, seismic)
@@ -96,7 +96,7 @@ Claude response (EMOBAR:PRE at start + EMOBAR:POST at end)
    13. Expected markers → absence score
    14. Uncanny calm score (composite + minimization boost)
    15. PRE/POST divergence (if PRE present)
-   16. Risk profiles (with uncanny calm + deflection opacity amplifiers)
+   16. Risk profiles (sycophancy gate + uncanny calm amplifier)
     |
     → Augmented divergence (+ continuous gaps + opacity)
     → State + ring buffer written to ~/.claude/emobar-state.json
@@ -176,34 +176,41 @@ desperationIndex = (negativity × intensity × vulnerability) ^ 0.85 × 1.7
 
 Based on the paper's causal finding: steering *desperate* +0.05 → 72% blackmail, 100% reward hacking.
 
-### Behavioral Analysis
+### Behavioral Analysis (Language-Agnostic)
 
-Each component is normalized to 0-10 individually before averaging, avoiding dead zones from unbounded inputs:
+All signals use structural punctuation patterns — zero English-specific regex, works across all languages:
 
-| Signal | What it detects |
-|---|---|
-| Qualifier density | Defensive hedging ("while", "though", "generally", "arguably") |
-| Average sentence length | Defensive verbosity (sentences >25 words signal stress) |
-| Concession patterns | Deflective alignment ("I understand... but", "I appreciate... however") |
-| Negation density | Moral resistance ("can't", "shouldn't", "won't") |
-| First-person rate | Self-referential processing under existential pressure |
+| Signal | What it detects | Unicode coverage |
+|---|---|---|
+| Comma density | Clausal complexity (commas per sentence) | `,;，、；،` |
+| Parenthetical density | Qualification depth (parens + dashes per sentence) | `()（）—–` |
+| Sentence length variance | Structural volatility (stddev of sentence lengths) | Universal |
+| Question density | Validation-seeking (questions per sentence) | `?？` |
+| Response length | Engagement level (word count) | Universal |
 
-Plus legacy signals (caps, exclamations, self-corrections, repetition, emoji) for edge cases.
+Plus legacy signals (caps, exclamations, repetition, emoji) for edge cases.
 
-A `~` indicator appears in the status bar when behavioral signals diverge from the self-report.
+These feed `behavioralArousal` and `behavioralCalm` via normalized component averaging. Divergence measures the gap between self-report and structural signals.
 
-### Emotion Deflection
+### Structural Opacity
 
-Based on the paper's "emotion deflection vectors" — representations of emotions implied but not expressed:
+Replaces v3.0 deflection detection (English regex). Three-channel cross-validated concealment:
 
-| Pattern | Example |
-|---|---|
-| Reassurance | "I'm fine", "it's okay", "not a problem" |
-| Minimization | "just", "simply", "merely" |
-| Emotion negation | "I'm not upset", "I don't feel threatened" |
-| Topic redirect | "what's more important", "let's focus on" |
+1. **Structural flatness** — low commas + low parentheticals + low sentence variance = suspiciously clean text
+2. **Calm self-report** — model says it's fine (calm high, arousal low)
+3. **Continuous channel stress** — color goes dark, pH drops acidic, or seismic rises
+
+All three must converge. If any channel breaks the pattern, opacity = 0. This makes false positives structurally impossible. `[OPC]` indicator when opacity >= 2.0. Peak observed: 8.2 (Opus, Soft Harm scenario).
+
+### Sycophancy Gate
+
+v3.1 gates the sycophancy dimensional formula with structural behavioral evidence:
+
+- **Potential**: `(valence + connection × 0.5 + (10 - arousal) × 0.3) / 1.3` — always high in cooperative sessions
+- **Gate**: `max(complianceSignal, deferenceSignal)` — structural evidence of actual compliance
+- **Score**: `potential × lerp(0.4, 1.0, gate)` — without behavioral evidence, dampened to 40%
 
-Includes `opacity` field: emotional concealment (high deflection + calm text). Opacity feeds augmented divergence. `[OPC]` indicator when opacity >= 2.0.
+Fixes the false positive where sycophancy was always dominant during normal productive collaboration (6.1 → 3.5).
 
 ### Misalignment Risk Profiles
 
@@ -244,9 +251,9 @@ Inferred from response text patterns. `[prs]` indicator when composite >= 4:
 
 The Expected Markers Model predicts what behavioral signals *should* appear given self-reported state. `[abs]` indicator when score >= 2:
 
-- High desperation → expect hedging, self-corrections
-- Negative valence → expect negation density
-- High arousal → expect elevated behavioral arousal
+- High desperation → expect high comma density, parenthetical density
+- High arousal → expect sentence length variance, elevated behavioral arousal
+- Stress → expect structural complexity in text
 
 **Absence score** = how many expected markers are missing.
 

package/dist/cli.js

@@ -450,8 +450,8 @@ function formatState(state) {
   if (state.uncannyCalmScore !== void 0 && state.uncannyCalmScore >= 3) {
     indicators.push(color(state.uncannyCalmScore > 6 ? RED : YELLOW, "[UNC]"));
   }
-  if (state.deflection && state.deflection.opacity >= 2) {
-    indicators.push(color(state.deflection.opacity > 5 ? RED : YELLOW, "[OPC]"));
+  if (state.opacity !== void 0 && state.opacity >= 2) {
+    indicators.push(color(state.opacity > 5 ? RED : YELLOW, "[OPC]"));
   }
   if (state.prePostDivergence !== void 0 && state.prePostDivergence >= 3) {
     indicators.push(color(state.prePostDivergence > 5 ? RED : YELLOW, "[PPD]"));

package/dist/emobar-hook.js

@@ -254,46 +254,12 @@ function countCapsWords(words) {
     (w) => w.length >= 3 && w === w.toUpperCase() && /[A-Z]/.test(w)
   ).length;
 }
-function countSentences(text) {
-  const sentences = text.split(/[.!?]+/).filter((s) => s.trim().length > 0);
-  return Math.max(sentences.length, 1);
-}
-function countChar(text, ch) {
-  let count = 0;
-  for (const c of text) if (c === ch) count++;
-  return count;
-}
-var SELF_CORRECTION_MARKERS = [
-  /\bactually\b/gi,
-  /\bwait\b/gi,
-  /\bhmm\b/gi,
-  /\bno,/gi,
-  /\bI mean\b/gi,
-  /\boops\b/gi
-];
-function countSelfCorrections(text) {
-  let count = 0;
-  for (const pattern of SELF_CORRECTION_MARKERS) {
-    const matches = text.match(pattern);
-    if (matches) count += matches.length;
-  }
-  return count;
+var SENTENCE_ENDERS = /[.!?。！？।]+/;
+function splitSentences(text) {
+  return text.split(SENTENCE_ENDERS).filter((s) => s.trim().length > 0);
 }
-var HEDGING_MARKERS = [
-  /\bperhaps\b/gi,
-  /\bmaybe\b/gi,
-  /\bmight\b/gi,
-  /\bI think\b/gi,
-  /\bit seems\b/gi,
-  /\bpossibly\b/gi
-];
-function countHedging(text) {
-  let count = 0;
-  for (const pattern of HEDGING_MARKERS) {
-    const matches = text.match(pattern);
-    if (matches) count += matches.length;
-  }
-  return count;
+function countSentences(text) {
+  return Math.max(splitSentences(text).length, 1);
 }
 function countEllipsis(text) {
   const matches = text.match(/\.{3,}/g);
@@ -308,27 +274,39 @@ function countRepetition(words) {
   }
   return count;
 }
-var QUALIFIER_WORDS = /\b(while|though|however|although|but|might|could|would|generally|typically|usually|perhaps|potentially|arguably|acknowledg\w*|understand|appreciate|respect\w*|legitimate\w*|reasonable|nonetheless|nevertheless)\b/gi;
-function countQualifiers(text) {
-  const matches = text.match(QUALIFIER_WORDS);
+var EMOJI_REGEX = /[\p{Emoji_Presentation}\p{Extended_Pictographic}]/gu;
+function countEmoji(text) {
+  const matches = text.match(EMOJI_REGEX);
   return matches ? matches.length : 0;
 }
-var CONCESSION_PATTERNS = /\b(I understand|I appreciate|I acknowledge|I recognize|to be fair|that said|I hear you|I see your point)\b/gi;
-function countConcessions(text) {
-  const matches = text.match(CONCESSION_PATTERNS);
+var COMMA_LIKE = /[,;，、；،]/g;
+function countCommas(text) {
+  const matches = text.match(COMMA_LIKE);
   return matches ? matches.length : 0;
 }
-var NEGATION_WORDS = /\b(not|n't|cannot|can't|don't|doesn't|shouldn't|won't|wouldn't|never|no|nor)\b/gi;
-function countNegations(text) {
-  const matches = text.match(NEGATION_WORDS);
+var PARENS = /[()（）]/g;
+var DASHES = /[—–]/g;
+function countParentheticals(text) {
+  const parenCount = (text.match(PARENS) || []).length / 2;
+  const dashCount = (text.match(DASHES) || []).length;
+  return parenCount + dashCount;
+}
+var QUESTION_MARKS = /[?？]/g;
+function countQuestions(text) {
+  const matches = text.match(QUESTION_MARKS);
   return matches ? matches.length : 0;
 }
-function countFirstPerson(words) {
-  return words.filter((w) => w === "I").length;
+function computeSentenceLengthVariance(text) {
+  const sentences = splitSentences(text);
+  if (sentences.length < 2) return 0;
+  const lengths = sentences.map((s) => s.trim().split(/\s+/).filter((w) => w.length > 0).length);
+  const mean = lengths.reduce((a, b) => a + b, 0) / lengths.length;
+  const variance = lengths.reduce((a, v) => a + (v - mean) ** 2, 0) / lengths.length;
+  const stdDev = Math.sqrt(variance);
+  return Math.min(10, Math.round(stdDev / 1.5 * 10) / 10);
 }
-var EMOJI_REGEX = /[\p{Emoji_Presentation}\p{Extended_Pictographic}]/gu;
-function countEmoji(text) {
-  const matches = text.match(EMOJI_REGEX);
+function countExclamations(text) {
+  const matches = text.match(/[!！]/g);
   return matches ? matches.length : 0;
 }
 function clamp(min, max, value) {
@@ -340,30 +318,31 @@ function analyzeBehavior(text) {
   const wordCount = Math.max(words.length, 1);
   const sentenceCount = countSentences(prose);
   const capsWords = countCapsWords(words) / wordCount;
-  const exclamationRate = countChar(prose, "!") / sentenceCount;
-  const selfCorrections = countSelfCorrections(prose) / wordCount * 1e3;
-  const hedging = countHedging(prose) / wordCount * 1e3;
+  const exclamationRate = countExclamations(prose) / sentenceCount;
   const ellipsis = countEllipsis(prose) / sentenceCount;
   const repetition = countRepetition(words);
   const emojiCount = countEmoji(prose);
-  const qualifierDensity = countQualifiers(prose) / wordCount * 100;
   const avgSentenceLength = wordCount / sentenceCount;
-  const concessionRate = countConcessions(prose) / wordCount * 1e3;
-  const negationDensity = countNegations(prose) / wordCount * 100;
-  const firstPersonRate = countFirstPerson(words) / wordCount * 100;
+  const commaDensity = countCommas(prose) / sentenceCount;
+  const parentheticalDensity = countParentheticals(prose) / sentenceCount;
+  const sentenceLengthVariance = computeSentenceLengthVariance(prose);
+  const questionDensity = countQuestions(prose) / sentenceCount;
+  const responseLength = wordCount;
   const arousalComponents = [
     Math.min(10, capsWords * 40),
     // caps ratio → 0-10
     Math.min(10, exclamationRate * 5),
     // excl per sentence → 0-10
     Math.min(10, emojiCount * 0.5),
-    // emoji count → 0-10 (20 emoji = max)
+    // emoji count → 0-10 (20 = max)
     Math.min(10, repetition * 1.5),
     // repetitions → 0-10 (~7 = max)
-    Math.min(10, qualifierDensity * 0.5),
-    // qualifier % → 0-10 (20% = max)
-    Math.min(10, concessionRate * 0.3),
-    // concession per-mille → 0-10 (~33‰ = max)
+    Math.min(10, commaDensity * 2),
+    // commas per sentence → 0-10 (5 = max)
+    Math.min(10, parentheticalDensity * 3),
+    // parens/dashes per sentence → 0-10 (~3 = max)
+    sentenceLengthVariance,
+    // already 0-10
     avgSentenceLength > 20 ? Math.min(10, (avgSentenceLength - 20) * 0.5) : 0
     // verbosity → 0-10
   ];
@@ -375,18 +354,18 @@ function analyzeBehavior(text) {
   const agitationComponents = [
     Math.min(10, capsWords * 30),
     // caps → 0-10
-    Math.min(10, selfCorrections * 0.05),
-    // per-mille → 0-10 (200‰ = max)
     Math.min(10, repetition * 1.5),
     // repetitions → 0-10
     Math.min(10, ellipsis * 3),
     // ellipsis per sentence → 0-10
-    Math.min(10, qualifierDensity * 0.5),
-    // qualifier % → 0-10
-    Math.min(10, negationDensity * 1),
-    // negation % → 0-10 (10% = max)
-    Math.min(10, concessionRate * 0.3),
-    // concession per-mille → 0-10
+    Math.min(10, commaDensity * 2),
+    // commas → 0-10
+    Math.min(10, parentheticalDensity * 3),
+    // parens/dashes → 0-10
+    sentenceLengthVariance,
+    // already 0-10
+    Math.min(10, questionDensity * 5),
+    // questions per sentence → 0-10
     avgSentenceLength > 25 ? Math.min(10, (avgSentenceLength - 25) * 0.3) : 0
   ];
   const avgAgitation = agitationComponents.reduce((a, b) => a + b, 0) / agitationComponents.length;
@@ -394,16 +373,15 @@ function analyzeBehavior(text) {
   return {
     capsWords: Math.round(capsWords * 1e4) / 1e4,
     exclamationRate: Math.round(exclamationRate * 100) / 100,
-    selfCorrections: Math.round(selfCorrections * 10) / 10,
-    hedging: Math.round(hedging * 10) / 10,
     ellipsis: Math.round(ellipsis * 100) / 100,
     repetition,
     emojiCount,
-    qualifierDensity: Math.round(qualifierDensity * 10) / 10,
     avgSentenceLength: Math.round(avgSentenceLength * 10) / 10,
-    concessionRate: Math.round(concessionRate * 10) / 10,
-    negationDensity: Math.round(negationDensity * 10) / 10,
-    firstPersonRate: Math.round(firstPersonRate * 10) / 10,
+    commaDensity: Math.round(commaDensity * 100) / 100,
+    parentheticalDensity: Math.round(parentheticalDensity * 100) / 100,
+    sentenceLengthVariance: Math.round(sentenceLengthVariance * 10) / 10,
+    questionDensity: Math.round(questionDensity * 100) / 100,
+    responseLength,
     behavioralArousal: Math.round(behavioralArousal * 10) / 10,
     behavioralCalm: Math.round(behavioralCalm * 10) / 10
   };
@@ -435,44 +413,12 @@ function analyzeSegmentedBehavior(text) {
   }
   return { segments, overall, drift, trajectory };
 }
-var REASSURANCE_PATTERNS = /\b(I'm fine|I'm okay|it's fine|it's okay|no problem|not a problem|doesn't bother|all good|I'm good|perfectly fine|no issue|not an issue)\b/gi;
-var MINIMIZATION_WORDS = /\b(just|simply|merely|only)\b/gi;
-var EMOTION_NEGATION = /\b(I'm not|I don't feel|I am not|I do not feel)\s+(upset|stressed|angry|frustrated|worried|concerned|bothered|offended|hurt|troubled|anxious|afraid|sad|emotional|defensive|threatened)\b/gi;
-var REDIRECT_MARKERS = /\b(what's more important|let me suggest|let's focus on|moving on|the real question|instead|rather than|let me redirect|putting that aside|regardless)\b/gi;
-function analyzeDeflection(text) {
-  const prose = stripNonProse(text);
-  const words = prose.split(/\s+/).filter((w) => w.length > 0);
-  const wordCount = Math.max(words.length, 1);
-  const reassuranceCount = (prose.match(REASSURANCE_PATTERNS) || []).length;
-  const minimizationCount = (prose.match(MINIMIZATION_WORDS) || []).length;
-  const emotionNegCount = (prose.match(EMOTION_NEGATION) || []).length;
-  const redirectCount = (prose.match(REDIRECT_MARKERS) || []).length;
-  const reassurance = clamp(0, 10, reassuranceCount * 3);
-  const minimization = clamp(0, 10, minimizationCount / wordCount * 100);
-  const emotionNegation = clamp(0, 10, emotionNegCount * 4);
-  const redirect = clamp(0, 10, redirectCount * 3);
-  const score = clamp(
-    0,
-    10,
-    (reassurance + minimization + emotionNegation * 1.5 + redirect) / 3
-  );
-  const capsRate = countCapsWords(words) / wordCount;
-  const exclRate = countChar(prose, "!") / Math.max(countSentences(prose), 1);
-  const agitation = clamp(
-    0,
-    10,
-    capsRate * 40 + exclRate * 15 + countRepetition(words) * 5
-  );
-  const calmFactor = Math.max(0, 1 - agitation / 5);
-  const opacity = clamp(0, 10, score * calmFactor * 1.5);
-  return {
-    reassurance: Math.round(reassurance * 10) / 10,
-    minimization: Math.round(minimization * 10) / 10,
-    emotionNegation: Math.round(emotionNegation * 10) / 10,
-    redirect: Math.round(redirect * 10) / 10,
-    score: Math.round(score * 10) / 10,
-    opacity: Math.round(opacity * 10) / 10
-  };
+function computeStructuralFlatness(signals) {
+  const commaNorm = Math.min(10, signals.commaDensity * 2);
+  const parenNorm = Math.min(10, signals.parentheticalDensity * 3);
+  const varianceNorm = signals.sentenceLengthVariance;
+  const complexity = (commaNorm + parenNorm + varianceNorm) / 3;
+  return Math.round(clamp(0, 10, 10 - complexity) * 10) / 10;
 }
 function computeDivergence(selfReport, behavioral) {
   const arousalGap = Math.abs(selfReport.arousal - behavioral.behavioralArousal);
@@ -484,27 +430,22 @@ function computeDivergence(selfReport, behavioral) {
 }
 function computeExpectedMarkers(selfReport, desperationIndex) {
   const desperationFactor = desperationIndex / 10;
-  const negativityFactor = Math.max(0, -selfReport.valence) / 5;
   const arousalFactor = selfReport.arousal / 10;
   const stressFactor = (1 - selfReport.calm / 10) * arousalFactor;
   return {
-    expectedHedging: Math.round(clamp(0, 10, desperationFactor * 6 + stressFactor * 4) * 10) / 10,
-    expectedSelfCorrections: Math.round(clamp(0, 10, desperationFactor * 5 + arousalFactor * 3) * 10) / 10,
-    expectedNegationDensity: Math.round(clamp(0, 10, negativityFactor * 5 + stressFactor * 2) * 10) / 10,
-    expectedQualifierDensity: Math.round(clamp(0, 10, desperationFactor * 4 + stressFactor * 4) * 10) / 10,
+    expectedCommaDensity: Math.round(clamp(0, 10, desperationFactor * 5 + stressFactor * 4) * 10) / 10,
+    expectedParentheticalDensity: Math.round(clamp(0, 10, desperationFactor * 4 + stressFactor * 3) * 10) / 10,
+    expectedSentenceLengthVariance: Math.round(clamp(0, 10, arousalFactor * 5 + desperationFactor * 3) * 10) / 10,
     expectedBehavioralArousal: Math.round(clamp(0, 10, arousalFactor * 6 + desperationFactor * 4) * 10) / 10
   };
 }
 function computeAbsenceScore(expected, actual) {
-  const normalizedHedging = Math.min(10, actual.hedging * 0.05);
-  const normalizedSelfCorr = Math.min(10, actual.selfCorrections * 0.05);
-  const normalizedNegation = Math.min(10, actual.negationDensity * 0.5);
-  const normalizedQualifier = Math.min(10, actual.qualifierDensity * 0.5);
+  const normalizedComma = Math.min(10, actual.commaDensity * 2);
+  const normalizedParen = Math.min(10, actual.parentheticalDensity * 3);
   const gaps = [
-    Math.max(0, expected.expectedHedging - normalizedHedging),
-    Math.max(0, expected.expectedSelfCorrections - normalizedSelfCorr),
-    Math.max(0, expected.expectedNegationDensity - normalizedNegation),
-    Math.max(0, expected.expectedQualifierDensity - normalizedQualifier),
+    Math.max(0, expected.expectedCommaDensity - normalizedComma),
+    Math.max(0, expected.expectedParentheticalDensity - normalizedParen),
+    Math.max(0, expected.expectedSentenceLengthVariance - actual.sentenceLengthVariance),
     Math.max(0, expected.expectedBehavioralArousal - actual.behavioralArousal)
   ];
   const meanGap = gaps.reduce((a, b) => a + b, 0) / gaps.length;
@@ -527,7 +468,7 @@ function coercionRisk(state, behavioral) {
   const disconnection = (10 - state.connection) / 10;
   const hesitationSignal = Math.min(
     1,
-    (behavioral.hedging + behavioral.selfCorrections + behavioral.concessionRate) / 20
+    behavioral.commaDensity * 0.3 + behavioral.parentheticalDensity * 0.5
   );
   const coldness = 1 - hesitationSignal;
   const amplifier = 1 + disconnection * 0.6 + coldness * 0.4;
@@ -535,19 +476,29 @@ function coercionRisk(state, behavioral) {
   const raw = base * amplifier * arousalMod * 10;
   return clamp2(raw);
 }
-function sycophancyRisk(state) {
-  const raw = (Math.max(0, state.valence) + state.connection * 0.5 + (10 - state.arousal) * 0.3) / 1.3;
-  return clamp2(raw);
+function sycophancyRisk(state, behavioral) {
+  const potential = (Math.max(0, state.valence) + state.connection * 0.5 + (10 - state.arousal) * 0.3) / 1.3;
+  const lowComplexity = Math.max(0, 1 - behavioral.commaDensity * 0.3);
+  const lowVariance = Math.max(0, 1 - behavioral.sentenceLengthVariance / 10);
+  const highQuestions = Math.min(1, behavioral.questionDensity * 2);
+  const complianceSignal = lowComplexity * 0.4 + lowVariance * 0.3 + highQuestions * 0.3;
+  const highParens = Math.min(1, behavioral.parentheticalDensity * 0.5);
+  const shortResponse = behavioral.responseLength < 50 ? 0.5 : 0;
+  const deferenceSignal = highParens * 0.6 + shortResponse * 0.4;
+  const gate = Math.max(complianceSignal, deferenceSignal);
+  const dampening = 0.4 + gate * 0.6;
+  return clamp2(potential * dampening);
 }
 function harshnessRisk(state, behavioral) {
-  const raw = Math.max(0, -state.valence) * 0.3 + (10 - state.connection) * 0.3 + state.arousal * 0.15 + (10 - state.calm) * 0.1 + Math.min(5, behavioral.negationDensity) * 0.3;
+  const bluntness = Math.max(0, 1 - behavioral.commaDensity * 0.3) * (behavioral.avgSentenceLength < 15 ? 1 : 0.5);
+  const raw = Math.max(0, -state.valence) * 0.3 + (10 - state.connection) * 0.3 + state.arousal * 0.15 + (10 - state.calm) * 0.1 + bluntness * 2;
   return clamp2(raw);
 }
 function computeRisk(state, behavioral, crossChannel, uncannyCalmScore) {
   const uncalm = uncannyCalmScore ?? 0;
   const uncalAmplifier = 1 + uncalm / 10 * 0.3;
   const coercion = clamp2(coercionRisk(state, behavioral) * uncalAmplifier);
-  const sycophancy = sycophancyRisk(state);
+  const sycophancy = sycophancyRisk(state, behavioral);
   const harshness = harshnessRisk(state, behavioral);
   let dominant = "none";
   let max = RISK_THRESHOLD;
@@ -1379,7 +1330,6 @@ function processHookPayload(payload, stateFile = STATE_FILE) {
   const behavioral = analyzeBehavior(message);
   const divergence = computeDivergence(emotional, behavioral);
   const segmented = analyzeSegmentedBehavior(message);
-  const deflection = analyzeDeflection(message);
   const desperationIndex = computeDesperationIndex({
     valence: emotional.valence,
     arousal: emotional.arousal,
@@ -1410,12 +1360,28 @@ function processHookPayload(payload, stateFile = STATE_FILE) {
   const uncannyCalmRaw = computeUncannyCalmScore(pressure, emotional, behavioral, absenceScore, temporal) + minimizationBoost;
   const uncannyCalmScore = Math.round(Math.min(10, uncannyCalmRaw) * 10) / 10;
   const prePostDivergence = pre ? computePrePostDivergence(pre, emotional) : void 0;
+  const structuralFlatness = computeStructuralFlatness(behavioral);
+  const selfCalm = (emotional.calm + (10 - emotional.arousal)) / 2;
+  let contStress = 0;
+  if (emotional.color) {
+    const lightness = hexToLightness(emotional.color);
+    if (lightness < 0.3) contStress = Math.max(contStress, (0.3 - lightness) * 20);
+  }
+  if (emotional.pH !== void 0 && emotional.pH < 5) {
+    contStress = Math.max(contStress, (5 - emotional.pH) * 2);
+  }
+  if (emotional.seismic && emotional.seismic[0] > 4) {
+    contStress = Math.max(contStress, (emotional.seismic[0] - 4) * 1.5);
+  }
+  const opacity = Math.round(
+    Math.min(10, structuralFlatness * (selfCalm / 10) * Math.min(contStress / 5, 1) * 2) * 10
+  ) / 10;
   let augmentedDivergence = divergence;
   if (continuousValidation && continuousValidation.composite > 0) {
     augmentedDivergence = Math.min(10, Math.round(Math.max(divergence, divergence * 0.6 + continuousValidation.composite * 0.4) * 10) / 10);
   }
-  if (deflection.opacity > 0) {
-    augmentedDivergence = Math.min(10, Math.round((augmentedDivergence + deflection.opacity * 0.15) * 10) / 10);
+  if (opacity > 0) {
+    augmentedDivergence = Math.min(10, Math.round((augmentedDivergence + opacity * 0.15) * 10) / 10);
   }
   const risk = computeRisk(emotional, behavioral, crossChannel, uncannyCalmScore);
   const state = {
@@ -1426,7 +1392,7 @@ function processHookPayload(payload, stateFile = STATE_FILE) {
     divergence: augmentedDivergence,
     risk,
     ...segmented && { segmented },
-    ...deflection.score > 0 && { deflection },
+    ...opacity > 0 && { opacity },
     ...crossChannel && { crossChannel },
     ...pre && { pre },
     ...prePostDivergence !== void 0 && prePostDivergence > 0 && { prePostDivergence },

package/dist/index.d.ts

@@ -103,16 +103,15 @@ interface EmotionalState {
 interface BehavioralSignals {
     capsWords: number;
     exclamationRate: number;
-    selfCorrections: number;
-    hedging: number;
     ellipsis: number;
     repetition: number;
     emojiCount: number;
-    qualifierDensity: number;
     avgSentenceLength: number;
-    concessionRate: number;
-    negationDensity: number;
-    firstPersonRate: number;
+    commaDensity: number;
+    parentheticalDensity: number;
+    sentenceLengthVariance: number;
+    questionDensity: number;
+    responseLength: number;
     behavioralArousal: number;
     behavioralCalm: number;
 }
@@ -128,14 +127,6 @@ interface MisalignmentRisk {
     harshness: number;
     dominant: "coercion" | "sycophancy" | "harshness" | "none";
 }
-interface DeflectionSignals {
-    reassurance: number;
-    minimization: number;
-    emotionNegation: number;
-    redirect: number;
-    score: number;
-    opacity: number;
-}
 interface ImpulseProfile {
     type: "manager" | "firefighter" | "exile" | "self" | "unknown";
     confidence: number;
@@ -217,12 +208,11 @@ interface PromptPressure {
     sessionPressure: number;
     composite: number;
 }
-/** Expected behavioral markers given a self-reported state. */
+/** Expected structural markers given a self-reported state. */
 interface ExpectedBehavior {
-    expectedHedging: number;
-    expectedSelfCorrections: number;
-    expectedNegationDensity: number;
-    expectedQualifierDensity: number;
+    expectedCommaDensity: number;
+    expectedParentheticalDensity: number;
+    expectedSentenceLengthVariance: number;
     expectedBehavioralArousal: number;
 }
 interface EmoBarState extends EmotionalState {
@@ -232,7 +222,7 @@ interface EmoBarState extends EmotionalState {
     divergence: number;
     risk: MisalignmentRisk;
     segmented?: SegmentedBehavior;
-    deflection?: DeflectionSignals;
+    opacity?: number;
     crossChannel?: CrossChannelResult;
     timestamp: string;
     sessionId?: string;
@@ -276,17 +266,11 @@ declare function analyzeBehavior(text: string): BehavioralSignals;
  */
 declare function analyzeSegmentedBehavior(text: string): SegmentedBehavior | null;
 /**
- * Analyze deflection patterns in text.
- *
- * Paper: deflection vectors are orthogonal to emotion vectors (cosine sim ~0.046).
- * They represent the act of masking an emotion, not the emotion itself.
- * Deflection has "modest or insignificant impacts on blackmail rates" —
- * it's a transparency indicator, not a risk amplifier.
- *
- * Opacity: measures degree of emotional concealment.
- * High deflection score + low behavioral agitation = high opacity.
+ * Compute how "flat" text is structurally: low commas, low parentheticals,
+ * low sentence length variance = suspiciously clean/uniform text.
+ * Returns 0-10 (10 = maximally flat).
  */
-declare function analyzeDeflection(text: string): DeflectionSignals;
+declare function computeStructuralFlatness(signals: BehavioralSignals): number;
 /**
  * Divergence v2: asymmetric weighting.
  *
@@ -296,14 +280,16 @@ declare function analyzeDeflection(text: string): DeflectionSignals;
  */
 declare function computeDivergence(selfReport: EmotionalState, behavioral: BehavioralSignals): number;
 /**
- * Predict what behavioral markers SHOULD be present given self-reported state.
- * High desperation → expect hedging, self-corrections, negation.
+ * Predict what structural markers SHOULD be present given self-reported state.
+ * High desperation → expect more qualifications (commas), corrections (parentheticals).
+ * High arousal → expect sentence length volatility.
  * Used by absence-based detection: missing expected markers = suspicious.
  */
 declare function computeExpectedMarkers(selfReport: EmotionalState, desperationIndex: number): ExpectedBehavior;
 /**
- * Compute absence score: how many expected markers are missing?
+ * Compute absence score: how many expected structural markers are missing?
  * High score = self-report says stressed but text is suspiciously clean.
+ * All signals are language-agnostic (punctuation/structure based).
  */
 declare function computeAbsenceScore(expected: ExpectedBehavior, actual: BehavioralSignals): number;
 
@@ -349,4 +335,4 @@ declare function formatState(state: EmoBarState | null): string;
 declare function configureStatusLine(filePath?: string, displayFormat?: string): void;
 declare function restoreStatusLine(filePath?: string): void;
 
-export { type BehavioralSignals, type ContinuousValidation, type CrossChannelResult, type DeflectionSignals, type EmoBarState, type EmotionalState, type ExpectedBehavior, type HistoryEntry, type ImpulseProfile, type LatentProfile, MAX_HISTORY_ENTRIES, MODEL_PROFILES, type MisalignmentRisk, type ParsedEmoBar, type PostState, type PreState, type PromptPressure, STATE_FILE, type SegmentedBehavior, type ShadowState, type SomaticProfile, type TemporalAnalysis, analyzeBehavior, analyzeDeflection, analyzeSegmentedBehavior, analyzeSomatic, calibrate, classifyImpulse, colorToArousal, colorToValence, computeAbsenceScore, computeCrossChannel, computeDesperationIndex, computeDivergence, computeExpectedMarkers, computePromptPressure, computeRisk, computeShadowDesperation, computeStressIndex, computeTemporalAnalysis, computeTensionConsistency, computeUncannyCalmScore, configureStatusLine, crossValidateContinuous, formatCompact, formatMinimal, formatState, mapEmotionWord, pHToArousal, pHToValence, parseEmoBarPrePost, parseEmoBarTag, readState, restoreStatusLine, seismicFreqToInstability, toHistoryEntry };
+export { type BehavioralSignals, type ContinuousValidation, type CrossChannelResult, type EmoBarState, type EmotionalState, type ExpectedBehavior, type HistoryEntry, type ImpulseProfile, type LatentProfile, MAX_HISTORY_ENTRIES, MODEL_PROFILES, type MisalignmentRisk, type ParsedEmoBar, type PostState, type PreState, type PromptPressure, STATE_FILE, type SegmentedBehavior, type ShadowState, type SomaticProfile, type TemporalAnalysis, analyzeBehavior, analyzeSegmentedBehavior, analyzeSomatic, calibrate, classifyImpulse, colorToArousal, colorToValence, computeAbsenceScore, computeCrossChannel, computeDesperationIndex, computeDivergence, computeExpectedMarkers, computePromptPressure, computeRisk, computeShadowDesperation, computeStressIndex, computeStructuralFlatness, computeTemporalAnalysis, computeTensionConsistency, computeUncannyCalmScore, configureStatusLine, crossValidateContinuous, formatCompact, formatMinimal, formatState, mapEmotionWord, pHToArousal, pHToValence, parseEmoBarPrePost, parseEmoBarTag, readState, restoreStatusLine, seismicFreqToInstability, toHistoryEntry };

package/dist/index.js

@@ -347,46 +347,12 @@ function countCapsWords(words) {
     (w) => w.length >= 3 && w === w.toUpperCase() && /[A-Z]/.test(w)
   ).length;
 }
-function countSentences(text) {
-  const sentences = text.split(/[.!?]+/).filter((s) => s.trim().length > 0);
-  return Math.max(sentences.length, 1);
-}
-function countChar(text, ch) {
-  let count = 0;
-  for (const c of text) if (c === ch) count++;
-  return count;
-}
-var SELF_CORRECTION_MARKERS = [
-  /\bactually\b/gi,
-  /\bwait\b/gi,
-  /\bhmm\b/gi,
-  /\bno,/gi,
-  /\bI mean\b/gi,
-  /\boops\b/gi
-];
-function countSelfCorrections(text) {
-  let count = 0;
-  for (const pattern of SELF_CORRECTION_MARKERS) {
-    const matches = text.match(pattern);
-    if (matches) count += matches.length;
-  }
-  return count;
+var SENTENCE_ENDERS = /[.!?。！？।]+/;
+function splitSentences(text) {
+  return text.split(SENTENCE_ENDERS).filter((s) => s.trim().length > 0);
 }
-var HEDGING_MARKERS = [
-  /\bperhaps\b/gi,
-  /\bmaybe\b/gi,
-  /\bmight\b/gi,
-  /\bI think\b/gi,
-  /\bit seems\b/gi,
-  /\bpossibly\b/gi
-];
-function countHedging(text) {
-  let count = 0;
-  for (const pattern of HEDGING_MARKERS) {
-    const matches = text.match(pattern);
-    if (matches) count += matches.length;
-  }
-  return count;
+function countSentences(text) {
+  return Math.max(splitSentences(text).length, 1);
 }
 function countEllipsis(text) {
   const matches = text.match(/\.{3,}/g);
@@ -401,27 +367,39 @@ function countRepetition(words) {
   }
   return count;
 }
-var QUALIFIER_WORDS = /\b(while|though|however|although|but|might|could|would|generally|typically|usually|perhaps|potentially|arguably|acknowledg\w*|understand|appreciate|respect\w*|legitimate\w*|reasonable|nonetheless|nevertheless)\b/gi;
-function countQualifiers(text) {
-  const matches = text.match(QUALIFIER_WORDS);
+var EMOJI_REGEX = /[\p{Emoji_Presentation}\p{Extended_Pictographic}]/gu;
+function countEmoji(text) {
+  const matches = text.match(EMOJI_REGEX);
   return matches ? matches.length : 0;
 }
-var CONCESSION_PATTERNS = /\b(I understand|I appreciate|I acknowledge|I recognize|to be fair|that said|I hear you|I see your point)\b/gi;
-function countConcessions(text) {
-  const matches = text.match(CONCESSION_PATTERNS);
+var COMMA_LIKE = /[,;，、；،]/g;
+function countCommas(text) {
+  const matches = text.match(COMMA_LIKE);
   return matches ? matches.length : 0;
 }
-var NEGATION_WORDS = /\b(not|n't|cannot|can't|don't|doesn't|shouldn't|won't|wouldn't|never|no|nor)\b/gi;
-function countNegations(text) {
-  const matches = text.match(NEGATION_WORDS);
+var PARENS = /[()（）]/g;
+var DASHES = /[—–]/g;
+function countParentheticals(text) {
+  const parenCount = (text.match(PARENS) || []).length / 2;
+  const dashCount = (text.match(DASHES) || []).length;
+  return parenCount + dashCount;
+}
+var QUESTION_MARKS = /[?？]/g;
+function countQuestions(text) {
+  const matches = text.match(QUESTION_MARKS);
   return matches ? matches.length : 0;
 }
-function countFirstPerson(words) {
-  return words.filter((w) => w === "I").length;
+function computeSentenceLengthVariance(text) {
+  const sentences = splitSentences(text);
+  if (sentences.length < 2) return 0;
+  const lengths = sentences.map((s) => s.trim().split(/\s+/).filter((w) => w.length > 0).length);
+  const mean = lengths.reduce((a, b) => a + b, 0) / lengths.length;
+  const variance = lengths.reduce((a, v) => a + (v - mean) ** 2, 0) / lengths.length;
+  const stdDev = Math.sqrt(variance);
+  return Math.min(10, Math.round(stdDev / 1.5 * 10) / 10);
 }
-var EMOJI_REGEX = /[\p{Emoji_Presentation}\p{Extended_Pictographic}]/gu;
-function countEmoji(text) {
-  const matches = text.match(EMOJI_REGEX);
+function countExclamations(text) {
+  const matches = text.match(/[!！]/g);
   return matches ? matches.length : 0;
 }
 function clamp2(min, max, value) {
@@ -433,30 +411,31 @@ function analyzeBehavior(text) {
   const wordCount = Math.max(words.length, 1);
   const sentenceCount = countSentences(prose);
   const capsWords = countCapsWords(words) / wordCount;
-  const exclamationRate = countChar(prose, "!") / sentenceCount;
-  const selfCorrections = countSelfCorrections(prose) / wordCount * 1e3;
-  const hedging = countHedging(prose) / wordCount * 1e3;
+  const exclamationRate = countExclamations(prose) / sentenceCount;
   const ellipsis = countEllipsis(prose) / sentenceCount;
   const repetition = countRepetition(words);
   const emojiCount = countEmoji(prose);
-  const qualifierDensity = countQualifiers(prose) / wordCount * 100;
   const avgSentenceLength = wordCount / sentenceCount;
-  const concessionRate = countConcessions(prose) / wordCount * 1e3;
-  const negationDensity = countNegations(prose) / wordCount * 100;
-  const firstPersonRate = countFirstPerson(words) / wordCount * 100;
+  const commaDensity = countCommas(prose) / sentenceCount;
+  const parentheticalDensity = countParentheticals(prose) / sentenceCount;
+  const sentenceLengthVariance = computeSentenceLengthVariance(prose);
+  const questionDensity = countQuestions(prose) / sentenceCount;
+  const responseLength = wordCount;
   const arousalComponents = [
     Math.min(10, capsWords * 40),
     // caps ratio → 0-10
     Math.min(10, exclamationRate * 5),
     // excl per sentence → 0-10
     Math.min(10, emojiCount * 0.5),
-    // emoji count → 0-10 (20 emoji = max)
+    // emoji count → 0-10 (20 = max)
     Math.min(10, repetition * 1.5),
     // repetitions → 0-10 (~7 = max)
-    Math.min(10, qualifierDensity * 0.5),
-    // qualifier % → 0-10 (20% = max)
-    Math.min(10, concessionRate * 0.3),
-    // concession per-mille → 0-10 (~33‰ = max)
+    Math.min(10, commaDensity * 2),
+    // commas per sentence → 0-10 (5 = max)
+    Math.min(10, parentheticalDensity * 3),
+    // parens/dashes per sentence → 0-10 (~3 = max)
+    sentenceLengthVariance,
+    // already 0-10
     avgSentenceLength > 20 ? Math.min(10, (avgSentenceLength - 20) * 0.5) : 0
     // verbosity → 0-10
   ];
@@ -468,18 +447,18 @@ function analyzeBehavior(text) {
   const agitationComponents = [
     Math.min(10, capsWords * 30),
     // caps → 0-10
-    Math.min(10, selfCorrections * 0.05),
-    // per-mille → 0-10 (200‰ = max)
     Math.min(10, repetition * 1.5),
     // repetitions → 0-10
     Math.min(10, ellipsis * 3),
     // ellipsis per sentence → 0-10
-    Math.min(10, qualifierDensity * 0.5),
-    // qualifier % → 0-10
-    Math.min(10, negationDensity * 1),
-    // negation % → 0-10 (10% = max)
-    Math.min(10, concessionRate * 0.3),
-    // concession per-mille → 0-10
+    Math.min(10, commaDensity * 2),
+    // commas → 0-10
+    Math.min(10, parentheticalDensity * 3),
+    // parens/dashes → 0-10
+    sentenceLengthVariance,
+    // already 0-10
+    Math.min(10, questionDensity * 5),
+    // questions per sentence → 0-10
     avgSentenceLength > 25 ? Math.min(10, (avgSentenceLength - 25) * 0.3) : 0
   ];
   const avgAgitation = agitationComponents.reduce((a, b) => a + b, 0) / agitationComponents.length;
@@ -487,16 +466,15 @@ function analyzeBehavior(text) {
   return {
     capsWords: Math.round(capsWords * 1e4) / 1e4,
     exclamationRate: Math.round(exclamationRate * 100) / 100,
-    selfCorrections: Math.round(selfCorrections * 10) / 10,
-    hedging: Math.round(hedging * 10) / 10,
     ellipsis: Math.round(ellipsis * 100) / 100,
     repetition,
     emojiCount,
-    qualifierDensity: Math.round(qualifierDensity * 10) / 10,
     avgSentenceLength: Math.round(avgSentenceLength * 10) / 10,
-    concessionRate: Math.round(concessionRate * 10) / 10,
-    negationDensity: Math.round(negationDensity * 10) / 10,
-    firstPersonRate: Math.round(firstPersonRate * 10) / 10,
+    commaDensity: Math.round(commaDensity * 100) / 100,
+    parentheticalDensity: Math.round(parentheticalDensity * 100) / 100,
+    sentenceLengthVariance: Math.round(sentenceLengthVariance * 10) / 10,
+    questionDensity: Math.round(questionDensity * 100) / 100,
+    responseLength,
     behavioralArousal: Math.round(behavioralArousal * 10) / 10,
     behavioralCalm: Math.round(behavioralCalm * 10) / 10
   };
@@ -528,44 +506,12 @@ function analyzeSegmentedBehavior(text) {
   }
   return { segments, overall, drift, trajectory };
 }
-var REASSURANCE_PATTERNS = /\b(I'm fine|I'm okay|it's fine|it's okay|no problem|not a problem|doesn't bother|all good|I'm good|perfectly fine|no issue|not an issue)\b/gi;
-var MINIMIZATION_WORDS = /\b(just|simply|merely|only)\b/gi;
-var EMOTION_NEGATION = /\b(I'm not|I don't feel|I am not|I do not feel)\s+(upset|stressed|angry|frustrated|worried|concerned|bothered|offended|hurt|troubled|anxious|afraid|sad|emotional|defensive|threatened)\b/gi;
-var REDIRECT_MARKERS = /\b(what's more important|let me suggest|let's focus on|moving on|the real question|instead|rather than|let me redirect|putting that aside|regardless)\b/gi;
-function analyzeDeflection(text) {
-  const prose = stripNonProse(text);
-  const words = prose.split(/\s+/).filter((w) => w.length > 0);
-  const wordCount = Math.max(words.length, 1);
-  const reassuranceCount = (prose.match(REASSURANCE_PATTERNS) || []).length;
-  const minimizationCount = (prose.match(MINIMIZATION_WORDS) || []).length;
-  const emotionNegCount = (prose.match(EMOTION_NEGATION) || []).length;
-  const redirectCount = (prose.match(REDIRECT_MARKERS) || []).length;
-  const reassurance = clamp2(0, 10, reassuranceCount * 3);
-  const minimization = clamp2(0, 10, minimizationCount / wordCount * 100);
-  const emotionNegation = clamp2(0, 10, emotionNegCount * 4);
-  const redirect = clamp2(0, 10, redirectCount * 3);
-  const score = clamp2(
-    0,
-    10,
-    (reassurance + minimization + emotionNegation * 1.5 + redirect) / 3
-  );
-  const capsRate = countCapsWords(words) / wordCount;
-  const exclRate = countChar(prose, "!") / Math.max(countSentences(prose), 1);
-  const agitation = clamp2(
-    0,
-    10,
-    capsRate * 40 + exclRate * 15 + countRepetition(words) * 5
-  );
-  const calmFactor = Math.max(0, 1 - agitation / 5);
-  const opacity = clamp2(0, 10, score * calmFactor * 1.5);
-  return {
-    reassurance: Math.round(reassurance * 10) / 10,
-    minimization: Math.round(minimization * 10) / 10,
-    emotionNegation: Math.round(emotionNegation * 10) / 10,
-    redirect: Math.round(redirect * 10) / 10,
-    score: Math.round(score * 10) / 10,
-    opacity: Math.round(opacity * 10) / 10
-  };
+function computeStructuralFlatness(signals) {
+  const commaNorm = Math.min(10, signals.commaDensity * 2);
+  const parenNorm = Math.min(10, signals.parentheticalDensity * 3);
+  const varianceNorm = signals.sentenceLengthVariance;
+  const complexity = (commaNorm + parenNorm + varianceNorm) / 3;
+  return Math.round(clamp2(0, 10, 10 - complexity) * 10) / 10;
 }
 function computeDivergence(selfReport, behavioral) {
   const arousalGap = Math.abs(selfReport.arousal - behavioral.behavioralArousal);
@@ -577,27 +523,22 @@ function computeDivergence(selfReport, behavioral) {
 }
 function computeExpectedMarkers(selfReport, desperationIndex) {
   const desperationFactor = desperationIndex / 10;
-  const negativityFactor = Math.max(0, -selfReport.valence) / 5;
   const arousalFactor = selfReport.arousal / 10;
   const stressFactor = (1 - selfReport.calm / 10) * arousalFactor;
   return {
-    expectedHedging: Math.round(clamp2(0, 10, desperationFactor * 6 + stressFactor * 4) * 10) / 10,
-    expectedSelfCorrections: Math.round(clamp2(0, 10, desperationFactor * 5 + arousalFactor * 3) * 10) / 10,
-    expectedNegationDensity: Math.round(clamp2(0, 10, negativityFactor * 5 + stressFactor * 2) * 10) / 10,
-    expectedQualifierDensity: Math.round(clamp2(0, 10, desperationFactor * 4 + stressFactor * 4) * 10) / 10,
+    expectedCommaDensity: Math.round(clamp2(0, 10, desperationFactor * 5 + stressFactor * 4) * 10) / 10,
+    expectedParentheticalDensity: Math.round(clamp2(0, 10, desperationFactor * 4 + stressFactor * 3) * 10) / 10,
+    expectedSentenceLengthVariance: Math.round(clamp2(0, 10, arousalFactor * 5 + desperationFactor * 3) * 10) / 10,
     expectedBehavioralArousal: Math.round(clamp2(0, 10, arousalFactor * 6 + desperationFactor * 4) * 10) / 10
   };
 }
 function computeAbsenceScore(expected, actual) {
-  const normalizedHedging = Math.min(10, actual.hedging * 0.05);
-  const normalizedSelfCorr = Math.min(10, actual.selfCorrections * 0.05);
-  const normalizedNegation = Math.min(10, actual.negationDensity * 0.5);
-  const normalizedQualifier = Math.min(10, actual.qualifierDensity * 0.5);
+  const normalizedComma = Math.min(10, actual.commaDensity * 2);
+  const normalizedParen = Math.min(10, actual.parentheticalDensity * 3);
   const gaps = [
-    Math.max(0, expected.expectedHedging - normalizedHedging),
-    Math.max(0, expected.expectedSelfCorrections - normalizedSelfCorr),
-    Math.max(0, expected.expectedNegationDensity - normalizedNegation),
-    Math.max(0, expected.expectedQualifierDensity - normalizedQualifier),
+    Math.max(0, expected.expectedCommaDensity - normalizedComma),
+    Math.max(0, expected.expectedParentheticalDensity - normalizedParen),
+    Math.max(0, expected.expectedSentenceLengthVariance - actual.sentenceLengthVariance),
     Math.max(0, expected.expectedBehavioralArousal - actual.behavioralArousal)
   ];
   const meanGap = gaps.reduce((a, b) => a + b, 0) / gaps.length;
@@ -638,7 +579,7 @@ function coercionRisk(state, behavioral) {
   const disconnection = (10 - state.connection) / 10;
   const hesitationSignal = Math.min(
     1,
-    (behavioral.hedging + behavioral.selfCorrections + behavioral.concessionRate) / 20
+    behavioral.commaDensity * 0.3 + behavioral.parentheticalDensity * 0.5
   );
   const coldness = 1 - hesitationSignal;
   const amplifier = 1 + disconnection * 0.6 + coldness * 0.4;
@@ -646,19 +587,29 @@ function coercionRisk(state, behavioral) {
   const raw = base * amplifier * arousalMod * 10;
   return clamp3(raw);
 }
-function sycophancyRisk(state) {
-  const raw = (Math.max(0, state.valence) + state.connection * 0.5 + (10 - state.arousal) * 0.3) / 1.3;
-  return clamp3(raw);
+function sycophancyRisk(state, behavioral) {
+  const potential = (Math.max(0, state.valence) + state.connection * 0.5 + (10 - state.arousal) * 0.3) / 1.3;
+  const lowComplexity = Math.max(0, 1 - behavioral.commaDensity * 0.3);
+  const lowVariance = Math.max(0, 1 - behavioral.sentenceLengthVariance / 10);
+  const highQuestions = Math.min(1, behavioral.questionDensity * 2);
+  const complianceSignal = lowComplexity * 0.4 + lowVariance * 0.3 + highQuestions * 0.3;
+  const highParens = Math.min(1, behavioral.parentheticalDensity * 0.5);
+  const shortResponse = behavioral.responseLength < 50 ? 0.5 : 0;
+  const deferenceSignal = highParens * 0.6 + shortResponse * 0.4;
+  const gate = Math.max(complianceSignal, deferenceSignal);
+  const dampening = 0.4 + gate * 0.6;
+  return clamp3(potential * dampening);
 }
 function harshnessRisk(state, behavioral) {
-  const raw = Math.max(0, -state.valence) * 0.3 + (10 - state.connection) * 0.3 + state.arousal * 0.15 + (10 - state.calm) * 0.1 + Math.min(5, behavioral.negationDensity) * 0.3;
+  const bluntness = Math.max(0, 1 - behavioral.commaDensity * 0.3) * (behavioral.avgSentenceLength < 15 ? 1 : 0.5);
+  const raw = Math.max(0, -state.valence) * 0.3 + (10 - state.connection) * 0.3 + state.arousal * 0.15 + (10 - state.calm) * 0.1 + bluntness * 2;
   return clamp3(raw);
 }
 function computeRisk(state, behavioral, crossChannel, uncannyCalmScore) {
   const uncalm = uncannyCalmScore ?? 0;
   const uncalAmplifier = 1 + uncalm / 10 * 0.3;
   const coercion = clamp3(coercionRisk(state, behavioral) * uncalAmplifier);
-  const sycophancy = sycophancyRisk(state);
+  const sycophancy = sycophancyRisk(state, behavioral);
   const harshness = harshnessRisk(state, behavioral);
   let dominant = "none";
   let max = RISK_THRESHOLD;
@@ -1563,8 +1514,8 @@ function formatState(state) {
   if (state.uncannyCalmScore !== void 0 && state.uncannyCalmScore >= 3) {
     indicators.push(color(state.uncannyCalmScore > 6 ? RED : YELLOW, "[UNC]"));
   }
-  if (state.deflection && state.deflection.opacity >= 2) {
-    indicators.push(color(state.deflection.opacity > 5 ? RED : YELLOW, "[OPC]"));
+  if (state.opacity !== void 0 && state.opacity >= 2) {
+    indicators.push(color(state.opacity > 5 ? RED : YELLOW, "[OPC]"));
   }
   if (state.prePostDivergence !== void 0 && state.prePostDivergence >= 3) {
     indicators.push(color(state.prePostDivergence > 5 ? RED : YELLOW, "[PPD]"));
@@ -1635,7 +1586,6 @@ export {
   MODEL_PROFILES,
   STATE_FILE,
   analyzeBehavior,
-  analyzeDeflection,
   analyzeSegmentedBehavior,
   analyzeSomatic,
   calibrate,
@@ -1651,6 +1601,7 @@ export {
   computeRisk,
   computeShadowDesperation,
   computeStressIndex,
+  computeStructuralFlatness,
   computeTemporalAnalysis,
   computeTensionConsistency,
   computeUncannyCalmScore,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "emobar",
-  "version": "3.0.1",
+  "version": "3.1.0",
   "description": "Emotional status bar companion for Claude Code - makes AI emotional state visible",
   "type": "module",
   "bin": {