cbrowser 18.78.0 → 18.80.0

package/dist/personas.js

@@ -14,11 +14,28 @@ import { join } from "path";
 import { homedir } from "os";
 import { applyTraitCorrelations } from "./persona-questionnaire.js";
 import { AGENT_PERSONAS, isAgentPersona, getAgentPersona, listAgentPersonas, isAgentPersonaObject } from "./agent-personas.js";
+import { scopedDataDir } from "./persona-scope.js";
+import { TRAIT_DEFINITIONS, canonicalizeTraits } from "./trait-reference.js";
 // ============================================================================
 // Custom Personas Storage
 // ============================================================================
 const DATA_DIR = process.env.CBROWSER_DATA_DIR || join(homedir(), ".cbrowser");
-const PERSONAS_DIR = join(DATA_DIR, "personas");
+/**
+ * Resolved per call, not once at module load.
+ *
+ * A hosted server process serves many accounts, so the directory cannot be a
+ * constant fixed when the file is imported. `scopedDataDir()` returns the
+ * account bound to the CURRENT request's async context, or undefined when
+ * nothing entered a scope — which is every CLI run, every stdio session and
+ * every local install, all of which keep the previous behaviour exactly.
+ *
+ * Deliberately NOT a module-level `let` reassigned per request: that is the
+ * shape that lets two overlapping requests read each other's accounts, and it
+ * is the shape this codebase already removed from the billing path. (2026-08-01)
+ */
+function personasDirFor() {
+    return join(scopedDataDir() ?? DATA_DIR, "personas");
+}
 // ============================================================================
 // Runtime Persona Registry (for Enterprise extensions)
 // ============================================================================
@@ -32,21 +49,34 @@ function ensurePersonasDir() {
     if (!existsSync(DATA_DIR)) {
         mkdirSync(DATA_DIR, { recursive: true });
     }
-    if (!existsSync(PERSONAS_DIR)) {
-        mkdirSync(PERSONAS_DIR, { recursive: true });
+    if (!existsSync(personasDirFor())) {
+        mkdirSync(personasDirFor(), { recursive: true });
     }
 }
 /**
  * Load all custom personas from disk.
  */
+/**
+ * Normalise a persona read off disk: deprecated trait keys to canonical ones.
+ *
+ * Without this a persona saved before the mentalModelRigidity rename loads with
+ * a key nothing recognises, so that trait silently stops contributing.
+ */
+function normalizePersonaOnRead(p) {
+    const t = p?.cognitiveTraits;
+    if (!t)
+        return p;
+    const canon = canonicalizeTraits(t);
+    return canon === t ? p : { ...p, cognitiveTraits: canon };
+}
 export function loadCustomPersonas() {
     ensurePersonasDir();
     const personas = {};
     try {
-        const files = readdirSync(PERSONAS_DIR).filter(f => f.endsWith(".json") || f.endsWith(".yaml") || f.endsWith(".yml"));
+        const files = readdirSync(personasDirFor()).filter(f => f.endsWith(".json") || f.endsWith(".yaml") || f.endsWith(".yml"));
         for (const file of files) {
             try {
-                const content = readFileSync(join(PERSONAS_DIR, file), "utf-8");
+                const content = readFileSync(join(personasDirFor(), file), "utf-8");
                 let persona;
                 if (file.endsWith(".json")) {
                     // Bare `as Persona` trusted arbitrary user-authored JSON to be complete.
@@ -54,7 +84,10 @@ export function loadCustomPersonas() {
                     // branch the same floor rather than letting a partial file reach
                     // getCognitiveProfile half-formed. (2026-07-28)
                     const parsed = JSON.parse(content);
-                    persona = { ...parsed, behaviors: parsed.behaviors ?? {} };
+                    // Deprecated trait keys resolved here, at the read boundary, so a
+                    // persona saved before a rename keeps contributing instead of
+                    // silently dropping the renamed trait.
+                    persona = normalizePersonaOnRead({ ...parsed, behaviors: parsed.behaviors ?? {} });
                 }
                 else {
                     // Simple YAML parsing for persona files
@@ -84,11 +117,31 @@ export function loadCustomPersonas() {
 /**
  * Save a custom persona to disk.
  */
+/**
+ * Remove run-scoped values a persona record must not carry.
+ *
+ * `siteFamiliarity` is a persona-SITE pair, not a disposition, so a stored one
+ * is always answering "familiar with which site?" about a site nobody named.
+ * Worse, the persona builder merged it over the caller's per-run parameter, so
+ * the tool discarded the input while reporting that it had applied it.
+ *
+ * Stripping on the way to disk rather than trusting every caller: this is the
+ * single chokepoint every custom persona passes through, and a rule enforced at
+ * one write is worth more than the same rule written in six call sites.
+ * (2026-08-02)
+ */
+export function stripStoredFamiliarity(persona) {
+    if (persona?.cognitiveTraits && "siteFamiliarity" in persona.cognitiveTraits) {
+        const { siteFamiliarity: _dropped, ...rest } = persona.cognitiveTraits;
+        return { ...persona, cognitiveTraits: rest };
+    }
+    return persona;
+}
 export function saveCustomPersona(persona) {
     ensurePersonasDir();
     const filename = `${persona.name.toLowerCase().replace(/[^a-z0-9-]/g, "-")}.json`;
-    const filepath = join(PERSONAS_DIR, filename);
-    writeFileSync(filepath, JSON.stringify(persona, null, 2));
+    const filepath = join(personasDirFor(), filename);
+    writeFileSync(filepath, JSON.stringify(stripStoredFamiliarity(persona), null, 2));
     return filepath;
 }
 /**
@@ -103,7 +156,7 @@ export function deleteCustomPersona(name) {
     const baseName = name.toLowerCase().replace(/[^a-z0-9-]/g, "-");
     const extensions = [".json", ".yaml", ".yml"];
     for (const ext of extensions) {
-        const filepath = join(PERSONAS_DIR, `${baseName}${ext}`);
+        const filepath = join(personasDirFor(), `${baseName}${ext}`);
         try {
             if (existsSync(filepath)) {
                 unlinkSync(filepath);
@@ -560,7 +613,7 @@ function generateCognitiveTraitsFromDescription(descLower, techLevel, isImpatien
     let emotionalContagion = 0.5;
     let fearOfMissingOut = 0.5;
     let socialProofSensitivity = 0.5;
-    let mentalModelRigidity = 0.5;
+    let mentalModelFlexibility = 0.5;
     // Adjust new traits by tech level
     if (techLevel === "expert") {
         informationForaging = 0.9; // Experts follow scent efficiently
@@ -575,7 +628,7 @@ function generateCognitiveTraitsFromDescription(descLower, techLevel, isImpatien
         emotionalContagion = 0.2; // Mood-stable
         fearOfMissingOut = 0.2; // Not swayed by urgency
         socialProofSensitivity = 0.3; // Evaluate on merits
-        mentalModelRigidity = 0.8; // Adapt to new patterns
+        mentalModelFlexibility = 0.8; // Adapt to new patterns
     }
     else if (techLevel === "beginner") {
         informationForaging = 0.3; // Exhaustive search
@@ -590,7 +643,7 @@ function generateCognitiveTraitsFromDescription(descLower, techLevel, isImpatien
         emotionalContagion = 0.7; // Mood influenced by UI
         fearOfMissingOut = 0.7; // Susceptible to urgency
         socialProofSensitivity = 0.8; // Rely on reviews
-        mentalModelRigidity = 0.3; // Struggle with new patterns
+        mentalModelFlexibility = 0.3; // Struggle with new patterns
     }
     // Adjust for impatience
     if (isImpatient) {
@@ -615,7 +668,7 @@ function generateCognitiveTraitsFromDescription(descLower, techLevel, isImpatien
         proceduralFluency = Math.min(proceduralFluency, 0.35); // Struggle with sequences
         transferLearning = Math.min(transferLearning, 0.25); // Each UI novel
         authoritySensitivity = Math.max(authoritySensitivity, 0.7); // Trust authority
-        mentalModelRigidity = Math.min(mentalModelRigidity, 0.2); // Rigid mental models
+        mentalModelFlexibility = Math.min(mentalModelFlexibility, 0.2); // Rigid mental models
     }
     // Adjust for vision issues (screen reader users)
     if (hasVisionIssues) {
@@ -647,7 +700,7 @@ function generateCognitiveTraitsFromDescription(descLower, techLevel, isImpatien
         emotionalContagion,
         fearOfMissingOut,
         socialProofSensitivity,
-        mentalModelRigidity,
+        mentalModelFlexibility,
     };
 }
 /**
@@ -691,8 +744,7 @@ export function createCognitivePersona(name, description, traits, options = {})
         emotionalContagion: traits.emotionalContagion ?? basePersona.cognitiveTraits?.emotionalContagion ?? 0.5,
         fearOfMissingOut: traits.fearOfMissingOut ?? basePersona.cognitiveTraits?.fearOfMissingOut ?? 0.5,
         socialProofSensitivity: traits.socialProofSensitivity ?? basePersona.cognitiveTraits?.socialProofSensitivity ?? 0.5,
-        mentalModelRigidity: traits.mentalModelRigidity ?? basePersona.cognitiveTraits?.mentalModelRigidity ?? 0.5,
-        siteFamiliarity: traits.siteFamiliarity ?? basePersona.cognitiveTraits?.siteFamiliarity ?? 0.5,
+        mentalModelFlexibility: traits.mentalModelFlexibility ?? basePersona.cognitiveTraits?.mentalModelFlexibility ?? 0.5,
     };
     // Backstop for the failure mode above. The enumeration is kept because it
     // documents each trait's fallback, but a hand-written list silently drops
@@ -738,7 +790,14 @@ export function createCognitivePersona(name, description, traits, options = {})
 export function getCognitiveProfile(persona) {
     // Derive attention pattern from humanBehavior.attention.pattern or behaviors
     let attentionPattern = "f-pattern";
+    let attentionPatternSource = "default";
+    let attentionPatternMargin;
+    let attentionPatternConflict;
+    let techLevel;
+    let techLevelSource = "derived";
+    let techLevelConflict;
     if (persona.humanBehavior?.attention?.pattern) {
+        attentionPatternSource = "declared";
         const pattern = persona.humanBehavior.attention.pattern;
         if (pattern === "skim")
             attentionPattern = "skim";
@@ -758,17 +817,41 @@ export function getCognitiveProfile(persona) {
     // it. (2026-07-28)
     if (persona.behaviors?.attentionPattern) {
         attentionPattern = persona.behaviors.attentionPattern;
+        attentionPatternSource = "declared";
     }
+    const declaredAttentionPattern = attentionPatternSource === "declared" ? attentionPattern : undefined;
     // Derive decision style from traits
     let decisionStyle = "cautious";
+    let decisionStyleSource = "default";
+    let decisionStyleMargin;
+    let decisionStyleConfidence;
+    let decisionStyleRunnerUp;
+    let scoresTop;
     const traits = persona.cognitiveTraits;
     // v16.7.1: Validate trait completeness
     if (traits) {
-        const expectedTraitCount = 25; // 7 required + 18 optional
-        const actualTraitCount = Object.keys(traits).filter(k => traits[k] !== undefined).length;
+        // Derived from the trait definitions rather than hardcoded. This was the
+        // literal 25 while the model had 26, so a persona missing exactly one trait
+        // counted as complete and the warning could never fire for it. A count that
+        // has to be updated by hand in step with a table is a count that will drift
+        // from it -- this file, mcp-server.ts, cognitive/index.ts, widget-kit.ts and
+        // cognitive-transport.ts had all drifted to 25 by 2026-08-01.
+        // Run-scoped and runtime-derived dimensions are excluded from the count.
+        //
+        // siteFamiliarity is deliberately not stored on personas (it is a
+        // persona-SITE pair supplied per run), and readingCapacity/motorCapacity/
+        // sustainedAttention are injected at analysis time -- the first two from the
+        // formal models, the third resolved from accessibilityTraits.attentionSpan.
+        // Counting them as "missing" made every persona in the roster warn on every
+        // load, which turns a real signal -- a persona that genuinely skipped traits
+        // -- into noise nobody reads. (2026-08-02)
+        const RUNTIME_SCOPED = new Set(["siteFamiliarity", "readingCapacity", "motorCapacity", "sustainedAttention"]);
+        const expectedTraitCount = Object.keys(TRAIT_DEFINITIONS).filter((t) => !RUNTIME_SCOPED.has(t)).length;
+        const actualTraitCount = Object.keys(traits)
+            .filter((k) => !RUNTIME_SCOPED.has(k) && traits[k] !== undefined).length;
         if (actualTraitCount < expectedTraitCount) {
             console.warn(`[CBrowser] Persona "${persona.name}" has ${actualTraitCount}/${expectedTraitCount} traits. ` +
-                `Missing traits will use defaults. Consider regenerating the persona for full 25-trait model.`);
+                `Missing traits will use defaults. Consider regenerating the persona for the full ${expectedTraitCount}-trait model.`);
         }
     }
     if (traits) {
@@ -777,24 +860,157 @@ export function getCognitiveProfile(persona) {
         const riskTolerance = traits.riskTolerance ?? 0.5;
         const comprehension = traits.comprehension ?? 0.5;
         const readingTendency = traits.readingTendency ?? 0.5;
-        if (patience < 0.3 && riskTolerance > 0.6) {
-            decisionStyle = "impulsive";
+        // Scored, not branched.
+        //
+        // An if/else chain is partial by construction: it answers only for the
+        // trait combinations someone thought to write a condition for, and every
+        // combination nobody covered silently inherits the initial literal. Six of
+        // twenty-one personas landed there -- first-timer among them -- reporting
+        // "cautious" because no branch matched, not because anything about them is
+        // cautious. Adding more branches moves the gap rather than closing it.
+        //
+        // Every style is scored from the trait vector and the highest wins, so a
+        // style is always chosen on evidence and the derivation is total: there is
+        // no combination left with no answer. (2026-07-31)
+        const t = (k, d = 0.5) => {
+            const v = traits[k];
+            return typeof v === "number" ? v : d;
+        };
+        const mean = (...xs) => xs.reduce((a, b) => a + b, 0) / xs.length;
+        const patienceV = t("patience");
+        const riskV = t("riskTolerance");
+        const satisficeV = t("satisficing");
+        const readingV = t("readingTendency");
+        const scores = [
+            // Acts before evaluating: tolerant of risk, short on patience, unplanned,
+            // and content with the first adequate option. Satisficing belongs here --
+            // without it someone with riskTolerance 0.8 and satisficing 0.25 scored
+            // impulsive, when refusing to settle is the opposite of impulsive.
+            ["impulsive", mean(riskV, 1 - patienceV, 1 - t("metacognitivePlanning"), satisficeV)],
+            // Knows the pattern and takes the shortest path that works.
+            ["efficient", mean(t("comprehension"), t("proceduralFluency"), satisficeV, 1 - readingV)],
+            // Checks before committing, and does not extend trust by default.
+            ["cautious", mean(1 - riskV, 1 - t("trustCalibration"), patienceV)],
+            // Reads, plans, and refuses to settle for the first adequate option.
+            ["deliberate", mean(readingV, patienceV, 1 - satisficeV, t("metacognitivePlanning"))],
+        ];
+        // Only reachable on a touch device, where it describes the input mode
+        // rather than the disposition.
+        if (persona.demographics.device === "mobile") {
+            // Weighted to win on mobile unless the trait evidence clearly says
+            // otherwise. Device was previously a terminal branch, so every mobile
+            // persona was quick-tap; letting it compete unweighted relabelled all of
+            // them at once, which is a bigger claim than closing a coverage gap
+            // should be making.
+            scores.push(["quick-tap", mean(1 - patienceV, satisficeV, riskV) + 0.2]);
         }
-        else if (comprehension > 0.8 && patience < 0.4) {
-            decisionStyle = "efficient";
+        scores.sort((a, b) => b[1] - a[1]);
+        decisionStyle = scores[0][0];
+        scoresTop = scores[0][0];
+        decisionStyleSource = "derived";
+        // A near-tie is a weak reading, and saying so beats presenting the winner
+        // as though it were clear.
+        decisionStyleMargin = Math.round((scores[0][1] - scores[1][1]) * 100) / 100;
+        // ...but exposing the margin was not enough on its own.
+        //
+        // The style STRING carried identical authority at a margin of 0.03 as at
+        // 0.90, and a consumer reading only `decisionStyle` -- which is most of
+        // them -- got a coin flip presented as a finding. A number nobody is
+        // required to look at does not qualify the claim sitting next to it.
+        //
+        // Below 0.05 the two top styles are indistinguishable given the precision of
+        // the inputs, so the label is withheld rather than guessed and the runner-up
+        // is named so the tie is inspectable. (2026-08-02)
+        decisionStyleConfidence = decisionStyleMargin >= 0.20 ? "clear"
+            : decisionStyleMargin >= 0.05 ? "weak"
+                : "indistinguishable";
+        if (decisionStyleConfidence === "indistinguishable") {
+            decisionStyleRunnerUp = scores[1][0];
         }
-        else if (riskTolerance < 0.3) {
-            decisionStyle = "cautious";
+        // Tech level, derived rather than trusted.
+        //
+        // demographics.tech_level is a creation-time literal that nothing ever
+        // revisits, and the default is "intermediate". Ten of twenty-one personas
+        // carried exactly that string while their fluency traits ranged from 0.28
+        // to 0.85 -- which is not ten independent judgements, it is one unset
+        // default. It is read by persona-comparison, so it affects output rather
+        // than only display.
+        //
+        // The stored value is left alone; someone may have meant it. The derived
+        // one is reported alongside, and a disagreement is named. (2026-08-01)
+        const fluency = mean(t("proceduralFluency"), t("transferLearning"), t("comprehension"));
+        const derivedTech = fluency >= 0.75 ? "expert" : fluency >= 0.5 ? "intermediate" : "beginner";
+        const declaredTech = persona.demographics?.tech_level;
+        techLevel = declaredTech ?? derivedTech;
+        techLevelSource = declaredTech ? "declared" : "derived";
+        if (declaredTech && declaredTech !== derivedTech) {
+            techLevelConflict =
+                `demographics.tech_level is "${declaredTech}", but this persona's fluency traits average ${Math.round(fluency * 100) / 100}, `
+                    + `which reads as "${derivedTech}" (proceduralFluency ${t("proceduralFluency")}, transferLearning ${t("transferLearning")}, comprehension ${t("comprehension")}). `
+                    + `The stored value is what persona-comparison uses.`;
         }
-        else if (readingTendency > 0.8 && patience > 0.7) {
-            decisionStyle = "deliberate";
+        // Attention pattern, derived the same way decisionStyle is.
+        //
+        // It was declared-or-default with no derivation at all, so a persona whose
+        // traits describe hard skimming could carry attentionPattern "thorough"
+        // and nothing anywhere would say the two disagreed -- and the declaration
+        // is what drives behaviour at runtime, so the contradiction changed how the
+        // persona read a page rather than merely how it was reported.
+        //
+        // Scored rather than branched, for the reason decisionStyle is: a branch
+        // chain answers only for combinations someone wrote a condition for, and
+        // everything else silently inherits the initial literal. Only the patterns
+        // the trait vector can actually speak to are scored. `sequential` is a
+        // navigation MODE (screen reader, tab order) and `z-pattern` is a property
+        // of sparse layouts rather than of a person, so neither is inferred from
+        // disposition -- they stay declarable and are never guessed. (2026-08-01)
+        const attnScores = [
+            // Reads everything, slowly: high reading, patient, refuses to settle.
+            ["thorough", mean(readingV, patienceV, 1 - satisficeV, t("comprehension"))],
+            // Big elements only: low reading, impatient, settles fast.
+            ["skim", mean(1 - readingV, 1 - patienceV, satisficeV)],
+            // Straight to the expected place: knows the conventions, forages by
+            // scent, does not need to read to locate.
+            ["targeted", mean(t("proceduralFluency"), t("informationForaging"), satisficeV, 1 - readingV)],
+            // Notices everything, follows what is interesting rather than a plan.
+            ["exploratory", mean(t("curiosity"), 1 - t("metacognitivePlanning"), t("informationForaging"))],
+            // The web-reading default: middling on the axes that separate the others.
+            // Scored rather than used as a fallback literal so it has to WIN to be
+            // chosen, which is what stops it becoming the silent catch-all the old
+            // default was.
+            ["f-pattern", mean(0.5 + (0.5 - Math.abs(readingV - 0.5)), 0.5 + (0.5 - Math.abs(patienceV - 0.5))) * 0.9],
+        ];
+        attnScores.sort((a, b) => b[1] - a[1]);
+        const derivedAttentionPattern = attnScores[0][0];
+        attentionPatternMargin = Math.round((attnScores[0][1] - attnScores[1][1]) * 100) / 100;
+        if (declaredAttentionPattern === undefined) {
+            attentionPattern = derivedAttentionPattern;
+            attentionPatternSource = "derived";
         }
-        else if (persona.demographics.device === "mobile") {
-            decisionStyle = "quick-tap";
+        else if (declaredAttentionPattern !== derivedAttentionPattern) {
+            // The declaration still WINS -- someone wrote it deliberately and it may
+            // encode something the traits cannot, like a screen reader's sequential
+            // pass. But a disagreement this size is reported rather than swallowed,
+            // because the runtime behaviour follows the declaration and the numbers
+            // say something else.
+            // A conflict is only as strong as the derivation behind it. When the
+            // top two derived patterns are themselves a near-tie, "the traits say X"
+            // overstates what the numbers support -- the honest claim is that the
+            // traits do not endorse the declaration, not that they pick a different
+            // winner. Same rule the Maslow tie reporting follows.
+            const decisive = (attentionPatternMargin ?? 0) > 0.05;
+            attentionPatternConflict = decisive
+                ? `declared "${declaredAttentionPattern}", but this persona's traits score "${derivedAttentionPattern}" highest by ${attentionPatternMargin} `
+                    + `(reading ${readingV}, patience ${patienceV}, satisficing ${satisficeV}). `
+                    + `The declaration is in force at runtime; the trait vector disagrees with it.`
+                : `declared "${declaredAttentionPattern}", which is not what the traits favour — though they do not clearly favour anything either `
+                    + `("${derivedAttentionPattern}" leads by only ${attentionPatternMargin}). `
+                    + `Read this as the declaration being unsupported rather than contradicted.`;
         }
     }
     if (persona.behaviors?.decisionStyle) {
         decisionStyle = persona.behaviors.decisionStyle;
+        decisionStyleSource = "declared";
     }
     // v16.14.1: Merge traits with defaults to ensure complete CognitiveTraits
     // This handles both Persona (full traits) and AccessibilityPersona (partial traits)
@@ -826,13 +1042,44 @@ export function getCognitiveProfile(persona) {
         emotionalContagion: 0.5,
         fearOfMissingOut: 0.5,
         socialProofSensitivity: 0.5,
-        mentalModelRigidity: 0.5,
-        siteFamiliarity: 0.5,
+        mentalModelFlexibility: 0.5,
     };
     return {
         traits: traits ? { ...defaultTraits, ...traits } : defaultTraits,
         attentionPattern,
-        decisionStyle,
+        // NULL when suppressed, not omitted.
+        //
+        // Omitting the key left consumers unable to tell suppression from "never
+        // computed", while decisionStyleSource still said "derived" -- claiming a
+        // derivation for a value that was not there. An explicit null plus a
+        // suppression flag and the threshold says what happened. (2026-08-02)
+        decisionStyle: decisionStyleConfidence === "indistinguishable" ? null : decisionStyle,
+        ...(decisionStyleConfidence === "indistinguishable"
+            ? { decisionStyleSuppressed: true, decisionStyleThreshold: 0.05 }
+            : {}),
+        attentionPatternSource,
+        attentionPatternMargin,
+        attentionPatternConflict,
+        techLevel,
+        techLevelSource,
+        techLevelConflict,
+        // Reports SUPPRESSION rather than claiming a derivation whose result was
+        // withheld. It said "derived" beside an absent decisionStyle.
+        decisionStyleSource: decisionStyleConfidence === "indistinguishable" ? "suppressed" : decisionStyleSource,
+        ...(decisionStyleMargin !== undefined ? { decisionStyleMargin } : {}),
+        ...(decisionStyleConfidence ? { decisionStyleConfidence } : {}),
+        // The LABEL is withheld at an indistinguishable margin, not just annotated.
+        // Annotating it leaves the string in place for every consumer that reads
+        // only `decisionStyle`, which is the failure this fixes. The candidate is
+        // still available as decisionStyleRunnerUp plus the winner below, so
+        // nothing is hidden -- it just stops being asserted.
+        ...(decisionStyleConfidence === "indistinguishable"
+            ? { decisionStyleCandidates: [scoresTop, decisionStyleRunnerUp].filter((x) => !!x) }
+            : {}),
+        ...(decisionStyleRunnerUp ? { decisionStyleRunnerUp } : {}),
+        ...(decisionStyleConfidence === "indistinguishable"
+            ? { decisionStyleNote: `Withheld: the top two styles differ by ${decisionStyleMargin}, which is inside the noise of the inputs. Runner-up was ${decisionStyleRunnerUp}. A label at this margin is a coin flip wearing a finding's clothes.` }
+            : {}),
         innerVoiceTemplate: persona.behaviors?.innerVoiceTemplate,
     };
 }
@@ -883,32 +1130,31 @@ export const BUILTIN_PERSONAS = {
         },
         cognitiveTraits: {
             patience: 0.3, // Low - expects things to work
-            riskTolerance: 0.9, // High - clicks confidently
-            comprehension: 0.95, // Expert - knows all conventions
+            riskTolerance: 0.8, // High - clicks confidently
+            comprehension: 0.9, // Expert - knows all conventions
             persistence: 0.4, // Low - switches approaches quickly
             curiosity: 0.2, // Low - stays focused on goal
             workingMemory: 0.9, // High - never repeats attempts
-            readingTendency: 0.1, // Low - scans for shortcuts
-            resilience: 0.85, // High - shrugs off errors quickly (BRS)
+            readingTendency: 0.2, // Low - scans for shortcuts
+            resilience: 0.8, // High - shrugs off errors quickly (BRS)
             selfEfficacy: 0.9, // High - trusts own expertise (Bandura)
-            satisficing: 0.3, // Low - knows quality, seeks optimal (Simon)
+            satisficing: 0.4, // Low - knows quality, seeks optimal (Simon)
             trustCalibration: 0.7, // High - trusts tech but verifies (Fogg)
-            interruptRecovery: 0.85, // High - uses env cues, resumes fast (Mark)
+            interruptRecovery: 0.9, // High - uses env cues, resumes fast (Mark)
             // New traits (v15.0.0)
             informationForaging: 0.9, // High - follows scent efficiently
-            changeBlindness: 0.2, // Low - notices UI changes
+            changeBlindness: 0.3, // Low - notices UI changes
             anchoringBias: 0.3, // Low - compares options freely
             timeHorizon: 0.3, // Long-term - invests in learning
             attributionStyle: 0.2, // Blames system - "this is broken"
             metacognitivePlanning: 0.8, // High - plans before acting
             proceduralFluency: 0.9, // High - follows procedures easily
-            transferLearning: 0.95, // High - applies knowledge across UIs
+            transferLearning: 0.9, // High - applies knowledge across UIs
             authoritySensitivity: 0.3, // Low - questions authority, verifies
-            emotionalContagion: 0.2, // Low - mood-stable
+            emotionalContagion: 0.3, // Low - mood-stable
             fearOfMissingOut: 0.2, // Low - not swayed by urgency
             socialProofSensitivity: 0.3, // Low - evaluates on merits
-            mentalModelRigidity: 0.8, // High - adapts to new patterns
-            siteFamiliarity: 0.9, // Very high - daily user, knows everything
+            mentalModelFlexibility: 0.8, // High - adapts to new patterns
         },
         context: {
             viewport: [1920, 1080],
@@ -962,33 +1208,32 @@ export const BUILTIN_PERSONAS = {
             },
         },
         cognitiveTraits: {
-            patience: 0.6, // Medium - willing to learn
+            patience: 0.7, // Medium - willing to learn
             riskTolerance: 0.3, // Low - hesitates before clicking
             comprehension: 0.3, // Low - doesn't know conventions
             persistence: 0.5, // Medium - tries a few times
-            curiosity: 0.7, // High - explores the interface
+            curiosity: 0.9, // High - explores the interface
             workingMemory: 0.4, // Medium - might repeat mistakes
-            readingTendency: 0.8, // High - reads tooltips and help
+            readingTendency: 0.6, // High - reads tooltips and help
             resilience: 0.4, // Low-medium - new users get discouraged (BRS)
             selfEfficacy: 0.4, // Low - uncertain of abilities (Bandura)
-            satisficing: 0.5, // Medium - unsure what's "good enough" (Simon)
+            satisficing: 0.6, // Medium - unsure what's "good enough" (Simon)
             trustCalibration: 0.4, // Low - skeptical, unsure what to trust (Fogg)
-            interruptRecovery: 0.35, // Low - easily loses place, restarts (Mark)
+            interruptRecovery: 0.4, // Low - easily loses place, restarts (Mark)
             // New traits (v15.0.0)
-            informationForaging: 0.3, // Low - exhaustive search, clicks everything
+            informationForaging: 0.4, // Low - exhaustive search, clicks everything
             changeBlindness: 0.6, // High - misses peripheral changes
             anchoringBias: 0.7, // High - anchored to first information
             timeHorizon: 0.7, // Short-term - wants quick results
             attributionStyle: 0.7, // Self-blaming - "I must be doing it wrong"
             metacognitivePlanning: 0.3, // Low - trial-and-error approach
-            proceduralFluency: 0.35, // Low - struggles with multi-step flows
+            proceduralFluency: 0.2, // Low - struggles with multi-step flows
             transferLearning: 0.2, // Low - each interface feels new
-            authoritySensitivity: 0.8, // High - trusts official-looking cues
-            emotionalContagion: 0.7, // High - influenced by UI tone
+            authoritySensitivity: 0.7, // High - trusts official-looking cues
+            emotionalContagion: 0.6, // High - influenced by UI tone
             fearOfMissingOut: 0.6, // Medium-high - susceptible to urgency
-            socialProofSensitivity: 0.8, // High - relies heavily on reviews
-            mentalModelRigidity: 0.3, // Low - struggles when patterns break
-            siteFamiliarity: 0.0, // Zero - brand new, never visited before
+            socialProofSensitivity: 0.7, // High - relies heavily on reviews
+            mentalModelFlexibility: 0.3, // Low - struggles when patterns break
         },
         context: {
             viewport: [1280, 800],
@@ -1036,33 +1281,32 @@ export const BUILTIN_PERSONAS = {
             },
         },
         cognitiveTraits: {
-            patience: 0.4, // Low - mobile = quick tasks
+            patience: 0.3, // Low - mobile = quick tasks
             riskTolerance: 0.6, // Medium - taps somewhat freely
             comprehension: 0.6, // Medium - knows mobile patterns
-            persistence: 0.3, // Low - gives up if fiddly
-            curiosity: 0.3, // Low - wants to complete and go
+            persistence: 0.4, // Low - gives up if fiddly
+            curiosity: 0.4, // Low - wants to complete and go
             workingMemory: 0.5, // Medium
-            readingTendency: 0.3, // Low - minimal reading on mobile
+            readingTendency: 0.2, // Low - minimal reading on mobile
             resilience: 0.5, // Medium - recovers if next try works (BRS)
             selfEfficacy: 0.6, // Medium - comfortable in mobile context (Bandura)
             satisficing: 0.8, // High - mobile context demands quick decisions (Simon)
             trustCalibration: 0.6, // Medium - quick decisions, moderate trust (Fogg)
-            interruptRecovery: 0.45, // Low-Med - constant phone interruptions (Mark)
+            interruptRecovery: 0.5, // Low-Med - constant phone interruptions (Mark)
             // New traits (v15.0.0)
             informationForaging: 0.7, // High - efficient mobile navigation
-            changeBlindness: 0.5, // Medium - small screen limits peripheral
+            changeBlindness: 0.6, // Medium - small screen limits peripheral
             anchoringBias: 0.5, // Medium
             timeHorizon: 0.8, // Very short-term - mobile is quick tasks
             attributionStyle: 0.4, // Medium - blames small screen/touch
             metacognitivePlanning: 0.4, // Low-medium - quick decisions
             proceduralFluency: 0.6, // Medium - knows mobile patterns
-            transferLearning: 0.7, // High - mobile UI conventions
+            transferLearning: 0.6, // High - mobile UI conventions
             authoritySensitivity: 0.5, // Medium
             emotionalContagion: 0.5, // Medium
-            fearOfMissingOut: 0.7, // High - push notifications, urgency
+            fearOfMissingOut: 0.6, // High - push notifications, urgency
             socialProofSensitivity: 0.6, // Medium-high - app store ratings
-            mentalModelRigidity: 0.6, // Medium - adapts to mobile patterns
-            siteFamiliarity: 0.4, // Low-medium - used desktop version before
+            mentalModelFlexibility: 0.6, // Medium - adapts to mobile patterns
         },
         context: {
             viewport: [375, 812], // iPhone X dimensions
@@ -1110,33 +1354,41 @@ export const BUILTIN_PERSONAS = {
             },
         },
         cognitiveTraits: {
-            patience: 0.8, // High - used to slow navigation
-            riskTolerance: 0.5, // Medium - careful but experienced
+            patience: 0.9, // High - used to slow navigation
+            // 0.5 -> 0.2 (2026-08-02). Screen-reader navigation carries a real
+            // disorientation cost: no visual context to recover from, high cognitive
+            // load from remembering shortcuts and prior steps, and interface changes
+            // that go unannounced. The documented behaviour is cautious exploration
+            // to avoid that cost, which is the low end of this scale ('only clicks
+            // obvious, labeled buttons') rather than the middle.
+            // Sources: Information Wayfinding of Screen Reader Users, ASSETS 2024
+            // (dl.acm.org/doi/fullHtml/10.1145/3663548.3688539); Context-Aware
+            // Guidance for Screen Reader Users, CHI 2026 (10.1145/3772318.3790661).
+            riskTolerance: 0.2, // Medium - careful but experienced
             comprehension: 0.8, // High - expert at a11y patterns
             persistence: 0.9, // High - determination required
             curiosity: 0.2, // Low - structured navigation
-            workingMemory: 0.9, // High - mental model essential
-            readingTendency: 1.0, // Full - ALL content is read aloud
+            workingMemory: 0.7, // High - mental model essential
+            readingTendency: 0.9, // Full - ALL content is read aloud
             resilience: 0.8, // High - experienced with setbacks (CD-RISC)
-            selfEfficacy: 0.5, // Medium - high in accessible sites, low otherwise (Bandura)
+            selfEfficacy: 0.7, // Medium - high in accessible sites, low otherwise (Bandura)
             satisficing: 0.6, // Medium - relies on structured review content (Simon)
-            trustCalibration: 0.5, // Medium - cautious from past a11y failures (Fogg)
-            interruptRecovery: 0.75, // High - strong mental model aids recovery (Mark)
+            trustCalibration: 0.6, // Medium - cautious from past a11y failures (Fogg)
+            interruptRecovery: 0.6, // High - strong mental model aids recovery (Mark)
             // New traits (v15.0.0)
-            informationForaging: 0.7, // High - sequential but efficient with landmarks
+            informationForaging: 0.6, // High - sequential but efficient with landmarks
             changeBlindness: 0.3, // Low - ARIA live regions announce changes
             anchoringBias: 0.4, // Medium - sequential exposure to info
             timeHorizon: 0.3, // Long-term - patient, thorough
             attributionStyle: 0.4, // Medium - blames inaccessible design
-            metacognitivePlanning: 0.85, // Very high - plans navigation carefully
+            metacognitivePlanning: 0.7, // Very high - plans navigation carefully
             proceduralFluency: 0.8, // High - follows heading structure
-            transferLearning: 0.75, // High - knows a11y patterns across sites
+            transferLearning: 0.7, // High - knows a11y patterns across sites
             authoritySensitivity: 0.5, // Medium - verifies with screen reader
             emotionalContagion: 0.3, // Low - focused on structure, not tone
-            fearOfMissingOut: 0.2, // Low - methodical, not impulsive
+            fearOfMissingOut: 0.4, // Low - methodical, not impulsive
             socialProofSensitivity: 0.5, // Medium - reviews are accessible info
-            mentalModelRigidity: 0.6, // Medium - adapts but needs time
-            siteFamiliarity: 0.3, // Low - has visited but forgets layout
+            mentalModelFlexibility: 0.6, // Medium - adapts but needs time
         },
         context: {
             viewport: [1280, 800],
@@ -1184,33 +1436,59 @@ export const BUILTIN_PERSONAS = {
             },
         },
         cognitiveTraits: {
-            patience: 0.9, // High - not in a rush
-            riskTolerance: 0.1, // Very low - afraid of mistakes
+            patience: 0.8, // High - not in a rush
+            riskTolerance: 0.2, // Very low - afraid of mistakes
             comprehension: 0.2, // Low - unfamiliar with modern UI
-            persistence: 0.7, // High - determined but confused
-            curiosity: 0.1, // Very low - just wants to finish
-            workingMemory: 0.3, // Low - may forget steps
-            readingTendency: 0.9, // High - reads everything carefully
+            persistence: 0.6, // High - determined but confused
+            curiosity: 0.4, // Very low - just wants to finish
+            workingMemory: 0.4, // Low - may forget steps
+            readingTendency: 0.8, // High - reads everything carefully
             resilience: 0.3, // Low - frustration lingers longer (BRS age-related)
-            selfEfficacy: 0.3, // Low - often blame themselves for tech issues (Bandura)
-            satisficing: 0.7, // High - values simplicity over optimization (Simon)
+            selfEfficacy: 0.4, // Low - often blame themselves for tech issues (Bandura)
+            satisficing: 0.6, // High - values simplicity over optimization (Simon)
             trustCalibration: 0.25, // Very low - skeptical of online requests (Fogg)
-            interruptRecovery: 0.3, // Low - difficulty resuming, forgets context (Mark)
+            interruptRecovery: 0.4, // Low - difficulty resuming, forgets context (Mark)
             // New traits (v15.0.0)
             informationForaging: 0.25, // Very low - exhaustive, uncertain search
-            changeBlindness: 0.7, // High - misses UI updates and changes
-            anchoringBias: 0.8, // High - sticks to first information
+            changeBlindness: 0.6, // High - misses UI updates and changes
+            anchoringBias: 0.7, // High - sticks to first information
             timeHorizon: 0.4, // Medium - patient but unsure of value
             attributionStyle: 0.85, // Very high - blames self for all errors
             metacognitivePlanning: 0.25, // Low - doesn't know what to plan for
-            proceduralFluency: 0.3, // Low - struggles with multi-step processes
+            // 0.5, corrected 2026-08-03 from 0.4, and it is a REASONED MIDPOINT rather
+            // than an unfilled default -- the two forces genuinely cancel.
+            //
+            // The old value came from the doc's rationale, "slower development of
+            // procedural skills with new interfaces". That describes ACQUISITION
+            // RATE. This trait measures EXECUTION: its own scale runs from "struggles
+            // with sequences, skips steps, confuses order" to "follows procedures
+            // precisely and efficiently". Third instance of the same defect class on
+            // these pages, after timeHorizon and attributionStyle -- a number set for
+            // the construct the rationale describes rather than the one the trait
+            // measures.
+            //
+            // It scored 0.4, IDENTICAL to cognitive-adhd, while carrying higher
+            // working memory (0.40 vs 0.30) and four times the patience (0.80 vs
+            // 0.20). The low end of this scale is impulsive error, and this persona's
+            // own page says older users "make fewer impulsive errors".
+            //
+            // Precision is preserved or improved with age and efficiency is reduced
+            // -- the speed-accuracy tradeoff shifts toward accuracy (Fisk et al.
+            // 2009, Designing for Older Adults). Those net to the middle: the trait
+            // reference's own `medium` band reads "follows most procedures with
+            // occasional errors", which is the honest description.
+            //
+            // Stays BELOW motor-impairment-tremor (0.6), which is cognitively intact
+            // with working memory 0.70 and metacognitive planning 0.70 against this
+            // persona's 0.40 and 0.25. Stays ABOVE first-timer (0.2), which has no
+            // procedural schema at all.
+            proceduralFluency: 0.5,
             transferLearning: 0.2, // Low - each new site is confusing
-            authoritySensitivity: 0.85, // Very high - trusts official-looking things
+            authoritySensitivity: 0.7, // Very high - trusts official-looking things
             emotionalContagion: 0.6, // Medium-high - affected by stern warnings
             fearOfMissingOut: 0.3, // Low - not driven by urgency, confused by it
             socialProofSensitivity: 0.7, // High - relies on grandchildren's recommendations
-            mentalModelRigidity: 0.2, // Very low - rigid, struggles with novel UIs
-            siteFamiliarity: 0.5, // Medium - knows the basics
+            mentalModelFlexibility: 0.2, // Very low - rigid, struggles with novel UIs
         },
         context: {
             viewport: [1280, 800],
@@ -1265,37 +1543,154 @@ export const BUILTIN_PERSONAS = {
         },
         cognitiveTraits: {
             patience: 0.1, // Very low - abandons instantly
-            riskTolerance: 0.8, // High - clicks first thing
-            comprehension: 0.5, // Medium
-            persistence: 0.1, // Very low - one strike and out
-            curiosity: 0.1, // Very low - no time to explore
+            riskTolerance: 0.6, // High - clicks first thing
+            comprehension: 0.6, // Medium
+            persistence: 0.2, // Very low - one strike and out
+            curiosity: 0.3, // Very low - no time to explore
             workingMemory: 0.6, // Medium
-            readingTendency: 0.05, // Almost none - scanning only
-            resilience: 0.2, // Very low - no recovery, abandons (BRS)
-            selfEfficacy: 0.5, // Medium - impatient doesn't mean incapable (Bandura)
+            readingTendency: 0.1, // Almost none - scanning only
+            resilience: 0.3, // Very low - no recovery, abandons (BRS)
+            selfEfficacy: 0.6, // Medium - impatient doesn't mean incapable (Bandura)
             satisficing: 0.9, // Very high - definition of satisficing (Simon)
             trustCalibration: 0.7, // High - clicks through without reading (Fogg)
             interruptRecovery: 0.2, // Very low - abandons rather than resumes (Mark)
             // New traits (v15.0.0)
             informationForaging: 0.9, // Very high - follows first scent, moves on
-            changeBlindness: 0.5, // Medium - focused on task
+            changeBlindness: 0.7, // Medium - focused on task
             anchoringBias: 0.6, // Medium-high - goes with first option
             timeHorizon: 0.95, // Maximum short-term - no time for later
             attributionStyle: 0.3, // Low - blames slow site
-            metacognitivePlanning: 0.1, // Very low - no planning, just clicking
-            proceduralFluency: 0.5, // Medium - can follow if fast
+            metacognitivePlanning: 0.3, // Very low - no planning, just clicking
+            proceduralFluency: 0.6, // Medium - can follow if fast
             transferLearning: 0.7, // High - recognizes patterns quickly
             authoritySensitivity: 0.4, // Medium - only if it speeds things up
             emotionalContagion: 0.4, // Medium - focused on goal, not tone
-            fearOfMissingOut: 0.9, // Very high - urgency works on them
-            socialProofSensitivity: 0.5, // Medium - if it's quick to evaluate
-            mentalModelRigidity: 0.7, // High - adapts quickly, expects conventions
-            siteFamiliarity: 0.2, // Low - site structure feels different each visit
+            fearOfMissingOut: 0.8, // Very high - urgency works on them
+            socialProofSensitivity: 0.4, // Medium - if it's quick to evaluate
+            mentalModelFlexibility: 0.6, // High - adapts quickly, expects conventions
         },
         context: {
             viewport: [1280, 800],
         },
     },
+    // Four personas that had a values profile in the Schwartz registry and no
+    // definition here, so nothing could be RUN as them: empathy_audit,
+    // cognitive_journey and every other persona-driven tool rejected the name
+    // while persona_values_lookup happily returned a full motivational profile.
+    // The values were authored and coherent -- explorer at selfDirection 0.9 /
+    // stimulation 0.9 / security 0.2 is a deliberate description, not a stub --
+    // so the missing half is written here rather than the authored half deleted.
+    // Traits are chosen to be consistent with each one's existing values.
+    // (2026-07-31)
+    "distracted-user": {
+        name: "distracted-user",
+        description: "Attention repeatedly pulled away mid-task; returns with context lost",
+        demographics: { age_range: "18-55", tech_level: "intermediate", device: "desktop" },
+        behaviors: { loses_place: true, task_switches: true, rereads: true },
+        humanBehavior: {
+            timing: {
+                reactionTime: { min: 250, max: 1400 }, clickDelay: { min: 120, max: 500 },
+                typeSpeed: { min: 25, max: 70 }, readingSpeed: 200,
+                scrollPauseTime: { min: 200, max: 1500 },
+            },
+            errors: { misClickRate: 0.09, doubleClickAccidental: 0.04, typoRate: 0.08, backtrackRate: 0.35 },
+            mouse: { curvature: 0.4, jitter: 8, overshoot: 0.15, speed: "normal" },
+            attention: { pattern: "skim", scrollBehavior: "jump", focusAreas: ["header", "cta"], distractionRate: 0.85 },
+        },
+        cognitiveTraits: {
+            patience: 0.4, riskTolerance: 0.5, comprehension: 0.5, persistence: 0.3,
+            curiosity: 0.6, workingMemory: 0.25, readingTendency: 0.3, resilience: 0.4,
+            selfEfficacy: 0.45, satisficing: 0.7, trustCalibration: 0.5,
+            interruptRecovery: 0.1, // The defining trait: context is not recovered
+            informationForaging: 0.6, changeBlindness: 0.8, anchoringBias: 0.6,
+            timeHorizon: 0.7, attributionStyle: 0.4, metacognitivePlanning: 0.2,
+            proceduralFluency: 0.5, transferLearning: 0.5, authoritySensitivity: 0.5,
+            emotionalContagion: 0.5, fearOfMissingOut: 0.7, socialProofSensitivity: 0.5,
+            mentalModelFlexibility: 0.5,
+        },
+        context: { viewport: [1280, 800] },
+    },
+    "careful-reader": {
+        name: "careful-reader",
+        description: "Reads before acting; verifies details and dislikes surprises",
+        demographics: { age_range: "25-65", tech_level: "intermediate", device: "desktop" },
+        behaviors: { reads_fully: true, verifies_before_submit: true, avoids_risk: true },
+        humanBehavior: {
+            timing: {
+                reactionTime: { min: 600, max: 2500 }, clickDelay: { min: 300, max: 900 },
+                typeSpeed: { min: 30, max: 80 }, readingSpeed: 130,
+                scrollPauseTime: { min: 800, max: 3000 },
+            },
+            errors: { misClickRate: 0.02, doubleClickAccidental: 0.01, typoRate: 0.02, backtrackRate: 0.08 },
+            mouse: { curvature: 0.2, jitter: 3, overshoot: 0.03, speed: "slow" },
+            attention: { pattern: "thorough", scrollBehavior: "continuous", focusAreas: ["text", "prices"], distractionRate: 0.1 },
+        },
+        cognitiveTraits: {
+            patience: 0.9, riskTolerance: 0.15, comprehension: 0.8, persistence: 0.8,
+            curiosity: 0.5, workingMemory: 0.7, readingTendency: 0.95, resilience: 0.7,
+            selfEfficacy: 0.6, satisficing: 0.1, trustCalibration: 0.25,
+            interruptRecovery: 0.7, informationForaging: 0.4, changeBlindness: 0.2,
+            anchoringBias: 0.3, timeHorizon: 0.3, attributionStyle: 0.6,
+            metacognitivePlanning: 0.8, proceduralFluency: 0.6, transferLearning: 0.6,
+            authoritySensitivity: 0.6, emotionalContagion: 0.3, fearOfMissingOut: 0.15,
+            socialProofSensitivity: 0.4, mentalModelFlexibility: 0.4,
+        },
+        context: { viewport: [1280, 800] },
+    },
+    "explorer": {
+        name: "explorer",
+        description: "Wanders the interface by choice, opening things to see what they do",
+        demographics: { age_range: "18-45", tech_level: "expert", device: "desktop" },
+        behaviors: { wanders: true, tries_everything: true, ignores_happy_path: true },
+        humanBehavior: {
+            timing: {
+                reactionTime: { min: 200, max: 900 }, clickDelay: { min: 100, max: 350 },
+                typeSpeed: { min: 45, max: 110 }, readingSpeed: 250,
+                scrollPauseTime: { min: 200, max: 1200 },
+            },
+            errors: { misClickRate: 0.05, doubleClickAccidental: 0.02, typoRate: 0.04, backtrackRate: 0.25 },
+            mouse: { curvature: 0.5, jitter: 6, overshoot: 0.12, speed: "fast" },
+            attention: { pattern: "z-pattern", scrollBehavior: "chunked", focusAreas: ["header", "images", "cta"], distractionRate: 0.5 },
+        },
+        cognitiveTraits: {
+            patience: 0.7, riskTolerance: 0.9, comprehension: 0.75, persistence: 0.7,
+            curiosity: 0.95, workingMemory: 0.65, readingTendency: 0.4, resilience: 0.8,
+            selfEfficacy: 0.85, satisficing: 0.2, trustCalibration: 0.6,
+            interruptRecovery: 0.7, informationForaging: 0.9, changeBlindness: 0.3,
+            anchoringBias: 0.25, timeHorizon: 0.35, attributionStyle: 0.7,
+            metacognitivePlanning: 0.5, proceduralFluency: 0.7, transferLearning: 0.85,
+            authoritySensitivity: 0.2, emotionalContagion: 0.4, fearOfMissingOut: 0.4,
+            socialProofSensitivity: 0.25, mentalModelFlexibility: 0.75,
+        },
+        context: { viewport: [1440, 900] },
+    },
+    "task-focused": {
+        name: "task-focused",
+        description: "Came to finish one thing; ignores everything that is not on the path",
+        demographics: { age_range: "25-55", tech_level: "intermediate", device: "desktop" },
+        behaviors: { goal_directed: true, ignores_promotions: true, resents_detours: true },
+        humanBehavior: {
+            timing: {
+                reactionTime: { min: 200, max: 800 }, clickDelay: { min: 100, max: 300 },
+                typeSpeed: { min: 40, max: 100 }, readingSpeed: 260,
+                scrollPauseTime: { min: 150, max: 600 },
+            },
+            errors: { misClickRate: 0.04, doubleClickAccidental: 0.02, typoRate: 0.04, backtrackRate: 0.12 },
+            mouse: { curvature: 0.2, jitter: 4, overshoot: 0.06, speed: "fast" },
+            attention: { pattern: "f-pattern", scrollBehavior: "jump", focusAreas: ["cta", "text"], distractionRate: 0.15 },
+        },
+        cognitiveTraits: {
+            patience: 0.6, riskTolerance: 0.5, comprehension: 0.7, persistence: 0.9,
+            curiosity: 0.2, workingMemory: 0.7, readingTendency: 0.35, resilience: 0.75,
+            selfEfficacy: 0.8, satisficing: 0.5, trustCalibration: 0.5,
+            interruptRecovery: 0.8, informationForaging: 0.7, changeBlindness: 0.6,
+            anchoringBias: 0.45, timeHorizon: 0.6, attributionStyle: 0.6,
+            metacognitivePlanning: 0.75, proceduralFluency: 0.8, transferLearning: 0.7,
+            authoritySensitivity: 0.4, emotionalContagion: 0.25, fearOfMissingOut: 0.3,
+            socialProofSensitivity: 0.2, mentalModelFlexibility: 0.6,
+        },
+        context: { viewport: [1280, 800] },
+    },
 };
 /**
  * Get a persona by name.
@@ -1329,7 +1724,7 @@ export function listCustomPersonas() {
  * Get the custom personas directory path.
  */
 export function getPersonasDir() {
-    return PERSONAS_DIR;
+    return personasDirFor();
 }
 /**
  * Built-in accessibility personas for empathy testing.
@@ -1394,37 +1789,49 @@ export const ACCESSIBILITY_PERSONAS = {
         cognitiveTraits: {
             // Research: Motor impairment + self-efficacy (Lorig et al., 2001 - Chronic Disease Self-Management)
             // Adapted users maintain moderate efficacy despite physical challenges
-            selfEfficacy: 0.5,
+            selfEfficacy: 0.6,
             // Research: Physical effort increases satisficing (Kool et al., 2010 - Effort discounting)
             // Motor strain leads to accepting "good enough" to minimize fatigue
-            satisficing: 0.7,
-            // Research: Chronic condition users develop moderate trust (Lorig 2001)
-            trustCalibration: 0.5,
+            // High = first adequate option. Each additional comparison means another
+            // set of pointing operations.
+            satisficing: 0.75,
+            trustCalibration: 0.6,
             // Research: Motor issues make resumption harder but adaptation helps (Mark)
-            interruptRecovery: 0.4,
-            // New traits (v15.0.0)
-            patience: 0.6,
-            riskTolerance: 0.4,
+            // Re-acquiring a small target after an interruption repeats the hardest
+            // part of the interaction.
+            interruptRecovery: 0.6,
+            // Every pointing action costs time and repeat attempts, so tolerance for
+            // slow interaction is built in -- the persona's own research note records
+            // 1-15Hz tremor and misclicks as the primary barrier.
+            patience: 0.8,
+            // Low = only clicks obvious, labelled controls. An accidental click is
+            // expensive to undo when re-pointing is itself hard.
+            riskTolerance: 0.3,
             comprehension: 0.7,
+            // A missed target is retried rather than abandoned; the alternative is not
+            // completing the task at all.
             persistence: 0.7,
             curiosity: 0.4,
             workingMemory: 0.7,
             readingTendency: 0.6,
             resilience: 0.6,
-            informationForaging: 0.5,
+            // Low = exhaustive rather than scent-following. Exploratory clicking is
+            // expensive, so paths are chosen carefully and few are sampled.
+            informationForaging: 0.35,
             changeBlindness: 0.4,
             anchoringBias: 0.5,
             timeHorizon: 0.4, // Patient due to motor constraints
             attributionStyle: 0.4, // Blames physical constraints, not self
-            metacognitivePlanning: 0.7, // Plans to avoid difficult interactions
+            // Target order and approach are planned to minimise pointing: keyboard
+            // where possible, large targets first.
+            metacognitivePlanning: 0.7,
             proceduralFluency: 0.6,
             transferLearning: 0.6,
             authoritySensitivity: 0.5,
             emotionalContagion: 0.4,
             fearOfMissingOut: 0.3, // Not swayed, focused on completing
             socialProofSensitivity: 0.5,
-            mentalModelRigidity: 0.5,
-            siteFamiliarity: 0.5, // Medium - regular user with motor challenges
+            mentalModelFlexibility: 0.5,
         },
     },
     "low-vision-magnified": {
@@ -1475,6 +1882,17 @@ export const ACCESSIBILITY_PERSONAS = {
             viewport: [1280, 800], // But effectively sees 1/9th at a time
         },
         accessibilityTraits: {
+            // Reading capacity, stated rather than inferred from the persona's NAME.
+            // These are the same values getReadingProfile returned for this persona
+            // via name matching, so behaviour is unchanged -- but they are now
+            // editable, which they were not: no trait change could move WPM, fixation
+            // span or phonological penalty. Source: Legge et al. (2007) -- central
+            // scotoma and crowding. (2026-08-02)
+            orthographicFluency: 0.6,
+            phonologicalDecoding: 0.75,
+            visualSpan: 3,
+            vocabularyBreadth: 0.75,
+            crowdingSensitivity: 0.7,
             visionLevel: 0.3,
             contrastSensitivity: 3.0,
             processingSpeed: 0.7,
@@ -1484,10 +1902,10 @@ export const ACCESSIBILITY_PERSONAS = {
         cognitiveTraits: {
             // Research: Visual impairment + self-efficacy (Brody et al., 2002 - AMD and depression)
             // Reduced vision correlates with lower confidence in task completion
-            selfEfficacy: 0.45,
+            selfEfficacy: 0.6,
             // Research: Limited information access increases satisficing (Pachur & Hertwig, 2006)
             // Seeing only 1/9th of screen at a time → accept accessible options
-            satisficing: 0.65,
+            satisficing: 0.4,
             // Research: Vision impairment increases caution (Brody 2002)
             trustCalibration: 0.4,
             // Research: Limited visible area makes context recovery harder (Mark)
@@ -1495,26 +1913,25 @@ export const ACCESSIBILITY_PERSONAS = {
             // New traits (v15.0.0)
             patience: 0.7,
             riskTolerance: 0.3,
-            comprehension: 0.6,
-            persistence: 0.7,
+            comprehension: 0.7,
+            persistence: 0.8,
             curiosity: 0.3,
             workingMemory: 0.6,
-            readingTendency: 0.8, // Reads carefully when visible
-            resilience: 0.5,
+            readingTendency: 0.9, // Reads carefully when visible
+            resilience: 0.7,
             informationForaging: 0.4, // Limited by zoom - systematic search
             changeBlindness: 0.8, // Very high - can't see peripheral
             anchoringBias: 0.6, // Anchored due to limited comparison
             timeHorizon: 0.4, // Patient, knows it takes time
             attributionStyle: 0.5, // Balanced - knows about limitations
             metacognitivePlanning: 0.7, // Plans navigation with zoom
-            proceduralFluency: 0.5, // Harder with limited view
-            transferLearning: 0.5, // Standard
+            proceduralFluency: 0.6, // Harder with limited view
+            transferLearning: 0.6, // Standard
             authoritySensitivity: 0.5,
             emotionalContagion: 0.4,
-            fearOfMissingOut: 0.3, // Not swayed - focused
+            fearOfMissingOut: 0.4, // Not swayed - focused
             socialProofSensitivity: 0.5,
-            mentalModelRigidity: 0.4, // Adapts but needs time
-            siteFamiliarity: 0.3, // Low - magnification makes layout unfamiliar
+            mentalModelFlexibility: 0.4, // Adapts but needs time
         },
     },
     "cognitive-adhd": {
@@ -1564,38 +1981,83 @@ export const ACCESSIBILITY_PERSONAS = {
             viewport: [1920, 1080],
         },
         accessibilityTraits: {
+            // Reading capacity, stated rather than inferred from the persona's NAME.
+            // These are the same values getReadingProfile returned for this persona
+            // via name matching, so behaviour is unchanged -- but they are now
+            // editable, which they were not: no trait change could move WPM, fixation
+            // span or phonological penalty. Source: Dhar et al. (2010) -- intact
+            // decoding, reduced sustained attention. (2026-08-02)
+            orthographicFluency: 0.8,
+            phonologicalDecoding: 0.75,
+            visualSpan: 6,
+            vocabularyBreadth: 0.7,
+            crowdingSensitivity: 0.2,
             processingSpeed: 0.6, // Variable - can be fast when interested
             attentionSpan: 0.3, // Short sustained attention
             fatigueSusceptibility: 0.4,
         },
+        // Reconciled against Persona-ADHD.md, the persona's own published trait
+        // table, on 2026-08-02. Nineteen of twenty-five values disagreed. Each was
+        // arbitrated against trait-reference.ts -- which names specific personas as
+        // typically high or low per trait and defines what each end of the scale
+        // MEANS -- rather than by preferring one file over the other.
+        //
+        // Fifteen adopted the doc's value; the doc was corrected on the other four,
+        // which are recorded in Persona-ADHD.md. The four are worth naming because
+        // they are all one failure: a number written against the plain-English
+        // reading of a trait's NAME instead of its documented scale.
+        //
+        //   timeHorizon           doc said 0.3, meaning "long-term focused"
+        //   attributionStyle      doc said 0.3, meaning "blames the system"
+        //   mentalModelFlexibility doc said 0.4, meaning "somewhat rigid"
+        //   trustCalibration      doc said 0.5, meaning "neutral"
+        //
+        // In each case the doc's own RATIONALE argued for the opposite number. It
+        // said "strong preference for immediate rewards" beside 0.3 on a scale
+        // whose 1.0 end is literally "Immediate gratification, wants results NOW",
+        // and "may blame self" beside 0.3 on a scale whose 1.0 end is "Blames
+        // self". The prose was right every time; the digit was inverted.
         cognitiveTraits: {
-            workingMemory: 0.25, // Research: large magnitude impairment (d=1.63-2.03)
-            patience: 0.25, // Impulsivity: low tolerance for waiting
-            persistence: 0.2, // Low task persistence before switching
-            curiosity: 0.8, // High novelty-seeking
+            workingMemory: 0.3, // Research: large magnitude impairment (d=1.63-2.03)
+            patience: 0.2, // Impulsivity: low tolerance for waiting
+            persistence: 0.3, // Low for unengaging tasks; can hyperfocus on interesting ones
+            curiosity: 0.9, // High novelty-seeking
             riskTolerance: 0.7, // Impulsive clicking
             readingTendency: 0.2, // Skims rather than reads
-            resilience: 0.55, // Medium - recovers emotionally but not focus (BRS)
+            resilience: 0.4, // Emotional dysregulation is common (doc; code had 0.55)
             selfEfficacy: 0.4, // Variable - can feel capable when engaged (Bandura)
-            satisficing: 0.85, // High - ADHD drives "good enough" decisions (Simon)
-            trustCalibration: 0.65, // Medium-high - impulsive clicks without reading (Fogg)
-            interruptRecovery: 0.2, // Very low - WM deficits make recovery difficult (Mark)
+            satisficing: 0.7, // High - ADHD drives "good enough" decisions (Simon)
+            // KEPT over the doc's 0.5: trait-reference lists cognitive-adhd under
+            // typicalScores.high, and 1.0 on this scale is "highly trusting, accepts
+            // claims at face value" -- which is what impulsive clicking without
+            // reading looks like.
+            trustCalibration: 0.65,
+            interruptRecovery: 0.3, // Very low - WM deficits make recovery difficult (Mark)
             // New traits (v15.0.0)
-            comprehension: 0.5,
-            informationForaging: 0.6, // Follows interesting scents, not systematic
-            changeBlindness: 0.7, // High - focus elsewhere when changes happen
-            anchoringBias: 0.4, // Lower - novelty-seeking overrides anchoring
-            timeHorizon: 0.9, // Very short-term - immediate gratification
-            attributionStyle: 0.6, // Slightly self-blaming
-            metacognitivePlanning: 0.15, // Very low - impulsive, no planning
-            proceduralFluency: 0.3, // Low - struggles with sequences
-            transferLearning: 0.6, // Medium - recognizes patterns when focused
+            comprehension: 0.6, // Capable when engaged; this is UI-convention grasp, not reading
+            informationForaging: 0.4, // Distractible; goes down tangential paths (doc; code had 0.6)
+            changeBlindness: 0.6, // High - focus elsewhere when changes happen
+            anchoringBias: 0.6, // First option favored, decision-fatigue avoidance (doc; code had 0.4)
+            // KEPT over the doc's 0.3. 1.0 on this scale is "Immediate gratification,
+            // wants results NOW" -- the name reads like "how far ahead do you look",
+            // and it means the opposite. trait-reference lists cognitive-adhd under
+            // typicalScores.high alongside impatient-user and mobile-user.
+            timeHorizon: 0.9,
+            // KEPT over the doc's 0.3. 1.0 is "Blames self (internal attribution)",
+            // and the doc's own rationale for 0.3 was "may blame self".
+            attributionStyle: 0.6,
+            metacognitivePlanning: 0.3, // Very low - impulsive, no planning
+            proceduralFluency: 0.4, // Inconsistent; routines help but are hard to establish
+            transferLearning: 0.5, // Variable; depends on engagement level
             authoritySensitivity: 0.4, // Low - impulsive, ignores authority
             emotionalContagion: 0.6, // Medium-high - emotional variability
-            fearOfMissingOut: 0.85, // Very high - ADHD correlates with FOMO
+            fearOfMissingOut: 0.7, // High - ADHD correlates with FOMO
             socialProofSensitivity: 0.5, // Medium - if interesting
-            mentalModelRigidity: 0.7, // High - adapts easily, novelty-seeking
-            siteFamiliarity: 0.3, // Low - doesn't retain site structure well
+            // KEPT over the doc's 0.4. 1.0 is "Highly adaptive - quickly forms new
+            // mental models"; the doc's rationale was "flexible thinking, sometimes
+            // too flexible". A fourth instance of the rigidity/flexibility inversion
+            // that commit 086d519 fixed in three other pages.
+            mentalModelFlexibility: 0.7,
         },
     },
     "dyslexic-user": {
@@ -1645,38 +2107,81 @@ export const ACCESSIBILITY_PERSONAS = {
             viewport: [1280, 800],
         },
         accessibilityTraits: {
+            // Reading capacity, stated rather than inferred from the persona's NAME.
+            // These are the same values getReadingProfile returned for this persona
+            // via name matching, so behaviour is unchanged -- but they are now
+            // editable, which they were not: no trait change could move WPM, fixation
+            // span or phonological penalty. Source: Perry, Zorzi & Ziegler (2019),
+            // Psychological Science. (2026-08-02)
+            orthographicFluency: 0.4,
+            phonologicalDecoding: 0.35,
+            visualSpan: 4,
+            vocabularyBreadth: 0.65,
+            crowdingSensitivity: 0.55,
             processingSpeed: 0.5, // Slower text processing
             attentionSpan: 0.5,
             fatigueSusceptibility: 0.6, // Reading requires more effort
         },
         cognitiveTraits: {
-            workingMemory: 0.6, // WM typically unaffected
-            patience: 0.5,
+            // WM deficit persists into adulthood across phonological AND visuospatial
+            // modalities; a cognitive marker distinguishing dyslexic from typical
+            // readers in every European orthography (de Jong 2006; n-back fNIRS,
+            // PMC3496727). Was 0.6 -- above average -- which inverted one of the best-
+            // established findings in the literature.
+            workingMemory: 0.3,
+            // Slower reading is one of three documented compensatory mechanisms,
+            // alongside text-structure use and verbal ability (Annals of Dyslexia,
+            // 10.1007/s11881-020-00195-w). Taking extra time with text is the
+            // strategy, not a failure of tolerance.
+            patience: 0.75,
             curiosity: 0.7,
-            readingTendency: 0.4, // Avoids heavy text, prefers visuals
-            resilience: 0.6, // Medium-high - adapted to text challenges (BRS)
-            selfEfficacy: 0.5, // Medium - effective with accommodations (Bandura)
-            satisficing: 0.6, // Medium - avoids text-heavy paths (Simon)
+            // Relies on structure and skimming over sustained prose. Note this is a
+            // STRATEGY, not decoding ability -- decoding now enters the chain via
+            // readingCapacity, bridged from getReadingProfile.
+            readingTendency: 0.35,
+            // Pulled down from 0.6 for documented secondary anxiety (Kapur & Agrawal
+            // 2026), but not lower: the compensation literature describes adults
+            // achieving academic success, so learned helplessness would overstate it.
+            resilience: 0.6,
+            // Dyslexia is a strong predictor of lower academic self-efficacy in higher
+            // education (PMC11488781; Nigerian undergraduate cohort).
+            selfEfficacy: 0.3,
+            // Reading is the expensive operation, so the first adequate option is
+            // strongly preferred over an exhaustive comparison.
+            satisficing: 0.7,
             trustCalibration: 0.5, // Medium - standard trust evaluation (Fogg)
-            interruptRecovery: 0.55, // Medium - visual cues help, text harder (Mark)
+            // Follows from the WM deficit: resuming a task means reloading context
+            // that is expensive to hold.
+            interruptRecovery: 0.35,
             // New traits (v15.0.0)
-            riskTolerance: 0.5,
+            riskTolerance: 0.4,
             comprehension: 0.5, // Comprehension fine for visuals
-            persistence: 0.6,
-            informationForaging: 0.5, // Standard - prefers visual cues
+            // The same compensation literature: strategies are built and sustained
+            // across years of schooling, and mastery experience feeds back into
+            // persistence.
+            persistence: 0.75,
+            // Text-structure sensitivity is the second documented compensator:
+            // headings and layout are used as scent rather than reading everything,
+            // which is the high end of this scale.
+            informationForaging: 0.65,
             changeBlindness: 0.4, // Lower - visual learner
-            anchoringBias: 0.5,
+            anchoringBias: 0.6,
             timeHorizon: 0.5,
-            attributionStyle: 0.5,
-            metacognitivePlanning: 0.5,
-            proceduralFluency: 0.5, // Fine with visual instructions
+            // High = blames self. Psychoeducation framing dyslexia as
+            // neurodevelopmental rather than personal inadequacy measurably REDUCES
+            // self-blame, which places the untreated baseline above neutral (Kapur &
+            // Agrawal 2026).
+            attributionStyle: 0.7,
+            // Deliberate compensatory strategy use IS metacognition -- choosing
+            // structure over linear prose is a planned approach, not a reflex.
+            metacognitivePlanning: 0.7,
+            proceduralFluency: 0.4,
             transferLearning: 0.6, // Recognizes visual patterns
             authoritySensitivity: 0.5,
             emotionalContagion: 0.5,
-            fearOfMissingOut: 0.5,
+            fearOfMissingOut: 0.4,
             socialProofSensitivity: 0.6, // Relies on visual reviews
-            mentalModelRigidity: 0.6, // Good adaptation
-            siteFamiliarity: 0.5, // Medium - knows sites but reading is hard
+            mentalModelFlexibility: 0.6, // Good adaptation
         },
     },
     "deaf-user": {
@@ -1729,33 +2234,32 @@ export const ACCESSIBILITY_PERSONAS = {
             attentionSpan: 0.8,
         },
         cognitiveTraits: {
-            patience: 0.7, // Accustomed to finding captions/text alternatives
-            riskTolerance: 0.5, // Standard caution
+            patience: 0.6, // Accustomed to finding captions/text alternatives
+            riskTolerance: 0.4, // Standard caution
             readingTendency: 0.9, // High - reads all text since can't hear audio
-            resilience: 0.75, // High - adapted to audio-free experience (BRS)
-            selfEfficacy: 0.6, // Medium-high - confident in visual navigation (Bandura)
-            satisficing: 0.55, // Medium - seeks captioned/text content (Simon)
-            trustCalibration: 0.55, // Medium - standard visual trust evaluation (Fogg)
-            interruptRecovery: 0.65, // Medium-high - visual cues assist recovery (Mark)
+            resilience: 0.7, // High - adapted to audio-free experience (BRS)
+            selfEfficacy: 0.7, // Medium-high - confident in visual navigation (Bandura)
+            satisficing: 0.5, // Medium - seeks captioned/text content (Simon)
+            trustCalibration: 0.6, // Medium - standard visual trust evaluation (Fogg)
+            interruptRecovery: 0.7, // Medium-high - visual cues assist recovery (Mark)
             // New traits (v15.0.0)
-            comprehension: 0.7,
+            comprehension: 0.8,
             persistence: 0.7,
-            curiosity: 0.5,
-            workingMemory: 0.7,
+            curiosity: 0.6,
+            workingMemory: 0.6,
             informationForaging: 0.6, // Good visual navigation
-            changeBlindness: 0.2, // Very low - visually attentive
+            changeBlindness: 0.3, // Very low - visually attentive
             anchoringBias: 0.5,
             timeHorizon: 0.5,
             attributionStyle: 0.4, // Blames lack of captions
-            metacognitivePlanning: 0.6, // Plans for captioned content
+            metacognitivePlanning: 0.7, // Plans for captioned content
             proceduralFluency: 0.7, // Good with visual instructions
             transferLearning: 0.7, // Recognizes visual patterns
             authoritySensitivity: 0.5,
-            emotionalContagion: 0.4, // Less affected by audio tone
+            emotionalContagion: 0.6, // Less affected by audio tone
             fearOfMissingOut: 0.4,
             socialProofSensitivity: 0.6, // Visual reviews/ratings
-            mentalModelRigidity: 0.65, // Good visual adaptation
-            siteFamiliarity: 0.5, // Medium - visual learner, remembers layout
+            mentalModelFlexibility: 0.65, // Good visual adaptation
         },
     },
     "elderly-low-vision": {
@@ -1806,6 +2310,21 @@ export const ACCESSIBILITY_PERSONAS = {
             viewport: [1280, 800],
         },
         accessibilityTraits: {
+            // This persona is BOTH elderly and low-vision, and the name lookup
+            // resolved it by accident: it tests 'low-vision' before 'elderly', so it
+            // silently returned the low-vision profile and the aging half was never
+            // applied. Rather than preserve an ordering accident, each dimension
+            // takes the MORE impaired of the two profiles -- both conditions impair
+            // reading -- except vocabulary, where age is an advantage: crystallized
+            // vocabulary grows across the lifespan (Rayner et al. 2006), so the
+            // elderly value of 0.85 is kept over low vision's 0.75.
+            // orthographic 0.6 (lowVision) / phonological 0.65 (elderly) /
+            // span 3 (lowVision) / vocabulary 0.85 (elderly) / crowding 0.7 (lowVision).
+            orthographicFluency: 0.6,
+            phonologicalDecoding: 0.65,
+            visualSpan: 3,
+            vocabularyBreadth: 0.85,
+            crowdingSensitivity: 0.7,
             motorControl: 0.5,
             tremor: false, // Not necessarily present
             reachability: 0.6,
@@ -1816,33 +2335,32 @@ export const ACCESSIBILITY_PERSONAS = {
             fatigueSusceptibility: 0.7,
         },
         cognitiveTraits: {
-            workingMemory: 0.35, // Research: inhibitory control decline limits WM
-            patience: 0.7, // High: more methodical than young users
-            persistence: 0.6, // Determined but confused
+            workingMemory: 0.3, // Research: inhibitory control decline limits WM
+            patience: 0.9, // High: more methodical than young users
+            persistence: 0.8, // Determined but confused
             curiosity: 0.3, // Low: prefers familiar patterns
-            riskTolerance: 0.15, // Very cautious with unfamiliar interfaces
+            riskTolerance: 0.1, // Very cautious with unfamiliar interfaces
             comprehension: 0.3, // Unfamiliar with modern UI conventions
-            resilience: 0.25, // Low - frustration compounds with physical challenges (BRS)
-            selfEfficacy: 0.25, // Very low - often blame selves for tech issues (Bandura)
-            satisficing: 0.75, // High - avoids complex decision trees (Simon)
-            trustCalibration: 0.2, // Very low - very skeptical of online requests (Fogg)
-            interruptRecovery: 0.25, // Very low - WM + vision decline makes recovery hard (Mark)
+            resilience: 0.6, // Low - frustration compounds with physical challenges (BRS)
+            selfEfficacy: 0.3, // Very low - often blame selves for tech issues (Bandura)
+            satisficing: 0.7, // High - avoids complex decision trees (Simon)
+            trustCalibration: 0.4, // Very low - very skeptical of online requests (Fogg)
+            interruptRecovery: 0.3, // Very low - WM + vision decline makes recovery hard (Mark)
             // New traits (v15.0.0)
-            readingTendency: 0.85,
-            informationForaging: 0.2, // Very low - exhaustive, uncertain
-            changeBlindness: 0.75, // High - vision issues
+            readingTendency: 0.9,
+            informationForaging: 0.4, // Very low - exhaustive, uncertain
+            changeBlindness: 0.7, // High - vision issues
             anchoringBias: 0.8, // High - sticks to first info
             timeHorizon: 0.4, // Medium - patient but unsure
             attributionStyle: 0.9, // Very high - blames self
             metacognitivePlanning: 0.2, // Low - doesn't know what to plan
-            proceduralFluency: 0.25, // Low - struggles with sequences
-            transferLearning: 0.15, // Very low - each UI is confusing
-            authoritySensitivity: 0.85, // Very high - trusts official things
+            proceduralFluency: 0.4, // Low - struggles with sequences
+            transferLearning: 0.4, // Very low - each UI is confusing
+            authoritySensitivity: 0.7, // Very high - trusts official things
             emotionalContagion: 0.6, // Medium-high
-            fearOfMissingOut: 0.25, // Low - not driven by urgency
+            fearOfMissingOut: 0.2, // Low - not driven by urgency
             socialProofSensitivity: 0.7, // High - relies on family advice
-            mentalModelRigidity: 0.15, // Very low - rigid, struggles
-            siteFamiliarity: 0.2, // Low - forgets between visits
+            mentalModelFlexibility: 0.15, // Very low - rigid, struggles
         },
     },
     "color-blind-deuteranopia": {
@@ -1900,31 +2418,46 @@ export const ACCESSIBILITY_PERSONAS = {
         cognitiveTraits: {
             patience: 0.6, // Accustomed to extra verification
             persistence: 0.7, // Used to working around color issues
-            riskTolerance: 0.5, // Cautious with color-only indicators
+            // Low = only clicks the obvious. A control whose state is colour-coded is
+            // ambiguous, so the unambiguous one is chosen.
+            riskTolerance: 0.4,
             resilience: 0.7, // High - adapted to workarounds, recovers quickly (BRS)
-            selfEfficacy: 0.7, // High - adapted strategies work well (Bandura)
+            // Deliberately ABOVE neutral. Deuteranopia is lifelong, common (about 8%
+            // of men of Northern European descent) and compensated from childhood; it
+            // is a perceptual difference, not a confidence deficit, and modelling it
+            // as one would be wrong.
+            selfEfficacy: 0.65,
             satisficing: 0.5, // Medium - needs extra verification (Simon)
-            trustCalibration: 0.55, // Medium - cautious with color-coded security indicators (Fogg)
+            // Low = sceptical. Interfaces routinely signal validity and error through
+            // colour alone, so a page's own status signals are not reliably readable
+            // and get verified rather than believed.
+            trustCalibration: 0.4,
             interruptRecovery: 0.6, // Medium-high - good adaptation, unaffected cognition (Mark)
             // New traits (v15.0.0)
             comprehension: 0.7,
-            curiosity: 0.5,
+            curiosity: 0.6,
             workingMemory: 0.7,
             readingTendency: 0.7, // Relies on text labels
-            informationForaging: 0.6, // Good - unaffected cognition
-            changeBlindness: 0.45, // Medium - misses color-only changes
+            // Scent-following depends on non-colour cues -- position, label, icon --
+            // which is a learned and efficient strategy.
+            informationForaging: 0.65,
+            // High = misses changes. A state change signalled by colour alone -- a
+            // green valid border, a red error border -- is invisible without a second
+            // cue (MDPI 13/2/53; simulated-interface study).
+            changeBlindness: 0.7,
             anchoringBias: 0.5,
             timeHorizon: 0.5,
             attributionStyle: 0.3, // Blames poor design, not self
-            metacognitivePlanning: 0.6, // Plans for color verification
+            // Compensation is deliberate: read the label, look for the icon, re-check
+            // the text rather than trusting the hue.
+            metacognitivePlanning: 0.7,
             proceduralFluency: 0.7, // Good with non-color cues
             transferLearning: 0.7, // Recognizes non-color patterns
             authoritySensitivity: 0.5,
             emotionalContagion: 0.5,
             fearOfMissingOut: 0.4,
             socialProofSensitivity: 0.5,
-            mentalModelRigidity: 0.65, // Adapts with workarounds
-            siteFamiliarity: 0.6, // Medium-high - functional user
+            mentalModelFlexibility: 0.65, // Adapts with workarounds
         },
     },
     // =========================================================================
@@ -2013,8 +2546,7 @@ export const ACCESSIBILITY_PERSONAS = {
             emotionalContagion: 0.3, // Low - less influenced by emotional UI tone
             fearOfMissingOut: 0.2, // Low - not driven by social urgency
             socialProofSensitivity: 0.25, // Low - evaluates independently
-            mentalModelRigidity: 0.2, // Very low (rigid) - needs consistent patterns
-            siteFamiliarity: 0.4, // Low-medium - remembers structure but needs consistency
+            mentalModelFlexibility: 0.2, // Very low (rigid) - needs consistent patterns
         },
     },
     "intellectual-disability": {
@@ -2100,8 +2632,7 @@ export const ACCESSIBILITY_PERSONAS = {
             emotionalContagion: 0.7, // High - influenced by UI emotional tone
             fearOfMissingOut: 0.3, // Low - not driven by urgency
             socialProofSensitivity: 0.6, // Medium - influenced by social cues
-            mentalModelRigidity: 0.1, // Extremely rigid - needs consistency
-            siteFamiliarity: 0.15, // Very low - limited retention between visits
+            mentalModelFlexibility: 0.1, // Extremely rigid - needs consistency
         },
     },
     "aphasia-receptive": {
@@ -2186,8 +2717,7 @@ export const ACCESSIBILITY_PERSONAS = {
             emotionalContagion: 0.5,
             fearOfMissingOut: 0.3,
             socialProofSensitivity: 0.4,
-            mentalModelRigidity: 0.4, // Medium - adapts visually, not linguistically
-            siteFamiliarity: 0.25, // Low - text-based nav makes sites feel unfamiliar each visit
+            mentalModelFlexibility: 0.4, // Medium - adapts visually, not linguistically
         },
     },
     "dyscalculia": {
@@ -2249,32 +2779,54 @@ export const ACCESSIBILITY_PERSONAS = {
             fatigueSusceptibility: 0.4, // Moderate - number-heavy pages are tiring
         },
         cognitiveTraits: {
-            workingMemory: 0.5, // Normal verbal, impaired numerical
+            // Working memory underpins number processing, and math anxiety reduces WM
+            // capacity further -- the two compound (PMC8946289, numerical activities
+            // of daily living).
+            workingMemory: 0.35,
             patience: 0.5, // Medium - frustrated by numbers, fine otherwise
             persistence: 0.6, // Medium-high - persistent on non-numerical tasks
             curiosity: 0.6, // Normal - explores freely until numbers appear
             riskTolerance: 0.4, // Low-medium - cautious with purchases/quantities
             readingTendency: 0.65, // Medium-high - reads text normally, skips number-heavy sections
             comprehension: 0.65, // Normal for text, low for numerical content
-            resilience: 0.5, // Medium - used to numerical difficulty
-            selfEfficacy: 0.5, // Medium - confident except with numbers
-            satisficing: 0.6, // Medium - avoids comparing numerical options
-            trustCalibration: 0.5, // Medium - can evaluate text-based trust, not numerical
+            // Math anxiety is documented as intense and self-reinforcing: errors raise
+            // anxiety, which lowers WM, which produces more errors.
+            resilience: 0.35,
+            // Years of number-task failure produce intense anxiety around bills,
+            // budgeting and any quantity judgement.
+            selfEfficacy: 0.3,
+            // High = takes the first adequate option. Comparison shopping means
+            // comparing numbers, which is the expensive operation.
+            satisficing: 0.75,
+            // High = trusting. A displayed total cannot be sanity-checked mentally, so
+            // it has to be taken on faith.
+            trustCalibration: 0.65,
             interruptRecovery: 0.6, // Normal
             informationForaging: 0.55, // Medium - normal except for numerical scent
             changeBlindness: 0.5, // Normal
-            anchoringBias: 0.6, // Medium - numerical anchoring is especially strong
-            timeHorizon: 0.5, // Medium
-            attributionStyle: 0.5, // Medium - knows it's a specific difficulty
-            metacognitivePlanning: 0.6, // Normal - good planning for non-numerical tasks
-            proceduralFluency: 0.55, // Medium - struggles when steps involve numbers
+            // High = the first number dominates. Unable to evaluate quantities
+            // independently, the first price or figure seen becomes the reference.
+            anchoringBias: 0.75,
+            // High = present-focused. Difficulty judging time and distance is a core
+            // symptom, so future consequences are hard to weigh against an immediate
+            // task.
+            timeHorizon: 0.75,
+            // High = blames self. Numeric failure is culturally read as personal
+            // deficiency rather than a processing difference, so self-attribution runs
+            // high.
+            attributionStyle: 0.7,
+            // Adults describe deliberate workarounds: calculators, written steps,
+            // avoiding mental arithmetic. That is planned strategy, not reflex.
+            metacognitivePlanning: 0.65,
+            // Multi-step numeric flows -- quantities, totals, percentages, dates --
+            // are exactly where slow and error-prone processing bites.
+            proceduralFluency: 0.4,
             transferLearning: 0.6, // Normal for non-numerical patterns
             authoritySensitivity: 0.5,
             emotionalContagion: 0.5,
             fearOfMissingOut: 0.5,
             socialProofSensitivity: 0.5,
-            mentalModelRigidity: 0.6, // Medium - adapts well to non-numerical patterns
-            siteFamiliarity: 0.5, // Medium - normal retention
+            mentalModelFlexibility: 0.6, // Medium - adapts well to non-numerical patterns
         },
     },
 };
@@ -2350,33 +2902,32 @@ export const EMOTIONAL_PERSONAS = {
             },
         },
         cognitiveTraits: {
-            patience: 0.3, // Low - anxiety makes waiting feel worse
-            riskTolerance: 0.15, // Very low - avoids uncertain outcomes
+            patience: 0.8, // Low - anxiety makes waiting feel worse
+            riskTolerance: 0.2, // Very low - avoids uncertain outcomes
             comprehension: 0.6, // Medium - understands but worries
-            persistence: 0.4, // Low-medium - gives up when anxious
+            persistence: 0.6, // Low-medium - gives up when anxious
             curiosity: 0.3, // Low - prefers safe known paths
             workingMemory: 0.5, // Medium - anxiety can impair WM
-            readingTendency: 0.8, // High - reads everything for reassurance
-            resilience: 0.2, // Very low - slow emotional recovery
-            selfEfficacy: 0.25, // Very low - doubts own abilities
-            satisficing: 0.4, // Low-medium - wants certainty
-            trustCalibration: 0.3, // Low - skeptical and worried
-            interruptRecovery: 0.3, // Low - anxiety compounds on interruption
+            readingTendency: 0.9, // High - reads everything for reassurance
+            resilience: 0.3, // Very low - slow emotional recovery
+            selfEfficacy: 0.3, // Very low - doubts own abilities
+            satisficing: 0.3, // Low-medium - wants certainty
+            trustCalibration: 0.2, // Low - skeptical and worried
+            interruptRecovery: 0.4, // Low - anxiety compounds on interruption
             // New traits (v15.0.0)
             informationForaging: 0.4, // Low - overthinks each path
             changeBlindness: 0.3, // Low - hypervigilant, notices changes
             anchoringBias: 0.7, // High - sticks to first safe option
             timeHorizon: 0.3, // Long-term - worries about consequences
-            attributionStyle: 0.8, // Very high - blames self for everything
+            attributionStyle: 0.7, // Very high - blames self for everything
             metacognitivePlanning: 0.6, // Medium-high - plans to avoid anxiety
             proceduralFluency: 0.5, // Medium - follows carefully
-            transferLearning: 0.5, // Medium
+            transferLearning: 0.4, // Medium
             authoritySensitivity: 0.7, // High - seeks reassurance from authority
-            emotionalContagion: 0.8, // Very high - absorbs UI emotional tone
+            emotionalContagion: 0.7, // Very high - absorbs UI emotional tone
             fearOfMissingOut: 0.6, // Medium-high - anxiety about missing out
-            socialProofSensitivity: 0.7, // High - seeks validation from others
-            mentalModelRigidity: 0.4, // Low-medium - struggles with unexpected
-            siteFamiliarity: 0.5, // Medium
+            socialProofSensitivity: 0.8, // High - seeks validation from others
+            mentalModelFlexibility: 0.4, // Low-medium - struggles with unexpected
         },
         context: {
             viewport: [1280, 800],
@@ -2428,33 +2979,32 @@ export const EMOTIONAL_PERSONAS = {
             },
         },
         cognitiveTraits: {
-            patience: 0.5, // Medium - not impatient but efficient
+            patience: 0.3, // Medium - not impatient but efficient
             riskTolerance: 0.8, // High - willing to try things
             comprehension: 0.7, // High - understands quickly
-            persistence: 0.8, // High - doesn't give up easily
+            persistence: 0.7, // High - doesn't give up easily
             curiosity: 0.6, // Medium-high - willing to explore
             workingMemory: 0.7, // High - focused and clear
-            readingTendency: 0.3, // Low - scans efficiently
-            resilience: 0.9, // Very high - bounces back immediately
+            readingTendency: 0.2, // Low - scans efficiently
+            resilience: 0.8, // Very high - bounces back immediately
             selfEfficacy: 0.9, // Very high - trusts own abilities
-            satisficing: 0.5, // Medium - balances speed and quality
+            satisficing: 0.7, // Medium - balances speed and quality
             trustCalibration: 0.7, // High - appropriate trust
-            interruptRecovery: 0.8, // High - resumes easily
+            interruptRecovery: 0.7, // High - resumes easily
             // New traits (v15.0.0)
             informationForaging: 0.8, // High - efficient scent-following
-            changeBlindness: 0.3, // Low - notices changes
+            changeBlindness: 0.6, // Low - notices changes
             anchoringBias: 0.3, // Low - evaluates options freely
             timeHorizon: 0.4, // Medium-long - thinks ahead
             attributionStyle: 0.2, // Low - blames system issues
-            metacognitivePlanning: 0.7, // High - plans efficiently
-            proceduralFluency: 0.8, // High - follows procedures well
-            transferLearning: 0.85, // High - applies knowledge
-            authoritySensitivity: 0.4, // Low-medium - independent judgment
-            emotionalContagion: 0.2, // Low - emotionally stable
+            metacognitivePlanning: 0.4, // High - plans efficiently
+            proceduralFluency: 0.7, // High - follows procedures well
+            transferLearning: 0.8, // High - applies knowledge
+            authoritySensitivity: 0.3, // Low-medium - independent judgment
+            emotionalContagion: 0.3, // Low - emotionally stable
             fearOfMissingOut: 0.3, // Low - not swayed by urgency
-            socialProofSensitivity: 0.4, // Low-medium - evaluates on merits
-            mentalModelRigidity: 0.8, // High - adapts easily
-            siteFamiliarity: 0.5, // Medium
+            socialProofSensitivity: 0.3, // Low-medium - evaluates on merits
+            mentalModelFlexibility: 0.8, // High - adapts easily
         },
         context: {
             viewport: [1920, 1080],
@@ -2511,28 +3061,27 @@ export const EMOTIONAL_PERSONAS = {
             comprehension: 0.5, // Medium
             persistence: 0.4, // Low-medium - gives up when frustrated
             curiosity: 0.6, // Medium-high - excitement drives exploration
-            workingMemory: 0.45, // Medium - emotions can interfere
-            readingTendency: 0.4, // Medium
-            resilience: 0.15, // Very low - emotions linger long
+            workingMemory: 0.5, // Medium - emotions can interfere
+            readingTendency: 0.6, // Medium
+            resilience: 0.4, // Very low - emotions linger long
             selfEfficacy: 0.5, // Medium - mood dependent
             satisficing: 0.6, // Medium-high - wants quick resolution
             trustCalibration: 0.5, // Medium - emotional judgment
-            interruptRecovery: 0.35, // Low - emotional state persists
+            interruptRecovery: 0.4, // Low - emotional state persists
             // New traits (v15.0.0)
             informationForaging: 0.5, // Medium - mood dependent
             changeBlindness: 0.5, // Medium - emotions distract
-            anchoringBias: 0.6, // Medium-high - first impressions stick
+            anchoringBias: 0.7, // Medium-high - first impressions stick
             timeHorizon: 0.7, // Short-term - emotional decisions
             attributionStyle: 0.5, // Medium - variable based on mood
-            metacognitivePlanning: 0.3, // Low - emotional impulsivity
-            proceduralFluency: 0.45, // Medium - emotions interfere
+            metacognitivePlanning: 0.4, // Low - emotional impulsivity
+            proceduralFluency: 0.5, // Medium - emotions interfere
             transferLearning: 0.5, // Medium
             authoritySensitivity: 0.5, // Medium - mood dependent
             emotionalContagion: 0.9, // Maximum - highly influenced by UI tone
-            fearOfMissingOut: 0.75, // High - emotional FOMO
-            socialProofSensitivity: 0.7, // High - seeks emotional validation
-            mentalModelRigidity: 0.5, // Medium
-            siteFamiliarity: 0.5, // Medium
+            fearOfMissingOut: 0.7, // High - emotional FOMO
+            socialProofSensitivity: 0.8, // High - seeks emotional validation
+            mentalModelFlexibility: 0.5, // Medium
         },
         context: {
             viewport: [1280, 800],
@@ -2584,33 +3133,32 @@ export const EMOTIONAL_PERSONAS = {
             },
         },
         cognitiveTraits: {
-            patience: 0.9, // Very high - doesn't rush
+            patience: 0.7, // Very high - doesn't rush
             riskTolerance: 0.5, // Medium - balanced judgment
             comprehension: 0.7, // High - clear thinking
-            persistence: 0.9, // Very high - never gives up
+            persistence: 0.7, // Very high - never gives up
             curiosity: 0.4, // Medium - methodical exploration
-            workingMemory: 0.8, // High - unaffected by emotion
-            readingTendency: 0.6, // Medium-high - thorough
-            resilience: 0.95, // Maximum - emotions barely register
-            selfEfficacy: 0.7, // High - quiet confidence
+            workingMemory: 0.7, // High - unaffected by emotion
+            readingTendency: 0.7, // Medium-high - thorough
+            resilience: 0.8, // Maximum - emotions barely register
+            selfEfficacy: 0.6, // High - quiet confidence
             satisficing: 0.4, // Low - willing to be thorough
-            trustCalibration: 0.6, // Medium-high - rational evaluation
-            interruptRecovery: 0.8, // High - returns to task easily
+            trustCalibration: 0.7, // Medium-high - rational evaluation
+            interruptRecovery: 0.7, // High - returns to task easily
             // New traits (v15.0.0)
-            informationForaging: 0.6, // Medium-high - methodical
-            changeBlindness: 0.3, // Low - attentive, focused
+            informationForaging: 0.8, // Medium-high - methodical
+            changeBlindness: 0.4, // Low - attentive, focused
             anchoringBias: 0.3, // Low - evaluates all options
             timeHorizon: 0.2, // Very long-term - patient deliberation
             attributionStyle: 0.4, // Medium - balanced evaluation
-            metacognitivePlanning: 0.8, // High - plans carefully
-            proceduralFluency: 0.8, // High - follows procedures precisely
+            metacognitivePlanning: 0.7, // High - plans carefully
+            proceduralFluency: 0.6, // High - follows procedures precisely
             transferLearning: 0.7, // High - applies experience
             authoritySensitivity: 0.4, // Low-medium - independent judgment
-            emotionalContagion: 0.1, // Minimum - emotionally unaffected
-            fearOfMissingOut: 0.1, // Minimum - not driven by urgency
-            socialProofSensitivity: 0.3, // Low - evaluates independently
-            mentalModelRigidity: 0.7, // High - adapts methodically
-            siteFamiliarity: 0.5, // Medium
+            emotionalContagion: 0.2, // Minimum - emotionally unaffected
+            fearOfMissingOut: 0.2, // Minimum - not driven by urgency
+            socialProofSensitivity: 0.4, // Low - evaluates independently
+            mentalModelFlexibility: 0.7, // High - adapts methodically
         },
         context: {
             viewport: [1280, 800],
@@ -2646,6 +3194,30 @@ export function listEmotionalPersonas() {
  * @param name Persona name to look up
  * @returns Persona object or undefined if not found
  */
+/**
+ * Resolve a persona name tolerantly: case and punctuation do not have to match.
+ *
+ * Resolution was exact-match only, so "POWER-USER" and "power user" both missed
+ * a persona that plainly exists. Before unknown names started throwing, a miss
+ * silently fabricated an all-midpoint persona named after whatever string was
+ * passed -- so a caller who capitalised a name got clean, plausible, invented
+ * data. Now it throws, which is better but still refuses a request anyone would
+ * call valid.
+ *
+ * Exact match is tried FIRST and always wins, so this can only ever add
+ * resolutions, never redirect one that already worked. (2026-08-02)
+ */
+function resolveLoosely(name) {
+    const norm = (v) => v.toLowerCase().replace(/[^a-z0-9]/g, "");
+    const target = norm(name);
+    if (!target)
+        return undefined;
+    for (const key of listAllPersonas()) {
+        if (norm(key) === target)
+            return key;
+    }
+    return undefined;
+}
 export function getAnyPersona(name) {
     // 1. Check runtime-registered personas first (Enterprise extensions)
     if (RUNTIME_PERSONAS[name]) {
@@ -2672,6 +3244,11 @@ export function getAnyPersona(name) {
     if (AGENT_PERSONAS[name]) {
         return AGENT_PERSONAS[name];
     }
+    // Exact match failed. Try again ignoring case and punctuation -- only after
+    // every exact lookup above has missed, so this can never shadow a real key.
+    const loose = resolveLoosely(name);
+    if (loose && loose !== name)
+        return getAnyPersona(loose);
     return undefined;
 }
 /**
@@ -2679,6 +3256,90 @@ export function getAnyPersona(name) {
  *
  * @returns Combined list of all available persona names
  */
+/**
+ * Raised when a persona name resolves to nothing.
+ *
+ * A distinct class so callers can tell "you asked for someone who does not
+ * exist" apart from "the page would not load", and so the message can carry
+ * suggestions instead of a bare failure.
+ */
+export class UnknownPersonaError extends Error {
+    requested;
+    suggestions;
+    code = "unknown_persona";
+    constructor(requested, suggestions) {
+        super(`Unknown persona "${requested}".` +
+            (suggestions.length ? ` Did you mean: ${suggestions.join(", ")}?` : "") +
+            ` Run list_cognitive_personas for the full roster. Nothing was measured — a persona that does not exist cannot be simulated.`);
+        this.requested = requested;
+        this.suggestions = suggestions;
+        this.name = "UnknownPersonaError";
+    }
+}
+/** Closest roster names to a miss, for the error message. */
+export function suggestPersonaNames(requested, limit = 3) {
+    const roster = listAllPersonas();
+    const q = requested.toLowerCase().replace(/[^a-z0-9]/g, "");
+    if (!q)
+        return roster.slice(0, limit);
+    const scored = roster.map((name) => {
+        const n = name.toLowerCase().replace(/[^a-z0-9]/g, "");
+        // Exact-after-normalising ranks first: it catches the case and punctuation
+        // misses ("POWER-USER", "power user") that are the commonest way to miss.
+        if (n === q)
+            return { name, score: 0 };
+        if (n.startsWith(q) || q.startsWith(n))
+            return { name, score: 1 };
+        if (n.includes(q) || q.includes(n))
+            return { name, score: 2 };
+        // Cheap edit distance, bounded — the roster is small enough that an exact
+        // Levenshtein is affordable and a typo is the other common miss.
+        const d = (a, b) => {
+            const prev = Array.from({ length: b.length + 1 }, (_, i) => i);
+            for (let i = 1; i <= a.length; i++) {
+                let last = prev[0];
+                prev[0] = i;
+                for (let j = 1; j <= b.length; j++) {
+                    const tmp = prev[j];
+                    prev[j] = Math.min(prev[j] + 1, prev[j - 1] + 1, last + (a[i - 1] === b[j - 1] ? 0 : 1));
+                    last = tmp;
+                }
+            }
+            return prev[b.length];
+        };
+        return { name, score: 3 + d(q, n) };
+    });
+    return scored.sort((a, b) => a.score - b.score).filter((x) => x.score < 3 + Math.max(4, q.length * 0.5))
+        .slice(0, limit).map((x) => x.name);
+}
+/**
+ * Resolve a persona name, or refuse.
+ *
+ * Every persona-accepting tool used to fall back to
+ * `createCognitivePersona(name, name, {})` on a miss, which builds a complete
+ * persona with every trait at its 0.5 default and names it after whatever
+ * string was passed. The result was a full six-layer breakdown, an abandonment
+ * risk, a rendered motor overlay and an interpretation sentence quoting the
+ * typo as if it were a subject — in range, plausible, and indistinguishable
+ * from a real measurement at analysis time.
+ *
+ * That is silent data corruption: it does not fail, it lies quietly, and a
+ * typo inside a batch run enters a results table having passed eyeballing and
+ * range checks. So a miss now throws.
+ *
+ * `syntheticTraits` is the one legitimate synthesis: a caller defining an
+ * ad-hoc persona by supplying its traits outright. Supplying NO traits and no
+ * known name is not defining a persona, it is a typo. (2026-08-02)
+ */
+export function resolvePersonaOrThrow(name, syntheticTraits) {
+    const found = getAnyPersona(name);
+    if (found)
+        return found;
+    if (syntheticTraits && Object.keys(syntheticTraits).length > 0) {
+        return createCognitivePersona(name, name, syntheticTraits);
+    }
+    throw new UnknownPersonaError(name, suggestPersonaNames(name));
+}
 export function listAllPersonas() {
     const runtimeNames = Object.keys(RUNTIME_PERSONAS);
     const builtinNames = Object.keys(BUILTIN_PERSONAS);