@ak--47/dungeon-master 1.3.1 → 1.4.1

package/lib/utils/utils.js

@@ -1156,21 +1156,14 @@ CORE
 */
 
 //the function which generates $distinct_id + $anonymous_ids, $session_ids, and created, skewing towards the present
-function generateUser(user_id, opts, amplitude = 1, frequency = 1, skew = 1) {
+function generateUser(user_id, opts) {
 	const chance = getChance();
-	const { numDays, isAnonymous, hasAvatar, hasAnonIds, hasSessionIds, datasetEndUnix } = opts;
-	// Uniformly distributed `u`, then skew applied
-	let u = Math.pow(chance.random(), skew);
-
-	// Sine function for a smoother curve
-	const sineValue = (Math.sin(u * Math.PI * frequency - Math.PI / 2) * amplitude + 1) / 2;
-
-	// Scale the sineValue to the range of days
-	let daysAgoBorn = Math.round(sineValue * (numDays - 1)) + 1;
-
-	// Clamp values to ensure they are within the desired range
+	const { numDays, isAnonymous, hasAvatar, hasAnonIds, hasSessionIds, datasetEndUnix, avgDevicePerUser } = opts;
+	// Birth date placement is overridden by bornRecentBias in user-loop.js;
+	// use a simple uniform distribution as the seed value.
+	let daysAgoBorn = Math.round(chance.random() * (numDays - 1)) + 1;
 	daysAgoBorn = Math.min(daysAgoBorn, numDays);
-	const props = person(user_id, daysAgoBorn, isAnonymous, hasAvatar, hasAnonIds, hasSessionIds, datasetEndUnix);
+	const props = person(user_id, daysAgoBorn, isAnonymous, hasAvatar, hasAnonIds, hasSessionIds, datasetEndUnix, avgDevicePerUser);
 
 	const user = {
 		distinct_id: user_id,
@@ -1275,9 +1268,13 @@ function TimeSoup(earliestTime, latestTime, peaks = 5, deviation = 2, mean = 0,
  * @param {boolean} hasAvatar
  * @param {boolean} hasAnonIds
  * @param {boolean} hasSessionIds
+ * @param {number} [datasetEndUnix]
+ * @param {number} [avgDevicePerUser] - Whole number ≥ 0. When ≥ 1 (and `hasAnonIds: true`),
+ *   sets the per-user device pool size. `1` = exactly one device. `>1` = drawn from
+ *   `chance.normal({mean: avgDevicePerUser, dev: avgDevicePerUser/2})`, clamped ≥ 1, integer.
  * @return {Person}
  */
-function person(userId, bornDaysAgo = 30, isAnonymous = false, hasAvatar = false, hasAnonIds = false, hasSessionIds = false, datasetEndUnix) {
+function person(userId, bornDaysAgo = 30, isAnonymous = false, hasAvatar = false, hasAnonIds = false, hasSessionIds = false, datasetEndUnix, avgDevicePerUser) {
 	const chance = getChance();
 	//names and photos
 	const l = chance.letter.bind(chance);
@@ -1320,9 +1317,25 @@ function person(userId, bornDaysAgo = 30, isAnonymous = false, hasAvatar = false
 
 	if (!hasAvatar) delete user.avatar;
 
-	//anon Ids
+	// Device pool ("anonymousIds" — name preserved for backwards compat).
+	// Phase 2 identity model: pool size is governed by `avgDevicePerUser` when set,
+	// else legacy 2–10 random pool size for any dungeon that just sets `hasAnonIds: true`
+	// without `avgDevicePerUser`. (Validator coerces `hasAnonIds: true` to an effective
+	// `avgDevicePerUser >= 1`, so this branch fires whenever there's a device pool to build.)
 	if (hasAnonIds) {
-		const clusterSize = integer(2, 10);
+		let clusterSize;
+		if (typeof avgDevicePerUser === 'number' && avgDevicePerUser >= 1) {
+			if (avgDevicePerUser === 1) {
+				clusterSize = 1;
+			} else {
+				const sd = avgDevicePerUser / 2;
+				const sample = chance.normal({ mean: avgDevicePerUser, dev: sd });
+				clusterSize = Math.max(1, Math.round(sample));
+			}
+		} else {
+			// Legacy fallback (pre-1.4 behavior): random pool of 2–10 devices.
+			clusterSize = integer(2, 10);
+		}
 		for (let i = 0; i < clusterSize; i++) {
 			// Use seeded chance, not ak-tools uid() (which uses Math.random).
 			const anonId = chance.string({ length: 42, pool: 'abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ0123456789' });
@@ -1584,6 +1597,15 @@ function assignSessionIds(events, timeoutMinutes = 30) {
 	return events;
 }
 
+async function deleteFile(filePath) {
+	if (filePath.startsWith('gs://')) {
+		const { bucket, file } = parseGCSUri(filePath);
+		await storage.bucket(bucket).file(file).delete({ ignoreNotFound: true });
+	} else {
+		await fs.promises.unlink(filePath).catch(() => {});
+	}
+}
+
 export {
 	pick,
 	date,
@@ -1638,4 +1660,5 @@ export {
 	assignSessionIds,
 	bunchIntoSessions,
 	setDatasetNow,
+	deleteFile,
 };

package/lib/verify/emulate-breakdown.js

@@ -0,0 +1,281 @@
+/**
+ * Mixpanel breakdown emulator.
+ *
+ * Best-effort approximation of the table shapes Mixpanel produces for the five
+ * analyses the Phase 4 hook patterns target. Used by `verify-hooks` to assert
+ * that engineered patterns actually produce the expected ratios in the data, AND
+ * by consumers who want to validate dungeons against expected business shapes
+ * outside of Mixpanel.
+ *
+ * Reference: Mixpanel Insights / Funnels / Flows reports, as of 2026-05.
+ *
+ * Caveats:
+ * - Mixpanel applies its own per-account UTC offset and time-bucketing rules. This
+ *   emulator uses raw event times unless the breakdown explicitly involves a window.
+ * - Mixpanel "users" are typically distinct profiles with at least one event in
+ *   the date range; this emulator counts unique `user_id` (falling back to
+ *   `distinct_id`) found across the events array.
+ * - This is not bit-exact — it's the *shape* check Phase 4 verification needs.
+ */
+
+/**
+ * @typedef {Object} EmulateOptions
+ * @property {'frequencyByFrequency'|'funnelFrequency'|'aggregatePerUser'|'timeToConvert'|'attributedBy'} type
+ *
+ * @property {string} [metricEvent]
+ * @property {string} [breakdownByFrequencyOf]
+ * @property {boolean} [perUser]
+ *
+ * @property {string[]} [steps]
+ *
+ * @property {string} [event]
+ * @property {string} [property]
+ * @property {'avg'|'sum'|'count'|'max'|'min'} [agg]
+ *
+ * @property {string} [fromEvent]
+ * @property {string} [toEvent]
+ * @property {string} [breakdownByUserProperty]
+ * @property {Array<Object>} [profiles]
+ *
+ * @property {string} [conversionEvent]
+ * @property {string} [attributionEvent]
+ * @property {string} [attributionProperty]
+ * @property {'firstTouch'|'lastTouch'} [model]
+ */
+
+/**
+ * Run a Mixpanel breakdown emulation against an events array.
+ * Routes to the type-specific implementation based on `config.type`.
+ *
+ * @param {Array<Object>} events
+ * @param {EmulateOptions} config
+ * @returns {Array<Object>} Breakdown table rows.
+ */
+export function emulateBreakdown(events, config) {
+	if (!Array.isArray(events)) throw new Error('emulateBreakdown: events must be an array');
+	if (!config || !config.type) throw new Error('emulateBreakdown: config.type is required');
+	switch (config.type) {
+		case 'frequencyByFrequency': return frequencyByFrequency(events, /** @type {*} */ (config));
+		case 'funnelFrequency':      return funnelFrequency(events, /** @type {*} */ (config));
+		case 'aggregatePerUser':     return aggregatePerUser(events, /** @type {*} */ (config));
+		case 'timeToConvert':        return timeToConvert(events, /** @type {*} */ (config));
+		case 'attributedBy':         return attributedBy(events, /** @type {*} */ (config));
+		default: throw new Error(`emulateBreakdown: unknown type "${config.type}"`);
+	}
+}
+
+// ── Frequency × Frequency (Insights, Frequency Distribution by per-user count of B) ──
+
+function frequencyByFrequency(events, { metricEvent, breakdownByFrequencyOf }) {
+	if (!metricEvent || !breakdownByFrequencyOf) {
+		throw new Error('frequencyByFrequency requires metricEvent and breakdownByFrequencyOf');
+	}
+	const userMetric = new Map();
+	const userBreakdown = new Map();
+	const uids = new Set();
+	for (const ev of events) {
+		const uid = userIdOf(ev);
+		if (!uid) continue;
+		uids.add(uid);
+		if (ev.event === metricEvent) userMetric.set(uid, (userMetric.get(uid) || 0) + 1);
+		if (ev.event === breakdownByFrequencyOf) userBreakdown.set(uid, (userBreakdown.get(uid) || 0) + 1);
+	}
+	const cell = new Map(); // `${m}|${b}` → user_count
+	for (const uid of uids) {
+		const m = userMetric.get(uid) || 0;
+		const b = userBreakdown.get(uid) || 0;
+		const key = `${m}|${b}`;
+		cell.set(key, (cell.get(key) || 0) + 1);
+	}
+	return [...cell.entries()].map(([k, count]) => {
+		const [m, b] = k.split('|').map(Number);
+		return { metric_freq: m, breakdown_freq: b, user_count: count };
+	}).sort((x, y) => x.breakdown_freq - y.breakdown_freq || x.metric_freq - y.metric_freq);
+}
+
+// ── Funnel Frequency Breakdown (Funnel report broken down by per-user count of X) ──
+
+function funnelFrequency(events, { steps, breakdownByFrequencyOf }) {
+	if (!Array.isArray(steps) || !steps.length) throw new Error('funnelFrequency requires steps[]');
+	if (!breakdownByFrequencyOf) throw new Error('funnelFrequency requires breakdownByFrequencyOf');
+	const userEvents = groupByUser(events);
+	const userBreakdown = new Map();
+	for (const [uid, evs] of userEvents) {
+		const c = evs.filter(e => e && e.event === breakdownByFrequencyOf).length;
+		userBreakdown.set(uid, c);
+	}
+	const result = [];
+	for (let s = 0; s < steps.length; s++) {
+		const stepName = steps[s];
+		const conversions = new Map(); // breakdown_freq → count
+		for (const [uid, evs] of userEvents) {
+			const sorted = sortByTime(evs);
+			let stepIdx = 0;
+			for (const ev of sorted) {
+				if (ev.event === steps[stepIdx]) stepIdx++;
+				if (stepIdx > s) break;
+			}
+			if (stepIdx > s) {
+				const b = userBreakdown.get(uid) || 0;
+				conversions.set(b, (conversions.get(b) || 0) + 1);
+			}
+		}
+		for (const [b, c] of conversions) {
+			result.push({ step: stepName, step_index: s, breakdown_freq: b, conversions: c, conversion_pct: 0 });
+		}
+	}
+	// Conversion % at each step relative to its own breakdown_freq's step-0 baseline.
+	const baseline = new Map();
+	for (const r of result) {
+		if (r.step_index === 0) baseline.set(r.breakdown_freq, r.conversions);
+	}
+	for (const r of result) {
+		const denom = baseline.get(r.breakdown_freq) || 0;
+		r.conversion_pct = denom ? (r.conversions / denom) * 100 : 0;
+	}
+	return result.sort((a, b) => a.step_index - b.step_index || a.breakdown_freq - b.breakdown_freq);
+}
+
+// ── Aggregate per user (Insights, sum/avg of property X by per-user count of B) ──
+
+function aggregatePerUser(events, { event, property, agg = 'avg', breakdownByFrequencyOf }) {
+	if (!event) throw new Error('aggregatePerUser requires event');
+	if (!breakdownByFrequencyOf) throw new Error('aggregatePerUser requires breakdownByFrequencyOf');
+	if (agg !== 'count' && !property) throw new Error('aggregatePerUser requires property unless agg is "count"');
+	const userVals = new Map();
+	const userBreakdown = new Map();
+	for (const ev of events) {
+		const uid = userIdOf(ev);
+		if (!uid) continue;
+		if (ev.event === event) {
+			// `agg: 'count'` → count occurrences of the event regardless of property type.
+			// All other aggs only consider numeric property values.
+			if (agg === 'count') {
+				if (!userVals.has(uid)) userVals.set(uid, []);
+				userVals.get(uid).push(1);
+			} else if (property && typeof ev[property] === 'number') {
+				if (!userVals.has(uid)) userVals.set(uid, []);
+				userVals.get(uid).push(ev[property]);
+			}
+		}
+		if (ev.event === breakdownByFrequencyOf) userBreakdown.set(uid, (userBreakdown.get(uid) || 0) + 1);
+	}
+	const userAgg = new Map();
+	for (const [uid, vals] of userVals) userAgg.set(uid, applyAgg(vals, agg));
+	const buckets = new Map(); // breakdown_freq → [aggregates]
+	for (const [uid, v] of userAgg) {
+		const b = userBreakdown.get(uid) || 0;
+		if (!buckets.has(b)) buckets.set(b, []);
+		buckets.get(b).push(v);
+	}
+	return [...buckets.entries()].map(([b, vs]) => ({
+		breakdown_freq: b,
+		user_count: vs.length,
+		avg_aggregate: vs.reduce((a, x) => a + x, 0) / vs.length,
+	})).sort((x, y) => x.breakdown_freq - y.breakdown_freq);
+}
+
+// ── Time to Convert (Funnel TTC, broken down by user property) ──
+
+function timeToConvert(events, { fromEvent, toEvent, breakdownByUserProperty, profiles = [] }) {
+	if (!fromEvent || !toEvent) throw new Error('timeToConvert requires fromEvent and toEvent');
+	const userEvents = groupByUser(events);
+	const profileByUid = new Map();
+	for (const p of profiles) {
+		if (!p) continue;
+		const uid = p.distinct_id || p.user_id;
+		if (uid) profileByUid.set(uid, p);
+	}
+	const buckets = new Map(); // segValue → [ttcMs]
+	for (const [uid, evs] of userEvents) {
+		const sorted = sortByTime(evs);
+		const a = sorted.find(e => e && e.event === fromEvent);
+		if (!a) continue;
+		const aIdx = sorted.indexOf(a);
+		const b = sorted.slice(aIdx + 1).find(e => e && e.event === toEvent);
+		if (!b) continue;
+		const ttcMs = toMs(b.time) - toMs(a.time);
+		if (!Number.isFinite(ttcMs) || ttcMs < 0) continue;
+		const profile = profileByUid.get(uid);
+		const segValue = breakdownByUserProperty
+			? (profile ? (profile[breakdownByUserProperty] ?? 'unknown') : 'unknown')
+			: 'all';
+		if (!buckets.has(segValue)) buckets.set(segValue, []);
+		buckets.get(segValue).push(ttcMs);
+	}
+	return [...buckets.entries()].map(([seg, ttcs]) => ({
+		segment_value: seg,
+		user_count: ttcs.length,
+		avg_ttc_ms: ttcs.reduce((a, x) => a + x, 0) / ttcs.length,
+		median_ttc_ms: median(ttcs),
+	})).sort((x, y) => String(x.segment_value).localeCompare(String(y.segment_value)));
+}
+
+// ── Attributed By (first-/last-touch attribution by event property value) ──
+
+function attributedBy(events, { conversionEvent, attributionEvent, attributionProperty, model = 'firstTouch' }) {
+	if (!conversionEvent || !attributionEvent || !attributionProperty) {
+		throw new Error('attributedBy requires conversionEvent, attributionEvent, attributionProperty');
+	}
+	const userEvents = groupByUser(events);
+	const counts = new Map();
+	for (const [uid, evs] of userEvents) {
+		const sorted = sortByTime(evs);
+		const conversion = sorted.find(e => e && e.event === conversionEvent);
+		if (!conversion) continue;
+		const conversionTime = toMs(conversion.time);
+		const touches = sorted.filter(e =>
+			e && e.event === attributionEvent && toMs(e.time) <= conversionTime
+		);
+		if (!touches.length) continue;
+		const touch = model === 'lastTouch' ? touches[touches.length - 1] : touches[0];
+		const v = touch[attributionProperty] ?? 'unknown';
+		counts.set(v, (counts.get(v) || 0) + 1);
+	}
+	return [...counts.entries()].map(([source, count]) => ({
+		attribution_value: source,
+		conversions: count,
+	})).sort((a, b) => b.conversions - a.conversions);
+}
+
+// ── shared helpers ──
+
+function userIdOf(ev) {
+	return ev && (ev.user_id || ev.distinct_id || ev.device_id);
+}
+
+function groupByUser(events) {
+	const userEvents = new Map();
+	for (const ev of events) {
+		const uid = userIdOf(ev);
+		if (!uid) continue;
+		if (!userEvents.has(uid)) userEvents.set(uid, []);
+		userEvents.get(uid).push(ev);
+	}
+	return userEvents;
+}
+
+function sortByTime(evs) {
+	return evs.slice().sort((a, b) => toMs(a && a.time) - toMs(b && b.time));
+}
+
+import { toMs } from '../hook-helpers/_internal.js';
+
+function applyAgg(vals, agg) {
+	if (!vals || !vals.length) return 0;
+	switch (agg) {
+		case 'sum': return vals.reduce((a, b) => a + b, 0);
+		case 'count': return vals.length;
+		case 'max': return Math.max(...vals);
+		case 'min': return Math.min(...vals);
+		case 'avg':
+		default: return vals.reduce((a, b) => a + b, 0) / vals.length;
+	}
+}
+
+function median(arr) {
+	if (!arr.length) return 0;
+	const sorted = arr.slice().sort((a, b) => a - b);
+	const mid = Math.floor(sorted.length / 2);
+	return sorted.length % 2 ? sorted[mid] : (sorted[mid - 1] + sorted[mid]) / 2;
+}

package/lib/verify/index.js

@@ -0,0 +1,12 @@
+/**
+ * @ak--47/dungeon-master/verify — Phase 4 verification surface.
+ *
+ * `emulateBreakdown` produces the table shapes Mixpanel shows for the five
+ * supported analyses (frequencyByFrequency, funnelFrequency, aggregatePerUser,
+ * timeToConvert, attributedBy). `verifyDungeon` is the higher-level wrapper
+ * that runs a dungeon and asserts emulator outputs match expected ratios — wire
+ * this into your CI to catch dungeon drift.
+ */
+
+export { emulateBreakdown } from './emulate-breakdown.js';
+export { verifyDungeon } from './verify-dungeon.js';

package/lib/verify/verify-dungeon.js

@@ -0,0 +1,61 @@
+/**
+ * verifyDungeon — run a dungeon (in-memory) and run a series of emulator checks
+ * against its output, returning a structured report. Designed for CI use:
+ *
+ *   const report = await verifyDungeon(dungeonConfig, [
+ *     {
+ *       name: 'engaged users do 2x purchases',
+ *       breakdown: { type: 'frequencyByFrequency',
+ *                    metricEvent: 'Purchase', breakdownByFrequencyOf: 'Browse' },
+ *       assert: (rows) => {
+ *         // custom assertion against the emulator's output table
+ *         return { pass: true, detail: 'looks good' };
+ *       }
+ *     }
+ *   ]);
+ *   if (!report.pass) process.exit(1);
+ *
+ * The dungeon is run via the same default-export entry point external consumers
+ * use, so this is a true end-to-end check.
+ */
+
+import DUNGEON_MASTER from '../../index.js';
+import { emulateBreakdown } from './emulate-breakdown.js';
+
+/**
+ * @typedef {Object} VerifyCheck
+ * @property {string} name - Human-readable name for the check.
+ * @property {Object} breakdown - Argument passed to `emulateBreakdown`.
+ * @property {(rows: Array<Object>, ctx: { events: Array<Object>, profiles: Array<Object> }) => { pass: boolean, detail?: string }} assert
+ */
+
+/**
+ * @param {Object} config - Dungeon config (or path; passed straight to DUNGEON_MASTER).
+ * @param {VerifyCheck[]} checks
+ * @returns {Promise<{ pass: boolean, results: Array<{ name: string, pass: boolean, detail?: string, rows?: Array<Object> }> }>}
+ */
+export async function verifyDungeon(config, checks) {
+	if (!checks || !checks.length) throw new Error('verifyDungeon: at least one check required');
+	let result = await DUNGEON_MASTER(config);
+	if (Array.isArray(result)) result = result[0];
+	const events = Array.isArray(result.eventData) ? result.eventData : Array.from(result.eventData);
+	const profiles = Array.isArray(result.userProfilesData) ? result.userProfilesData : Array.from(result.userProfilesData);
+	const ctx = { events, profiles };
+	const results = [];
+	for (const check of checks) {
+		try {
+			const breakdownArgs = { ...check.breakdown };
+			// timeToConvert + attributedBy may want profiles; auto-inject if not provided.
+			if (breakdownArgs.type === 'timeToConvert' && !breakdownArgs.profiles) {
+				breakdownArgs.profiles = profiles;
+			}
+			const rows = emulateBreakdown(events, breakdownArgs);
+			const verdict = check.assert(rows, ctx);
+			results.push({ name: check.name, pass: !!verdict.pass, detail: verdict.detail, rows });
+		} catch (err) {
+			results.push({ name: check.name, pass: false, detail: `error: ${err.message}` });
+		}
+	}
+	const pass = results.every(r => r.pass);
+	return { pass, results };
+}

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@ak--47/dungeon-master",
-	"version": "1.3.1",
+	"version": "1.4.1",
 	"description": "generate fancy datasets",
 	"type": "module",
 	"main": "index.js",
@@ -11,7 +11,10 @@
 			"types": "./types.d.ts"
 		},
 		"./utils": "./lib/utils/utils.js",
-		"./text": "./lib/generators/text.js"
+		"./text": "./lib/generators/text.js",
+		"./hook-helpers": "./lib/hook-helpers/index.js",
+		"./hook-patterns": "./lib/hook-patterns/index.js",
+		"./verify": "./lib/verify/index.js"
 	},
 	"files": [
 		"index.js",
@@ -21,7 +24,6 @@
 		"!dungeons/customers/",
 		"!dungeons/user/",
 		"scripts/",
-		"!scripts/experiments/",
 		"package.json",
 		"README.md",
 		"CHANGELOG.md"
@@ -72,7 +74,7 @@
 		"dotenv": "^16.4.5",
 		"hyparquet-writer": "^0.6.1",
 		"mixpanel": "^0.18.0",
-		"mixpanel-import": "^3.2.9",
+		"mixpanel-import": "^3.3.1",
 		"p-limit": "^3.1.0",
 		"pino": "^9.0.0",
 		"pino-pretty": "^11.0.0",