@ak--47/dungeon-master 1.5.4 → 1.6.0

package/dungeons/vertical/fintech/fintech.sql

@@ -0,0 +1,175 @@
+-- ============================================================
+-- fintech.js — v1.6 human-inspection queries (DuckDB)
+--
+-- Every query is keyed to a story id in fintech.js's `stories` export;
+-- the machine-checked verdicts come from:
+--   node scripts/verify-stories.mjs dungeons/vertical/fintech/fintech.js --data-prefix verify-fintech
+-- Generate first:
+--   node scripts/verify-runner.mjs dungeons/vertical/fintech/fintech.js verify-fintech
+-- Run this file:
+--   duckdb -c ".read dungeons/vertical/fintech/fintech.sql"
+-- ============================================================
+
+-- ── identity-resolution prelude ─────────────────────────────
+-- avgDevicePerUser: 2 — resolve device_id → canonical distinct_id via the
+-- USERS shards so uniques are identity-correct even for device-only rows.
+CREATE OR REPLACE VIEW users AS
+SELECT * FROM read_json_auto('data/verify-fintech-USERS*.json', sample_size=-1, union_by_name=true);
+
+CREATE OR REPLACE VIEW device_map AS
+-- profiles store the device pool under the legacy "anonymousIds" key
+-- (buildIdentityMap in lib/verify/identity.js reads the same field)
+SELECT unnest("anonymousIds") AS device_id, distinct_id FROM users;
+
+CREATE OR REPLACE VIEW ev AS
+-- ::VARCHAR casts — user_id sniffs as UUID, device_id as VARCHAR; DuckDB
+-- refuses to coalesce mixed types
+SELECT coalesce(m.distinct_id::VARCHAR, e.user_id::VARCHAR, e.device_id::VARCHAR) AS uid,
+       e.time::TIMESTAMP AS t,
+       e.*
+FROM read_json_auto('data/verify-fintech-EVENTS*.json', sample_size=-1, union_by_name=true) e
+LEFT JOIN device_map m ON e.device_id = m.device_id;
+
+
+-- ── H1-business-txn-4x ──────────────────────────────────────
+-- user mix ~ 80/20 personal/business; business avg txn amount ~ 4x personal
+SELECT account_segment, count(*) AS users FROM users GROUP BY 1 ORDER BY 1;
+
+SELECT u.account_segment, count(*) AS txns, round(avg(e.amount), 0) AS avg_amount
+FROM ev e JOIN users u ON e.uid = u.distinct_id
+WHERE e.event = 'transaction completed' AND e.amount IS NOT NULL
+GROUP BY 1 ORDER BY avg_amount DESC;
+
+
+-- ── H2-payday-amounts ───────────────────────────────────────
+-- direct deposits 3x on the 1st/15th; transfers ~1.6x on days 1-3 / 15-17
+SELECT EXTRACT(DAY FROM t) AS dom, count(*) AS deposits, round(avg(amount), 0) AS avg_amount
+FROM ev WHERE event = 'transaction completed' AND transaction_type = 'direct_deposit'
+GROUP BY 1 ORDER BY 1;
+
+SELECT CASE WHEN EXTRACT(DAY FROM t) IN (1, 2, 3, 15, 16, 17) THEN 'payday_window' ELSE 'other' END AS bucket,
+       count(*) AS transfers, round(avg(amount), 0) AS avg_amount
+FROM ev WHERE event = 'transfer sent' GROUP BY 1 ORDER BY 1;
+
+
+-- ── H3-fraud-cohort-share ───────────────────────────────────
+-- ~3% of SUPPORT-HISTORY users carry the full burst signature: 3+ rapid
+-- credit purchases, then suspicious lock + unauthorized dispute within 1h.
+-- The bare lock+dispute pair is NOT the detector (organic Support-funnel
+-- reason collisions outnumber the hook ~4:1).
+WITH locks AS (SELECT uid, epoch(t) AS ts FROM ev WHERE event = 'card locked' AND reason = 'suspicious_activity'),
+disputes AS (SELECT uid, epoch(t) AS ts FROM ev WHERE event = 'dispute filed' AND reason = 'unauthorized'),
+txns AS (SELECT uid, epoch(t) AS ts FROM ev WHERE event = 'transaction completed' AND transaction_type = 'purchase' AND payment_method = 'credit'),
+burst_locks AS (
+  SELECT l.uid, l.ts FROM locks l JOIN txns x ON x.uid = l.uid AND x.ts BETWEEN l.ts - 3900 AND l.ts
+  GROUP BY 1, 2 HAVING count(*) >= 3
+),
+sig AS (SELECT DISTINCT b.uid FROM burst_locks b JOIN disputes d ON d.uid = b.uid AND d.ts - b.ts BETWEEN 0 AND 3600),
+hist AS (SELECT count(*) AS n FROM (
+  SELECT uid FROM ev WHERE event = 'card locked' GROUP BY 1
+  INTERSECT
+  SELECT uid FROM ev WHERE event = 'dispute filed' GROUP BY 1
+))
+SELECT (SELECT count(*) FROM sig) AS fraud_users,
+       (SELECT n FROM hist) AS support_history_users,
+       round((SELECT count(*) FROM sig)::DOUBLE / (SELECT n FROM hist), 4) AS fraction;
+
+
+-- ── H4-lowbal-churn-suppression ─────────────────────────────
+-- 3+ balance checks under $8K → post-d30/pre-d30 event ratio ~ 0.5x healthy
+WITH low AS (
+  SELECT uid, count(*) FILTER (WHERE event = 'balance checked' AND account_balance < 8000) AS low_checks
+  FROM ev GROUP BY 1
+),
+cutoff AS (SELECT min(t) + INTERVAL 30 DAY AS c FROM ev)
+SELECT CASE WHEN l.low_checks >= 3 THEN 'lowbal' ELSE 'healthy' END AS cohort,
+       count(DISTINCT e.uid) AS users,
+       round((count(*) FILTER (WHERE e.t >= (SELECT c FROM cutoff)))::DOUBLE
+             / nullif(count(*) FILTER (WHERE e.t < (SELECT c FROM cutoff)), 0), 2) AS post_pre
+FROM ev e JOIN low l USING (uid) GROUP BY 1 ORDER BY 1;
+
+
+-- ── H5-budget-discipline ────────────────────────────────────
+-- disciplined budgeters (3+ budgets, ~77% of users): 2x savings
+-- contributions, 1.5x investment amounts vs light budgeters (0-2, ~14%)
+WITH cohort AS (SELECT uid, count(*) FILTER (WHERE event = 'budget created') >= 3 AS disciplined FROM ev GROUP BY 1)
+SELECT CASE WHEN c.disciplined THEN 'disciplined' ELSE 'light' END AS cohort, count(*) AS n,
+       round(avg(e.monthly_contribution), 0) AS avg_contribution
+FROM ev e JOIN cohort c USING (uid)
+WHERE e.event = 'savings goal set' AND e.monthly_contribution IS NOT NULL
+GROUP BY 1 ORDER BY 1;
+
+WITH cohort AS (SELECT uid, count(*) FILTER (WHERE event = 'budget created') >= 3 AS disciplined FROM ev GROUP BY 1)
+SELECT CASE WHEN c.disciplined THEN 'disciplined' ELSE 'light' END AS cohort, count(*) AS n,
+       round(avg(e.amount), 0) AS avg_amount
+FROM ev e JOIN cohort c USING (uid)
+WHERE e.event = 'investment made' AND e.amount IS NOT NULL
+GROUP BY 1 ORDER BY 1;
+
+
+-- ── H6-autopay-missed-share ─────────────────────────────────
+-- missed/paid ~ 0.22 (60% manual x 30% missed → 0.18/0.82); missed rows are
+-- always auto_pay = false
+SELECT event, auto_pay, count(*) AS n
+FROM ev WHERE event IN ('bill paid', 'bill payment missed')
+GROUP BY 1, 2 ORDER BY 1, 2;
+
+
+-- ── H7-premium-tier-value ───────────────────────────────────
+-- premium 3x reward value, plus 1.5x; premium 2x investment-sell amount
+SELECT account_tier, count(*) AS redemptions, round(avg("value"), 1) AS avg_value
+FROM ev WHERE event = 'reward redeemed' GROUP BY 1 ORDER BY avg_value DESC;
+
+SELECT account_tier, count(*) AS sells, round(avg(amount), 0) AS avg_amount
+FROM ev WHERE event = 'investment made' AND action = 'sell'
+GROUP BY 1 ORDER BY avg_amount DESC;
+
+
+-- ── H8-month-end-anxiety ────────────────────────────────────
+-- days >= 28: session durations 1.4x, reported balances 0.7x
+SELECT CASE WHEN EXTRACT(DAY FROM t) >= 28 THEN 'monthEnd' ELSE 'other' END AS bucket,
+       round(avg(session_duration_sec), 0) AS avg_session_sec, count(*) AS sessions
+FROM ev WHERE event = 'app session' GROUP BY 1 ORDER BY 1;
+
+SELECT CASE WHEN EXTRACT(DAY FROM t) >= 28 THEN 'monthEnd' ELSE 'other' END AS bucket,
+       round(avg(account_balance), 0) AS avg_balance, count(*) AS checks
+FROM ev WHERE event = 'balance checked' GROUP BY 1 ORDER BY 1;
+
+
+-- ── H9-onboarding-ttc-by-tier ───────────────────────────────
+-- cross-event MIN→MIN TTC (account opened → first balance checked after it):
+-- MEDIAN basic ~ 2x premium (factors 1.33 vs 0.67), plus between. Median,
+-- not avg — TTC is heavy-tailed and a single multi-day straggler dominates
+-- the mean of these small cohorts (only born-in users have "account opened").
+WITH ao AS (SELECT uid, min(t) AS t0 FROM ev WHERE event = 'account opened' GROUP BY 1),
+bc AS (
+  SELECT e.uid, min(e.t) AS t1
+  FROM ev e JOIN ao ON e.uid = ao.uid AND e.t >= ao.t0
+  WHERE e.event = 'balance checked' GROUP BY 1
+)
+SELECT u.account_tier, count(*) AS converters,
+       round(median(epoch(t1 - t0)) / 60, 1) AS median_ttc_min,
+       round(avg(epoch(t1 - t0)) / 60, 1) AS avg_ttc_min
+FROM ao JOIN bc USING (uid) JOIN users u ON u.distinct_id = ao.uid
+GROUP BY 1 ORDER BY median_ttc_min DESC;
+
+
+-- ── H10-txn-magic-number ────────────────────────────────────
+-- sweet band (12-19 txns, measured: median 12 / p75 17 / p90 20) invests
+-- 1.4x the 1-11 band; over band (20+) has 0.8x the premium-upgrade share of
+-- NON-TXN events (raw per-user counts RISE with activity, and the over band
+-- is selected for high txn counts — normalizing by non-txn events removes
+-- both distortions)
+WITH txn AS (SELECT uid, count(*) AS n FROM ev WHERE event = 'transaction completed' GROUP BY 1),
+bands AS (SELECT uid, CASE WHEN n BETWEEN 12 AND 19 THEN 'sweet' WHEN n < 12 THEN 'low' ELSE 'over' END AS band FROM txn)
+SELECT b.band, count(DISTINCT e.uid) AS users, round(avg(e.amount), 0) AS avg_invest_amount
+FROM ev e JOIN bands b USING (uid)
+WHERE e.event = 'investment made' AND e.amount IS NOT NULL
+GROUP BY 1 ORDER BY 1;
+
+WITH txn AS (SELECT uid, count(*) AS n FROM ev WHERE event = 'transaction completed' GROUP BY 1),
+bands AS (SELECT uid, CASE WHEN n BETWEEN 12 AND 19 THEN 'sweet' WHEN n < 12 THEN 'low' ELSE 'over' END AS band FROM txn)
+SELECT b.band, count(DISTINCT e.uid) AS users,
+       round((count(*) FILTER (WHERE e.event = 'premium upgraded'))::DOUBLE
+             / (count(*) FILTER (WHERE e.event != 'transaction completed')), 5) AS upgrade_share
+FROM ev e JOIN bands b USING (uid) GROUP BY 1 ORDER BY 1;

package/dungeons/vertical/fintech/fintech.verify.mjs

@@ -0,0 +1,70 @@
+#!/usr/bin/env node
+/**
+ * fintech.verify.mjs — thin wrapper around the story runner.
+ *
+ * All verification logic lives in the `stories` export of ./fintech.js;
+ * this script just streams the shards and delegates. It is equivalent to:
+ *
+ *   node scripts/verify-stories.mjs dungeons/vertical/fintech/fintech.js --data-prefix verify-fintech
+ *
+ * Generate first:
+ *   node scripts/verify-runner.mjs dungeons/vertical/fintech/fintech.js verify-fintech
+ * Run:
+ *   node dungeons/vertical/fintech/fintech.verify.mjs [prefix]   # default verify-fintech
+ */
+import fs from 'node:fs';
+import path from 'node:path';
+import readline from 'node:readline';
+import { execFile } from 'node:child_process';
+import { promisify } from 'node:util';
+import { buildIdentityMap, evaluateStories, VERDICT_RANK } from '@ak--47/dungeon-master/verify';
+import config, { stories } from './fintech.js';
+
+const PREFIX = process.argv[2] || 'verify-fintech';
+const prefixPath = path.join('data', PREFIX);
+
+async function loadShards(suffix) {
+	// streaming load: full-fidelity EVENTS shards exceed the readFileSync cap
+	const dir = path.dirname(prefixPath), base = path.basename(prefixPath);
+	const out = [];
+	if (!fs.existsSync(dir)) return out;
+	for (const f of fs.readdirSync(dir).filter(f => f.startsWith(`${base}-${suffix}`) && f.endsWith('.json')).sort()) {
+		const rl = readline.createInterface({ input: fs.createReadStream(path.join(dir, f)), crlfDelay: Infinity });
+		for await (const line of rl) {
+			if (line.trim()) out.push(JSON.parse(line));
+		}
+	}
+	return out;
+}
+
+const events = await loadShards('EVENTS');
+const profiles = await loadShards('USERS');
+if (!events.length) {
+	console.error(`no shards at ${prefixPath}-EVENTS*.json — run: node scripts/verify-runner.mjs dungeons/vertical/fintech/fintech.js ${PREFIX}`);
+	process.exit(1);
+}
+console.log(`fintech — events=${events.length} users=${profiles.length} (${prefixPath})`);
+
+const execFileP = promisify(execFile);
+const runSql = async (sql) => {
+	const { stdout } = await execFileP('duckdb', ['-json', '-c', sql.replaceAll('{{PREFIX}}', prefixPath)], { maxBuffer: 64 * 1024 * 1024 });
+	return stdout.trim() ? JSON.parse(stdout) : [];
+};
+
+// funnels passed raw (unvalidated) — every timeToConvert story carries an
+// explicit conversionWindowMs, so no funnel-default threading is needed
+const results = await evaluateStories(stories, events, {
+	profiles,
+	funnels: config.funnels,
+	identityMap: buildIdentityMap(profiles),
+	runSql,
+});
+
+let worst = 'NAILED';
+for (const r of results) {
+	console.log(`${r.verdict.padEnd(7)} ${r.id}`);
+	for (const a of r.assertions) console.log(`        ${a.verdict.padEnd(7)} ${a.detail}`);
+	if (VERDICT_RANK[r.verdict] < VERDICT_RANK[worst]) worst = r.verdict;
+}
+console.log(`\nworst verdict: ${worst}`);
+process.exit(VERDICT_RANK[worst] >= VERDICT_RANK.STRONG ? 0 : 1);