@ak--47/dungeon-master 1.5.4 → 1.6.1

package/dungeons/vertical/fitness/fitness.sql

@@ -0,0 +1,206 @@
+-- ============================================================
+-- fitness.js — v1.6 human-inspection queries (DuckDB)
+--
+-- Every query is keyed to a story id in fitness.js's `stories` export;
+-- the machine-checked verdicts come from:
+--   node scripts/verify-stories.mjs dungeons/vertical/fitness/fitness.js --data-prefix verify-fitness
+-- Generate first:
+--   node scripts/verify-runner.mjs dungeons/vertical/fitness/fitness.js verify-fitness
+-- Run this file:
+--   duckdb -c ".read dungeons/vertical/fitness/fitness.sql"
+-- ============================================================
+
+-- ── identity-resolution prelude ─────────────────────────────
+-- avgDevicePerUser: 3 + account created is both isAuthEvent and
+-- isFirstEvent, so born users auth on their first event; the device-pool
+-- resolve is belt-and-braces for any device-only edge.
+CREATE OR REPLACE VIEW users AS
+SELECT * FROM read_json_auto('data/verify-fitness-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
+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-fitness-EVENTS*.json', sample_size=-1, union_by_name=true) e
+LEFT JOIN device_map m ON e.device_id = m.device_id;
+
+-- Per-user workout counts. H3/H4/H10 classify on counts taken after H8's
+-- progress-checked drop and H5's resolver thinning; for non-resolver users
+-- no later hook deletes "workout completed", so output counts rebuild the
+-- hook cohorts exactly (resolver-sensitive queries exclude that segment).
+CREATE OR REPLACE VIEW workout_ct AS
+SELECT uid, count(*) AS w FROM ev WHERE event = 'workout completed' GROUP BY 1;
+
+
+-- ── H1-morning-calorie-boost ────────────────────────────────
+-- workouts 05:00-09:00 UTC carry calories_burned × 1.3. Nothing else
+-- touches calories_burned, so avg AND median ratios read the knob.
+SELECT CASE WHEN extract(hour FROM t) >= 5 AND extract(hour FROM t) < 9 THEN 'morning' ELSE 'other' END AS grp,
+  count(*) AS workouts, round(avg(calories_burned), 1) AS avg_cal, median(calories_burned) AS med_cal
+FROM ev WHERE event = 'workout completed' GROUP BY 1 ORDER BY 1;
+
+
+-- ── H2-ai-coaching-lift ─────────────────────────────────────
+-- after day 35 (2026-02-05), workouts flip to coaching_mode='ai_assisted'
+-- at 40% per-event, and ai_assisted duration × 1.2. Pre-launch ai rows
+-- must be ZERO (declared pool is the single value 'self_guided').
+SELECT count(*) FILTER (WHERE coaching_mode = 'ai_assisted' AND t <= TIMESTAMP '2026-02-05') AS pre_launch_ai,
+  count(*) FILTER (WHERE coaching_mode = 'ai_assisted') AS ai_total
+FROM ev WHERE event IN ('workout completed', 'workout planned');
+
+SELECT coaching_mode, count(*) AS workouts, round(avg(duration_minutes), 1) AS avg_dur
+FROM ev WHERE event = 'workout completed' AND t > TIMESTAMP '2026-02-05' GROUP BY 1 ORDER BY 1;
+
+
+-- ── H3-streak-achievements ──────────────────────────────────
+-- ≥2-workout users: profile streak_days OVERWRITTEN to hook-time workout
+-- count, plus C(w) = min(w−1, 3) + 4·max(w−4, 0) cloned achievements.
+-- Contract is ONE-SIDED for non-resolvers: sd < w impossible (nothing adds
+-- workouts after H3); sd > w happens when the silent future-time guard
+-- deletes a counted workout post-hook (~0.5% of users) — so expect
+-- below_w = 0 and eq_share ≥ 0.99, NOT perfect equality.
+WITH j AS (SELECT u.distinct_id::VARCHAR AS uid, u.segment, u.streak_days AS sd, coalesce(w.w, 0) AS w
+  FROM users u LEFT JOIN workout_ct w ON w.uid = u.distinct_id::VARCHAR
+  WHERE u.segment <> 'resolver')
+SELECT count(*) FILTER (WHERE w >= 2 AND sd < w) AS below_w,
+  count(*) FILTER (WHERE segment <> 'coach' AND sd = 1) AS unreachable_one,
+  round(count(*) FILTER (WHERE w >= 2 AND sd = w)::DOUBLE / nullif(count(*) FILTER (WHERE w >= 2), 0), 4) AS eq_share,
+  count(*) FILTER (WHERE w >= 2) AS streak_users
+FROM j;
+
+-- implied organic achievements (total − C(w)) — median should sit in the
+-- low single digits (ach weight 2 of 68); a drifted clone formula would
+-- push it negative or huge.
+WITH ac AS (SELECT uid, count(*) AS a FROM ev WHERE event = 'achievement unlocked' GROUP BY 1),
+j AS (SELECT u.distinct_id::VARCHAR AS uid, coalesce(w.w, 0) AS w, coalesce(a.a, 0) AS a
+  FROM users u LEFT JOIN workout_ct w ON w.uid = u.distinct_id::VARCHAR LEFT JOIN ac a ON a.uid = u.distinct_id::VARCHAR
+  WHERE u.segment <> 'resolver')
+SELECT count(*) AS cohort,
+  round(count(*) FILTER (WHERE a - (LEAST(w - 1, 3) + GREATEST(w - 4, 0) * 4) >= 1)::DOUBLE / count(*), 4) AS ok_share,
+  median(a - (LEAST(w - 1, 3) + GREATEST(w - 4, 0) * 4)) AS med_organic
+FROM j WHERE w BETWEEN 2 AND 14 AND a >= 1;
+
+
+-- ── H4-social-challenge-completion ──────────────────────────
+-- ≥3-friend users get max(1, floor(cc × 0.5)) cloned challenge completions.
+-- out = cc + max(1, floor(cc/2)) skips {5, 8, 11, …} = {n≥5 : n≡2 mod 3} —
+-- gap hits on the clean cohort only come from the future-time guard (~2%).
+WITH fr AS (SELECT uid, count(*) AS f FROM ev WHERE event = 'friend added' GROUP BY 1),
+ch AS (SELECT uid, count(*) AS c FROM ev WHERE event = 'challenge completed' GROUP BY 1),
+j AS (SELECT u.distinct_id::VARCHAR AS uid, coalesce(w.w, 0) AS w, coalesce(f.f, 0) AS f, coalesce(c.c, 0) AS c
+  FROM users u LEFT JOIN workout_ct w ON w.uid = u.distinct_id::VARCHAR
+  LEFT JOIN fr f ON f.uid = u.distinct_id::VARCHAR LEFT JOIN ch c ON c.uid = u.distinct_id::VARCHAR
+  WHERE u.segment <> 'resolver')
+SELECT count(*) AS clean_cohort, count(*) FILTER (WHERE c >= 5 AND c % 3 = 2) AS gap_hits
+FROM j WHERE w <= 14 AND f >= 3 AND c >= 2;
+
+-- within-social-segment gradient (composite: clone lift × activity correlation)
+WITH fr AS (SELECT uid, count(*) AS f FROM ev WHERE event = 'friend added' GROUP BY 1),
+ch AS (SELECT uid, count(*) AS c FROM ev WHERE event = 'challenge completed' GROUP BY 1),
+j AS (SELECT u.distinct_id::VARCHAR AS uid, coalesce(f.f, 0) AS f, coalesce(c.c, 0) AS c
+  FROM users u LEFT JOIN fr f ON f.uid = u.distinct_id::VARCHAR LEFT JOIN ch c ON c.uid = u.distinct_id::VARCHAR
+  WHERE u.segment = 'social')
+SELECT CASE WHEN f >= 3 THEN 'friend_heavy' ELSE 'friend_light' END AS grp,
+  count(*) AS users, round(avg(c), 3) AS avg_challenge_completions
+FROM j WHERE f >= 3 OR f <= 1 GROUP BY 1 ORDER BY 1;
+
+
+-- ── H5-resolver-churn-cliff ─────────────────────────────────
+-- resolvers with <30 hook-time events lose 70% of post-day-14
+-- (2026-01-15) events. Persona churnRate/activeWindow are deprecated
+-- engine no-ops — the cliff is the hook alone, so the estimator targets
+-- the 0.30 keep-rate. Deletions-only ⇒ eligible ⟺ output n < 30.
+-- Double-difference: birth pinned to first event < day 2 (removes
+-- birth-composition), lo/hi volume split normalized by the same split
+-- inside casual (removes n-selection; casual lo/hi ≈ 1.22 measured).
+-- DD = (res_lo/res_hi) ÷ (cas_lo/cas_hi) ≈ 0.30.
+WITH tot AS (SELECT uid, count(*) AS n, min(t) AS first_t,
+  count(*) FILTER (WHERE t < TIMESTAMP '2026-01-15') AS pre,
+  count(*) FILTER (WHERE t >= TIMESTAMP '2026-01-15') AS post
+  FROM ev GROUP BY 1),
+j AS (SELECT u.segment AS seg, CASE WHEN t.n < 30 THEN 'lo' ELSE 'hi' END AS arm, t.pre, t.post
+  FROM users u JOIN tot t ON t.uid = u.distinct_id::VARCHAR
+  WHERE t.first_t < TIMESTAMP '2026-01-03' AND u.segment IN ('resolver', 'casual'))
+SELECT seg, arm, count(*) AS users,
+  round(sum(post)::DOUBLE / nullif(sum(pre), 0), 4) AS rho
+FROM j GROUP BY 1, 2 ORDER BY 1, 2;
+-- read: (rho[resolver,lo] / rho[resolver,hi]) / (rho[casual,lo] / rho[casual,hi])
+-- ≈ 0.30; the casual lo/hi ratio itself is the placebo (~1, far from 0.3).
+
+
+-- ── H6-coach-session-quality ────────────────────────────────
+-- every coach-session satisfaction_score redrawn uniform [4.0, 5.0]
+-- (unconditional — a single sub-4.0 score is a hook bug; avg = median = 4.5).
+SELECT count(*) FILTER (WHERE satisfaction_score < 4.0) AS below_min,
+  count(*) AS sessions, round(avg(satisfaction_score), 3) AS avg_sat, median(satisfaction_score) AS med_sat
+FROM ev WHERE event = 'coach session';
+
+
+-- ── H7-coach-profile-enrichment ─────────────────────────────
+-- user hook: coaches get total_workouts uniform [200, 500] (avg 350);
+-- everyone else keeps the declared 0. Ranges deterministic. (streak_days
+-- 60-365 also seeded here, but H3 overwrites it for ≥2-workout users —
+-- total_workouts is the durable coach signature.)
+SELECT CASE WHEN segment = 'coach' THEN 'coach' ELSE 'other' END AS grp,
+  count(*) AS users, min(total_workouts) AS min_tw, max(total_workouts) AS max_tw,
+  round(avg(total_workouts), 1) AS avg_tw
+FROM users GROUP BY 1 ORDER BY 1;
+
+
+-- ── H8-annual-follow-through ────────────────────────────────
+-- 30% of free/monthly users lose ALL progress-checked events (per-user
+-- cliff). Tier assigned BY segment → raw cross-tier reads are confounded
+-- by persona composition BY CONSTRUCTION; only athlete and social contain
+-- both an affected and a control tier, so standardize within those.
+WITH pc AS (SELECT uid, count(*) AS ct FROM ev WHERE event = 'progress checked' GROUP BY 1),
+j AS (SELECT u.segment AS seg, CASE WHEN u.subscription_tier IN ('annual', 'family') THEN 'ctl' ELSE 'aff' END AS arm,
+  coalesce(p.ct, 0) AS ct
+  FROM users u LEFT JOIN pc p ON p.uid = u.distinct_id::VARCHAR
+  WHERE u.segment IN ('athlete', 'social'))
+SELECT seg, arm, count(*) AS users,
+  round(count(*) FILTER (WHERE ct = 0)::DOUBLE / count(*), 4) AS zero_share,
+  round(avg(ct) FILTER (WHERE ct > 0), 2) AS avg_ct_survivors
+FROM j GROUP BY 1, 2 ORDER BY 1, 2;
+-- read: (z_aff − z_ctl) / (1 − z_ctl) per segment ≈ 0.30; survivor avgs
+-- match within segment (per-user cliff, not thinning).
+
+
+-- ── H9-workout-loop-ttc ─────────────────────────────────────
+-- funnel-post scales EVERY funnel instance's gaps by tier: annual/family
+-- × 0.77, free × 1.25, monthly = 1.0.
+-- CAUTION: cross-event TTC SQL here would be censored by its lookback
+-- window (the ai-platform H9 lesson — censoring can invert the measured
+-- direction). The story asserts TTC through the Mixpanel-aligned emulator
+-- at a 60h window (= 48h generative window × 1.25 max stretch, covering
+-- the stretched support); trust the story verdict, not ad-hoc pair SQL.
+SELECT subscription_tier, count(*) AS planned_events
+FROM ev WHERE event = 'workout planned' GROUP BY 1 ORDER BY 1;
+
+
+-- ── H10-workout-magic-number ────────────────────────────────
+-- sweet 12-14 workouts → ALL workout durations × 1.35; over 15+ → 65% of
+-- post-day-30 (2026-01-31) events dropped EXCEPT workout completed /
+-- progress checked. Duration read restricted to pre-AI-launch (≤2026-02-05)
+-- where H2 never touched durations — AVG ratio reads the 1.35 knob exactly;
+-- median atom-snaps on the discrete duration pool (reads high, ~1.48).
+WITH coh AS (SELECT uid, CASE WHEN w BETWEEN 12 AND 14 THEN 'sweet'
+  WHEN w BETWEEN 2 AND 11 THEN 'low' WHEN w >= 15 THEN 'over' END AS grp FROM workout_ct)
+SELECT c.grp, count(DISTINCT c.uid) AS users, round(avg(e.duration_minutes), 2) AS avg_dur_prelaunch
+FROM coh c JOIN ev e ON e.uid = c.uid AND e.event = 'workout completed' AND e.t <= TIMESTAMP '2026-02-05'
+WHERE c.grp IS NOT NULL GROUP BY 1 ORDER BY avg_dur_prelaunch;
+
+-- over-drop: within-user-normalized double ratio of non-preserved volume
+WITH coh AS (SELECT uid, CASE WHEN w >= 15 THEN 'over' WHEN w BETWEEN 12 AND 14 THEN 'sweet' END AS grp
+  FROM workout_ct WHERE w >= 12),
+per AS (SELECT c.grp, c.uid,
+  count(*) FILTER (WHERE e.event NOT IN ('workout completed', 'progress checked') AND e.t < TIMESTAMP '2026-01-31') AS pre,
+  count(*) FILTER (WHERE e.event NOT IN ('workout completed', 'progress checked') AND e.t >= TIMESTAMP '2026-01-31') AS post
+  FROM coh c JOIN ev e ON e.uid = c.uid GROUP BY 1, 2)
+SELECT grp, count(*) AS users, round(sum(post)::DOUBLE / nullif(sum(pre), 0), 4) AS post_pre
+FROM per WHERE grp IS NOT NULL GROUP BY 1 ORDER BY 1;

package/dungeons/vertical/fitness/fitness.verify.mjs

@@ -0,0 +1,71 @@
+#!/usr/bin/env node
+/**
+ * fitness.verify.mjs — thin wrapper around the story runner.
+ *
+ * All verification logic lives in the `stories` export of ./fitness.js;
+ * this script just streams the shards and delegates. It is equivalent to:
+ *
+ *   node scripts/verify-stories.mjs dungeons/vertical/fitness/fitness.js --data-prefix verify-fitness
+ *
+ * Generate first:
+ *   node scripts/verify-runner.mjs dungeons/vertical/fitness/fitness.js verify-fitness
+ * Run:
+ *   node dungeons/vertical/fitness/fitness.verify.mjs [prefix]   # default verify-fitness
+ */
+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 './fitness.js';
+
+const PREFIX = process.argv[2] || 'verify-fitness';
+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/fitness/fitness.js ${PREFIX}`);
+	process.exit(1);
+}
+console.log(`fitness — events=${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) — the H9 stories carry their own explicit
+// conversion window (48h generative window × the 1.25 max stretch factor,
+// covering the stretched support), 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);