Music Machinery

SmarterPlaylists launched on July 25, 2015 as a side project — a visual programming tool that let anyone wire together Spotify sources, filters, and combiners to build playlists that no single Spotify feature could create on its own. No ads, no business model, no growth team. Just a tool I thought was cool, shared on a few subreddits, and then left running on a single server for over ten years.

Now, as I’m retiring the old system and replacing it with a ground-up rewrite, I cracked open the database one last time to see what a decade of quiet, organic usage actually looks like. The data only goes back to early 2016 (the first ~6 months predate the current database), but even so, the numbers surprised me.

The headline stats: 262,000 people logged in over the years. Nearly 70,000 of them built something. They created 278,600 programs and ran them over 9 million times. More than 50,000 scheduled jobs were still marked “active” when I pulled the plug — playlists that were meant to refresh themselves daily, weekly, or monthly, faithfully updating for years.

Not bad for a weekend project that never got a v1.0.

---

The Big Picture

Metric	Value
Total programs created	278,600
Total unique users who built programs	69,562
Users who ever authenticated	262,181

That gap between “authenticated” (262K) and “built something” (70K) is interesting — nearly 75% of visitors logged in with Spotify, poked around, and left without saving a program. The visual programming interface was powerful but had a steep learning curve. That’s one of the biggest things the rewrite aims to fix.

---

Run Statistics

Metric	Value
Total program runs (all time)	9,029,034
Total errors	643,645
Error rate	7.1%
Programs run at least once	205,984 (74%)
Programs never run	72,616 (26%)
Mean runs per program (of those run)	43.8
Median runs per program	4
Max runs (single program)	68,351

Nine million runs. That’s roughly 2,500 playlist generations per day on average, every day, for ten years — all from a single server.

The median of 4 runs tells the real story though: most people built a program, ran it a few times to get it right, then either scheduled it or moved on. Meanwhile the mean of 43.8 is dragged way up by power users running programs tens of thousands of times via the scheduler. The gap between mean and median is the signature of a long-tail product.

A 7.1% error rate sounds high, but most errors were transient Spotify API issues — rate limits, expired tokens, playlists that got deleted. The system was designed to shrug these off and retry on the next scheduled run.

---

Sharing & Discovery

Metric	Value
Shared programs (marked public)	4,707
Programs imported at least once	2,488
Total imports across all programs	53,879

Sharing was always a secondary feature — there was no feed, no recommendations, just a “shared programs” page and the occasional Reddit post. Despite that, 53,879 imports means the average shared program that got any traction was imported about 22 times. The most popular one was imported over 5,000 times. Word of mouth carried the whole thing.

---

Scheduler

Metric	Value
Total scheduled jobs ever created	56,629
Currently active scheduled jobs	50,234
Total scheduled runs	1,914,294

This is the feature that kept SmarterPlaylists alive. Nearly 2 million automated playlist updates — your “Discover Weekly Archive” growing every Monday, your “Daily Mix minus songs I’ve already heard” refreshing every morning. Over 50,000 jobs were still marked “active” when I shut things down, though many of those had stale tokens. The scheduler was the heartbeat of the app.

---

User Distribution

Programs per User	Users	%
1	47,031	68%
2–5	18,595	27%
6–10	2,474	4%
11–50	1,350	2%
50+	112	0.2%

A classic power-law distribution. Two thirds of users built exactly one program. But that top 0.2% — 112 users with 50+ programs each — they were prolific. The most prolific user created 693 programs. That’s not a user, that’s a hobbyist playlist factory.

---

Activity Timeline

SmarterPlaylists launched July 25, 2015. The database only goes back to early 2016, so the first ~6 months of activity aren’t captured here.

Earliest recorded run	2016-01-03
Latest recorded run	2026-03-05

Programs Last Active by Year

Year	Programs	% of Total
2016	4,751	1.7%
2017	9,185	3.3%
2018	13,016	4.7%
2019	12,176	4.4%
2020	21,032	7.5%
2021	16,290	5.8%
2022	15,463	5.6%
2023	13,094	4.7%
2024	11,740	4.2%
2025	12,721	4.6%
2026	5,302	1.9%

The COVID bump is unmistakable — 2020 saw a 73% jump over 2019. People stuck at home, listening to more music, and apparently building more elaborate playlists. Usage never quite returned to that peak, but it stabilized at a healthy plateau from 2021 onward rather than falling off a cliff. Even in 2025, over 12,000 programs were still actively running — ten years after launch.

Monthly Active Users (last 36 months)

Note: “MAU” here represents users whose program’s most recent run falls in that month. True monthly active usage was likely higher, since this only captures the last run timestamp per program — a program run daily all year only shows up in its final month.

Month	Users	Programs
2023-04	761	1,086
2023-05	884	1,324
2023-06	725	1,137
2023-07	780	1,338
2023-08	660	981
2023-09	694	1,069
2023-10	717	1,051
2023-11	713	1,028
2023-12	383	611
2024-01	638	1,134
2024-02	662	1,046
2024-03	586	847
2024-04	669	941
2024-05	656	1,010
2024-06	664	1,005
2024-07	579	960
2024-08	652	959
2024-09	639	1,022
2024-10	674	938
2024-11	627	885
2024-12	693	993
2025-01	706	1,266
2025-02	587	961
2025-03	693	1,096
2025-04	683	966
2025-05	691	939
2025-06	724	1,633
2025-07	691	1,112
2025-08	638	900
2025-09	658	1,012
2025-10	643	1,096
2025-11	509	926
2025-12	351	814
2026-01	232	547
2026-02	908	2,917
2026-03	539	1,838

The remarkably steady ~600-700 MAU through 2024-2025 is striking for a project with zero marketing. December always dips (holiday lull?), and February 2026 spikes because I started publicly talking about the retirement and migration, which brought a wave of people back.

---

Top 25 Most Prolific Users (by program count)

Rank	User	Programs	Total Runs
1	User A	693	2,115
2	User B	387	7,183
3	Julio	379	4,690
4	User C	367	36,068
5	User D	353	59,331
6	User E	268	25,662
7	User F	221	661
8	User G	196	11,901
9	Hunter	190	4,393
10	User H	172	477
11	Nipun	163	10,078
12	Thomas	154	11,994
13	User I	148	8,149
14	User J	144	18,153
15	User K	142	9,867
16	User L	142	229,198
17	User M	139	800
18	User N	136	32,002
19	User O	135	4,583
20	User P	135	1,054
21	User Q	134	5,682
22	plamere	131	1,404
23	User R	127	169,251
24	User S	126	9,436
25	User T	119	38,211

There’s a fascinating split here between “builders” and “runners.” User A created 693 programs but only ran them 2,115 times — about 3 runs each on average. They were a tinkerer, endlessly experimenting. User L, on the other hand, had 142 programs but ran them 229,198 times — these were workhorse playlists, scheduled and running daily for years. Two very different ways to love the same tool.

Top 25 Power Users (by total runs)

Rank	User	Total Runs	Programs	Errors
1	Tim	330,997	110	35,122
2	User L	229,198	142	3,567
3	User R	169,251	127	7,963
4	User D	59,331	353	3,622
5	User U	49,040	20	1,090
6	Kjell	39,445	42	1,397
7	User T	38,211	119	1,517
8	User V	37,386	109	2,940
9	User W	37,248	33	1,374
10	User C	36,068	367	928
11	User X	35,765	29	1,851
12	User N	32,002	136	3,520
13	User Y	31,507	83	1,206
14	User Z	29,753	64	721
15	User AA	28,050	14	1,103
16	User BB	27,167	41	381
17	User CC	27,023	42	796
18	Fredrik	26,578	115	1,782
19	User DD	26,177	92	567
20	User EE	25,725	31	908
21	User E	25,662	268	1,260
22	User FF	25,201	12	1,421
23	John	24,378	71	1,208
24	Roger	24,024	19	733
25	User GG	22,456	33	1,120

Tim is in a league of his own: 330,997 total runs across 110 programs. That’s roughly 90 program runs per day sustained over years. His top 5 programs alone account for nearly 280,000 runs. Whatever Tim was doing with his music, he was serious about it.

User AA is the efficiency champion — only 14 programs, but 28,050 runs. That’s 2,004 runs per program on average. Build it once, run it forever.

Top 25 Most Run Programs

Rank	Program	Owner	Runs
1	#random.select	Tim	68,351
2	#random.bucket	Tim	68,265
3	#random.listening.year	Tim	66,119
4	#nine	Tim	39,852
5	#random.release.year	Tim	36,953
6	(unnamed)	(unknown)	29,850
7	(unnamed)	(unknown)	27,741
8	moetjehoren002	User U	24,510
9	(unnamed)	(unknown)	22,160
10	(unnamed)	(unknown)	19,337
11	(unnamed)	(unknown)	16,626
12	(unnamed)	(unknown)	16,067
13	(unnamed)	(unknown)	13,020
14	Today Artists	User R	11,908
15	Songs V2a	User HH	10,617
16	A list	User GG	10,399
17	(unnamed)	(unknown)	10,203
18	Combined Lists	User AA	10,131
19	(unnamed)	(unknown)	10,048
20	(unnamed)	(unknown)	9,660
21	Made For You	User AA	9,526
22	(unnamed)	(unknown)	9,206
23	(unnamed)	(unknown)	8,823
24	(unnamed)	(unknown)	8,150
25	my super mix	Rodolfo	8,123

Tim’s top 5 programs owned the leaderboard so completely that the next contender had less than half his #1’s run count. The #random.* naming convention suggests he built a suite of randomization programs — different ways to shuffle his listening by bucket, era, or mood.

A lot of the heavy hitters are “(unnamed) / (unknown)” — these are programs from early in the app’s life, before I started tracking names and owners in metadata. The programs themselves still exist, but the metadata was lost. Ghost playlists, faithfully running for years with no name attached.

Top 25 Most Popular Shared Programs (by imports)

Rank	Program	Owner	Imports	Runs
1	My forgotten tracks	User II	5,054	8
2	My Discovery Weekly Archiver	plamere	4,853	60
3	yesterday and today	plamere	3,400	27
4	The daily dozen	plamere	3,292	1
5	All Time Top Tracks	Abhishek	2,326	97
6	Combine two playlists	plamere	2,186	20
7	True Release Radar	James	1,529	244
8	Ultimate Coffee House	plamere	1,287	7
9	Less Teen-Oriented New Music	plamere	1,096	1
10	My forgotten tracks	plamere	987	116
11	My Top Played Tracks of All-Time	User JJ	787	149
12	Shuffler	User KK	769	5
13	BigMix	User LL	743	351
14	Recommended Daily (posted)	User J	684	29
15	DailyRec	User LL	663	469
16	Daily Discover	User MM	641	277
17	Gothic Metal front-loaded with Ravenscry	plamere	615	4
18	RELAXING MUSIC PLAYLIST GENERATOR	User NN	553	16
19	My Discovery Weekly Archiver	User OO	506	51
20	Bot-Mix	User J	502	0
21	Will they make it?	User PP	455	448
22	My Top Tracks	User J	422	1
23	Workout	User QQ	415	3
24	recently played	plamere	411	2
25	Cleaning List	User RR	391	81

The sharing leaderboard tells you what people actually wanted from Spotify that they couldn’t get natively. The top hits are all utilities: archive your Discover Weekly before it disappears, combine playlists, deduplicate, filter out songs you’ve already heard. “My forgotten tracks” — a program that surfaces songs from your library you haven’t played in a long time — was the single most imported program at 5,054 imports, and User II only ever ran it 8 times themselves. They built it, shared it, and the community ran with it.

I’m a little proud that so many of my example programs (plamere) made it into the top 25. “Gothic Metal front-loaded with Ravenscry” at #17 is a personal favorite — 615 people apparently also wanted their gothic metal playlist to lead with Ravenscry. Niche appeal is still appeal.

25 Biggest Programs (by component count)

Rank	Program	Owner	Components
1	bing-short	Dylan	401
2	h10	User SS	305
3	h80	User SS	305
4	h90	User SS	305
5	hOld	User SS	305
6	h20	User SS	305
7	h00	User SS	305
8	h70	User SS	305
9	Nirvana Radio v26	User TT	289
10	Nirvana Radio v26	User UU	287
11	import of Nirvana Radio v26	User VV	287
12	Nirvana Radio v24	User UU	283
13	Nirvana Radio v22	User UU	282
14	Nirvana Radio v23	User UU	270
15	Nirvana Radio v20	User UU	270
16	Nirvana Radio v27.02	User TT	268
17	Nirvana Radio v18	User UU	267
18	Nirvana Radio v21	User UU	266
19	Nirvana Radio v25	User UU	265
20	Nirvana Radio v17 saturday	User UU	261
21	Nirvana Radio v17	User UU	261
22	Nirvana Radio v19	User UU	257
23	Rock Favourites	Tom	254
24	Groovy Artists	User WW	252
25	Die drei Schlafezeichen	User XX	249

The biggest programs are a testament to user dedication (or obsession). Dylan’s “bing-short” tops the chart at 401 components — that’s a visual program with 401 nodes wired together on a canvas. I can barely imagine what that looks like.

But the real star here is User UU and their “Nirvana Radio” saga. They iterated through at least 27 versions of the same program, each one with 250-290 components, growing and refining it over time. Versions 17 through 27, spanning years. It was popular enough that User TT and User VV imported copies of it. That’s someone who turned SmarterPlaylists into an art form.

Most Used Component Types (across all programs)

Rank	Component	Usage Count
1	SpotifyPlaylist	352,807
2	Sample	59,253
3	Shuffler	59,138
4	PlaylistSave	53,270
5	MyTopTracks	52,261
6	Concatenate	49,843
7	SpotifyArtistRadio	44,560
8	DeDup	43,162
9	ArtistTopTracks	41,066
10	MySavedTracks	38,764
11	AlbumSource	37,538
12	Sorter	35,645
13	comment	34,670
14	TrackFilter	30,325
15	Alternate	28,736
16	RelativeDatedSpotifyPlaylist	28,338
17	First	25,567
18	Mixer	22,342
19	AttributeRangeFilter	20,625
20	PlaylistSaveToNew	17,150
21	MyFollowedArtists	15,442
22	MixIn	13,035
23	RandomSelector	12,129
24	Energy	11,675
25	ReleaseDateFilter	11,501
26	ArtistDeDup	11,219
27	SeparateArtists	9,974
28	Tempo	9,850
29	Weighted Shuffler	9,728
30	DatedSpotifyPlaylist	9,286

SpotifyPlaylist dominates at 352K uses — almost every program starts with “take this playlist as input.” The rest of the top 10 reads like a recipe: take a playlist, sample some tracks, shuffle them, remove duplicates, and save. That’s the core loop.

The “comment” component at #13 with 34,670 uses warms my heart. It does nothing — it’s just a sticky note on the canvas. But 34,000 times, people felt the need to annotate their programs with explanations. They were documenting their work, even in a visual tool, even when nobody else would see it.

---

Looking Back

Ten years, nine million runs, seventy thousand users, zero revenue. SmarterPlaylists was never a business — it was a proof of concept that escaped into the wild and found an audience of people who cared about their music enough to program their playlists.

The data tells a story of a long-tail product: most people tried it once, but those who got it really got it. Tim with his 330,000 runs. User UU with 27 versions of Nirvana Radio. The 50,000 scheduled jobs humming away in the background. The person who imported “My forgotten tracks” and rediscovered songs they’d loved and lost.

The new version aims to keep everything that made the old one special — the power, the flexibility, the “I can’t believe I can do this” moments — while making it approachable enough that more than 27% of visitors actually build something.

Here’s to the next decade.

(Thanks to Claude and Gemini for helping me write this post)