SLEEPER MARKETS · PROOF
Does taste beat “just play the biggest cities”?
We checked 162 artists’ real 2024–25 US tours before trusting Qloo’s taste data with anyone’s tour. Qloo taste alone lost. Taste plus city size won, but only from about 7 stops up. Here is all of it, artist by artist.
Beyond the big cities: on the 66 held-out artists, the tested plan’s 8-stop picks included 176 cities outside the 12 biggest (by city population, counted per artist); 107 of them (61%) were cities the artist really played. A taste signal, not a ticket forecast.
Why trust this result?
A bigger number from one split, tuned and scored on the same data, is easy to get. These five steps are what make ours hard to bend:
- Pre-registered. The protocol and decision rule were committed before we requested any new data from Qloo. Protocol commit 140d811
- Frozen weights. 70% taste + 30% size was chosen on 96 artists and frozen in a commit before the held-out artists were scored. Freeze commit 50b3d46
- Held out, scored once. 66 artists the weights never saw, scored a single time: 28 better, 16 worse at 8 stops. Scoring commit 384a293
- Replicated on new artists. 42 artists outside the first study, on their 2026 tours, protocol committed first, same weights: 22 better, 9 worse. 2026 confirmation report
- Predictions sealed. The next-tour picks for 204 artists are locked by SHA-256 before any 2027 tour was read. Check the hash below
Loading the backtest…
Then we tested it again, on tours it had never seen
After choosing the weights, we pre-registered a second test: artists with 2026 US tours on Wikipedia who were not in the first study (42 matched in Qloo), scored once with the weights unchanged. Verdict, by the rule we fixed in advance: confirmed.
| Stops | Tour-tested (the app) | Biggest cities | Qloo taste alone | Artists better / worse with tour-tested |
|---|---|---|---|---|
| 5 | 68% | 72% | 57% | 7 / 16 |
| 8 (primary) | 64% | 58% | 51% | 22 / 9 |
| 12 | 58% | 50% | 46% | 25 / 6 |
Same pattern as the first study: the biggest cities still win at 5 stops. 2026 tours include announced future dates; Qloo’s maps are from October 2026. Protocol and report.
Exploratory: when Qloo says a big city is weak for an artist
Not pre-registered. We asked this after seeing the results above, so treat it as a lead, not a test. For each artist we took the 12 biggest US cities and checked where Qloo’s taste map for that artist ranks each one among our 103 cities (the same Qloo score the app uses), then whether the artist really played it.
| Tours | Qloo ranks it in the bottom half: played | Top half: played | Artists with both: top half played more / less |
|---|---|---|---|
| 2024–25 tours (162 artists) | 49 of 125 (39%) | 1,150 of 1,806 (64%) | 44 / 9 |
| 2026 tours, new artists (42 artists) | 7 of 32 (22%) | 244 of 469 (52%) | 12 / 1 |
| 2026 tours, all artists (90 artists) | 26 of 76 (34%) | 604 of 991 (61%) | 27 / 4 |
Read with care. Much of the gap is which cities: San Antonio, Fort Worth, Houston, Austin and Jacksonville are 93 of the 125 bottom-half cases in 2024–25, and many tours skip them anyway. Comparing the same city across artists, the gap shrinks to 7 points in 2024–25 (95% interval 0 to 15) and 16 points in 2026 (3 to 30). It is also uncommon: 56 of 162 artists had any of the 12 biggest cities in Qloo’s bottom half. A taste signal, not a ticket forecast. Exploratory notes.
Next: predictions we can’t take back
On 29 Oct 2026 we locked, for 204 artists, the 12 cities the app’s ranking predicts for their next US tour, before any of those 2027 tours were read. The file is sealed with its SHA-256 and a commit; the scoring rule is written in advance (precision at 8 stops against the biggest cities, on tour pages created after the lock).
SHA-256: 04c0352d590359dcaabd3710d3ffa6cedcabf266b74c2c5f73bb39aae82adff7
How we tested it
- Ground truth. Every artist on Wikipedia’s 2024 and 2025 concert-tour pages who played at least 5 of our 103 US cities (192 artists). 162 matched exactly one artist in Qloo.
- Pre-registration. We committed the protocol before requesting any data from Qloo, froze the weights tuned on 96 artists in a second commit, and scored the other 66 artists once.
- Decision rule, fixed in advance. At 8 stops, adopt a mix only if it beats the biggest cities on the held-out artists in mean precision and in wins vs losses. 70% taste + 30% size passed (74% vs 70%; 28 wins, 16 losses), and became the app’s ranking.
- Exploratory, after the decision. The same frozen weights at every other tour size (the chart above). Not an independent confirmation.
Limits. This measures agreement with past booking decisions of mostly established acts, not demand, tickets or causality. Qloo’s heatmaps are from 2026 and may reflect earlier tours. Population is the Census 2025 estimate for the city proper.
Protocol, code and results: protocol · report · all files. Our first, smaller backtest went against Qloo and is published too.
Tour cities are derived from Wikipedia tour pages (each linked to the revision we read) and published under CC BY-SA 4.0. Taste rankings from Qloo’s Insights API; raw Qloo responses are not published.
Tour traffic (supply proxy)
Next to each sleeper market the plan shows how many of 246 tracked touring artists stopped in that city in 2024–26: the eligible artists of the 2024–25 and 2026 Wikipedia tour-page tables above (at least 5 of our 103 cities), each counted once even if they appear in both. A stop with top 10% Qloo taste and tour traffic at or below the 103-city median (23), not within 60 miles of a city with 100+ tracked tours (a nearby hub serves that audience), is flagged “Whitespace: strong taste, few big tours come through”.
It is a proxy for how crowded a city’s calendar is, not a count of all shows, and says nothing about attendance or sales. City-name matching undercounts some cities: tour tables often write just “Portland”, so Portland, OR and Portland, ME show no count and never get the Whitespace flag. Display only: it does not change the ranking, the agent’s picks or the backtest above. Built offline from the same CC BY-SA 4.0 tables: script · data.
Browse these whitespace cities for every real touring artist in the backtest: Whitespace Atlas.