For fans
Tell us the artists you love, even ones not playing. Qloo scores every act on the bill for your taste, and Verdict plans your best path through the festival day, and nights out with food and drinks to match your taste.
Try
Why Qloo: it links your favourite artists to every act on the bill, including acts you have never heard of, so the plan finds you new acts, not only the names you typed.
Your route
Every act on the bill, scored for you. Where two acts you'd love play at the same time, the plan picks the better one and files the clash with the court.
Type your artists above and press Plan my day.
With friends
Add each friend and the artists they love, pick a city and month, and press Plan our month. You get a month of nights out on a calendar: the shows the whole group would enjoy, and for each night a place to eat before and a bar for after, near the venue, matched by Qloo to your music taste and open at those hours.
Festival timetables Lollapalooza Chicago
A clash is two acts that share many fans, playing at the same time on different stages: someone misses one. Verdict uses Qloo's taste data to find every clash, and AI agents put each one on trial and fix it without breaking the festival's rules.
What the number means: how many fans are forced to choose between two acts they love (shared fans times hours of overlap, over every pair). Lower is better.
Four AI agents argue each clash with Qloo's fan data: one for each act's fans, a booker who checks the rules, a judge who rules. You get the final say on the first case.
28 pre-registered tests over 22 days found
OpenAI's frontier reasoning model
The same frontier model, asked the same questions, alone and with Qloo's data, scored against real-world truth. Pre-registered tests; Gemini 3.8 Flash gains the same way.
See every test and chartAgentic, not a chatbot
Ask in plain words. The agent decides which tools to call (Qloo's fan overlap, the live timetable, Ticketmaster listings, the booker's rules), shows every call as it runs, and checks each number in its answer against the tool results. Orders like a cancellation really change the timetable.
Why it matters
Every timetable, every booking date and every marketing plan in live music depends on which acts draw the same people. Today nobody puts a number on it before the tickets go on sale. Verdict does, with Qloo.
Sources: Omdia (2026), Live Nation 2025 results, Lollapalooza 2025 coverage, our own runs.
In session
Left, the timetable: red sets share many fans with a set at the same time. Right, the courtroom: each case is one clash, argued and ruled, and the timetable moves as it is decided.
Timetable Drag any set to move it. You decide; the court advises.
Court is not yet in session
The bill as Qloo sees it
Acts that share fans sit close together. A pink line joins two acts on at the same time: their fans must choose. Switch to "After the court" to see which clashes the court removed. Click an act for its details.
How the court works
Code builds the timetable; the agents argue, object and explain. Nothing an agent says can move a set in a way that breaks a rule.
Qloo scores how much every pair of acts shares fans.
Each red clash becomes a case; each act's agent proposes a move.
The booker checks the festival's rules and offers a legal move.
The judge rules; code checks and makes the move.
Type "Tyla just cancelled" and the day is rebuilt.
Clashes fans can't avoid weigh on the next ruling.
The same engine
The fan overlap that defines a clash also plans a fan's day and reads a city's month for a promoter.
For promoters and booking agents
Every music show on sale in a city this month (Ticketmaster), scored by how much each act shares fans with yours (Qloo). The map shows where those fans are being pulled.
On this demo: New York, Chicago, Los Angeles, Boston and San Francisco, November 2026, loaded ahead so every answer is quick. Any US city works the same way.
City desk
Pick the city and month, then a tool. Where our tests found a language model better (openers and tour order), the model leads, labelled, and Qloo checks its picks against real fans.
Why Qloo
We ran 28 pre-registered tests: each protocol and decision rule was written down and committed before any result. Against real-world data, GPT-6 Astra and Gemini 3.8 Flash both came closer to the truth with Qloo than without it.
The same model, with and without Qloo
Each bar is the same model answering the same questions, once alone and once with Qloo's data in its prompt, scored against a source outside Qloo. Pink is Qloo. Hover anything for the exact numbers.
Test 20: when does this restaurant close on Friday?
30 real New York restaurants, checked against Google Maps by a person. Each ring is one model's 30 answers. Hover a ring for the counts.
The same model, alone and with Qloo
Grey dot: the model alone. Pink dot: the same model with Qloo's data. Whiskers: 95% intervals. Every score is against real-world data outside Qloo; higher is better.
Qloo against popularity, six tests
Booking by fame alone is what most tools do. Qloo's taste graph beats it in every test below.
Identity: does Qloo find the right one?
Places a model named in its plans (Test 22), and acts on a real Lollapalooza bill.
Ask the same question twice: does the answer change?
Each square is one pair of acts on a real festival day. Lit squares are pairs whose shared-fan score moved by 10 points or more between two identical requests. The count is measured; where the lit squares sit is for illustration. Pick a day and a source.
Every test where Qloo helped
Each test was registered before it ran (PREREGISTRATION files, committed first), with its data source, metric and decision rule fixed in advance. 95% intervals by bootstrap.
Question. Asked when a real New York restaurant closes on Friday, is a model right more often with Qloo's place record?
Data. 30 places, mostly small independents, whose Friday closing time a person read off Google Maps.
Question. Does a model rank pairs of acts by the fans they really share better with Qloo's scores?
Data. 10 bills of 30 real artists each, 4,350 pairs, against the Last.fm listeners each pair shares.
Question. Build a timetable from each scorer's numbers: how much of the real clash does it remove, from a random timetable (0) to the best possible one (1)?
Data. The same 10 bills, judged by real shared listeners.
Question. Among 35 small artists, can a scorer find the 3 to 5 a real listener actually played?
Data. 40 real Last.fm listeners (GroupLens HetRec 2011). AUC, higher is better.
Question. Score every pair of acts on a real Lollapalooza day for shared fans, twice. How many answers move by 10 points or more?
Data. Five real days, 903 to 946 pairs each.
Question. When a model names a real place, does Qloo find the right one?
Data. 120 New York places named in model-written plans; 44 acts on a Lollapalooza bill.
Question. On a real published Lollapalooza day, how much clash does the court remove without breaking a festival rule?
Data. Lollapalooza Chicago, Thursday 2024: 44 acts, 7 stages, 946 pairs scored by Qloo.
Question. What does it cost to score every pair of acts on a day?
Data. Lollapalooza Thursday 2024, 946 pairs.
On the festival day you picked
Pick a day in the bar above: the numbers below come from that day.
What Qloo adds on this festival day
How we tested
Each protocol, metric and decision rule was committed to git before any code or result existed. Deviations are recorded in the log. The clash test is PREREGISTRATION_28.md in this repo.
Google Maps hours read by a person, real listeners' plays (Last.fm, GroupLens) and real published set times. Never Qloo graded by Qloo.
A frontier model and a small fast one, each run alone and with Qloo on the same questions, so the gain is Qloo's and not one model's quirk.
Qloo alone and GPT-6 Astra alone are level on taste (Test 28: 0.723 and 0.697, not a significant gap), which is why Verdict uses them together. Full protocols and readings: PREREGISTRATION_28.md and PROJECT_LOG.md.