Should you move your launch? Quiet Week measures whether the big release next to yours shares your players.
GTA VI lands on November 19, 2026, and studios are emptying the month around it. But dodging only helps if the big release has your players. Quiet Week answers that question for one indie game, week by week, with Qloo's audience data: for an action western, GTA's players overlap at 0.90; for a cozy farming game, 0.55. Same week, opposite advice.
Indie studios and small publishers picking a launch week between November 2026 and March 2027. They usually have one shot at a launch week and today they guess: release calendars count games, but none says whether a big release shares your audience.
The home page (/) says what Quiet Week is, who it is for and how it works. The tool lives at /check, in three
numbered steps:
- Your game. Name 1 to 3 well-known games whose players look like yours, and your launch window. No game yet? Open one of the ready-made examples from the same screen.
- Confirm the matches. An AI agent finds the games in Qloo, fixes typos and checks Qloo has audience data for each one; you confirm each match (with Qloo's cover images) before anything is measured. A plain checklist and a live log show what the agent is doing.
- Your weeks. The answer comes first: one sentence ("Move away from GTA VI's launch" or "No need to dodge GTA VI"),
the GTA VI player overlap on its 0 to 1 scale, the three best weeks and the two to avoid. Then:
- Week by week: 21 weeks, one bar each; height and colour show how much that week's big releases (games and films) share your players. Click a bar to see each release, its score, the earlier game it was measured through, and the source of its date.
- How to read this: the 0 to 1 score, the colour rule and the limits, explained once.
- Sources & method: every release with its score and date source, the exact Qloo calls (entity ids and parameters), and every step the agent took.
A bar at the top of the result always says what you are looking at (an example or your game, and its audience).
Results have their own URL (/check?example=action-western); your own result is kept in the tab, so reload and the
browser's back button work.
Three examples are precomputed and load instantly: cozy farming game (like Stardew Valley), action western (like Red Dead Redemption 2), metroidvania (like Hollow Knight).
Base URL https://hackathon.api.qloo.com, header X-Api-Key, server side only.
| Step | Call |
|---|---|
| Find a game | GET /search?query=<title>&types=urn:entity:videogame&take=5 |
| Check Qloo has audience data for it | GET /v2/insights?filter.type=urn:entity:videogame&signal.interests.entities=<id>&filter.results.entities=<4 well-known games> |
| Score every upcoming game | GET /v2/insights?filter.type=urn:entity:videogame&signal.interests.entities=<your 1-3 games>&filter.results.entities=<up to 25 release ids>&take=25 |
| Score every upcoming film | same call with filter.type=urn:entity:movie |
Qloo returns a query.affinity between 0 and 1 for each entity it can score and leaves the others out. A full
analysis takes 4 insights calls (games and films, own entities then stand-ins), about 1 second each.
Stand-ins. Qloo does not score games or films before they come out (GTA VI has an entity but no affinity), so each upcoming release in the calendar declares a stand-in, usually the previous game or film of the series (GTA V for GTA VI, Metroid Dread for Metroid Ravenous, Dune: Part Two for Dune: Part Three). The release's own score is always used when Qloo has one (Pikmin 4, Metaphor: ReFantazio, Avengers: Doomsday). The stand-in is named next to every number it produces.
A Gemini model (gemini-3.5-flash-lite, falling back to gemini-3.5-flash) drives the work through function calling:
- Resolver tools:
search_games,check_audience_data,propose_matches. It retries with a corrected title when a search fails and explains each pick in a short note. - Analyst tools:
get_release_calendar,measure_overlap,get_stand_ins,measure_stand_ins,build_calendar,submit_verdict. - Its steps stream to the screen as it works (search, check, stand-in, measure, compute, write).
- It may only quote Qloo's numbers.
submit_verdictrejects any decimal not returned by Qloo, any week or date the tools did not return, and any word about sales, revenue, wishlists, percentages or forecasts. The agent then rewrites; if it still fails, the app shows a verdict assembled from the numbers instead. - Privacy. The model only sees game and film titles, dates and Qloo numbers. Your one-sentence pitch is never sent to it. Quiet Week uses Gemini's free tier, so Google may use these requests to improve its models.
The colours and the week picks are computed by plain code (lib/calendar.ts), not by the model, so they are
reproducible and tested.
A week turns red when a game released that week has a Qloo affinity of 0.75 or more with your players, yellow from 0.68 (or a film from 0.78; films never turn a week red), green below that, and grey when no tracked release lands that week.
Chosen from the real scores of the three examples: games ranged from 0.44 to 0.90 with half between 0.55 and 0.67, so 0.68 marks the top quarter and 0.75 the clear outliers. Blockbuster films score 0.63 to 0.81 for every audience tried, so their score says more about their reach than about a shared niche.
Recommended weeks have no Steam storewide sale or Next Fest and no game at 0.68 or more within two weeks either side.
data/releases.json: 34 releases (23 games, 11 films) from November 2026 to March 2027, plus three Steam events
(Winter Sale, February Next Fest, Spring Sale) from Valve's Steamworks documentation. Every entry was typed by hand
from a public source (publisher site, platform store, press, Cinema United's release calendar) and has its link and
check date. Nothing is scraped. Edit the three source files in data/ and run python3 -I scripts/merge-releases.py.
- Affinity is not sales. It is Qloo's 0 to 1 score of how much two audiences like the same things, not a share of players. Quiet Week makes no sales forecast.
- Stand-ins are approximations, and one is a genre stand-in (Graveyard Keeper for Grave Seasons, a new game).
- Dates move. They were checked on October 23, 2026.
- Small games have little Qloo data, which is why you describe your players with well-known games.
- Football Manager 27 has no score (Qloo does not score its predecessor either).
cp .env.example .env.local # then fill QLOO_API_KEY and GEMINI_API_KEY
npm install
npm run dev # http://localhost:3410
npm test # unit tests: colour rule, calendar, week picks, stand-ins, agent text check, result wording
node scripts/ux-shots.mjs http://localhost:3410 # screenshots of every screen + layout checks (examples only, no API call)
npm run check:live # runs the 3 examples end to end against the real APIs
npm run precompute # regenerates data/examples/*.jsonNext.js (App Router) and TypeScript, no database. Qloo answers are cached on disk (.cache/qloo). Keys stay on
the server (API routes app/api/resolve and app/api/analyze). Ready for Vercel; on Vercel the cache lives in /tmp.
MIT. Game and film titles are cited as facts; cover images come from the Qloo API.