| name | steam-launch-forecast |
| description | Use when a user needs a Steam game launch forecast, wishlist-to-sales read, comparable-game analysis, regional demand split, or market signal review for first-week or first-month performance. |
Steam Launch Forecast
Use this skill to produce a market-research style forecast for Steam launch
performance. It is strongest when the user provides a Steam page, wishlist
count, launch date, price, publisher, social links, trailer data, or a list of
comparable games.
Workflow
- Parse the target game or game set.
- Names, release dates, prices, wishlist counts, publishers, regions, notes.
- Treat relative dates using the current date and verify fast-moving data
online.
- Set the forecast boundary before interpreting signals.
- Declare
Forecast mode: blind_prelaunch, retrospective_backcast, or
post_launch_nowcast.
- Record a
Knowledge cutoff: the latest instant at which a permitted input
was public or available. Only blind_prelaunch is eligible to be described
or scored as a blind forecast.
- A date-only source captured on release day is not blind-prelaunch evidence
unless its order relative to the release is demonstrated with a timestamp
and IANA time zone.
- Build a game profile.
- Genre, tags, production scale, IP status, localization, demo/Next Fest
history, publisher credibility, and main audience regions.
- Collect permitted signals.
- Steam page, SteamDB followers, wishlist rank when visible, discussions,
reviews/CCU if released, launch discount, supported languages.
- YouTube, Bilibili, Reddit, Discord, creator coverage, press, and community
heat when relevant.
- Mark each material input as verified, provisional, unverified, or disputed;
exclude disputed inputs from the estimate.
- In
blind_prelaunch mode, do not use post-launch reviews, CCU, sales, or
any signal published after the knowledge cutoff.
- Build a dynamic comparable set.
- Use 3-7 recent comparables, preferably within 6-24 months.
- Match genre, price, scale, audience region, visibility path, and launch
condition.
- Model the funnel.
- Wishlists are an input, not the answer.
- Adjust by wishlist freshness, genre conversion, price, review risk,
Steam visibility, creator coverage, localization, and regional split.
- If calculating
first-week units / prelaunch wishlist snapshot, call it a
snapshot-normalized sales ratio, not cohort conversion.
- Output a forecast.
Required Output
Game:
Launch status:
Forecast mode:
Knowledge cutoff:
Evidence tier:
Calibration eligibility:
Known inputs:
Comparable logic:
Funnel read:
First-week forecast:
Confidence:
Main risks / upside triggers:
For multiple games, add a ranking table first.
Use the calibration contract when the
request compares forecasts with outcomes, audits a prior call, or adds a result
to a calibration dataset.
Rules
- Prefer ranges over false precision.
- Say what data is missing.
- Separate China-facing and Western/global demand when evidence supports it.
- Do not make investment or publishing claims from wishlist count alone.
- Keep the forecast issue time separate from the knowledge cutoff and cite the
source and capture time for material inputs.
- Call a released-game analysis
post_launch_nowcast or
retrospective_backcast; never present it as a blind forecast.
- Do not call an estimate unbiased without a stated benchmark and error method;
a post-launch nowcast is never an unbiased blind forecast.
- Label calibration provisional when actuals are not verified or the sample is
too small to generalize.
Common Failure Modes
- Converting wishlists to sales with one fixed multiplier.
- Using old comparables when the genre or Steam visibility environment changed.
- Treating Western press silence as global demand weakness for China-facing games.
- Inventing private wishlist rank, revenue, or publisher data.
- Calling a same-day, date-only snapshot pre-launch without proving the time
order.
- Letting post-launch reviews or CCU leak into a claimed blind forecast.
- Calling a snapshot-normalized sales ratio a conversion rate for a defined
wishlist cohort.
