Answers for studios
evaluating Anchored Horizon.

Anchored Horizon is a hosted Bayesian commercial forecasting system for game studios. It retrieves comparable titles, builds ESS-scaled priors from their outcomes, fits a hierarchical hit-mixture model, and returns a posterior predictive distribution—draws, quantiles, hit probability, and exceedance chances—scoped to a planning horizon, not a single midpoint.

Producers, finance, analytics, portfolio, and strategy leads who need commercial expectations they can defend: greenlights, wishlist targets, delay and scope tradeoffs, publisher conversations, and capital planning. It fits premium and on-platform gross planning windows; live-service F2P and off-platform revenue need different treatment and are called out honestly in scope.

Studios still plan against comps decks and false-precision spreadsheets that hide uncertainty, age immediately, and cannot answer “what must be true?” Anchored Horizon keeps a living, time-scoped belief about outcomes—anchored in comparable games, honest about evidence strength, and usable for scenarios and inverse planning.

No. It runs stand-alone with local or static similarity modes. Latent Spark deepens comps, Subtle Beacon feeds experiment evidence, and Patient Cartographer consumes path horizons—but none are required to produce a distribution-first forecast.

Comps slides pick titles by narrative; spreadsheets often invent ±% bands. Anchored Horizon retrieves comps in game-embedding space, weights them by similarity, recency, and quality, and returns a calibrated posterior—with ESS and explanations—so the range means something statistically.

Point models optimize a midpoint and bury uncertainty. Anchored Horizon’s primary output is a predictive distribution (including hit-regime mass), so planning questions map to probabilities—“how likely is ≥ $X?”—not a single expected value that looks more precise than the evidence supports.

BI tools excel at historical reporting and simple extrapolations. Anchored Horizon is built for pre- and early-launch commercial belief under sparse outcomes: comparable-game priors, hit mixtures, horizon anchors, and inverse solves for planning targets.

No. It supplies the uncertainty-aware commercial belief those packs should start from—distributions, scenarios, and evidence strength—while finance still owns contractual waterfalls, accounting definitions, and board narrative.

Posterior predictive draws, planning quantiles, hit probability, and exceedance probabilities for a scoped outcome (entity × horizon × anchor). You also get evidence diagnostics (ESS, reliability flags) and explain payloads (comp influence, prior vs features).

Because the evidence is weak, sparse, conflicting, or the title is unusual in embedding space. Anchored Horizon would rather show a wide honest band than a tight false one. Stronger comps, signals, and outcomes narrow the range over time.

Premium / on-platform gross over a fixed post-launch window (profiles such as 12- or 24-month). F2P live-service economies and off-platform revenue are out of scope for that primary model—ask about alternate profiles if your slate needs them.

By asymmetric KL retrieval over Latent Spark Gaussian embeddings (or stand-alone similarity modes). Top-k comps are weighted by similarity × recency × quality, with diversity guards so one cluster cannot dominate the prior.

Every forecast is tied to a horizon window and an anchor (launch, as-of date, or fiscal year end)—not a timeless number. The same title can carry multiple scoped beliefs as planning questions change.

Time-stamped wishlist, engagement, and revenue signals build trajectory features—slope, momentum, volatility, velocity—in pre- and post-launch modes. As panels refresh, features stay current and belief can be re-forecast under the same session controls.

Yes. Inverse constraint solve finds inputs (for example wishlist level) that achieve an exceedance, median, or quantile target at a stated probability—so plans negotiate against the distribution, not a hoped-for base case.

Rolling-origin backtests with CRPS, log score, coverage, and PIT—plus per-horizon segments. Models promote from candidate to active only when statistical gates pass; rollback remains available.

Subject features, a clear revenue definition, and access to a comparable-game inventory (via Latent Spark or stand-alone maps). Signal ingest deepens trajectories; historical outcomes improve calibration over time.

The primary premium/on-platform windowed model is not a full F2P live-ops economy simulator. Use Anchored Horizon for scoped commercial questions it is calibrated for, and pair with experiments and pathway planning for live-service levers.