Subtle Beacon is Iridae’s adaptive experimentation system. It helps studios test ideas, learn from player behavior, and roll out changes with confidence, even when the data is incomplete and the right decision is still taking shape.
Most experimentation platforms assume a clean workflow: define a test, split traffic, wait, decide. In practice, features are already live, behavior is continuously observable, and decisions need to evolve before the test feels “finished.” Subtle Beacon is designed for that reality.
What Subtle Beacon Is
Subtle Beacon is a decision system for continuous learning under uncertainty.
It treats player preferences and experiment outcomes as evolving probability distributions rather than fixed point estimates. As new evidence arrives, it updates what the system believes about each variant, how likely each option is to be best, and how much uncertainty still remains around that conclusion.
That gives studios a much more useful basis for action than a binary “significant” or “not significant” result. Instead of waiting for artificial certainty, teams can see when a signal is becoming strong, when it is still fragile, and when the expected cost of waiting is higher than the expected cost of moving.
How It Works in Practice
Subtle Beacon assumes that products are already instrumented and that behavior is the most honest expression of preference.
That means learning can happen continuously through rollout decisions, traffic allocation, feature exposure, pricing changes, offer design, onboarding variants, and other live product choices. The system updates as those signals arrive, helping studios decide whether to keep exploring, shift traffic toward stronger options, stop an experiment, or hold off because the current evidence is still too ambiguous.
In practice, it supports both straightforward testing and deeper preference modeling. It can compare variants in live experiments, but it can also model tradeoffs, helping teams understand not just which option is winning, but which attributes players actually value and where preferences differ across segments.
What It Enables
Because Subtle Beacon is built as a continuous learning layer, it supports more than traditional A/B testing.
Studios can use it to:
- Evaluate feature, UX, and onboarding changes while data is still arriving
- Adapt traffic allocation toward stronger options without giving up exploration too early
- Measure whether a result is truly decision-ready, not just numerically ahead
- Learn which combinations of product attributes players prefer most
- Understand where different player groups are aligned, divided, or shifting over time
That makes it useful both for fast-moving live decisions and for more deliberate product and design work upstream.
How It Powers the Iridae Stack
Subtle Beacon is one of the core decision systems in the Iridae platform.
It draws on Latent Spark for richer context about what a game or concept is, how comparable different titles really are, and how much uncertainty surrounds that representation. Its outputs can then inform the rest of the stack, feeding stronger preference signals into forecasting, planning, and strategic evaluation.
Because it lives inside a shared uncertainty-aware architecture, experimentation does not get isolated from the rest of decision-making. Preference learning, forecasting, and strategic planning all operate from the same language of evidence and confidence.
Why It Matters
Studios rarely fail because they are not testing enough. More often, they struggle because evidence arrives unevenly, decisions cannot wait for perfect clarity, and legacy experimentation tools force a false separation between learning and doing.
Subtle Beacon closes that gap. It helps studios treat action as a way of asking questions, treat behavior as a way the market answers, and make decisions that reflect both momentum and uncertainty.
That is what makes it more than an experimentation dashboard. It is a live decision layer for studios that want to learn faster without pretending the world becomes certain just because the sample size got bigger.