Raman Horse

Raman Horse is a solo final-year research project by Raman Prasad (Bachelor of Science Computer Science & Data Science, Massey University). The application functions simultaneously as a programming demonstration and as a controlled behavioural experiment in sequential choice under uncertainty.

Core research question

How do people allocate limited resources across sequential risky options when the environment is transparent (visible probabilities, multi-attribute form, recent performance signals) yet still contains irreducible noise and a structural house edge?

What is being observed

• Selection among alternatives relative to displayed probability and current form
• Stake sizing as a proportion of remaining bankroll
• Thresholds at which participants elect to risk ruin (all-in)
• Behavioural response to winning and losing streaks
• Whether strategy shifts when race distance re-weights the importance of different horse attributes (speed vs stamina framing)

Experimental design rationale

Each horse is defined by six latent attributes (speed, acceleration, stamina, consistency, late finish, early start). Race distance alters the relative weighting of those attributes, creating genuine skill–luck trade-offs rather than pure noise. Displayed odds are not arbitrary: they are the empirical win frequencies obtained from 3,000 Monte-Carlo realisations of the generative model under the chosen distance, then scaled by a controlled 10% house edge. Form is a slowly drifting latent state so that recent outcomes carry predictive information. Photo finishes deliberately introduce near-miss experiences. Permanent termination after an all-in loss creates a true absorbing state analogous to the classic gambler's-ruin problem.

Theoretical anchors

The design is informed by Prospect Theory and probability weighting (Kahneman & Tversky), the house-money effect (Thaler & Johnson), loss aversion, and ecological approaches to risk. A central interest is whether these regularities still appear when participants know the currency is fictional and most sessions can be restarted — except after the irreversible all-in outcome.

Kahneman & Tversky (1979); Thaler & Johnson (1990); subsequent work on dynamic risk-taking and near-miss effects.

Data handling & ethics

If consent was given, only anonymous behavioural traces from the present browser session (choice sequences, relative stake sizes, streak context, final outcomes) may later be aggregated for analysis. The application itself collects no names, emails, IP addresses, device identifiers, or any other personal data. This is educational research conducted as a solo project; it is not a commercial product and involves no real-money gambling.

You may close this panel and continue playing at any time. The study exists to understand human choice; the game exists so that those choices feel consequential.