Research simulation · fictional money · no deposits, prizes or visitor wagers
Virtual Casino ResearchDownload data ↓

Protocol · Code · Ledger · Replay

How the experiment was built and checked

The definitions were fixed before the results were interpreted. Any change to a rule, pay table, cohort, bug or displayed definition requires a new version.

Release identity

v1.0.0

Experiment
greyscienx-virtual-casino-v1.0-local
Configuration SHA-256
0fa5fcd22a919694dbfb3d91b01462d73418a817aab845912e037ebf955a461d
Manifest SHA-256
1e3dc6ed3d16fffe9cd4d79f330a1c9aa2dcf6c35d06b5bbaef77c74b57f053e
Final ledger hash
cd7660df0ce7ac2f22b56c222e9a247aa8f8746441ea4f55a448fc8ee94cf290
Final state hash
be45c461564d6532bfaa55be47c56898a8424aa85bd6408eedac7761564d82da

Seed integrity

Committed, then revealed

Commitment
d08c24f5d9ba3a720099b17a1917e0a5774b5fdf9c8da52d0008b66a3732c163
Revealed seed
greyscienx-vc-visible-world-0001-revealed
Status
revealed after local run
Code revision
ba0e521dd6232895f01398894a8a41ee890f6a5a+dirty

Terms used in the study

The definitions stayed fixed

Every percentage identifies its denominator. Expected values are calculated wager by wager and stored in integer cents.

Handle
Sum of every accepted stake.
Player payouts
Sum of gross amounts returned after settlement.
GGR
Handle minus gross payouts; allocation injections are excluded.
Actual RTP
Gross payouts divided by handle in the realised ledger.
Theoretical RTP
Stake-weighted expected gross payout from each exact wager choice.
Net P&L
Current bankroll minus all R10,000 allocations received.
Ruin
After settlement, bankroll is below the R1 minimum legal stake.
Bankroll month
A numbered allocation episode, separate from calendar month.

Questions set in advance

Nine hypothesis families

Exploratory comparisons are identified as such and require confirmation on held-out data.

  1. H1Exposure
  2. H2Frequency convergence
  3. H3Stake-size ruin
  4. H4Loss-chasing
  5. H5Limit-setting
  6. H6Equal-RTP volatility
  7. H7Winner persistence
  8. H8Replication
  9. H9Roulette rules

Deterministic streams

Concurrency cannot alter a draw.

Each schedule, game choice, stake variation, bet choice, and game outcome stream is addressed by stable world, player, game, and purpose identifiers. The xoshiro128** generator advances locally; SHA-256 protects commitments, manifests, checkpoints, and ledger integrity.

Changing a dashboard refresh or worker count therefore cannot shift the sequence used by another player.

Replay procedure

101 422 settlements rechecked.

npm run casino:test
npm run casino:generate
npm run casino:replay

The replay validates sequence and predecessor hashes, duplicate bet IDs, integer stakes, bankroll arithmetic, non-negative balances, allocation caps, final ledger/state hashes, and casino-to-player reconciliation.

Limits of the evidence

What this model cannot establish

  1. Synthetic agents are simplified behavioural models, not people.
  2. Results depend on the frozen policies, game abstractions, stake limits, and six-month horizon.
  3. Bankroll month is an allocation episode, not a calendar month.
  4. The resolved-hand blackjack model is auditable but is not a card-by-card shoe implementation.
  5. Rare jackpot estimates remain unstable even across 1,000 worlds.

Download the protocol, manifest, schema, aggregates, ledger and checksums →