How They Work (RNG + skill check)

PA Skill Machines • Chapter 2

How They Work: RNG + Skill Check

This chapter shows the full life cycle of a round: seeding and sampling by the RNG, display of a resolvable state, the player’s skill input, and deterministic settlement. It also covers logging, fairness, latency, anti-tamper, and operator configuration.

Read time: 25–30 min
Technical
v1.0

Overview

PA Skill outcomes are a function of two inputs: a random sample and a player action. The system must keep these inputs separable, auditable, and stable across time.

Principle 1: Independence

Each round’s sample is independent of past rounds except where a title explicitly encodes state. No “hot/cold” biasing.

Principle 2: Determinism

Given the base draw and a specific player action, the settlement must be deterministic and reproducible.

Principle 3: Transparency

Players can see how their action changed the base outcome. Logs capture enough to prove it.

RNG Fundamentals

The RNG produces uniformly distributed samples over a defined state space. Implementation details vary, but the goals are uniformity, unpredictability, and testability.

Common Implementations

  • Cryptographic PRNGs for unpredictability and resistance to prediction.
  • PCG/Xoroshiro families for speed with proven statistical qualities.
  • Hardware TRNG mixed into seeds to improve entropy.
A compliant implementation documents period, seeding, sampling method, and test batteries used during certification.
RNG pipeline diagram
Sampling maps RNG outputs to symbols, tiles, or numbers without bias.

Seeding & Entropy

Seeds initialize PRNG state. Good seeding prevents repetition and prediction.

Recommended Practices

  • Combine TRNG bytes, monotonic clock, and device-unique IDs through a KDF.
  • Reseed periodically and on critical events (boot, firmware update, maintenance door).
  • Never expose raw seeds. Expose hash commitments for audit if desired.

Commit-Reveal (Optional)

System logs Note: Seed predictability or reuse across machines is a critical red flag.

From Sample to Display State

RNG output is transformed into a visible, interactive state. The mapping must be one-to-one or many-to-one but never biased.

State TypeSampling MethodBias PitfallMitigation
Symbol GridsIndex via cumulative distributionRemainder bias in moduloUse rejection sampling or 128-bit range
Hidden TilesFisher–Yates shufflePoor shufflesUniform shuffle, test with dieharder
Number DrawsReservoir samplingDuplicates or order skewDistinct sampling without replacement

Skill Check Taxonomy

Recognition

Find the optimal move: swaps, merges, or target selection within a timer.

Precision

Tapping or dragging with tolerance. Hit-boxes and debounce prevent randomness.

Planning

Choose among options with different EV/variance trade-offs.

The skill prompt must be outcome-relevant. If correct actions rarely change settlement, skill is cosmetic.

Round Lifecycle

  1. Stake: Player selects wager. Limits visible.
  2. Sample: RNG draws base state. Hash commitment optional.
  3. Render: State is displayed. Timer and controls appear.
  4. Resolve: Player action transforms state according to rules.
  5. Settle: Deterministic payout. Ticket/credit updated.
  6. Log: Seed hash, input, timestamps, state A→B, payout.
Lifecycle base state
Base state
Lifecycle resolved state
Resolved state

EV, Mastery, and Bounds

Skill raises expected value within a bounded range. Designs must prevent trivial infinite-EV loops.

EV Model

Let E₀ be EV without action, Δ be improvement from optimal action, and p be probability the player performs the optimal action. Then EV = E₀ + p·Δ.

Design Guards

  • Single action per round or diminishing returns.
  • Timer tuned to human ability, not superhuman macros.
  • Payout tables calibrated to maintain house margin at mastery cap.
Operator tip: Track aggregate p over time. Rising mastery should shift results within expected bounds, not break them.

Latency, Input Windows, Accessibility

Latency Budget

  • Touchscreen scan: <10 ms
  • Game loop + render: <20 ms
  • Input processing + settle: <10 ms

Target <50 ms end-to-end for perceived immediacy.

Accessibility

  • Adjustable timers and speed.
  • Clear focus outlines and large tap targets.
  • Audio-visual cues for success/failure.
If human reaction limits dominate outcomes, the game drifts from skill into reflex lottery. Balance timers for cognition, not twitch.

Audit Logs & Verification

FieldDescriptionPurpose
Round IDMonotonic unique identifierTraceability
Seed HashH(seed) at sample timeAnti-manipulation proof
State ACompact encoding of base drawReconstruction
ActionPlayer input and timestampDeterminism check
State BResolved state after actionOutcome proof
PayoutCredits or ticket amountAccounting
Store logs securely. Use append-only or signed ledgers for high-assurance venues.

Anti-Tamper & Integrity

  • Secure boot and code signing on firmware.
  • Encrypted storage of RNG state and logs.
  • Intrusion sensors on doors with event logs.
  • Remote attestation of hashes after updates.
  • Watchdogs for clock drift and frame pacing anomalies.
Red flags: Non-signed updates, shared seeds across units, inconsistent clocks, or missing round logs.

Operator Configuration

Site Setup

  • Stable power, UPS, and grounded outlets.
  • Network hardening if remote reporting is enabled.
  • Low-glare placement and clear signage.

Settings

  • Wager tiers and limits per venue policy.
  • Timer presets matched to audience skill.
  • Maintenance windows with automatic integrity checks.
Document every change. Configuration drift is a common cause of disputes.

FAQ

Does the RNG change based on player wins?

No. Proper systems maintain independent, stationary distributions. Perceived streaks are variance.

How do I verify the skill is determinative?

Replay a round from logs. Apply different actions to State A and confirm settlement changes as specified.

Can latency make outcomes random?

Excess latency can blur skill. Keep the end-to-end input pipeline under ~50 ms and provide larger hit-boxes.

Is commit-reveal mandatory?

No, but it adds assurance. Hash commitments enable third-party audits without exposing seeds.

Resource Pack

Place assets under /wp-content/uploads/paskill/. Keep filenames consistent with links above.

Next Chapters