Swiv Privacy Prediction Protocol

1. Overview

Swiv is a privacy-preserving, precision-based prediction market built on Solana using Anchor and MagicBlock Private Ephemeral Rollups. Unlike binary prediction markets (Yes/No), Swiv allows users to submit continuous numerical predictions (e.g. price levels, values, metrics), producing richer and more accurate market signals.

The protocol separates private execution from public settlement:

• Sensitive user actions happen inside a private ephemeral rollup
• Only finalized, aggregated results are committed to Solana L1

Frontend Repo:

https://github.com/swiv-labs/Frontend

Backend Repo:

https://github.com/swiv-labs/Backend

Demo Video:

https://youtu.be/IB5Ptia9SeI?si=g1wYQDfLLjf_di2s

---

2. Core Architecture

2.1 On-chain Program

• Written in Rust (Anchor framework)

• Deployed on Solana

• Manages:

  • Pool lifecycle
  • PDA ownership
  • Vault accounting
  • Reward distribution
  • Protocol fees

2.2 Private Execution Layer (MagicBlock)

• Enabled via #[ephemeral]

• Handles:

  • Private bet placement
  • Private prediction updates
  • Weight calculations

• Prevents:

  • Front-running
  • Wallet tracking
  • Strategy leakage

2.3 Public Settlement Layer

• Final states are flushed back to Solana via:

  • Pool undelegation
  • Bet undelegation

• Enables permissionless reward claiming

---

3. Protocol Lifecycle (End-to-End Flow)

3.1 Protocol Initialization (Admin)

initialize_protocol(protocol_fee_bps)

• Creates the global protocol config

• Sets:

  • Admin authority
  • Treasury wallet
  • Protocol fee (basis points)

• Executed once

---

3.2 Pool Creation

create_pool(
  pool_id,
  name,
  metadata,
  start_time,
  end_time,
  max_accuracy_buffer,
  conviction_bonus_bps
)

Each pool defines:

• A prediction window (start_time → end_time)
• A numerical outcome range
• Accuracy tolerance (max_accuracy_buffer)
• Conviction incentives

The pool PDA:

• Owns a token vault
• Tracks total stake, weights, and resolution state

---

3.3 Pool Delegation to TEE (Privacy Activation)

delegate_pool(pool_id)

• Transfers pool PDA authority to MagicBlock TEE

• From this point:

  • Pool state becomes privately mutable
  • No public on-chain writes for user actions

This step is required before resolution.

---

3.4 Private Bet Lifecycle (Users)

3.4.1 Initialize Bet

init_bet(amount, request_id)

• Creates a bet PDA
• Locks user stake into pool vault
• Stores encrypted metadata in the rollup

3.4.2 Place / Update Prediction

place_bet(prediction, request_id)
update_bet(new_prediction)

• Prediction value remains private
• Users may update predictions before pool expiry
• Update count affects conviction bonus

---

4. Pool Resolution & Settlement

4.1 Resolve Pool (TEE-only)

resolve_pool(final_outcome)

• Called inside MagicBlock TEE
• Marks pool as resolved
• Stores the final numerical outcome
• Enables weight computation

No user data is revealed on-chain at this stage.

---

4.2 Weight Calculation (Batch, Private)

batch_calculate_weights()

• Admin passes all bet PDAs as remaining accounts
• Each bet weight is calculated privately
• Results are written into bet accounts

This avoids per-user transactions and preserves privacy.

---

5. Weight Calculation Model (Core Math)

Swiv’s reward system is stake-weighted and accuracy-driven.

5.1 Accuracy Score

calculate_accuracy_score(prediction, result, buffer)

• Measures closeness to final outcome
• Linear decay inside buffer
• Outside buffer → score = 0

Formula:

accuracy = 1 - (|prediction - result| / buffer)

Scaled by MATH_PRECISION = 1_000_000

---

5.2 Time Bonus

calculate_time_bonus(start, end, entry_time)

• Rewards earlier participation
• Longer commitment → higher multiplier

Earlier bets receive higher influence.

---

5.3 Conviction Bonus

calculate_conviction_bonus(update_count)

• No updates → higher conviction bonus
• Multiple updates → neutral weight

Encourages confidence, not constant adjustment.

---

5.4 Final Weight Formula

weight =
stake
× accuracy
× time_bonus
× conviction

Scaled down by precision constants:

final_weight = raw_product / P³

This produces a fair, manipulation-resistant influence score.

---

6. Finalization & Public Settlement

6.1 Finalize Weights (On-chain)

finalize_weights()

Requirements:

• Pool must be resolved
• Weights must not be finalized already

Actions:

• Deduct protocol fee
• Lock distributable vault balance
• Emit final settlement event

Once called:

• Pool becomes immutable
• Claiming is enabled

---

6.2 Flush State Back to L1

batch_undelegate_bets()
undelegate_pool()

• Writes finalized bet + pool data to Solana
• Ends private execution phase

---

7. Reward Claiming (Users)

claim_reward()

• Permissionless

• Based on:

  user_weight / total_pool_weight
  

• Transfers tokens directly from pool vault

No admin trust required.

---

8. Emergency Handling

emergency_refund()

• Used if pool cannot be resolved
• Returns stakes proportionally
• Prevents fund lockups

---

9. Key Guarantees

• Privacy: Predictions never hit public mempool
• Fairness: No front-running or copy trading
• Precision: Not limited to binary outcomes
• Verifiability: Final state is fully on-chain
• Scalability: Batch operations reduce gas costs

---

10. Summary

Swiv introduces a new category of prediction markets by combining:

• Precision-based numerical forecasting
• Private ephemeral execution
• Trustless on-chain settlement

By leveraging MagicBlock Private Ephemeral Rollups and Solana’s PDA model, Swiv delivers privacy without sacrificing decentralization or verifiability.