Skip to main content

The End of the Casino. The Beginning of the Truth.

This whitepaper maps the structural rotation from leveraged perpetual futures into prediction markets, quantifies why manual retail trading fails under latency-driven alpha decay, and specifies BETTER’s deterministic trading engine and bounded inference stack.
All monetary amounts are denominated in United States dollars unless stated otherwise.

Executive summary

Crypto is undergoing its most significant structural rotation since the decentralised finance cycle of 2020. For the past half-decade, retail capital has concentrated in leveraged perpetual futures. That era rewarded leverage, liquidation mechanics, and microstructure games. Capital is migrating toward prediction markets where settlement is anchored to verifiable outcomes. The venue changes, but the structural disadvantage persists. Settlement can be fair while trading remains unequal. The advantage shifts from liquidation engines to latency. On-chain analysis indicates that approximately 89 per cent of retail participants on Polymarket realise net losses. The primary reason is not settlement. It is alpha decay. BETTER is a liquidity abstraction layer that packages institutional-grade execution into a retail-accessible system:
  • Terminal: real-time signal feed, execution context, and copy-trading tooling.
  • Vaults: pooled USD Coin on Base that executes strategies on behalf of depositors.
  • Execution rails: Rust-first, in-region and co-located routing.
  • OpenServ agents (BRAID): bounded reasoning DAGs and machine-checkable decision records.

Key quantitative claims (internal telemetry)

  • More than 10 million signals collected in approximately five months.
  • 10,000 to 40,000 elite-wallet trade setups per day.
  • Approximately 0.6 milliseconds tick-to-trade, measured from mempool read to transaction hash on Base, then to order finality on Polymarket (Polygon).
  • The system filters approximately 40,000 distinct on-chain signals per day.
These figures vary with market conditions, venue availability, and network performance.
Earlier drafts of this thesis referenced a December 18 and January 8 token generation event and an access gate of 1,500 tokens. The current live parameters are documented in this repository, including a 21 January 2026 token generation event and a fully diluted valuation ratchet gate. The older figures remain included here as historical references.

Part I: The erosion of trust and the failure of the perpetual casino

1.1 The illusion of price action

Perpetual futures promise price action. Retail traders believe chart reading, support and resistance, and macro narratives are sufficient to profit. In practice, they trade inside a venue optimised for professional liquidity providers.

1.2 The mechanics of liquidity pulling

The dominant extraction pattern is liquidity pulling. In this pattern, professional participants target the weighted average liquidation price of a leveraged cohort. The process operates in four phases:
  1. The bait (accumulation): price drifts into a technical buy zone. Retail enters with leverage, typically 10x to 50x.
  2. The trap (identification): algorithms estimate the cohort liquidation band.
  3. The pull (withdrawal): buy-side liquidity is removed in a synchronised event.
  4. The harvest (the wick): price snaps into liquidation levels, triggering stops and forced liquidations, then mean reverts.
One observed proxy for this behaviour is the frequency of price wicks greater than 3 per cent within 1-minute candles coinciding with liquidation volume.

1.3 The regulatory smoking gun

In late 2024, United States law enforcement ran a public operation using a decoy token (NexFundAI) to test whether market participants would offer manipulation services. The result validated that market manipulation can be offered as a service.

1.4 Casino fatigue

Over time, participants experience casino fatigue. They become less willing to accept microstructure outcomes that feel disconnected from fundamentals. This creates demand for markets that settle on facts.

Part II: The migration to truth and the rise of prediction markets

2.1 Binary settlement

A prediction market contract pays US$1.00 if an event resolves YES and US$0.00 otherwise. Settlement is enforced by an oracle resolution system. Unlike perpetual futures, a market maker cannot wick a real-world outcome.

2.2 The hard floor of value

If a trader holds a winning share to maturity, the payout is mathematically fixed at US$1.00. This creates a hard floor of value that does not exist in floating-index derivatives.

2.3 The brutal reality check

Despite fair settlement, retail traders still lose. Approximately 89 per cent of retail participants on Polymarket realise net losses. The explanation is the gap between settlement fairness and microstructure fairness.

Part III: The physics of alpha decay

3.1 Alpha is a melting ice cube

In prediction markets, alpha is a short-lived pricing error created by information asymmetry. In the digital age, information propagates quickly. Alpha decay is measured in milliseconds.

3.2 The timeline of a trade: human versus machine

Scenario: a breaking-news message confirms that an outcome has become far more likely.
  1. The event (T = 0.000s): information is published.
  2. Machine reaction (T = 0.010s): automated systems ingest the event, compute the new probability, and place orders.
  3. Price discovery (T = 0.100s): the order book reprices and the alpha window collapses.
  4. Human reaction (T = 30.000s): the retail trader receives a notification.
  5. Retail execution (T = 45.000s): the retail trader opens the venue and buys at the post-reprice level.
The retail trader buys stale odds. The machine sells into that demand.

3.3 Co-location and execution priority

The gap is not a skill gap. It is physics. Professional execution uses:
  • co-located or in-region routing
  • direct smart contract interaction and sequencer-aware submission
  • priority mechanisms, including MEV-style ordering, where applicable

3.4 The fallacy of manual copy trading

Manual copy trading degrades under slippage. Example:
  1. Master trader buys at US$0.40.
  2. Copy system reacts 2 to 5 seconds later.
  3. Copier buys at US$0.45.
  4. Both sell at US$0.50.
Profit comparison:
  • Master: US$0.10 per share, 25 per cent.
  • Copier: US$0.05 per share, 11 per cent.
Fees and gas consume the degraded edge.

3.5 Behavioural taxes

Retail participants also pay behavioural taxes:
  • Long-shot bias: buying contracts priced at 1 cent or 2 cents hoping for 100x, overweighting small probabilities (treating 1 per cent like 5 per cent).
  • Wishcasting: betting on desired outcomes rather than evidence.
  • Duration mismanagement: holding to maturity and absorbing tail risk and lock-up opportunity cost.

Part IV: The elite tier

If the majority loses, a minority captures the alpha window. On-chain behaviour clusters into three archetypes.

4.1 Information-driven flow

Case study: the 2024 Nobel Peace Prize market.
  • A wallet identified as 6741 placed a US$50,000 position approximately nine hours before the official announcement.
  • The implied probability shifted from 3 per cent to 70 per cent shortly after.
The BETTER approach does not rely on possessing inside information. It detects the footprint of informed flow and replicates it within the same latency regime.

4.2 Endgame sweeping

When an outcome becomes near-certain, contracts often trade at a discount (for example US$0.96 or US$0.97) due to temporary liquidity gaps. Automated systems sweep available shares and redeem at US$1.00, capturing an approximately 4 per cent return over minutes.

4.3 Cross-venue probability alignment

Sports and macro markets exhibit an odds-consensus across external venues. A simple alignment trade exists when one venue implies 66 per cent and the prediction market trades at US$0.60.

Part V: BETTER, a structural solution to the inequality of speed

5.1 What BETTER is

BETTER is not a signal group. It is not a chat room. It is a liquidity abstraction layer between user capital and prediction market execution. The thesis is simple: the alpha window is not accessible manually. Execution must be machine-mediated.

5.2 Technology: OpenServ agents and BRAID

BETTER uses OpenServ agents running BRAID (Bounded Reasoning for Autonomous Inference and Decisions). Each trade is gated by a machine-checkable reasoning DAG. Example gate:
  1. Ingest: wallet trigger detected with an Insider Score.
  2. Liquidity gate: can the trade enter without more than 2 per cent slippage?
  3. Risk gate: does correlation exceed 0.8 with existing exposure?
  4. Execute: submit via deterministic execution paths.
If the logic does not hold, the trade does not execute.

5.3 Infrastructure: speed as the product

The Terminal targets approximately 0.6 milliseconds tick-to-trade, defined as mempool read to transaction hash on Base, then to order finality on Polymarket (Polygon).

5.4 Filtering engine: separating signal from noise

Speed without curation is a fast way to lose money. BETTER filters approximately 40,000 daily signals and maintains a rotating pool of elite directional wallets. The Z-score engine weights features beyond simple profit and loss:
MetricWhy it matters
Sharpe ratioRisk-adjusted return; high volatility is penalised
Sortino ratioDownside-aware risk-adjusted return
Win rateConsistency across large samples
Profit per tradeScalping versus trend capture
Wallet age and interaction patternsInformed-flow detection and burner suppression

5.5 Vault abstraction

The Vault product packages this stack into one action: deposit USD Coin on Base, meet the $BETTER access gate, and allow the engine to rotate across qualified opportunities.

Part VI: Deterministic trading engine specification (current implementation)

The engine is split into two deterministic routes:
  • FAST15M: Rust-only for short-horizon up and down markets.
  • LONG: bounded inference via BRAID workflows and multi-LLM consensus, with deterministic execution and sizing.
Inference is advisory and bounded. Execution is deterministic and authoritative.

6.1 Contract model and notation

A binary contract pays US$1.00 if an event resolves YES and US$0.00 otherwise.
  • Let (\pi\in(0,1)) be the displayed executable ask price for YES.
  • Let (p\in(0,1)) be the engine’s subjective probability of YES.
  • Let (B) be the bankroll allocated to the engine.
  • Deployed notional is (x\in[0,B]), with fraction (f=x/B).

6.2 Core invariant: bounded inference, deterministic execution

The inference layer can reason freely, but it may only emit a bounded record with strict types. The deterministic core recomputes all numeric quantities and refuses any action that violates invariants. Malformed inference outputs are treated as abstentions.

6.3 Routing and persistent state

Markets route deterministically: Route(k) ∈ { FAST15M, LONG } Per-market state tracks book snapshots, spreads, depth proxies, time-to-expiry, wallet caches, inference bookkeeping, and current exposure.

6.4 Deterministic admissibility predicate

Admissibility is evaluated before any inference call, sizing, or order placement: Adm(k,t) = 1{Δt ≤ Tmax(Route(k))} · 1{SpreadBps ≤ Smax(Route(k))} · 1{Depth ≥ Dmin(Route(k))}

6.5 Wallet behaviour filter

Wallets are filtered deterministically to suppress non-directional flow. A wallet is flagged two-sided if it trades both outcomes inside a short window. Only directional wallets contribute to triggers.

6.6 Costs, effective price, and net edge discipline

All decisions use effective prices including conservative costs: πeff = πask + cfee + cslip(k,t) Net edge for buying YES is eYES = p − πeff. Trades require e ≥ τmin(Route(k)).

6.7 Kelly sizing with conservative multipliers and hard caps

For a YES buy, the cost-adjusted Kelly fraction is: f* = max(0, (p − πeff) / (1 − πeff)) Deployed fraction uses fractional Kelly and caps: fdeploy = λ · min(f*, fmax) Notional is x = B · fdeploy, then clipped by per-trade, per-market, and portfolio caps. No inference output can bypass caps.

6.8 FAST15M engine

FAST15M markets are handled exclusively by Rust logic.

Probability anchor

The engine computes a conservative driftless probability anchor (with uncertainty inflation): pup(t) = 1 − Φ( log(Sref/St) / (η·bσ·sqrt(Δt)) ), with η ≥ 1.

Permission and sizing

Trades require admissibility and edge. Sizing uses ultra-small fractional Kelly and a deterministic jump-risk haircut J ∈ (0,1): fdeploy = λ15m · J · min(f*, f15m_max)

Execution state machine

Orders follow a deterministic state machine:
  • Join: maker posting under stable depth and sufficient time.
  • Chase: bounded cancel-repost with deterministic cooldown.
  • Cross: taker execution only when edge exceeds a stricter threshold and time-to-expiry is small.
  • Abort: do nothing, log reason.
Client order ids are deterministic and idempotent.

6.9 LONG engine: BRAID graphs and multi-LLM consensus

LONG markets call bounded inference under strict throttles and budgets.

Bounded decision record

Each model emits a strict record. Narrative content is ignored. Example record fields:
ACTION=BUY|SELL|HOLD
OUTCOME=YES|NO
P_TRUE=0.xx
UNCERTAINTY=LOW|MED|HIGH
SIZE_MULT=0.xx
FLAGS=...
RATIONALE_HASH=...
Malformed records are treated as abstentions.

Consensus gate

A trade is permitted only with 3-of-4 consensus on direction.

Aggregation and dispersion penalty

The deterministic core aggregates probabilities using calibration weights, then applies a dispersion penalty that reduces deployed risk when agreeing models disagree.

Debouncing and budgeting

Inference is per-market debounced with deterministic triggers. A strict global budget is enforced. Budget exhaustion yields abstention rather than weakened constraints.

6.10 Exit and de-risking

As prices approach extremes, remaining upside is small while dispute risk, ambiguity, and capital lock-up dominate. The engine implements deterministic scale-out bands. One reference schedule uses bands at 0.90 and 0.95. This is designed to manage the common path where an entry near 0.40 compresses toward 0.95.

6.11 Storage, audit, and monitoring

Hot-path trading uses compact storage (no large JSON parsing). Every decision and fill is logged for replay. Monitoring includes realised versus assumed slippage, fill rates, abstention reasons, edge distributions, calibration curves, and exposure concentration.

Part VII: In-house open-source LLM and OpenRouter API credits

BETTER trains an in-house, open-source LLM exclusively on prediction market data and makes it available via OpenRouter. Access is sold as API credits through OpenRouter. This is distinct from BETTER application telemetry endpoints.

Part VIII: Token economics and alignment

Token summary

  • Token: $BETTER
  • Network: Base
  • Total supply: 1,000,000,000
  • Initial liquidity: Base BETTER/WETH pool (liquidity provider tokens locked)

Allocation and vesting

AllocationPer centAmountVesting and lock-up
Public sale + liquidity (Base BETTER/WETH)40%400,000,000Public sale unlocked at the token generation event; liquidity deployed to the Base BETTER/WETH pool with liquidity provider tokens locked permanently
Team20%200,000,00015-month vesting with a 6-month cliff, then 9-month linear unlock
Treasury25%250,000,00012-month linear unlock from the token generation event
OpenServ SERV token drop5%50,000,000Unlocked (claimed by SERV token stakers via tasks)
Programmatic funding10%100,000,000Released across fully diluted valuation bands
Initial circulating supply is approximately 400,000,000 tokens (40 per cent), representing the combined public sale plus liquidity allocation.

Access gating

Hold $BETTER to view the Terminal. Stake the same quantity to deposit into Vaults. The current access gate ratchets by fully diluted valuation. See Tokenomics for phase thresholds, worked examples, and Lite Mode behaviour.

Fees

  • Vault performance fee: 20 per cent on profits only, charged on withdrawal.
  • Lite Mode: 2 per cent of notional volume (in USD Coin terms) traded via the Terminal.

Value accrual

$BETTER is designed to accrue value through routine buybacks and burns funded by BETTER revenue. Recurring funding sources include:
  • vault performance fees
  • the business-to-business prediction market data ingestion product (sold to prop firms)
  • OpenRouter API credits for BETTER’s in-house open-source LLM trained exclusively on prediction market data
In addition, 30 per cent of net arbitrage profits from vBETTER arbitrage and $TRUTH-PERP arbitrage are used to buy and burn $BETTER.

Trading taxes

On buys and sells of $BETTER, a two per cent (2%) buy tax and two per cent (2%) sell tax apply. These taxes route to the protocol treasury and fund capital expenditure such as infrastructure and hosting.

Roadmap

  • 21 January 2026: token generation event; Terminal live within 48 hours.
  • January 2026: Vaults live (staged rollout).
  • First quarter of 2026: tokenised vault shares (vBETTER); Kalshi and Opinion integrations (March 2026); business-to-business Rust data ingestion.
  • Second quarter of 2026: in-house open-source LLM trained exclusively on prediction market data; API credits sold via OpenRouter; arbitrage flywheel.
  • Fourth quarter of 2026: $TRUTH-PERP on Hyperliquid.

Risks and disclaimers

This document is not investment, legal, or tax advice. Key risks include:
  • smart contract risk
  • market risk
  • liquidity risk
  • execution and latency risk
  • oracle and settlement risk
  • cross-chain and integration risk
  • regulatory risk
Protocol mechanics, parameters, and timelines may change.