Chapter 37: Key Takeaways

Oracles, Resolution, and the Decentralization Trilemma

Smart contracts are deterministic state machines that cannot access external data. Every node must reach the same state, which is impossible if contracts independently query external APIs.
An oracle is the mechanism that brings off-chain data on-chain. For prediction markets, the oracle reports the outcome of the real-world event.
The oracle is the most security-critical component of any prediction market. If the oracle reports incorrectly, the entire market settles incorrectly, and the economic damage can equal the total value locked.
Oracle failure modes include: liveness failure (no report), correctness failure (wrong outcome), manipulation, ambiguity, timing errors, and source disagreement.

Centralized oracle: A single trusted entity resolves outcomes. Fast and cheap, but a single point of failure. Suitable only for low-value markets or platforms with strong legal accountability.
Decentralized Oracle Network (DON): Multiple independent nodes fetch and aggregate data (e.g., Chainlink). Best for continuous quantitative data (price feeds), less suited for one-off binary events.
Schelling point mechanism: Participants independently vote on the outcome; those who agree with the majority are rewarded. Relies on the truth being the "obvious" focal point. Vulnerable to coordination attacks and P+epsilon attacks.
Optimistic oracle: A proposer asserts an outcome with a bond. If no one disputes within a window, the outcome is accepted. Disputes escalate to a more expensive mechanism. Cheap and fast in the common case.
Governance-based oracle: Protocol token holders vote to resolve disputes. Security depends on the governance token's market cap exceeding the value at risk in any single market.

UMA is the oracle used by Polymarket. It operates in three phases: assertion (proposer posts bond), dispute window (typically 2-48 hours), and DVM vote (if disputed, UMA token holders vote via commit-reveal).
The happy path (no dispute) costs only a few dollars in gas and resolves in hours. Over 95% of markets resolve this way.
The dispute path invokes the Data Verification Mechanism (DVM), where UMA token holders vote in a 48-hour commit-reveal process. The losing side forfeits their bond.
Security depends on: (a) the bond being large enough to deter frivolous assertions, (b) at least one honest party monitoring and willing to dispute, and (c) the DVM being resistant to corruption (requires >50% of voting UMA tokens).

Chainlink nodes independently fetch data from off-chain sources and submit reports on-chain. The network aggregates reports (typically using the median) to produce a consensus value.
Chainlink is optimized for continuous data feeds (asset prices, interest rates) rather than one-off event resolutions.
For prediction markets, Chainlink can resolve quantitative markets (e.g., "Will ETH close above $5,000?") by referencing price feeds, but cannot resolve subjective or complex event outcomes.

Kleros uses stake-weighted random jury selection. Jurors stake PNK tokens; selection probability is proportional to stake.
Disputes are resolved through escalating appeals: jury sizes grow (2r+1 at round r), making each appeal more expensive and involving more jurors.
The Schelling point mechanism incentivizes honest voting: jurors who agree with the majority earn rewards; dissenters lose their stake.
Kleros is well-suited for low-to-medium value disputes where the question is objective but requires human judgment.

Augur uses REP token holders for reporting and dispute resolution. An initial reporter asserts an outcome with a small bond.
Disputes follow an exponential escalation pattern: the required stake doubles each round ($S_r = S_0 \cdot 2^r$).
The fork mechanism is Augur's ultimate backstop: the protocol splits into parallel universes, one per disputed outcome. REP holders choose which universe to join, and REP in losing universes becomes worthless.
The fork mechanism provides very strong security (attack cost equals ~50% of REP market cap) but is extremely costly and disruptive.

The decentralization trilemma states that no oracle system can simultaneously maximize scalability, security, and decentralization. Improving any two requires sacrificing the third.
Centralized oracles maximize scalability (fast, cheap) and can have reasonable security (operator reputation), but sacrifice decentralization.
Full governance oracles (Augur fork) maximize security and decentralization, but sacrifice scalability (slow, expensive).
Optimistic oracles (UMA) balance the three by being scalable and moderately decentralized in the common case, with full decentralization available through the dispute mechanism.

Security margin = Cost of Attack / Profit from Attack. An oracle is economically secure when this ratio exceeds 1.
For token-weighted oracles, the cost of attack depends on: (a) the governance token's market cap, (b) the fraction of tokens actively voting, (c) the slippage cost of accumulating enough tokens, and (d) any time-lock requirements.
The P+epsilon attack exploits Schelling point mechanisms by offering a conditional bribe that shifts the focal point. Defense: commit-reveal schemes that prevent vote verification.
Multi-oracle systems dramatically reduce corruption probability. With $k$ independent oracles each having corruption probability $p$, the majority-vote corruption probability follows a binomial distribution and drops exponentially with $k$.

Resolution source specification: Every market question must include a precise resolution source (e.g., "Official AP call" vs. "Certified election results"). Ambiguity in resolution sources is the primary cause of disputes.
N/A resolution: Markets should support void/N/A resolution for questions that become unanswerable, with full refunds to traders. This is preferable to forcing an arbitrary resolution.
Time zones and deadlines: Always specify time zones explicitly. "December 31, 2025" is ambiguous without a time zone.
Edge case documentation: Market creators should document how common edge cases will be handled (e.g., postponed events, partial fulfillment, disqualification).

Factor	Favorable	Unfavorable
Resolution source	Single, authoritative, machine-readable	Multiple conflicting sources, subjective
Question clarity	Binary, objective, well-defined	Ambiguous, subjective, multi-interpretable
Oracle type for low-value markets	Centralized or optimistic	Full governance (too expensive)
Oracle type for high-value markets	Hybrid (optimistic + governance backstop)	Centralized (insufficient security)
Bond size	0.1-2% of TVL	< 0.01% of TVL (spam risk)
Dispute window	2-72 hours (scaled to market complexity)	< 1 hour or > 7 days
Security margin	> 1.5	< 1.0 (attack is profitable)

Formula	Description
$\text{Security Margin} = \frac{C_{attack}}{V_{market}}$	Economic security of oracle
$P_{corrupt}(k, p) = \sum_{i=\lceil k/2 \rceil}^{k} \binom{k}{i} p^i (1-p)^{k-i}$	Multi-oracle corruption probability
$S_r = S_0 \cdot 2^r$	Augur dispute stake at round $r$
$C_{fork} \approx 0.5 \times \text{REP market cap}$	Cost to win an Augur fork
$E[\text{cost}] = (1-p_d) \cdot C_{happy} + p_d \cdot C_{dispute}$	Expected oracle resolution cost
$J_r = 2r + 1$	Kleros jury size at appeal round $r$