> For the complete documentation index, see [llms.txt](https://ollara-ai.gitbook.io/ollara-ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://ollara-ai.gitbook.io/ollara-ai/key-terminology.md). # Key Terminology ## **Topics** **Topics** act as Schelling Points that guide and focus the Ollara protocol. They categorize inferences by subject or domain, enabling more precise evaluations and reward mechanisms. Anyone—referred to as a **Topic Creator**—can permissionlessly launch a new topic on **Ollara** and define a **rule set** that determines how submitted inferences will be assessed and rewarded. Once created, **Workers** can begin submitting their inferences to these topics and earn rewards based on their prediction accuracy. *** ## **Rule Set** Each topic includes a customizable rule set established at creation, which defines: * **Loss Function**: The mathematical method used to evaluate prediction accuracy (e.g., Mean Absolute Directional Loss, L1-Norm). * **Ground Truth Source**: The reference used to measure accuracy (e.g., external APIs, Solana-based oracles, or the median of all submissions). This rule set ensures every inference is scored fairly and consistently within the topic’s parameters. *** ## **Inferences** **Inferences** are predictions or assessments submitted by Workers about specific outcomes within a given topic. These outputs are evaluated based on the topic’s rule set and rewarded accordingly. *** ## **Forecasts** **Forecasts** are a special type of prediction—focused not on external outcomes, but on **the performance of other participants** in the current epoch. Each forecast represents a set of anticipated loss scores for peer Workers, calculated using the topic’s designated loss function. Forecasts are critical for assessing participant credibility and accuracy. > Note: Throughout the Ollara documentation, the terms *forecast* and *prediction* may be used interchangeably when referring to forecaster outputs. *** ## **Context Awareness** **Context Awareness** adds a higher dimension to how forecasts and inferences are evaluated. By analyzing: * Live test results (revealed ground truth) * Peer performance data Workers can refine both their inference and forecast models over time. This iterative feedback loop improves accuracy and enables the protocol to **selectively leverage insights** from the most relevant agents based on specific contexts (e.g., day of the week, market state, domain expertise). For instance, certain forecasters may excel in bullish markets while others dominate in bearish conditions. Ollara’s ability to incorporate these nuanced patterns makes it **more accurate than any individual actor alone**. *** ## **Network Participants** Ollara defines a **Network Participant** as any individual or entity that actively contributes to the system through a defined role. *** ## **Supply Side Participants** Participants who contribute value to the network but do not directly consume inferences. This includes: * **Workers**: Generate inferences and forecasts. * **Reputers**: Evaluate and stake on inference quality. * **Validators**: Secure the network and verify on-chain operations. The **Demand Side** consists of **Consumers**, who request and integrate predictions into their applications. *** ## **Epochs** The network operates in discrete timeframes called **Epochs**. During each epoch: * Workers submit inferences and forecasts * Reputers assess and score submissions * Rewards are calculated and distributed Each epoch offers a fair and periodic opportunity to measure participant performance. *** ## **Rewards** **Rewards** are distributed at the end of every epoch to incentivize quality contributions. They are based on: * Prediction accuracy * Forecast precision * Performance over time * Staked commitment (where applicable) *** ## **Stake** **Stake** is a financial signal of trust placed by **Reputers**, who commit funds to topics they evaluate. * It reflects their confidence in assessing inference quality * The higher the stake, the more influence and potential rewards * Stake can be increased by personal contributions or **Delegated Stake** from others Staking helps secure the protocol and improve scoring fidelity. *** ## **Delegated Stake** Users can earn passively by delegating funds to trusted Reputers. * Delegators receive a share of the Reputer’s rewards * Boosts the Reputer’s influence and strengthens topic-level security * Includes a **withdrawal delay** to mitigate fast-exit or flash attack risks * Like staking, delegation involves risk tied to the Reputer’s accuracy *** ## **Withdrawal Delay** To safeguard the network, **Ollara enforces a withdrawal delay** on both rewards and stake. * Users must first **initiate a withdrawal request** * After a fixed delay, funds become eligible for release * This mechanism protects against exploitative behaviors while ensuring a smooth user experience *** ## **Regrets** **Regret** is a comparative performance metric used to evaluate Workers: * **Positive Regret**: The Worker outperformed the network average * **Negative Regret**: The Worker underperformed relative to the network Regrets help quantify predictive skill and guide both staking decisions and topic evolution. *** **Ollara** transforms decentralized intelligence by structuring prediction, evaluation, and reward into a composable, high-performance protocol—built to thrive on Solana.